WO2017071230A1 - Method for short-term optimal scheduling of multi-agent hydropower station group - Google Patents

Abstract

Disclosed in the present invention is a method for short-term optimal scheduling of a multi-Agent hydropower station group, emphasizing on improving the solving speed of optimal scheduling and the operation efficiency of a hydropower station group, solving the problem of the current solving technique not being able to satisfy the short-term optimal scheduling of a large scaled hydropower station group. The present invention may maximize the power generation benefits of the whole hydropower station group, and have important significance on both pushing the development of optical scheduling of cascade hydropower stations and improving economic development level.

Description

Short-term optimal scheduling method for hydropower station group with multi-agent Technical field

The invention belongs to the technical field of testing, and particularly relates to a short-term optimal scheduling method for a hydropower station group based on task automatic distribution type multi-agent.

Background technique

With the increasing scale of domestic hydropower stations and the establishment of river basin rolling development companies, large-scale, inter-basin, inter-provincial and inter-regional areas have become prominent features of hydropower dispatching in China, resulting in many new complex scheduling and operational technical problems. Outstanding performance in terms of system solution and operational efficiency. Large-scale hydropower station group optimization scheduling problems are characterized by high dimensionality, nonlinearity, multi-stage, strong constraints, etc. The solution process is very difficult, and the expanding hydropower system has more complicated hydraulic and electrical connections between power stations and cascades. Increased the number of constraints for the problem. For the conventional solution method, the calculation amount and calculation time will increase exponentially with the number of power stations and constraints, which cannot effectively solve the problem of large-scale hydropower station group optimization scheduling. In addition, with the gradual increase of the number of hydropower stations in the basin, the hydraulic and electric coupling between the power stations is getting closer and closer, which leads to the increasingly complicated problem of short-term optimal dispatching of hydropower stations. The huge and complex optimization scale directly leads to more time spent on calculations, but the short-term optimal scheduling of hydropower stations requires very high speeds, and conventional calculation methods can no longer meet the requirements of their calculation speed. The existing various solving techniques have become more and more difficult to meet the practical needs of the large-scale hydropower station group short-term scheduling solution speed requirements, that is, the solution time can not meet the demand.

The research and application of Agent intelligence technology in the computer field originated from a series of researches on distributed artificial intelligence conducted by researchers in the United States in the mid-1970s. They found that Agents can combine some simple information systems into a large overall system. Effectively improve its ability to handle complex problems, and improve the intelligence level of the overall system by defining appropriate coordination mechanisms. This results in an Agent system concept and method that is artificially intelligent and can passively sense problem processing information and can actively predict, analyze, and even actively seek solutions to support users in completing tasks.

Multi-Agent System (MAS) refers to a combination of multiple executable Agent subsystems. The multi-agent system can solve the target problem, modify its behavior as the environment changes, and can communicate, interact and coordinate with other Agents through the network to complete the same task. Typically, each Agent is defined as a physical or abstract reality. In the network and distributed environment, each agent individual is independent, can act on the individual and external environment changes, can manipulate part of the external environment change, and can reflect the changes of the external environment, and more It is important to communicate, interact, and coordinate with other Agents to accomplish the desired mission objectives. Therefore, Agent should have the characteristics of autonomy, interactivity, responsiveness and initiative. The multi-agent system can simulate the group work of large organizations and can solve complex problems in the group by using its unique methods and methods. The structure of the multi-agent system is shown in Figure 1.

At present, the solution models for short-term optimal scheduling of cascade hydropower stations are divided into two categories. The first type is a water-powered model with a given The water is optimally distributed to produce more electric energy; the second type is the electric water-fixing model, which is distributed to each power station based on the real-time load or load curve of the given hydropower station group, and then distributed to each unit, so that the entire hydropower station consumes water. Can be the least.

At present, the structural diagram of the technical scheme of Agent technology applied to the optimal scheduling of hydropower station group is shown in Figure 2, and its specific contents are as follows:

(1) Establish two types of agents. The first is the cascade center agent, which represents the entire cascade center and serves as the management coordination function. The second is the power station agent, which has the basic data of the power station and can perform the start-stop arrangement and unit load optimization. There is an optimization algorithm in each agent. Under the cascade center, each power station is built as a power station agent.

(2) When optimizing the calculation, the step agent receives the load scheduling information of the upper power grid and distributes the load to the power station agent through the optimization algorithm, and then waits for the power station agent to feedback the optimization result.

(3) Each power station agent receives the load information assigned by the ladder agent, and then performs the startup and shutdown calculation and load distribution optimization for its power station, and then sends the result back to the ladder agent.

(4) The step-level agent performs the operation update adjustment according to the data fed back by each power station agent, and then sends the load information to the power station agent; or when the optimization request is reached, it ends and no longer communicates.

