Evolution of Central Pattern Generators for Bipedal Walking in a Real-Time Physics Environm

Evolution of Central Pattern Generators for Bipedal Walking in a Real-Time Physics Environment

Torsten Reil and Phil Husbands

Abstract—We describe an evolutionary approach to the control problem of bipedal https://www.sodocs.net/doc/7b10791589.html,ing a full rigid-body simulation of a biped,it was possible to evolve recurrent neural networks that con-trolled stable straight-line walking on a planar surface.No propri-oceptive information was necessary to achieve this task.Further-more,simple sensory input to locate a sound source was integrated to achieve directional walking.To our knowledge,this is the first work that demonstrates the application of evolutionary optimiza-tion to three-dimensional physically simulated biped locomotion. Index Terms—Bipedal walking,evolutionary algorithms,evolu-tionary robotics,physics,recurrent neural networks.

I.I NTRODUCTION

B IPEDAL walking is a difficult task due to its intrinsic insta-

bility and developing successful controller architectures for this mode of locomotion has proved substantially more dif-ficult than for other types of walking[1].

There is considerable interest in this matter from disciplines as diverse as robotics,computer graphics,virtual reality,and bi-ology.However,previous approaches have been based on con-ventional control strategies.As will be discussed shortly,this brings about considerable complications and limitations.In ad-dition,past work has been constrained by only limited available means to simulate the physics of the body to be controlled,thus, making it either necessary to build robots or resort to simplified models in simulation.

Given the inherent difficulties in designing stable con-trollers for natural-looking bipedal walking,it was decided to investigate the use of evolutionary robotics techniques[2]in de-veloping recurrent dynamical neural-network-based controllers for the task.This paper describes successful experiments in evolving controllers for a realistically simulated biped.

The structure of the paper is as follows.We first review previous work on controller architectures for bipedal walking. These are subsequently contrasted with the approach taken here:evolutionary robotics.Section II describes the implemen-tation of both the biped and the neural controller,as well as the evolutionary algorithm(EA)used in this research.As shown in the subsequent results section,this combination succeeded in producing natural-looking bipedal walking.Section IV

Manuscript received July11,2000;revised March15,2001and July23,2001. This work was supported by the Engineering and Physical Sciences Research Council and by the Guy Newton Scholarship in Zoology with Wolfson College, Oxford.

T.Reil is with the Department of Zoology,University of Oxford,Oxford OX1 3PS,U.K.(e-mail:torsten.reil@https://www.sodocs.net/doc/7b10791589.html,).

P.Husbands is with the School of Cognitive and Computing Sciences,Uni-versity of Sussex,Falmer,Brighton BN19QH,U.K.

Publisher Item Identifier S1089-778X(02)02977-6.addresses the integration of sensory input and describes a corresponding successful experiment.The paper closes with a discussion of the research.

A.Related Work

The major thrust of research on bipedal walking has come from computer graphics and robotics.In the case of the former, animation techniques such as motion capture[3]have come to dominate the area.Motion capture essentially implies filming the desired human behavior and using the obtained data to an-imate a computer generated equivalent.The advantage of this approach is clearly the ability to immediately generate realistic bipedal motion dynamics.However,Laszlo et al.[4]make it clear that motion capture does not provide us with sufficient un-derstanding to create more general walking motions,especially when conditions are unpredictable,when new motions need to be generated,or when dealing with nonhuman characters. These shortcomings can be overcome by a second approach, which is based on a semiphysical representation of the biped and a controller to create movement patterns.With techniques such as inverse kinematics and inverse dynamics,virtual limbs can be placed at the desired positions and the required forces are computed accordingly.

Several workers[4],[5]have followed this approach to create computer animations of humans.The equations of motion are either produced specifically for the model to be animated or are generated with available packages[5],[6].Typically,a fi-nite-state machine determines the control actions(with cyclic states such as heel contact,toe contact,unloading,or flight[4], [5],[7])and special forms of limit cycle control may be applied to achieve the necessary stability[4].The animated end results of these efforts closely resemble natural motion patterns,but may nevertheless fail to convince the human eye in specifically designed“motion Turing tests”[5](these tests confront human subjects with computer generated and real-life animations and ask them to discern between the two).This lack of realism is a direct consequence of the controller architecture employed;a state machine does not readily produce the fluctuations typical of real locomotion.More significantly,it cannot easily be ex-tended to integrate sensory input.Furthermore,the creation of state machines can be a cumbersome process,as states have to be identified,implemented and fine-tuned by hand for each type of gait to be modeled[5].

Bipedal locomotion in robots is subject to the physical laws of the natural world and hence short cuts like motion capture are not available.Bipedal robots with varying complexities have been produced and controlled by several researchers[8]–[16].

1089-778X/02$17.00?2002IEEE

Fig.1.MathEngine implementation of the biped.Note that although two bodies are used to implement the knee,only one body is necessary.

TABLE I

W ALKER D IMENSIONS AND M

ASSES

MathEngine bodies are combined to composite

bodies.

Fig.2.Recurrent neural network used to control bipedal walking.Shaded nodes are motor neurons.Connections are bidirectional and asymmetric.

As with computer graphics models,the corresponding controller architectures are typically based on state machines with special algorithms added on top to provide the necessary stability.Most recently,Pratt et al.[14]have used Virtual Model Controllers for planar bipedal robots.Here,virtual mechanical components are attached to the robot and exert real actuator torques or forces.For example,a virtual dog track bunny is used to maintain a de-sired velocity in a planar biped robot.A state machine changes the virtual component connections or parameters at each state transition.Together with a set of simple rules for,e.g.,height,pitch,and speed stabilization,this allows a more intuitive devel-opment of stable controller architectures and eases the problem

TABLE II

A NGLE L IMITS OF

B IPED J

OINTS

Values were obtained heuristically.

of mathematical tractability encountered in previous attempts [13].In addition to testing and optimising control strategies on the real robot,Pratt and Pratt [17]have used a rigid-body simu-lation [6]to create a realistic model of a biped.This allowed ef-ficient experimentation with the robot’s natural dynamics (such as passively swinging legs).

In summary,with few exceptions,such as Miller [18],who utilizes reinforcement learning for training a neural net,pre-vious approaches to bipedal walking have been based on engi-neering techniques like state machines and conventional control theory.As remarked on earlier,this causes a number of prob-lems:1)mathematical tractability;2)manual optimization;3)limited extendability;and 4)limited biological plausibility.It is argued here that the evolutionary robotics approach presented below has the potential to overcome the first three constraints by improving on the last one,biological plausibility.B.Evolutionary Robotics

Evolutionary robotics was introduced as an alternative to the hand design of robot controllers,especially for autonomous robots acting in uncertain and noisy domains [19],[20].EAs are used to search spaces of controllers (and potentially body and sensor layouts too)described by a set of variables encoded on the artificial genotype.The fitness function is usually task-based,i.e.,high scores are achieved by controllers that enable the robot to perform the desired task well.These con-

REIL AND HUSBANDS:EVOLUTION OF CENTRAL PATTERN GENERATORS

161

Fig.3.Motor connections between controller and walker.Hips have two DOFs each(sagittal,i.e.,front to back,and lateral);knees have one DOF each

(sagittal).

Fig.4.Encoding scheme.Chromosome index is shown in general form(n is the number of nodes)and for special case of n=10(in brackets).Parameters are encoded as real values.Ranges in the boxes represent upper and lower bounds of the respective parameter types.

trollers are nearly always in the form of some kind of artificial neural network(ANN).The evolutionary search algorithm’s job ranges from optimizing the parameters of a fixed-archi-tecture ANN[21]–[23]to exploring complex network spaces where the architecture and many properties of the nodes and connections are under evolutionary control[24],[25].

