An intelligent scheduling method and system of a wind-water-fire integrated energy system

Abstract

The application discloses an intelligent scheduling method and system of a wind-water-fire integrated energy system, which is used for coping with the influence of new energy uncertainty on the system, applying rolling optimization to the wind-water-fire integrated energy system, formulating a scheduling scheme according to global information, using local optimization instead of global optimization, and feeding back and correcting according to the latest information; the rolling optimization is constructed as a Markov decision process to ensure that the mathematical mechanism is applicable to deep reinforcement learning. In order to ensure the solving time and quality of the scheduling scheme, a hybrid enhanced intelligent scheduling algorithm combining deep reinforcement learning and evolutionary computation is provided, deep reinforcement learning is used to mine the value of historical data, and interactive learning is carried out with the energy system, the control strategy is optimized, and the fast giving of the preliminary scheduling scheme is realized; and the evolutionary computation is further used to optimize the preliminary scheduling scheme again, so that the economy and stability of the wind-water-fire integrated energy system are ensured.

Description

Technical Field

[0001] This invention belongs to the field of integrated energy management and intelligent scheduling technology, and more specifically, relates to an intelligent scheduling method and system for a wind-water-fire integrated energy system. Background Technology

[0002] Thermal power is one of the most important energy sources in my country's power industry, characterized by stability and controllability, but it suffers from high operating costs and severe environmental pollution. In contrast, wind power has lower operating costs and less environmental impact, but its characteristics of peak flow inversion, randomness, and intermittency may have some impact on the power grid. Hydropower has the advantages of rapid start-up and shutdown and strong regulation capabilities, which can improve the power grid's ability to cope with uncertainties. However, due to its significant dependence on water inflow, over-reliance on peak shaving can lead to substantial water wastage losses and capacity safety risks. To enable wind, hydro, and thermal power to complement each other's characteristics, establishing a joint optimization scheduling model that comprehensively analyzes the characteristics of wind, hydro, and thermal power is of great significance for improving the operating efficiency of multi-energy power industrial systems and promoting the consumption of new energy sources.

[0003] Compared to scheduling a single energy system, solving scheduling schemes for integrated wind-water-thermal energy systems is more difficult. This is primarily due to the increased scale and variety of energy systems, leading to greater problem size and complexity. Furthermore, the spatiotemporal coupling characteristics between different energy subsystems further complicate the solution process. Existing computational methods can be broadly categorized into two types: mathematical computational methods and evolutionary computational methods. For mathematical methods, the most commonly used algorithms include mixed-integer linear programming (MILP) and mixed-integer nonlinear programming (MINLP). While these methods can theoretically yield optimal solutions, they struggle to provide solutions in polynomial time for large-scale non-convex nonlinear problems. Moreover, the accuracy of these methods is often highly dependent on the relaxation accuracy of the problem. The most commonly used evolutionary computation methods include Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Evolutionary Predator and Prey Strategy (EPPS). While these methods have been widely applied in many fields, their drawbacks, such as premature convergence and high computational cost, remain significant. Therefore, addressing this problem by considering the uncertainty of wind power output and designing flexible, fast, and stable scheduling methods to ensure the economy and safety of the wind-water-thermal integrated dispatch system and improve its absorption capacity, is a pressing issue that needs to be resolved. Summary of the Invention

[0004] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides an intelligent scheduling method and system for a wind-water-thermal integrated energy system. This system enables coordinated and complementary scheduling of wind power, hydropower, and thermal power systems, thereby improving the responsiveness of the integrated energy system, reducing system power generation costs, enhancing the system's wind power absorption capacity, and ensuring stable and safe system operation.

[0005] To achieve the above objectives, according to a first aspect of the present invention, an intelligent scheduling method for a wind-water-fire integrated energy system is provided, comprising:

[0006] S1, establish the objective function and constraints of the integrated wind-water-fire energy system;

[0007] S2, the process of solving the objective function is transformed into a Markov decision process. Based on the historical data of the wind-water-fire integrated energy system, a rolling optimization strategy is adopted to enable the agent to interact with the system, and the interaction data is collected to update the control strategy of the agent in order to train the agent.