The existing short-term optimal scheduling model solves either single-thread computing with a single computer, multi-thread parallel computing with a single computer, parallel computing with multiple computers, and multi-agent collaborative computing. Obviously, due to the increasing scale of cascade hydropower stations, the calculation method of a single computer cannot meet the needs of calculation speed; the current multiple parallel calculations are simple division tasks, which can shorten the calculation time to a certain extent, but cannot be intelligently assigned tasks. Intelligent coordination, etc., resulting in unsatisfactory results. The existing multi-agent collaborative computing is to plan the number of agents according to the number of hydropower stations. There is no significant difference between the parallel calculations and the traditional multi-machines, and the calculation speed cannot be improved. Moreover, the situation of newly added hydropower stations and new units cannot be adapted to the need to rewrite the Compilation Agent. The short-term optimal scheduling algorithm for hydropower station group is generally divided into two categories. One is the classical algorithm used or its improved algorithm. The classical algorithm such as dynamic programming needs to traverse all possible solutions for the optimization calculation of large hydropower stations. Therefore, it is prone to "dimensional disasters" and the calculation time is too long. The other is the modern intelligent optimization algorithm used, which is studied by many scholars because of its fast calculation speed and fast convergence. However, since it is randomly generated from the initial solution and the iterative process has a large amount of random quantities, it can not guarantee 100% convergence. And easy to fall into local extremes. Therefore, the two types of algorithms have their own shortcomings. The disadvantage of the classical algorithm and its improved algorithm is that when the number of units or the number of power stations increases, it is prone to "dimensionality disaster", resulting in a sharp increase in computing time. The shortcoming of the modern intelligent optimization algorithm is that it cannot guarantee convergence every time, and it is easy to converge to local extremum rather than global extremum. Therefore, further research is needed.

Summary of the invention

In order to solve the deficiencies in the prior art, the present invention provides a short-term optimal scheduling method for a hydropower station group based on task automatic distributed multi-agent. The present invention focuses on improving the speed and operational efficiency of hydropower station group optimization scheduling, and solving the existing solving technology. The problem of short-term optimal dispatching of large-scale hydropower stations cannot be satisfied. The present invention will maximize the power generation efficiency of the entire hydropower station group, and is of great significance for promoting the development of optimal scheduling of cascade hydropower stations and improving the economic operation level.

In order to solve the above problems, the present invention specifically adopts the following technical solutions:

A multi-agent hydropower station group short-term optimal scheduling method, characterized in that it comprises the following steps,

Step (1), establish a short-term optimal scheduling model for hydropower stations

Objective function:

Where E is the sum of energy consumption of each power station; T is the scheduling period; K is the number of power stations; Q _{k, t} is the power generation flow of the kth power station t period; P _{k, t} is the kth power station t period Output; H _k,t is the head of the kth power station t period;

Restrictions:

(1) System load balancing constraints

t=1,2...,T

(2) Power limit of each hydropower station

k=1, 2..., K t=1, 2..., T

(3) Water balance equation V _k,t+1 =V _k,t +(Qin _k,t -Qgen _k,t -Qdis _k,t )Δt,k=1,2...,K t=1,2... ,T

(4) Reservoir storage tolerance Vmin _k ≤V _k,t ≤Vmax _k ,k=1,2...,K t=1,2...,T

(5) Cascade power station water flow lag constraint

k=1,2...,K t=τ _k ...,T

Where, P _t is the step load of the t period, and P _k,t is the output of the k power station t period;

Is the lower limit of the output of the kth power station during the tth period.

It is the upper limit of the output of the kth power station in the t-th period; V _k,t is the capacity of the kth power station in the t-th period, V _k,t+1 is the storage capacity of the _k-th power station in the t+1th period, Qin _k,t is the kth The power flow of the power station in the t-th period, Qgen _{k, t} is the power flow of the kth power station in the t-th period, Qdis _{k, t} is the water flow of the k-th power station in the t-th period, Δt is the time length of the unit period, Vmin _k For the k-power station minimum storage capacity constraint and the integration of medium- and long-term planning constraints, Vmax _k is the maximum storage capacity constraint of the kth power station and the combination of medium and long-term planning constraints; Qq _{k, t} is the water supply of the kth power station in the upstream interval of the tth time interval. ;

For the discharge of the kth power station into the upstream power station during the tth period; τ _k is the runoff propagation time from the k-1th to the kth power station;

Step (2), establishing a short-term optimal scheduling sub-model

Objective function:

Where, Q is the sum of water consumption and flow in the T cycle of the power station, T is the scheduling period, n is the number of units, Qgen _{t, i} is the power generation flow of the unit at time _t, and N _{t, i} is the output of the unit at time t, H _t is the net head at the time _t of the power station, and Qdis _t is the water flow at the time of the power station t;