There have been many successful applications of evolu-tionary robotics to date,ranging from simple reactive behaviors in wheeled robots with infrared proximity sensors[22],[26], through visually guided behaviors in simple wheeled robots [27],[28],to fairly complex nonreactive behaviors in simple wheeled robots[29]and a variety of locomotion controllers for six-and eight-legged robots[30]–[34].For far more detailed reviews of the field,see[2]and[37].

To date,evolutionary robotics techniques have not been applied to a task as dynamically unstable as controlling bipedal locomotion.1It is this inherent instability(generally, two-legged walkers will fall over without continuous active control)that provides severe challenges to the hand design of such controllers,especially if smooth natural walking is required.However,given the success of evolved locomotion controllers for relatively stable hexapod and octopod robots [30]–[34],it was deemed appropriate to investigate the use of such techniques for developing bipedal locomotion controllers. 1While other researchers such as Rodrigues[35]and de Garis[36]did use evolutionary optimization in the context of bipedal walking,to the authors’knowledge,no research has so far demonstrated the applicability of evolved recurrent neurocontrollers for a real-time and physically realistic biped simula-tion.As will be seen later in this paper,evolutionary robotics methods were indeed successful in finding stable controllers for bipedal walking.

II.I MPLEMENTATION

A.The Biped

1)MathEngine Bodies and Joints:Unlike previous,em-bodied approaches,the agent to be controlled here is modeled using the rigid body dynamics simulation software developer’s kit(SDK)of MathEngine.This allows the evaluations to be run significantly faster than real time[36]and,thus,greatly increases the efficiency of the evolutionary approach. MathEngine’s Fast Dynamics Toolkit was developed to overcome the two most pressing problems in the simulation of physics:complexity and speed.Programmed in C,it supports bodies,joints,contacts,and forces,the attributes of which can be set by the user[39].Once set up,a physical world is integrated by the engine over time in user-defined intervals. The SDK used here(1.0.5)is shipped with an OpenGL and Direct3D renderer to visualize the scene.

The implementation of the bodies for this research is char-acterized by the need to capture the fundamental features of a biped while limiting the body’s complexity and degrees of freedom.Thus,the model used here consists of two articulated legs connected by a link.Thirteen MathEngine bodies and11 joints are used to implement these structures,as illustrated in Fig.1(because each leg consist of two composite bodies with two spheres and one connecting link each,the present imple-mentation uses two bodies for each knee).

162IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION,VOL.6,NO.2,APRIL2002 TABLE III

I NITIALIZATION(I)AND M UTATION(M)D ISTRIBUTIONS FOR D IFFERENT

P ARAMETERS T YPES

Standard deviation and separate means of Gaussian distribution are shown.

The degrees of freedom(DOFs)of the joints are:1)hip joint:

two DOFs(pitch/roll)and2)knee:one DOF(pitch),giving a

total of six DOFs.

Although important for real walking(for example in birds

or humans),feet and ankle joints are not implemented.They

impose additional DOF and would,therefore,considerably in-

crease the controller’s search space.In addition,the capabil-

ities of MathEngine SDK1.0.5make realistic foot-floor con-

tact a computationally expensive endeavour(due to the need for

multiple contact points).Sphere-plane contacts(sphere radius:8

cm)provide an uncomplicated and fast alternative and are used

instead.As will become clear later,this simplification does not

come at a noticeable cost in terms of the overall body dynamics.

2)Actuators:Muscle action is modeled by proportional

derivative(PD)controllers[5],[40],which are essentially

equivalent to damped torsional springs.Their modus operandi

is characterized by the following:

is torque force,

is the current angle.

Rather than directly defining the strength of actuator forces,

the controller updates the natural orientation of the PD con-

troller(the desired angle of the limb).Equation(1)is then used

to compute the force necessary to move the limb to that po-

sition.The spring constant and damping value determine the

strength and the tendency to oscillate.Their values therefore

significantly influence the realism of movements.By means of

manual experimentation,values of

th neuron is com-

puted according to

th neuron,

th neuron,and is the weight from

the th neuron.

The corresponding output is calculated as follows:

is the bias of the

REIL AND HUSBANDS:EVOLUTION OF CENTRAL PATTERN GENERATORS

163

Fig.6.Fitness graph of representative stable controller evolution.Top fitness (black)and average fitness (grey)are

shown.

Fig.7.Motor neuron activation levels of top individual of generation 120(see Fig.6).

C.Evolutionary Algorithm

1)Encoding Scheme and Population Parameters:The pa-rameters to be optimized are weights,time constants and biases.The encoding scheme spatially separates the three types in the chromosome (see Fig.4).

Parameter values are coded as real numbers,with different ranges for each data type.Following [42]and [32],these are [

4.0,4.0]in the case of the biases.The assignment of these ranges is simplified by the spatial separation of the types;similarly,different mutation rates and sizes can be applied.While the former remains constant throughout the chromo-some,the differential implementation of the latter is necessary due to the varying parameter value ranges.This is achieved by using Gaussian distributions.Table III shows the mutation sizes in form of standard deviations from mean zero.Values exceeding the allowed range are clipped to the maximally permitted level.(This is known to create disproportionate accumulations around the clipping points [43],which were,however,found to be negligible in this work.)

The mutation rate is calculated so as to cause on average one change per chromosome.Thus,for larger networks,an accord-ingly lower rate per locus is applied.Together with typically small mutation sizes (see Table III),this ensures that the evo-lutionary search is local and gradual.Each population consists of 50individuals and its individual controllers are initialized with randomized values using the initialization distributions of Table III.Rank-based selection is used for reproduction with a fittest fraction of 0.5(this essentially means culling the bottom half of the population and replacing it with a copy of the top half [44]).No crossover operations are applied both on theoretical (no identifiable functional units in the genotype and phenotype structure [45])and empirical grounds (recent experimental evi-dence on lack of efficiency of crossover in this problem domain [25]).

2)Evaluation of Controllers:Despite the complexity of bipedal locomotion,it is possible to reduce the fitness function to the following two components:

1)maximize distance travelled from origin;

2)do not lower center of gravity below a certain height.The first objective implicitly includes the locomotion com-ponent,while at the same time rewarding walking in a straight line rather than in circles (note that this would not be true for maximize overall distance travelled).The second goal com-bines two further factors:it penalizes falling down as well as grotesque movements.(Much of the second point is al-ready prohibited by constraining the joint angles in the physical model.)To improve efficiency evaluations are terminated early

164IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION,VOL.6,NO.2,APRIL2002

Fig.8.Motion sequence of biped controlled by top individual of generation

120(see Fig.6).Frame order is from left to right and top to bottom.

if they are unpromising,i.e.,as soon as the second objective

is not met.Hence,the fitness function for a biped on an

walking plane can be expressed as follows:

is the fitness,are the planar components of the

walker position,and

REIL AND HUSBANDS:EVOLUTION OF CENTRAL PATTERN GENERATORS

165

Fig.11.Fitness graph of bipedal walking with sensory integration.Population was seeded with individuals from the run depicted in Fig.7.

majority of unsuccessful controllers,analysis showed that the little distance they did cover was controlled by the settling phase of the recurrent net.We,therefore,added an additional fitness criterion that actively rewarded cyclic activity.This markedly increased the proportion of successful runs(to80%),but was not reflected in a proportionate improvement of the overall time ef-ficiciency(i.e.,successful controllers per processor cycle).The reason for this lies in the fact that,even in the original configura-tion,unsuccessful controller evaluations are aborted early(see Section II-C2),thus,taking up only limited computational re-sources.