[0008] S3. The trained agent is used to solve the actual scheduling problem to obtain a preliminary scheduling scheme. The particle swarm optimization algorithm is then used to optimize the preliminary scheduling scheme to obtain the optimal scheduling scheme.

[0009] According to a second aspect of the present invention, an intelligent scheduling system for a wind-water-fire integrated energy system is provided, comprising: a computer-readable storage medium and a processor;

[0010] The computer-readable storage medium is used to store executable instructions;

[0011] The processor is configured to read executable instructions stored in the computer-readable storage medium and execute the method as described in the first aspect.

[0012] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0013] 1. Compared to traditional single-energy dispatching, the wind-water-thermal synergistic model constructed in this invention can fully leverage the complementary advantages of wind, water, and thermal power, thereby improving the economy and stability of the integrated energy system. Based on this, a rolling optimization method is used to comprehensively consider global information, replacing global optimization with local optimization, and continuously using the latest information for feedback correction, which is beneficial for addressing the impact of wind power uncertainties on the system. Simultaneously, the proposed hybrid enhanced intelligent algorithm fully utilizes deep reinforcement learning algorithms to mine the value of data and historical information, enabling rapid generation of dispatching schemes. Furthermore, due to the inclusion of evolutionary computation, the algorithm can adjust the dispatching scheme based on current information, achieving further optimization of the dispatching scheme. The wind-water-thermal intelligent control system designed in this way can flexibly and quickly generate dispatching schemes according to environmental changes, maximizing wind power absorption capacity, reducing system operating costs, and increasing system stability.

[0014] 2. The intelligent scheduling method for the wind-water-fire integrated energy system provided by this invention adds an evolutionary computation stage to the online application phase, building upon offline learning and online application. This allows the generalization capability of offline learning to be combined with the personalized capability of online evolutionary computation, forming a complementary advantage and significantly increasing the scheduling and adaptability of the wind-water-fire system. Most existing research first utilizes deep reinforcement learning to train the agent offline based on historical data, and then directly applies it to the online phase. Additionally, some research directly applies evolutionary computation online to solve scheduling problems. However, considering the complex spatiotemporal coupling relationship of the wind-water-fire joint scheduling system, the randomness and uncertainty of wind speed, and the finiteness of historical data, training the agent solely with historical data cannot fully cover the vast system state space, potentially leading to unsatisfactory scheduling results. Simultaneously, relying solely on evolutionary computation to calculate the scheduling plan online would undoubtedly create a huge computational burden and solution time pressure. Therefore, unlike the above two methods, this invention uses deep reinforcement learning for offline generalization learning, while simultaneously utilizing evolutionary computation for further personalized optimization in the online phase, overcoming the aforementioned problems.

[0015] 3. The intelligent scheduling method for a wind-water-fire integrated energy system provided by this invention represents the rolling optimization decision-making process as an MDP in deep reinforcement learning, enabling deep reinforcement learning to interact with the environment using rolling optimization strategies. Existing research on using deep reinforcement learning to solve energy scheduling problems has not considered the use of rolling optimization strategies. However, the wind-water-fire joint scheduling model constructed in this invention has a very obvious time coupling characteristic, that is, the current scheduling plan will have a certain impact on the future state of the system. Considering the uncertainty of wind speed, it is necessary to appropriately consider the impact of the current scheduling plan on the future when formulating the scheduling plan. Therefore, the rolling optimization strategy is considered when constructing the scheduling process as an MDP.

[0016] 4. The intelligent scheduling method for the integrated wind-water-fire energy system provided by this invention combines deep reinforcement learning with evolutionary computation (PSO). The designed dynamic guidance method can balance the exploration and utilization capabilities of the algorithm during the optimization process. When using deep reinforcement learning to guide the PSO algorithm, this invention designs a dynamic guidance method. This method can compare the optimal value in the PSO population with the value given by deep reinforcement learning, and adjust the guidance capability of the PSO population based on the difference between the two. Attached Figure Description

[0017] Figure 1 A schematic diagram of a wind-water-fire integrated energy system provided in an embodiment of the present invention:

[0018] Figure 2A flowchart of a wind-water-fire hybrid enhanced intelligent scheduling system based on deep reinforcement learning and evolutionary computation is provided for embodiments of the present invention.