Restrictions:

(1) Power balance

i=1,2...,n

(2) Water balance constraint V _t+1 =V _t +(Qin _t -Qgen _t -Qdis _t )Δt,t=1,2...,T

(3) Output constraint, which determines the turbine efficiency and head N _i,min ≤N _i ≤N _i,max ,i=1,2...,n

(4) Reference flow constraint Qout _min ≤ Q _i ≤ Qout _max , i = 1, 2..., n

(5) Unit inoperable area

i=1,2...,n

(6) Storage tolerance Vmin ≤ V _t ≤ Vmax, t = 1, 2..., T

Where, N _s is the output of the cascade to the power station, N _i is the output of the power station i, V _t is the capacity of the power station in the t-th period, Qin _t is the water flow of the power station in the t-th period, Qgen _t is the power station in the t-th During the period of power generation, Qdis _t is the abandoned water flow of the power station during the t-th period, Δt is the length of time in the unit period, N _i,min is the minimum output of the i unit, N _i,max is the maximum output of the i unit, and Qout _max is the maximum output of the unit. Flow, Qout _min is the minimum over flow of the unit.

It is the lower limit of the operating zone of the i unit.

It is the upper limit of the operating area of the i unit, Vmin is the minimum storage capacity constraint of the power station and the comprehensive minimum storage capacity constrained by the medium and long-term planning. Vmax is the maximum storage capacity constraint of the power station and the comprehensive maximum storage capacity constrained by the medium and long-term planning;

Step (3), establishing an agent model according to the model in step (1) and step (2), the agent model includes an inter-station load distribution agent, an in-station start-stop agent, and an in-station load optimization agent.

The task of the inter-station load distribution agent is to monitor whether the step center receives a new next-day load curve, and once the new next-day load curve is monitored, Np=(Np ₁ , Np ₂ , Np ₃ , . . . , Np ₉₆ The inter-station load distribution agent immediately assigns the new next-day load curve to each power station according to a predetermined algorithm. The station-initiated shutdown agent and the station-in-load optimization agent respectively calculate the start-stop plan for the power station unit, distribute the load, and distribute the load between stations. The agent stores the result and continues to repeat the allocation until the result is satisfactory. The inter-station load distribution agent is also used to delete the redundant agent and increase the agent to achieve the purpose of adapting to the optimization calculation;

The task of the start-stop agent in the station is to formulate a start-stop plan and find a feasible area of the unit that passes through the vibration zone at least. The station-starting agent monitors whether each power station has a new load curve, and if the power station within the monitoring range is allocated new The load curve is immediately calculated and expanded;

The task of the station load optimization agent is to optimize the distribution load to each unit according to the minimum water consumption criterion according to the start-stop plan and the feasible area of the unit, and then the station load optimization agent sends the water consumption energy in the scheduling period to the station load distribution. Agent

Step (4), the Agent algorithm library is built. The algorithm library contains two kinds of algorithms, a classical algorithm and a heuristic intelligent optimization algorithm. Each calculation randomly selects a classical algorithm and an intelligent optimization algorithm, and performs parallel calculation to calculate the calculation result. Compare and then feed back The Agent algorithm library and the Agent algorithm library make good and bad statistics of the algorithm. The probability that the excellent algorithm is selected increases, the probability of the low efficiency algorithm is small, and finally it is gradually eliminated.

In step (5), the Agent model established in the step (3) is invoked by the Agent algorithm library established in the step (4) to realize short-term optimal scheduling of the multi-agent hydropower station group.

The foregoing multi-agent short-term optimal scheduling method for a hydropower station group is characterized in that: the inter-station load distribution agent is set to one, and is set in an agent main container of a computer, wherein an agent platform is provided in the computer, and an agent is provided. The main container computer is connected to several other computers, and the other computers have an Agent platform, and a multi-agent platform is formed with the Agent platform in the computer with the Agent main container, and the Agent main container also has a management agent. The management agent is responsible for state management, agent management and startup of the main program of the entire multi-agent platform. The corresponding agent container of the other computer has a sub-agent container, and the sub-agent container has an on-site shutdown agent and an internal load. Optimize the Agent.