IV.I NTEGRATING S ENSORY I NPUT

The controllers described in Section III are purely rhythm-generating structures.Although sufficient to pro-duce stable walking behavior in a nonfluctuating environment, they are not capable of dealing with rough terrain or responding to external stimuli.In order to achieve this,sensory input must be integrated and the CPG activity modified accordingly.

A simple set of experiments was carried out to explore the potential to integrate basic sensory input and will be described now.

The biped is to walk toward the equivalent of a sound source. It is equipped with two“ears,”the inputs of which are prepro-cessed to give a single signal which becomes stronger with de-creasing distance from the source.2In addition,the signal van-ishes when the biped is directly facing the source(see Fig.9). The signal is fed into the CPG as depicted in Fig.10.The pop-ulation is initialized uniformly with clones of the top individual from the run depicted in Fig.6.The CPG weights are clamped, but the weights of the ten connections between the sensory node and the RNN nodes are under the control of the EA.At each evaluation,the biped starts from its default position and is pre-sented successively with two sound source locations.Because the task is to approach the signal as closely as possible,the fit-ness function is the negative distance of the walker to the sound 2This approximation does not hold true when the biped is very close to the sound source(i.e.,when the distance to the source is comparable with the dis-tance between the agent’s

ears).Fig.12.Trajectories of biped with sensory integration(two runs are shown). Signals are located at[03;010]and[3,010].Run one:left.Run two:right. source at the time of termination(as caused by the conditions outlined in Section II-C2)or the natural end of the evaluation (after50s).

A.Results

Fig.11depicts an evolutionary run with the population seeded with individuals from the run depicted in Fig.6.The graph is characterized by an initially strong increase in fitness, but it fails to reach the maximum fitness value of zero.

As illustrated in Fig.12,the controller succeeded in walking toward the respective signal positions in the two runs.However, visual analysis of the walkers made clear that the gait becomes unstable close to the respective signal sources.This is particu-larly true for the second(right)run.

To further investigate the ability of the sensors to modulate the net’s activity pattern as well as to examine the reasons for the eventual instability,the neuronal activation patterns of the two runs were recorded and are represented in Fig.13.

The activation graphs indicate that the turning behavior is at least partly achieved by modulating the amplitude of the right hip sagittal motor neuron(which controls the front to back movement),with a decrease resulting in right(top

166IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION,VOL.6,NO.2,APRIL

2002

(a)

(b)

Fig.13.Neural activation graph of bipedal walking toward (a)right and (b)left signal.Only activation of right hip sagittal (i.e.,front to back)neuron is shown.(Other neural activations did not differ markedly from those of Fig.7).

graph)and an increase resulting in left (bottom graph)turning.Additional experiments were carried out,including evolution of the behavior from scratch (i.e.,unseeded populations)and evolution with seeded populations but with evolvable CPG weights.However,neither of these additional experimental series produced results superior to those documented above.

V .D ISCUSSION

Despite an extremely simple fitness function,the EA employed in this research was capable of producing stable straight-line walkers without the use of proprioceptive sensory input.EAs rely on evolvable systems and the gradual nature of the fitness graphs (e.g.,Fig.6)indicates that the recurrent networks used here can indeed be optimized in a continuous gradual manner.This notion has recently been corroborated by Rendel [47],who has shown that the fitness landscape underlying the current controller architectures is very smooth.For example,it was possible to gradually modify the amplitude and period of specific motor neuron cycles without affecting those of others.

A further characteristic of the current setup is the ability to create a large diversity of gaits,both in terms of speed and the use of limbs.Several gaits showed considerable similarity to human walking,although this was not specifically selected for.A potential way to further increase the realism of the motion is to select for minimum energy expenditure (a simple mea-sure for this would be the average actuator activity).It is ex-pected that this fitness component will particularly reward the use of knees.In the current implementation,several controllers walked with extended legs because this is concomitant with a large stride length.Humans,however,use the momentum of a forward-swinging lower leg,which is energetically more favor-able [17].

The evolved controllers were further characterized by non-repetitive activity cycles;instead,small fluctuations were ob-served (see Fig.7).Similar fluctuations have elsewhere been found to contribute to the perceived realism of simulated lo-comotive behavior [48].Real bipedal walking contains fluctu-ations in successive cycles and it is the lack of these that the human eye picks upon in other artificial walking bipeds.

The preliminary experiments on the integration of sensory input indicate that CPG activity can indeed be modified by ex-ternal stimuli in a meaningful way (Fig.13).However,it is clear that the current sensory architecture is insufficient to modulate the biped’s behavior and retain stability.For example,the simple preprocessing function causes large destabilizing fluctuations if the biped is close to the signal source.This problem is further intensified by the lack of active balancing mechanisms.The in-tegration of proprioceptive (e.g.,limb positions and velocities)and vestibular (balance)input is,therefore,a necessary next step to achieve more interactive and robust behavior.

A question that was not systematically explored is in how far the network size affects the efficiency of the approach,both in terms of search space as well as internal dynamics of the net.With the current architecture,a linear increase in the number of nodes leads to an quadratic increase of the corresponding search space.A possible way to circumvent this problem is to employ identical subnetworks for each leg.Such a constellation seems to reflect the natural arrangement of coupled oscillators more accurately [49],[50]and has been successfully used elsewhere in the context of multilegged locomotion [42],[51].

REIL AND HUSBANDS:EVOLUTION OF CENTRAL PATTERN GENERATORS167

We would like to reiterate that the stable walkers arrived at did not require proprioceptive input to achieve stable walking in a straight line.This corroborates results obtained elsewhere [52]that show that mechanical walkers can attain stable straight-line walking on a planar surface without active balance control.While those bipeds were mechanically fine tuned to exploit gravity as an energy source,the implementation presented here relies on evolutionary optimization to fine tune active actuation.

We believe that the neuroevolutionary approach described here brings about several major benefits:1)it is fully automated; hence,changes in morphology or actuator implementations can be easily accommodated by reevolving the controllers;2)the diversity of locomotive behaviors is large because the system does not require a priori knowledge as to how to solve the con-trol problem;and3)the evolved controllers are computationally very cheap(typically taking up0.5%of the processing power re-quired by graphics and physics).

VI.C ONCLUSION

We have demonstrated the suitability of an evolutionary robotics approach to the problem of stable three-dimensional bipedal walking in simulation.The current implementation is capable of walking in a straight line on a planar surface without the use of proprioceptive input.However,the use of the latter will become necessary to stabilize the biped on uneven terrain or in response to directional changes.The neural controller employed in this research lends itself to the incorporation of such additional input.

The quality of the results is expected to further improve by a refined fitness function,as well as a shift toward coupled neural oscillators instead of a single network.Furthermore,it is de-sirable to incorporate biomechanical knowledge about human walking in order to make maximum use of the passive dynamics of the bodies.These aspects are currently being implemented. In theory,the results obtained here are directly transferable to embodied robots.In practice,however,there are likely to be complications due to a possible lack of accuracy of the physics engine.It remains to be seen whether this“reality gap”can be crossed with appropriate techniques such as noise envelopes

[26].

A CKNOWLEDGMENT

The authors would like to thank N.Pattinson,J.Worby,and D.Raubenheimer for helpful discussions and the anonymous reviewers of a previous version of this paper for their helpful and constructive comments.The authors would also like to thank C. Massey and W.Wray(MathEngine)for support concerning the physics engine.

R EFERENCES

[1]R.D.Beer,H.J.Chiel,R.D.Quinn,and R.E.Ritzmann,“Biorobotic

approaches to the study of motor systems,”Current Opinion Neurobiol., vol.8,no.6,pp.777–782,1998.

[2]J.-A.Meyer,P.Husbands,and I.Harvey,“Evolutionary robotics:A

survey of applications and problems,”in Evolutionary Robotics,P.

Husbands and J.-A.Meyer,Eds.Berlin,Germany:Springer-Verlag, 1998,vol.1468,Lecture Notes Computer Science,pp.1–21.

[3] A.Bruderlin and L.Williams,“Motion signal processing,”in Proc.Int.

https://www.sodocs.net/doc/7b10791589.html,puter Graphics and Interactive Techniques(SIGGRAPH), Aug.1995,pp.97–104.

[4]https://www.sodocs.net/doc/7b10791589.html,szlo,M.van de Panne,and E.Fiume,“Limit cycle control and

its application to the animation of balancing and walking,”in Proc.

https://www.sodocs.net/doc/7b10791589.html,puter Graphics and Interactive Techniques(SIGGRAPH), Aug.1996,pp.155–162.

[5]J.K.Hodgins,“Three-dimensional human running,”in Proc.IEEE Int.

Conf.Robotics and Automation,vol.4,1996,pp.3271–3276.

[6]MIT Creatures[Online].Available:https://www.sodocs.net/doc/7b10791589.html,/projects/

leglab/simulations/simulations.html

[7]M.van de Panne,R.Kim,and E.Fiume,“Virtual wind-up toys for ani-

mation,”in Proc.Graphics Interface,1996,pp.208–215.

[8]K.Ogo,A.Ganse,and L.Kato,“Dynamic walking of biped walking

machine aiming at completion of steady walking,”in Proc.3rd Symp.

Theory and Practice of Robots and Manipulators,ch.5,1978.

[9]T.Kato,A.Takanishi,H.Jishikawa,and I.Kato,“The realization of

the quasidynamic walking by the biped walking machine,”in Proc.

4th Symp.Theory and Practice of Robots and Manipulators,1983,pp.

341–351.

[10] A.Takanishi,M.Tochizawa,H.Karaki,and I.Kato,“Dynamic biped

walking stabilized with optimal trunk and waist motion,”in Proc.

IEEE/RSJ Int.Workshop Intelligent Robots and Systems,1989,pp.

187–192.

[11]H.Miura and I.Shimoyama,“Dynamic walk of a biped,”Int.J.Robot.

Res.,vol.3,no.2,pp.30–74,1984.

[12]J.Furusho and A.Sano,“Sensor-based control of a nine-link biped,”Int.

J.Robot.Res.,vol.9,no.2,pp.83–98,1990.

[13]J.Pratt and G.Pratt,“Intuitive control of a planar bipedal walking robot,”

in Proc.IEEE Int.Conf.Robotics and Automation.,Leuven,Belgium, 1998,pp.2014–2022.

[14]J.Pratt,P.Dilworth,and G.Pratt,“Virtual model control of a bipedal

walking robot,”in Proc.IEEE Int.Conf.Robotics and Automations,vol.

1,Albuquerque,NM,1997,pp.193–198.

[15]T.Mita,T.Yamaguchi,T.Kashiwase,and T.Kawase,“Realization of a

high speed biped using modern control theory,”Int.J.Control,vol.40, no.1,pp.107–119,1994.

[16]Honda P3Humanoid Robot[Online].Available:

http://www.honda.co.jp/english/technology/robot/index.html

[17]J.Pratt and G.Pratt,“Exploiting natural dynamics in the control of a

planar bipedal walking robot,”in Proc.36th Annu.Allerton https://www.sodocs.net/doc/7b10791589.html,-munication,Control,and Computing,Monticello,IL,Sept.1998. [18]https://www.sodocs.net/doc/7b10791589.html,ler,“Real-time neural network control of a biped walking

robot,”IEEE Control Syst.Mag.,vol.14,pp.41–48,Feb.1994. [19]R.Brooks,“Artificial life and real robots,”in Proceedings of the First

European Conference on Artificial Life,F.Varela and P.Bourgine, Eds.Cambridge,MA:MIT Press,1992,pp.3–10.

[20]P.Husbands and I.Harvey,“Evolution versus design:Controlling au-

tonomous robots,”in Proc.3rd Annu.Conf.Artificial Intelligence,Sim-ulation,and Planning,1992,pp.139–146.

[21]R.Beer and J.Gallagher,“Evolving dynamical neural networks for

adaptive behavior,”Adapt.Behav.,vol.1,no.1,pp.91–122,1992. [22] D.Floreano and F.Mondada et al.,“Automatic creation of an au-

tonomous agent:Genetic evolution of a neural network driven robot,”

in From Animals to Animats3,D.C.Cliff et al.,Eds.Cambridge, MA:MIT Press,1994.

[23]S.Nolfi,D.Floreano,O.Miglino,and F.Mondada,“How to evolve

autonomous robots:Different approaches in evolutionary robotics,”in Artificial Life IV,R.Brooks and P.Maes,Eds.Cambridge,MA:MIT Press,1994.

[24] D.Floreano and F.Mondada et al.,“Evolution of plastic neurocon-

trollers for situated agents,”in From Animals to Animats4,P.Maes et al.,Eds.Cambridge,MA:MIT Press,1996.

[25]P.Husbands,T.Smith,N.Jakobi,and M.O’Shea,“Better living through

chemistry:Evolving gasNets for robot control,”Connection Sci.,vol.10, no.3–4,pp.185–210,1998.

[26]N.Jakobi,P.Husbands,and I.Harvey,“Noise and the reality gap:The

use of simulation in evolutionary robotics,”in Advances in Artificial Life:Proceedings of the Third European Conference on Artificial Life, J.Moran,J.Merelo,and P.Chancon,Eds.Berlin,Germany:Springer-Verlag,1995,Lecture Notes Computer Science.

[27]I.Harvey and P.Husbands et al.,“Seeing the light:Artificial evo-

lution,real vision,”in From Animals to Animats3,D.Cliff et al., Eds.Cambridge,MA:MIT Press,1994.

[28] D.Floreano,S.Nolfi,and F.Mondada,“Competitive co-evolutionary

robotics:From theory to practice,”in From Animals to Animats

5.Cambridge,MA:MIT Press,1998,pp.515–524.

168IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION,VOL.6,NO.2,APRIL2002

[29]S.Nolfi,“Evolving nontrivial behavior in autonomous robots:Adapta-

tion is more powerful than decomposition and integration,”in Evolu-tionary Robotics:From Intelligent Robots to Artificial Life,T.Gomi, Ed.Kanata,ON,Canada:AAI Books,1997.

[30]J.Gallagher,R.Beer,K.Espenschiel,and R.Quinn,“Application of

evolved locomotion controllers to a hexapod robot,”Robot.Autonomous Syst.,vol.19,no.1,pp.95–103,1996.

[31] F.Gruau,“Cellular encoding for interactive evolutionary robotics,”in

Proceedings of the Fourth European Conference on Artificial Life,P.

Husbands and I.Harvey,Eds.Cambridge,MA:MIT Press,1997. [32]N.Jakobi,“Running across the reality gap:Octopod locomotion evolved

in a minimal simulation,”in Evolutionary Robotics,O.Husbands and J.

Meyer,Eds.Berlin,Germany:Springer-Verlag,1998,Lecture Notes Computer Science.

[33]T.Gomi and K.Ide,“Emergence of gaits of a legged robot by collab-

oration through evolution,”in Proceedings of the International Sym-posium on Artificial Life and Robotics.Berlin,Germany:Springer-Verlag,1997.