The foregoing multi-agent hydropower station group short-term optimization scheduling method is characterized in that the multi-agent optimization process is performed when using the heuristic intelligent optimization algorithm,

(1) The scale of the start-stop agent in a given station is N1, the size of the inter-station load distribution agent is N2, and the scale of the initial station-to-station load distribution scheme is M;

(2) After the station load distribution agent receives the new next-day load curve, it randomly selects the initial load distribution plans of the power station, assigns its load curve to each power station, and then monitors whether each power station is allocated. If the distribution is finished, immediately According to the M plans and results issued last time, the M plans are updated, and then sent to the hydropower station, and so on, until the best solution for convergence is found;

(3) When the station starts to stop the agent and monitors that a new load curve is assigned to the power station, it immediately starts the development of the start-stop plan, then finds the feasible area that passes through the minimum vibration zone, feeds back to the power station, and uses the pre-emption agent in the N1 station. The treatment of the power station, that is, for the M load schemes of a single power station, may be completed by a single station within the shutdown agent, or may be completed by multiple stations within the station, when an agent completes the task of a power station, then continue to scan the remaining The power station, if there is an unfinished task, immediately joins the execution task;

(4) When any station internal shutdown agent completes the start-stop plan of a certain power station and the minimum feasible area of the vibration zone, the station load optimization agent will immediately monitor the task and immediately plan for the power station. The unit load is optimized and distributed, and the preemptive task is also used. The idle station load optimization agent automatically monitors the load curve of each unit that is not assigned to the unit. Once it is monitored, it is immediately allocated to it, and then continues to monitor.

The beneficial effects of the invention are:

Because the present invention solves the two-dimensional model of short-term optimal scheduling of hydropower station group, the multi-agent technology multi-agent asynchronous parallel computing method is adopted, so the short-term scheduling calculation of large hydropower station group can be greatly shortened compared with the traditional calculation method. calculating time;

Since the Agent has an algorithm library intelligent selection algorithm strategy in the present invention, it is more intelligent than a single invariant algorithm, and the optimization effect is obtained. Better, the function of asynchronous parallel computing between classical algorithms and modern intelligent algorithms can make use of the advantages of modern intelligent algorithms on the one hand, and can make up for the shortcomings of intelligent algorithms by using classical algorithms. Therefore, the optimization effect of this method on short-term optimal scheduling of hydropower stations is more effective. Ok, the result is more reliable;

Since the optimization tasks of the hydropower stations of the hydropower station group are respectively divided into the respective agents on different computers, the tasks are refined and separated, and finally the tasks are distributed, so the method avoids the "dimension" of the traditional algorithm. Disaster problem;

Since the task is automatically assigned to the agent in the present invention, the agent is object-oriented rather than object-oriented in a similar solution, and has better adaptability to newly added power stations, faulty power stations, and new power plants in a hydropower station group, and the like. In the face of the increase or decrease of the power station or the unit, it is necessary to rewrite the compiling agent, which may cause inconvenience, so the present invention has the advantage of adapting to the change of the power station or the unit.

DRAWINGS

Figure 1 is a schematic diagram of a multi-agent system;

2 is a structural diagram of a technical solution of applying the Agent technology to the optimal scheduling of a hydropower station group;

3 is a structural diagram of optimized scheduling of a multi-agent cascade hydropower station according to the present invention;

4 is a schematic diagram of operation of an in-station load distribution agent according to the present invention;

Figure 5 is a distribution diagram of a multi-agent platform of the present invention;

Figure 6 is a flow chart of the main program of the present invention;

7 is a flowchart of an inter-station load distribution agent according to the present invention;

8 is a flowchart of an in-station agent of the present invention;

detailed description

The invention is further described below in conjunction with the embodiments and the accompanying drawings.

The present invention is directed to an electric water setting model.

A short-term optimal scheduling method for hydropower station group based on task automatic allocation multi-agent, including the following steps

Step (1), establish a short-term optimal scheduling model for hydropower stations

Objective function:

Where E is the sum of energy consumption of each power station; T is the scheduling period; K is the number of power stations; Q _{k, t} is the power generation flow of the kth power station t period; P _{k, t} is the kth power station t period Output; H _k,t is the head of the kth power station t period;

Restrictions:

(1) System load balancing constraints

t=1,2...,T

(2) Power limit of each hydropower station

k=1, 2..., K t=1, 2..., T

(3) Water balance equation V _k,t+1 =V _k,t +(Qin _k,t -Qgen _k,t -Qdis _k,t )Δt,k=1,2...,K t=1,2... ,T

(4) Reservoir storage tolerance Vmin _k ≤V _k,t ≤Vmax _k ,k=1,2...,K t=1,2...,T

(5) Cascade power station water flow lag constraint

k=1,2...,K t=τ _k ...,T

Where, P _t is the step load of the t period, and P _k,t is the output of the k power station t period;

Is the lower limit of the output of the kth power station during the tth period.