[34]J.Kodjabachian and J.-A.Meyer,“Evolution and development of neural

networks controlling locomotion,gradient following and obstacle avoid-ance in artificial insects,”IEEE Trans.Neural Networks,vol.9,pp.

796–812,Sept.1998.

[35]L.Rodgrigues,M.Prado,T.Tavares,K.da Silva,and A.Rosa,“Sim-

ulation and control of biped locomotion—GA optimization,”in Proc.

IEEE Int.Conf.Evolutionary Computation,Nagoya,Japan,1996,pp.

390–395.

[36]H.de Garis,“Genetic programming:Evolution of time dependent neural

network modules which teach a pair of stick legs to walk,”in Proc.9th.

Eur.Conf.Artificial Intelligence,Stockholm,Sweden,Aug.1990,pp.

204–206.

[37]M.Mataric and D.Cliff,“Challenges in evolving controllers for physical

robots,”Robot.Autonomous Syst.,vol.19,no.1,pp.67–83,1996.

[38]MathEngine[Online].Available:https://www.sodocs.net/doc/7b10791589.html,

[39]MathEngine SDK1.x Code Book,MathEngine PLC.

[40]M.van de Panne and E.Fiume,“Sensor-actuator networks,”in Proc.

https://www.sodocs.net/doc/7b10791589.html,puter Graphics and Interactive Techniques(SIGGRAPH), 1993,pp.335–342.

[41]R.Calabrese,“Cellular,synaptic,network and modulatory mechanisms

involved in rhythm generation,”Current Opinion Neurobiol.,vol.8,no.

6,pp.710–717,1998.

[42]R.D.Beer and J.C.Gallagher,“Evovlving dynamic neural networks for

adaptive behavior,”Adapt.Beh.,vol.1,no.1,pp.91–122,1992. [43]S.Bullock,“Are artificial mutation biases unnatural?,”in Proceed-

ings5th European Conference on Artificial Life(ECAL’99), D.

Floreano,J.-D.Nicoud,and F.Mondada,Eds.Berlin,Germany: Springer-Verlag,1999,pp.64–78.

[44]W.D.Hillis et al.,“Co-evolving parasites improve simulated evolution

as an optimization procedure,”in Artificial Life II,https://www.sodocs.net/doc/7b10791589.html,ngton et al., Eds.Reading,MA:Addison-Wesley,1992,pp.313–324.

[45]R.Watson,G.Hornby,and J.Pollack,“Modeling building-block inter-

dependency,”in Parallel Problem Solving from Nature V,Eiben,Back, Schoenauer,and Schweffel,Eds.Berlin,Germany:Springer-Verlag, 1998.

[46]P.Getting and M.Dekin,“Tritonia swimming.A model system for inte-

gration within rhythmic motor systems,”in Model Neural Networks and Behavior,A.I.Selverston,Ed.New York:Plenum,1985,pp.3–19.

[47]M.Rendel,“The simulated evolution of central pattern generators,”Un-

dergraduate Project,Univ.Oxford,Oxford,U.K.,2001.

[48] B.Bodenheimer,A.V.Shlevman,and J.K.Hodgins,“The effects of

noise on the perception of animated human running,”in https://www.sodocs.net/doc/7b10791589.html,-puter Animation and Simulation’99,Eurographics Animation Work-shop,1999,pp.53–63.

[49] F.K.Skinner and B.Mulloney,“Intersegmental coordination in inver-

tebrates and vertebrates,”Current Opinion Neurobiol.,vol.9,no.6,pp.

725–732,1998.

[50]S.Grillner,P.Wallen,and G.Viana di Prisco,“Cellular network under-

lying locomotion as revealed in a lower vertebrate model:Transmitters, membrane properties,circuitry,and simulation,”in Proc.Cold Spring Harbor Symp.Quant.Biol.,vol.55,1990,pp.779–789.

[51] A.Ijspeert,J.Hallam,and D.Willshaw et al.,“From lampreys to sala-

manders:Evolving neural controllers for swimming and walking,”in From Animals to Animats5,R.Pfeifer et al.,Eds.Cambridge,MA: MIT Press,1998,pp.390–399.

[52]J.Adolfsson,Dankowicz,and A.Nordmark,“3-D stable gait in passive

bipedal mechanisms,”in Proc.357Euromech,

1998.

Torsten Reil received the B.A.(honors)degree in

biological sciences from the University of Oxford,

U.K.,in1998and the M.Sc.degre in evolutionary

and adaptive systems from the University of Sussex,

U.K.,in1999.He is currently working toward the

D.Phil.degree in theoretical biology with the Depart-

ment of Zoology,University of Oxford.

He currently teaches courses in Artificial Life and

Neural Networks at the University of Oxford.His

current research interests include the dynamics and

evolution of complex biological systems such as gene regulation networks and central pattern

generators.

Phil Husbands received the B.Sc.degree in physics

from Manchester University,Manchester,U.K.,

in1981,the M.Sc.degree in computer systems

engineering from South Bank Polytechnic,London,

U.K.,in1985,and the Ph.D.degree in com-

puter-aided engineering from Edinburgh University,

Edinburgh,U.K.,in1990.

He is currently Head of the Evolutionary and

Adaptive Systems Group at the University of

Sussex and Co-Director of the Sussex Centre

for Computational Neuroscience and Robotics. His current research interests include evolutionary robotics,biologically inspired neurocontrol systems,computational neuroscience,and evolutionary computing.

青岛市重点用能企业名单

南车四方机车车辆股份有限公司 青岛喜盈门集团公司 青岛广源发集团有限公司 青岛美高集团有限公司 济南山水集团有限公司青岛水泥分公司青岛正进集团有限公司 青岛大农服装有限公司 山东黄岛发电厂 青岛金晶股份有限公司 青岛恒源热电有限公司 青岛浮法玻璃有限公司 青岛压花玻璃有限公司 青岛市圣戈班韩洛玻玻璃有限公司 青岛高合有限公司 青岛浦项不锈钢有限公司 青岛北海船舶重工有限责任公司 青岛经济技术开发区热电燃气总公司 青岛赛轮子午线轮胎信息化生产示范基地 1 即墨市热电厂 青岛即发集团控股有限公司 青岛新源热电有限公司 青岛三湖制鞋有限公司 青岛正大有限公司 青岛高丽钢线有限公司 青岛北汇玻璃有限公司 即墨市双春水泥有限公司 青岛红领服饰股份有限公司 青岛恒光热电有限公司 青岛恒源化工有限公司 青岛天元化工股份有限公司 青岛海王纸业股份有限公司 青岛琅琊台酒业（集团）股份有限公司青岛胶南明月海藻工业有限责任公司 胶南易通热电有限责任公司 青岛泰发集团股份有限公司 青岛东亚轮胎有限公司

青岛康大外贸集团有限公司 胶南供电公司 胶南市水泥厂 2 胶南市海龙福利板纸有限公司 青岛振华工业集团有限公司 青岛德固萨化学有限公司 青岛龙发热电有限公司 青岛恒祥化肥有限公司 青岛世原鞋业有限公司 青岛华威建材有限公司 青岛广源发玻璃有限公司 青岛大明皮革有限公司 青岛昌新鞋业有限公司 青岛衣东纺织有限公司 青岛海尔金塑制品有限公司 山东金湖水泥有限公司青岛分公司 青岛福生食品有限公司 青岛信五皮革有限公司 青岛多福康食品有限公司 胶州天成玻璃工艺品厂 胶州市新纪元帘子布有限公司 青岛昌华集团股份有限公司 青岛热电集团金莱热电有限公司 青岛金浪热电有限公司 3 青岛泰光制鞋有限公司 青岛现代人热力发展有限公司 青岛金浪化工集团有限公司 青岛凤凰东翔印染有限公司 青岛九联集团股份有限公司 青岛海升果业有限责任公司 青岛交河技工塑料有限公司 青岛东方化工股份有限公司 海尔集团公司 青岛崂山玻璃有限公司 青岛啤酒第五有限公司