It is the upper limit of the output of the kth power station in the t-th period; V _k,t is the capacity of the kth power station in the t-th period, V _k,t+1 is the storage capacity of the _k-th power station in the t+1th period, Qin _k,t is the kth The power flow of the power station in the t-th period, Qgen _{k, t} is the power flow of the kth power station in the t-th period, Qdis _{k, t} is the water flow of the k-th power station in the t-th period, Δt is the time length of the unit period, Vmin _k For the k-power station minimum storage capacity constraint and the integration of medium- and long-term planning constraints, Vmax _k is the maximum storage capacity constraint of the kth power station and the combination of medium and long-term planning constraints; Qq _{k, t} is the water supply of the kth power station in the upstream interval of the tth time interval. ;

For the discharge of the kth power station into the upstream power station during the tth period; τ _k is the runoff propagation time from the k-1th to the kth power station;

Step (2), establishing a short-term optimal scheduling sub-model

Objective function:

Where, Q is the sum of water consumption and flow in the T cycle of the power station, T is the scheduling period, n is the number of units, Qgen _{t, i} is the power generation flow of the unit at time _t, and N _{t, i} is the output of the unit at time t, H _t is the net head at the time _t of the power station, and Qdis _t is the water flow at the time of the power station t;

Restrictions:

(1) Power balance

i=1,2...,n

(2) Water balance constraint V _t+1 =V _t +(Qin _t -Qgen _t -Qdis _t )Δt,t=1,2...,T

(3) Output constraint, which determines the turbine efficiency and head N _i,min ≤N _i ≤N _i,max ,i=1,2...,n

(4) Reference flow constraint Qout _min ≤ Q _i ≤ Qout _max , i = 1, 2..., n

(5) Unit inoperable area

i=1,2...,n

(6) Storage tolerance Vmin ≤ V _t ≤ Vmax, t = 1, 2..., T

Where, N _s is the output of the cascade to the power station, N _i is the output of the power station i, V _t is the capacity of the power station in the t-th period, Qin _t is the water flow of the power station in the t-th period, Qgen _t is the power station in the t-th During the period of power generation, Qdis _t is the abandoned water flow of the power station during the t-th period, Δt is the length of time in the unit period, N _i,min is the minimum output of the i unit, N _i,max is the maximum output of the i unit, and Qout _max is the maximum output of the unit. Flow, Qout _min is the minimum over flow of the unit.

It is the lower limit of the operating zone of the i unit.

It is the upper limit of the operating area of the i unit, Vmin is the minimum storage capacity constraint of the power station and the comprehensive minimum storage capacity constrained by the medium and long-term planning. Vmax is the maximum storage capacity constraint of the power station and the comprehensive maximum storage capacity constrained by the medium and long-term planning;

Step (3), according to the model in step (1) and step (2), the agent model is established, and the multi-agent architecture of the short-term optimal scheduling of the cascade hydropower station is designed as the agent network structure. The model for the solution similar to the present invention is for the hydropower station, and one hydropower station establishes an agent. If the new power station is newly installed, it is necessary to newly compile and compile the cascade agent and the new power station agent. Similarly, if a new unit is added to a power station, the whole program will also be caused. In addition, the computing power of a single power station agent is limited, and the speed of the single power station agent cannot be completed. The present invention aims at the problem that the cascade center and each hydropower station are multi-agent environments, and establish an inter-station load distribution agent, an in-station start-stop agent, and an in-station load optimization. Agent. This not only facilitates the increase and decrease of the power station and the unit without affecting the deployment of the agent, but also increases the progress of the task completion by adding an agent, so that the total calculation time is reduced.

The task of the inter-station load distribution agent is to monitor whether the step center receives a new next-day load curve, and once the new new next-day load curve Np=(Np ₁ , Np ₂ , Np ₃ , ..., is monitored, Np ₉₆ ), the inter-station load distribution agent immediately assigns the new next-day load curve to each power station according to a predetermined algorithm. The station-initiated shutdown agent and the station-in-load optimization agent respectively calculate the start-stop plan for the power station unit and distribute the load, and then the station The load distribution agent stores the result and continues to repeat the allocation until the result is satisfactory. The inter-station load distribution agent is also used to delete some redundant agents and add some agents to achieve the purpose of adapting to the optimization calculation; The task is to distribute load between stations, the pursuit of minimum water consumption, the less water consumption, the better the effect, so the agent is designed as an effect agent;

The two tasks that need to be completed in the station start-stop agent, formulate the start-stop plan and find the feasible domain that traverses the vibration zone at least. The task of the start-stop agent in the station is to monitor whether each power station has a new load curve. If the power station within the monitoring range is assigned a new load curve, it will be immediately calculated. The specific process is to formulate a start-stop plan based on the start-stop rule, and then find the optimal feasible domain according to the different feasible domain combinations of each unit. If there is no feasible domain, re-establish the start-stop plan;

The task of the station load optimization agent is to optimize the distribution load to each unit according to the minimum water consumption criterion according to the start-stop plan and the feasible area of the unit, and then the station load optimization agent sends the water consumption energy in the scheduling period to the station load distribution. Agent; the stop and stop agent in the station is reactive, and the load optimization agent in the station is effective;

Step (4), establishing an agent algorithm library, the algorithm library includes two types of algorithms, a classical algorithm and a heuristic intelligent optimization algorithm, each The second calculation randomly selects a classical algorithm and an intelligent optimization algorithm, performs parallel calculation, compares the calculation results, and then feeds back to the Agent algorithm library. The Agent algorithm library performs the algorithm's pros and cons statistics, and the excellent algorithm is selected. Large, inefficient algorithms are less likely to be selected and eventually phase out;

In step (5), the Agent model established in the step (3) is invoked by the Agent algorithm library established in the step (4) to realize short-term optimal scheduling of the multi-agent hydropower station group.

In the short-term optimal dispatch of hydropower station group, the multi-agent model structure is shown in Figure 3. It consists of an inter-station load distribution agent, multiple in-station shutdown agents, multiple in-station load optimization agents, and a power station environment. Each agent communicates with each other through the network. The power station and the cascade center are multi-agent monitoring environments, and each agent completes the corresponding task by monitoring the corresponding targets and collaborative cooperation.

In the joint optimization operation, under normal circumstances, there is only one station-to-station load distribution agent, which is responsible for the load distribution and coordination of the entire hydropower station group. There are multiple station-initiated shutdown agents and station-in-load optimization agents, and the agents in each station are distributed on different computers. Take the heuristic intelligent optimization algorithm as an example to illustrate the multi-agent optimization process as follows:

(1) The scale of the start-stop agent in a given station is N1, the size of the inter-station load distribution agent is N2, and the scale of the initial station-to-station load distribution scheme is M;

(2) After the station load distribution agent receives the new next-day load curve, it randomly selects the initial load distribution plans of the power station, assigns its load curve to each power station, and then monitors whether each power station is allocated. If the distribution is finished, immediately According to the M plans and results issued last time, the M plans are updated, and then sent to the hydropower station, and so on, until the best solution for convergence is found;

(3) When the station starts to stop the agent and monitors that a new load curve is assigned to the power station, it immediately starts the development of the start-stop plan, then finds the feasible area that passes through the minimum vibration zone, feeds back to the power station, and uses the pre-emption agent in the N1 station. The treatment of the power station, that is, for the M load schemes of a single power station, may be completed by a single station within the shutdown agent, or may be completed by multiple stations within the station, as shown in Figure 4, when an agent completes the task of a power station , then continue to scan the remaining power stations, if there are unfinished tasks, immediately join the execution task;

(4) When any station internal shutdown agent completes the start-stop plan of a certain power station and the minimum feasible area of the vibration zone, the station load optimization agent will immediately monitor the task and immediately plan for the power station. The unit load is optimized and distributed, and the preemptive task is also used. The idle station load optimization agent automatically monitors the load curve of each unit that is not assigned to the unit. Once it is monitored, it is immediately allocated to it, and then continues to monitor. As shown in Figure 4, the calculation of each agent to the power station is random.

The joint optimization operation of multi-agent is similar to a group of bee feeding processes. The queen bee coordinates and publishes the task. The worker bees search for flowers in a certain area, and once they are found, they immediately eat pollen. It is possible that many bees pick a flower or there may be only one bee. If you finish eating flowers, look for the next flower. Repeat the loop all the time.

The inter-station load distribution agent is set to one, and is set in the agent main container of the computer. The computer has an agent platform, and the computer with the agent main container is connected to several other computers, and the other computers are provided with an agent. Platform, and The Agent platform in the computer with the Agent main container forms a multi-agent platform, and the Agent main container also has a management agent, which is responsible for state management, Agent management and startup of the main program of the entire multi-agent platform, and the like. The corresponding agent container in the computer's Agent platform is provided, and the sub-agent container is provided with an in-site shutdown agent and an in-station load optimization agent. In fact, the multi-agent system can run on any computer operating system, as long as the Agent platform is installed on the system, the Agent can be run. The Agent runs on the platform and does not depend on the operating system, so it only needs to be programmed once. There is one Agent container in each Agent platform. The entire Multi-Agent system needs a main container. The main container has a management agent for managing and coordinating Agent containers in other computers. The hardware structure of the multi-agent joint operation is shown in Figure 5. In Figure 5, the management agent is responsible for state management, agent management, and startup of the main program of the entire multi-agent platform. The startup flowchart of the main program is shown in Fig. 6. In the main container of the agent, an inter-station load distribution agent is arranged, and in the agent container of other platforms, an in-station agent, that is, an in-station stop agent and an in-station load optimization agent are arranged. The calculation process of the main program and each agent is shown in Fig. 6, Fig. 7, and Fig. 8. In particular, the Agent is always there, so it should constantly monitor the target.