青岛恒源热电

注意：以下内容请进一步总结！ 青岛恒源热电有限公司 目标公司主要从事蒸汽、热水的生产及供应、蒸汽余热发电业务，同时提供供热管道及设施维修、安装业务。据介绍，目标公司开发了循环水供热工程项目，该项目是青岛市获批的第一个清洁发展机制(CDM)项目；前处该项目处于施工建设阶段，预计将于2009年上半年内正式投产。据介绍，目标公司主要负责临港工业区辖区内的蒸汽供应及热网管理，发电业务，对居民的用热服务。 公司成立于2001年，主要从事蒸汽、热水的生产及供应、蒸汽余热发电业务。 青岛恒源热电有限公司位于开发区B区供热范围，拥有12MW的抽凝式汽轮发电机组1台及12MW的背压机组1台，75t/h循环流化床锅炉3台和150t/h锅炉1台，最大供热能力是355t/h，担负着B区的生产、民用供热负荷，主要满足热电厂东部居民小区供热和山东科技大学供热。 青岛恒源热电有限公司位于青岛经济技术开发区临港工业区的中北部，海尔大道与渭河路交界处东北角，渭河路777号。厂区所在地东侧隔宽约100m绿化地为鑫龙物流公司，该公司东侧、距离本项目最近300m处为澳柯玛人才公寓；厂区南侧隔渭河路、绿化带100m处为东小庄村（原村庄平房已搬迁，现建有多座两层复式楼房），该村庄南侧、距离本项目约420m处为山孚日水食品有限公司；项目隔渭河路东南方向约200m处为澳柯玛工业园；西及西南方向隔海尔大道、渭河路均为浦项制铁有限公司；北侧与开发区消防大队以及正友砼业相邻。 企业所在地厂址东南距市中心约8km，东面距前湾港区约4.5km。 现有工程内容：青岛恒源热电有限公司主要服务于黄岛供热分区B 区（齐长城路以北、疏港高速以南、镰湾河以西、柳花泊和珠山以东片区（包括柳花泊），总占地面积约60平方公里）。企业现有锅炉规模为3×75t/h+1×130t/h 循环流化床蒸汽锅炉，总计约355t/h锅炉容量；发电机组规模为1×12MW C12-34.9/0.98（抽凝）+1×12MW B12-4.9/0.98（背压），总计发电装机容量24 MW。 近几年，恒源热电强化能源管理，合理调整运行方式，加强节能技术改造，企业能源管理工作上了一个新台阶，先后通过了“企业能源审计”、“热电联产机组认定”等审核认证工作，被评为“青岛市清洁生产企业”，2007年度“山东省节能先进企业”。 为进一步加强企业能源管理，完善优化企业节能减排工作，公司在本年度开始推行循环经济试点工作。目前，作为试点工作重点项目之一的企业冷渣机改造项目已基本完成，初步具备投运条件，预计本年度六月份正式投入运行。该项目是将循环流化床锅炉的人工排渣（温度一般在900℃），通过加装冷渣机把炉渣余热加热除盐水，将锅炉效率提高1-3%，同时解决人工放渣存在安全隐患、能源浪费以及不环保等问题，项目投资为85万元，年可节标煤700吨。

认识实习报告(青岛东亿热电厂)

热能与动力工程专业制热方向认识 实习报告 学院：机电工程学院 班级：热能一班 姓名：徐国庆 学号：201240502013

一．认识实习的目的和任务 1.认识实习的目的： （1）认识实习是四年制高等学校教学活动的实践环节之一； （2）认识实习是对学生进行火力发电厂主机（锅炉、汽轮机）、辅机（换热器、风机、水泵）及其制造厂的设备系统、生产工艺进行认识性训 练，对发电厂热力系统进行整体初步了解。 2.认识实习的任务： （1）对火力发电厂主机的认识实习 实习对象：锅炉本体、汽轮发电机本体。锅炉形式包括煤粉锅炉、循 环流化床锅炉、链条炉、余热锅炉等。汽轮机形式包括凝气式汽轮机、 背压式汽轮机、调节抽汽式汽轮机。 认识内容：设备外形特点、摆放位置、主要性能参数、安全生产常识。 （2）对火力发电厂辅助机械设备的认识实习 实习对象：制粉系统、除尘除灰系统、烟风系统、回热系统、润滑冷 却系统、水油净化系统等。 认识内容：设备外形特点、摆放位置、主要性能参数、安全生产常识。 （3）对火力发电厂设备系统的认识实习 实习对象：火力发电厂主机和辅机工程的系统。 认识内容：设备之间的空间关系、安全生产常识。 3.认识实习的意义 （1）强化学生对专业基础课程的理解 （2）国内火力发电厂的技术发展出现了新进展 CFB锅炉、燃气轮机、余热锅炉、超临界机组、烟气脱硫、布袋除尘、集中控制运行等新技术。 （3）认识实习有利于培养学生的职业精神 （4）认识实习有利于了解机组 （5）认识实习有利于了解机组建设过程 二．捷能汽轮机厂 （1）简介：汽轮机是火力发电厂三大主要设备之一。它是以蒸汽为工质，将热能转变为机械能的高速旋转式原动机。它为发电机的能量转换提供机 械能。 青岛捷能汽轮机集团股份有限公司始建于1950年，是我国汽轮机行业重 点骨干企业。拥有各种数控、数显等机械加工设备2200余台，以200MW 及以下“捷能牌”汽轮机为主导产品，拥有电站汽轮机和工业拖动汽轮 机两大系列产品，能够满足发电、石化、水泥、冶金、造纸、垃圾处理、燃气-蒸汽联合循环、城市集中供热等领域需求，年产能达500台/600万 千瓦以上。中小型汽轮机市场占有率居国内同行业首位，是目前国内中 小型汽轮机最大最强的设计制造供应商和电站成套工程总包商。 公司积极推进品牌战略，率先在汽轮机行业内取得了美国FMRC公司双重 ISO9001国际质量体系认证和ISO1400环境管理体系认证，率先在汽轮机 行业内第一个获得了“中国名牌产品”称号，先后获得了“全国AAA级 信用企业”、“中国优秀诚信企业”、“全国用户满意产品”、“山东

供热管网检修作业指导手册[青岛热电集团]

供热管网检修作业指导手册[青岛热电集团] 供热管网检修作业指导手册[青岛热电集团] 供热管网检修作业指导手册[青岛热电集团] 作者:佚名更新时间:2008-12-5 15:55:38 字体: 供热管网检修作业指导手册 1 总则 1.1 为使公司供热管网的维护、检修工作更为规范和科学合理，确保安全运行，制定作业指导手册。 1.2 本作业指导手册适用于公司供热管网的维护、检修及事故抢修。 本作业指导手册供热管网的工作压力限定为: 工作压力不大于1.6MPa(表压)，介质温度不大于300?的蒸汽供热管网。 1.3 管网的检修工作应符合原设计要求。 1.4 执行本作业指导手册时，尚应符合国家现行有关标准的规定。 2 术语 2.1 热网维修 热网的维护和检修。本作业指导手册中简称维修。 2.2 热网维护 供热运行期间，在不停热条件下对热网进行的维护工作。本作业指导手册中简称维护。 2.3 热网检修 在停热条件下对热网进行的检修工作。本作业指导手册中简称检修。 2.4 热网抢修