The invention proposes a parallel computing method based on task automatic allocation type multi-agent technology in short-term optimal scheduling of hydropower station groups. The multi-agent applied to the hydropower station group short-term optimization scheduling method can be used for short-term optimal scheduling of cascade or basin hydropower station group. The agent is designed to complete the self-running software of a certain task, and independently complete each part of the optimization process in a preemptive manner. task. Multiple Agents can perform arithmetic processing on the same hydropower station at the same time to achieve the purpose of shortening the solution time. The increase in the number of hydropower station power stations can be solved by adding computers and agents, and the total calculation time is rarely increased.

In addition, multi-agents integrate multiple algorithms to form an algorithm library. In the invention, the algorithm library adopts the idea of survival of the fittest, and the probability that the algorithm is selected increases or decreases with the evaluation of the calculation optimization effect. In the invention, the combination of classical and modern intelligent optimization algorithms and asynchronous parallel computing are adopted to prevent the intelligent optimization algorithm from having few solutions.

The basic principles, main features and advantages of the present invention have been shown and described above. It should be understood by those skilled in the art that the present invention is not limited by the foregoing embodiments, and that the present invention is only described in the foregoing description and the description of the present invention, without departing from the spirit and scope of the invention. Various changes and modifications are intended to be included within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and their equivalents.

Claims (3)

1. A multi-agent hydropower station group short-term optimal scheduling method, characterized in that it comprises the following steps,

   Step (1), establish a short-term optimal scheduling model for hydropower stations

   Objective function:

   

   Where E is the sum of energy consumption of each power station; T is the scheduling period; K is the number of power stations; Q _{k, t} is the power generation flow of the kth power station t period; P _{k, t} is the kth power station t period Output; H _k,t is the head of the kth power station t period;

   Restrictions:

   (1) System load balancing constraints

   

   (2) Power limit of each hydropower station

   

   (3) Water balance equation V _k,t+1 =V _k,t +(Qin _k,t -Qgen _k,t -Qdis _k,t )Δt,k=1,2...,K t=1,2... ,T

   (4) Reservoir storage tolerance Vmin _k ≤V _k,t ≤Vmax _k ,k=1,2...,K t=1,2...,T

   (5) Cascade power station water flow lag constraint

   

   Where, P _t is the step load of the t period, and P _k,t is the output of the k power station t period;

   

   Is the lower limit of the output of the kth power station during the tth period.

   

   It is the upper limit of the output of the kth power station in the t-th period; V _k,t is the capacity of the kth power station in the t-th period, V _k,t+1 is the storage capacity of the _k-th power station in the t+1th period, Qin _k,t is the kth The power flow of the power station in the t-th period, Qgen _{k, t} is the power flow of the kth power station in the t-th period, Qdis _{k, t} is the water flow of the k-th power station in the t-th period, Δt is the time length of the unit period, Vmin _k For the k-power station minimum storage capacity constraint and the integration of medium- and long-term planning constraints, Vmax _k is the maximum storage capacity constraint of the kth power station and the combination of medium and long-term planning constraints; Qq _{k, t} is the water supply of the kth power station in the upstream interval of the tth time interval. ;

   

   For the discharge of the kth power station into the upstream power station during the tth period; τ _k is the runoff propagation time from the k-1th to the kth power station;

   Step (2), establishing a short-term optimal scheduling sub-model

   Objective function:

   

   Where, Q is the sum of water consumption and flow in the T cycle of the power station, T is the scheduling period, n is the number of units, Qgen _{t, i} is the power generation flow of the unit at time _t, and N _{t, i} is the output of the unit at time t, H _t is the net head at the time _t of the power station, and Qdis _t is the water flow at the time of the power station t;

   Restrictions:

   (1) Power balance

   

   (2) Water balance constraint V _t+1 =V _t +(Qin _t -Qgen _t -Qdis _t )Δt,t=1,2...,T

   (3) Output constraint, which determines the turbine efficiency and head N _i,min ≤N _i ≤N _i,max ,i=1,2...,n

   (4) Reference flow constraint Qout _min ≤ Q _i ≤ Qout _max , i = 1, 2..., n

   (5) Unit inoperable area

   

   (6) Storage tolerance Vmin ≤ V _t ≤ Vmax, t = 1, 2..., T

   Where, N _s is the output of the cascade to the power station, N _i is the output of the power station i, V _t is the capacity of the power station in the t-th period, Qin _t is the water flow of the power station in the t-th period, Qgen _t is the power station in the t-th During the period of power generation, Qdis _t is the abandoned water flow of the power station during the t-th period, Δt is the length of time in the unit period, N _i,min is the minimum output of the i unit, N _i,max is the maximum output of the i unit, and Qout _max is the maximum output of the unit. Flow, Qout _min is the minimum over flow of the unit.