供热管道设备突发故障引起蒸汽大量泄漏，危及管网安全运行或对周边环境、人身安全造成威胁时进行的紧急检修工作。本作业指导手册中简称抢修。 2.5 供热管网 由热源向热用户输送和分配供热介质的管线系统。本作业指导手册中简称热网。 3 维护、检修机构设置、检修人员及设备 3.1 维护、检修机构设置及人员要求 3.1.1客户服务中心是公司高新区内供热管网运行、调度、维护、检修的责任机构，负责高新区内供热管网的维护、检修工作。 3.1.2 供热管冈的维护、检修人员必须经过培训和专业资格考 试合格后，方可独立进行维护、检修工作。供热管网维护、检修人员必须熟悉管辖范围内的管道分布情况、设备及附件位置。维护、检修人员必须掌握管辖范国内供热管线各种附件的作用、性能、构造以及安装操作和维护、检修方法。 3.1.3检修人员出门检修时应穿公司工作服，配戴上岗证，注意礼貌用语，维护公司形象。 3.2 维护、检修用主要设备与器材 3.2.1 供热管网的维护检修部门，应备有维护、检修及故障抢修时常用的设备与器材。 3.2.2检修设备、工具平时摆放在规定位置，检修设备和专用工具要有专人保管，所有设备、工具应保证完好，须保证检修时能够立即投入使用。检修物资也应分门别类码放整齐，方便查找，以保证检修、抢修时不会因为寻找物资配件而耽误时间。每次检修完后都应检查备品备件数量，发现不够时要及时与物质采购部联系进行必要地补充，确保检修时不会因无备品备件而影响检修时间与质量。

青岛西海岸公用事业集团易通热电有限公司新能源分公司_中标190922

招标投标企业报告 青岛西海岸公用事业集团易通热电有限公司新 能源分公司

本报告于 2019年9月22日 生成 您所看到的报告内容为截至该时间点该公司的数据快照 目录 1. 基本信息：工商信息 2. 招投标情况：中标/投标数量、中标/投标情况、中标/投标行业分布、参与投标 的甲方排名、合作甲方排名 3. 股东及出资信息 4. 风险信息：经营异常、股权出资、动产抵押、税务信息、行政处罚 5. 企业信息：工程人员、企业资质 * 敬启者：本报告内容是中国比地招标网接收您的委托，查询公开信息所得结果。中国比地招标网不对该查询结果的全面、准确、真实性负责。本报告应仅为您的决策提供参考。

一、基本信息 1. 工商信息 企业名称：青岛西海岸公用事业集团易通热电有限公司新能 源分公司 统一社会信用代码：91370211334195493K 工商注册号：370211120004502组织机构代码：334195493法定代表人：赵军田成立日期：2015-04-23 企业类型：有限责任公司分公司(非自然人投资或控股的法人 独资) 经营状态：注销 注册资本：/ 注册地址：山东省青岛市黄岛区相公山路723号 营业期限：2015-04-23 至 / 营业范围：为上级公司联系业务。（依法须经批准的项目，经相关部门批准后方可开展经营活动）联系电话：*********** 二、招投标分析 2.1 中标/投标数量 企业中标/投标数： 个 （数据统计时间：2017年至报告生成时间）

2.2 中标/投标情况（近一年） 截止2019年9月22日，根据国内相关网站检索以及中国比地招标网数据库分析，未查询到相关信息。不排除因信息公开来源尚未公开、公开形式存在差异等情况导致的信息与客观事实不完全一致的情形。仅供客户参考。 2.3 中标/投标行业分布（近一年） 截止2019年9月22日，根据国内相关网站检索以及中国比地招标网数据库分析，未查询到相关信息。不排除因信息公开来源尚未公开、公开形式存在差异等情况导致的信息与客观事实不完全一致的情形。仅供客户参考。 2.4 参与投标的甲方前五名（近一年） 截止2019年9月22日，根据国内相关网站检索以及中国比地招标网数据库分析，未查询到相关信息。不排除因信息公开来源尚未公开、公开形式存在差异等情况导致的信息与客观事实不完全一致的情形。仅供客户参考。 2.5 合作甲方前五名（近一年） 截止2019年9月22日，根据国内相关网站检索以及中国比地招标网数据库分析，未查询到相关信息。不排除因信息公开来源尚未公开、公开形式存在差异等情况导致的信息与客观事实不完全一致的情形。仅供客户参考。 三、股东及出资信息 截止2019年9月22日，根据国内相关网站检索以及中国比地招标网数据库分析，未查询到相关信息。不排除因信息公开来源尚未公开、公开形式存在差异等情况导致的信息与客观事实不完全一致的情形。仅供客户参考。 四、风险信息 4.1 经营异常() 截止2019年9月22日，根据国内相关网站检索以及中国比地招标网数据库分析，未查询到相关信息。不排除因信息公开来源尚未公开、公开形式存在差异等情况导致的信息与客观事实不完全一致的情形。仅供客户参考。 4.2 股权出资() 截止2019年9月22日，根据国内相关网站检索以及中国比地招标网数据库分析，未查询到相关信息。不排除因信息公开来源尚未公开、公开形式存在差异等情况导致的信息与客观事实不完全一致的情形。仅供客户参考。 4.3 动产抵押() 截止2019年9月22日，根据国内相关网站检索以及中国比地招标网数据库分析，未查询到相关信息。不排除因信息公开来源尚未公开、公开形式存在差异等情况导致的信息与客观事实不完全一致的情形。仅供客户参考。 4.4 税务信息() 截止2019年9月22日，根据国内相关网站检索以及中国比地招标网数据库分析，未查询到相关信息。不排除因信息公开来源尚未公开、公开形式存在差异等情况导致的信息与客观事实不完全一致的情形。仅供客户参考。

青岛热电集团有限公司简介

青岛热电集团有限公司成立于1993年，属于国有独资大型热电联产企业，主要担负着青岛市企、事业单位和居民供热及部分发电任务，同时，供热市场辐射黄岛、平度、莱西、即墨、城阳等县市区域。集团公司先后成立了工程公司和具有甲级设计资质的设计院，逐步形成了热电联产、区域锅炉、热网输配等多种供热形式并存，集供热、发电、热力设计、工程施工、热力产品制造经营为一体的完整产业链。 目前，热电集团为全省地方最大供热企业。企业资产总额48亿元，年销售收入16.2亿元，所属企业16个，职工2200余人，年供蒸汽312万吨，年发电能力9.3万千瓦，已建成蒸汽管网145.43公里，热水管网1552.93公里，供（换）热站294座，供热面积3561万平方米，拥有单位用户292家，居民用户28.8万余户。 集团公司先后被评为全国AAA级信用企业、全国建设系统文明服务示范窗口单位、思想政治工作先进单位、企业文化建设先进单位、精神文明建设先进单位；山东省文明单位、节能先进企业、思想政治工作优秀企业；青岛市和工商年度免检企业、安全生产先进单位、廉洁勤政先进单位；山东省供热协会副理事长单位。 自成立以来，公司始终秉承“关爱社会、服务民生”的企业宗旨和“励精图治、锲而不舍”的企业精神，贯彻科学发展，创新经营管理，实现了企业快速发展。1996年在全国供热行业首家推出社会服务责任赔偿制度，1997年在山东省供热行业首家进行了股份制改造，1998年在山东省供热行业首家成功地进行了集团产权制度改革，1999年在全国同行业中首家通过了ISO9001国际质量认证，并先后通过了ISO14001环境管理体系和GB/T28001-2001职业健康安全管理体系认证，2001年公司成为全国供热行业中首家申请注册服务商标的企业，推出“暖到家”服务品牌，并被评为山东省著名商标和服务名牌。“青岛热电”正在逐步步入标准化、规范化、品牌化的发展轨道。 招聘专业及人数: 1、结构专业1人(研究生)； 2、建筑专业1人(研究生)； 3、技经专业1人(研究生)； 4、焊接技术与工程1人； 5、无损检测专业1人；