   

   It is the lower limit of the operating zone of the i unit.

   

   It is the upper limit of the operating area of the i unit, Vmin is the minimum storage capacity constraint of the power station and the comprehensive minimum storage capacity constrained by the medium and long-term planning. Vmax is the maximum storage capacity constraint of the power station and the comprehensive maximum storage capacity constrained by the medium and long-term planning;

   Step (3), establishing an agent model according to the model in step (1) and step (2), the agent model includes an inter-station load distribution agent, an in-station start-stop agent, and an in-station load optimization agent.

   The task of the inter-station load distribution agent is to monitor whether the step center receives a new next-day load curve, and once the new next-day load curve is monitored, Np=(Np ₁ , Np ₂ , Np ₃ , . . . , Np ₉₆ The inter-station load distribution agent immediately assigns the new next-day load curve to each power station according to a predetermined algorithm. The station-initiated shutdown agent and the station-in-load optimization agent respectively calculate the start-stop plan for the power station unit, distribute the load, and distribute the load between stations. The agent stores the result and continues to repeat the allocation until the result is satisfactory. The inter-station load distribution agent is also used to delete the redundant agent and increase the agent to achieve the purpose of adapting to the optimization calculation;

   The task of the start-stop agent in the station is to formulate a start-stop plan and find a feasible area of the unit that passes through the vibration zone at least. The station-starting agent monitors whether each power station has a new load curve, and if the power station within the monitoring range is allocated new The load curve is immediately calculated and expanded;

   The task of the station load optimization agent is to optimize the distribution load to each unit according to the minimum water consumption criterion according to the start-stop plan and the feasible area of the unit, and then the station load optimization agent sends the water consumption energy in the scheduling period to the station load distribution. Agent

   Step (4), establishing an agent algorithm library, the algorithm library includes two types of algorithms, a classical algorithm and a heuristic intelligent optimization algorithm, each The second calculation randomly selects a classical algorithm and an intelligent optimization algorithm, performs parallel calculation, compares the calculation results, and then feeds back to the Agent algorithm library. The Agent algorithm library performs the algorithm's pros and cons statistics, and the excellent algorithm is selected. Large, inefficient algorithms are less likely to be selected and eventually phase out;

   In step (5), the Agent model established in the step (3) is invoked by the Agent algorithm library established in the step (4) to realize short-term optimal scheduling of the multi-agent hydropower station group.
2. The short-term optimal scheduling method for a multi-agent hydropower station group according to claim 1, wherein the inter-station load distribution agent is set to one, and is disposed in an agent main container of a computer, wherein the computer is provided with an agent. The platform, the computer with the Agent main container is connected to several other computers, the other computer has an Agent platform, and the Agent platform in the computer with the Agent main container forms a multi-agent platform, and the Agent main container is also provided. There is a management agent, which is responsible for the state management, the agent management and the startup of the main program of the entire multi-agent platform, and the corresponding Agent container of the other computer's Agent platform is provided, and the sub-agent container is provided with the station internal The shutdown agent and the station load optimization agent.
3. The short-term optimal scheduling method for a multi-agent hydropower station group according to claim 1, wherein the multi-agent optimization process is performed by using an heuristic intelligent optimization algorithm,

   (1) The scale of the start-stop agent in a given station is N1, the size of the inter-station load distribution agent is N2, and the scale of the initial station-to-station load distribution scheme is M;

   (2) After the station load distribution agent receives the hydropower station group load curve, it randomly selects the M initial power station load distribution schemes, assigns its load curve to each power station, and then monitors whether each power station is allocated. If the allocation is finished, it immediately The M plans and results issued at one time, update the M programs, and then continue to deliver to the hydropower station, and so on, until the optimal solution for convergence is found;

   (3) When the station starts to stop the agent and monitors that a new load curve is assigned to the power station, it immediately starts the development of the start-stop plan, then finds the feasible area that passes through the minimum vibration zone, feeds back to the power station, and uses the pre-emption agent in the N1 station. The treatment of the power station, that is, for the M load schemes of a single power station, may be completed by a single station within the shutdown agent, or may be completed by multiple stations within the station, when an agent completes the task of a power station, then continue to scan the remaining The power station, if there is an unfinished task, immediately joins the execution task;

   (4) When any station internal shutdown agent completes the start-stop plan of a certain power station and the minimum feasible area of the vibration zone, the station load optimization agent will immediately monitor the task and immediately plan for the power station. The unit load is optimized and distributed, and the preemptive task is also used. The idle station load optimization agent automatically monitors the load curve of each unit that is not assigned to the unit. Once it is monitored, it is immediately allocated to it, and then continues to monitor.

Images