五大电力发电厂及下属详细

华能集团所属电厂： 华能丹东电厂华能大连电厂华能上安电厂华能德州电厂华能威海电厂华能济宁电厂华能日照电厂华能太仓电厂华能淮阴电厂华能南京电厂华能南通电厂华能上海石洞口第一电厂华能上海石洞口第二电厂华能长兴电厂华能福州电厂华能汕头燃煤电厂华能汕头燃机电厂华能玉环电厂华能沁北电厂华能榆社电厂华能辛店电厂华能重庆分公司华能井冈山电厂华能平凉电厂华能岳阳电厂华能营口电厂华能邯峰电厂 大唐集团所属： 长山热电厂湖南省石门电厂鸡西发电厂洛阳首阳山电厂洛阳热电厂三门峡华阳发电公司河北马头电力公司唐山发电总厂北京大唐张家口发电总厂兰州西固热电有限公司合肥二电厂田家庵发电厂北京大唐高井发电厂永昌电厂北京大唐陡河电厂南京下关发电厂安徽淮南洛河发电厂保定热电厂略阳发电厂微水发电厂峰峰发电厂含岳城电站天津大唐盘山发电公司内蒙大唐托克托发电公司保定余热电厂华源热电有限责任公司阳城国际发电有限公司辽源热电有限责任公司四平发电运营中心长春第二热电有限公司晖春发电有限责任公司鸡西热电有限责任公司佳木斯第二发电厂台河第一电厂江苏徐塘发电有限公司安徽省淮北发电厂安徽淮南洛能发电公司安阳华祥电力有限公司许昌龙岗发电有限公司华银电力株洲发电厂华银株洲发电公司金竹山电厂华银金竹山火力发电厂湘潭发电有限责任公司湖南省耒阳发电厂灞桥热电有限责任公司灞桥热电厂陕西渭河发电厂陕西延安发电厂陕西韩城发电厂永昌发电厂甘肃甘谷发电厂甘肃八0三发电厂甘肃连城发电厂甘肃兰西热电有限公司广西桂冠电力股份公司桂冠大化水力发电总厂广西岩滩水电厂陈村水力发电厂王快水电厂张家界水电开发公司贺龙水电厂鱼潭水电厂陕西石泉水力发电厂石泉发电有限责任公司甘肃碧口水电厂百龙滩电厂华电所属： 1中国华电工程(集团)有限公司2华电煤业集团有限公司3华电财务有限公司4华电招标有限公司5华信保险经纪有限公司6北京华信保险公估有限公司7河北热电有限责任公司8包头东华热电有限公司（在建）9内蒙古华电乌达热电有限公司（在建）10华电国际电力股份有限公司11华电国际电力股份有限公司邹县发电厂（扩建）12华电国际电力股份有限公司莱城发电厂13华电国际电力

(集团发布)青岛热电集团有限公司关于实施供热计量收费工作的意见

青热电〔2010〕121号 青岛热电集团有限公司 关于实施供热计量收费工作的意见 各单位、处室： 为全面贯彻《山东省物价局、山东省住房和城乡建设厅关于推进供热计量改革的指导意见》，根据青热办【2010】25号文件要求，自2010年开始，新供热建筑及完成供热计量改造的既有居住建筑，取消以面积计价收费，实行按用热量计价收费，为做好供热计量收费工作，经研究确定以下实施意见： 一、实施计划 （一）对已经改造完成的既有居住建筑实施供热计量收费，明细如下：第一热力海信慧谷、丰华园、弘信花园、都市名家小区；第二热力公司天宝苑小区；金河热力公司荣馨苑小区。

（二）对新竣工非居住建筑全面按用热量计量收费。 二、实施措施： （一）加强组织领导，责任到位。 职责明确，责任到人。集团成立以董事长为组长，总经理为副组长的供热计量工作领导小组，工程开发处、生安处、财务处、服务处各司其职，全力做好供热计量收费实施工作。各所属生产单位必须成立工作领导小组，将宣传、收费、数据公示、政策答疑等工作落实到位，各单位要有专门的供热计量工作负责人。 （二）措施到位，计划周密。 责任部门要全力做好实施热计量收费工作的计划安排，配合相关科室做好用户协调、宣传、合同签订、数据公示以及收、退费工作。 （三）做好新建、竣工项目供热计量设施的管理工作。各相关部门要严格新建、在建、新竣工项目供热计量设施的审核、把关、验收和问题汇总工作。 三、工作要求 （一）做好用户宣传、解释工作。在张贴用户通知进行宣传的同时，相关人员要明确供热计量工作实施相关要求规定，收费方式以及供热调节方式等，做好对用户的宣传解释工作，让用户明白调节方式和收退费方式、时间等。 （二）做好用户结算工作。用户的供热计量数据要真实、准确，各单位要认真做好用户仪表底数（正式供热时间和停止供热时间）确认工作，并定时张贴通知公开热量数据。 （三）做好数据分析和总结。对供热数据按周期进行定

青岛钢铁公司城市钢厂环保搬迁项目环境影响报告书

1建设项目概况 1.1建设项目背景及建设地点 1.1.1建设项目背景 青岛钢铁有限公司（以下简称青钢）始建于1958年，位于青岛市北李沧区，属城市钢铁厂。厂区东邻重庆路、南渠村；西围墙距胶济铁路约85m；北距流亭国际机场约3.8km；南靠遵义路。距市中心约15km，厂内铁路专线与胶济铁路娄山站接轨，全厂总占地面积不足1.3km2。 青钢是集焦化、烧结、炼铁、炼钢、轧钢、发电等为一体的钢铁联合企业，是山东省重要的优质棒线材生产基地，山东省三大钢铁支柱企业。 目前，青钢已形成年产铁、钢、材各400×104t的生产能力。其主要生产设施有：60×104t焦化厂；2×50m2烧结机，2×105m2烧结机；5×500+1×625m3高炉；一炼钢有4×35t转炉，5座30tLF精炼炉，3台4机4流R5m小方坯连铸机和1台4机4流R8m连铸机；二炼钢有2×80t 顶底复吹转炉，3座90tLF精炼炉，1座90tRH精炼炉，2台6机6流R9m 连铸机；轧钢车间有1＃、2＃、3＃高速线材车间，复二重线材车间，半连续小型车间，横列式小型车间，还有与之配套的相应公辅设施。 青钢产品有：热轧盘条，热轧带肋钢筋、圆钢、扁钢等型钢。主要品种有：焊接用钢盘条、汽车用弹簧扁钢、硬线盘条、冷镦钢、PC钢棒用线材、拉丝线材、易切削钢、优质碳素结构圆钢、建筑用线材与螺纹钢等。2011年，青钢生产生铁330.69×104t、钢318.27×104t、钢材301.99×104t。2011年工业总产值297.79亿元，工业销售产值299.47亿元。青钢现有产品以优特钢为主，品种结构具有特色，产品附加值高。青钢现有职工1万余人，其中各类工程技术人员约1500余人。 自1997年以来，青钢经过不断的技术改造，其生产能力和经济效益均有了较大幅度的提高，但由于历史原因，还存在许多问题，如产品结构性矛盾仍