[0019] Figure 3 A flowchart of the rolling optimization scheme provided in an embodiment of the present invention;

[0020] Figure 4 A schematic diagram illustrating the construction of rolling optimization as a Markov decision process provided for embodiments of the present invention;

[0021] Figure 5 This is a schematic diagram of the internal structure of an intelligent agent provided in an embodiment of the present invention;

[0022] Figure 6 This is a flowchart illustrating the training process of an agent in deep reinforcement learning, provided in an embodiment of the present invention.

[0023] Figure 7 A flowchart illustrating how evolutionary computation is used to further adjust the scheduling scheme, as provided in an embodiment of the present invention; Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0025] This invention provides an intelligent scheduling method for a wind-water-fire integrated energy system, such as... Figure 1-2 As shown, it includes:

[0026] S1, establish the objective function and constraints of the integrated wind-water-fire energy system.

[0027] Specifically, based on the actual physical characteristics of wind power, cascade hydropower, and thermal power, a system is constructed as follows: Figure 1 The wind-water-fire integrated energy system shown in the figure also characterizes the system's supply and demand balance, the spatiotemporal coupling characteristics of reservoir capacity, and the physical characteristics of hydropower units, thermal power units, and wind power units.

[0028] S2, the process of solving the objective function is transformed into a Markov decision process. Based on the historical data of the wind-water-fire integrated energy system, a rolling optimization strategy is adopted to enable the agent to interact with the system, and the interaction data is collected to update the control strategy of the agent in order to train the agent.

[0029] Specifically, based on the spatiotemporal coupling relationship of the wind-water-fire integrated energy system, a rolling optimization strategy is used to solve the objective function, determine the optimization scheduling cycle and execution cycle, and design the alternation relationship between the two.

[0030] Furthermore, in the rolling optimization strategy, the optimization scheduling period is longer than the execution period.

[0031] Specifically, the optimized alternation between scheduling and execution cycles is as follows:

[0032] In rolling optimization, considering the prediction errors of data such as wind power and the spatiotemporal coupling of the system, a relatively long time scale is usually considered when formulating the scheduling scheme, but only the scheduling scheme corresponding to a portion of the time scale is executed in actual practice. In this invention, the optimized scheduling cycle refers to the size of the time scale considered when formulating the scheduling scheme, and the execution cycle refers to the size of the time scale corresponding to the specific scheduling scheme being executed. In this invention, the optimized scheduling cycle only needs to be greater than the execution cycle.

[0033] State variables are characterized based on system information such as upstream water inflow, reservoir capacity, thermal power unit output, and net load. Action variables are characterized based on decision information such as cascade hydropower generation flow and thermal power unit ramp-up. Reward mapping rules are established based on the system scheduling optimization objective function, and the rolling optimization model is then constructed as a Markov decision process.

[0034] Based on historical data, the agent in deep reinforcement learning interacts with the system, and the interaction data is collected to update the agent's policy, thereby improving the agent's generalization ability and enabling the rapid provision of a preliminary scheduling scheme.

[0035] S3. The trained agent is used to solve the actual scheduling problem to obtain a preliminary scheduling scheme. The particle swarm optimization algorithm is then used to optimize the preliminary scheduling scheme to obtain the optimal scheduling scheme.

[0036] Specifically, by combining the evolutionary computation algorithm—particle swarm optimization—with deep reinforcement learning, the scheduling scheme is further adjusted based on current real-world data. This ensures the algorithm's adaptability and optimizes the system's operating costs and wind power absorption capacity while guaranteeing stable and safe operation.

[0037] Furthermore, to ensure the stable and efficient operation of the integrated wind-hydro-thermal energy system, constraints need to be established based on the actual physical models and power conversion relationships of wind turbines, hydropower units, and thermal power units. These constraints mainly include power supply and demand balance constraints, cascade reservoir capacity balance constraints, and power generation constraints for hydropower units, wind turbines, and thermal power units. The specific constraints are as follows:

[0038]

[0039]

[0040]

[0041] V h min ≤V h,t ≤V h max (4)

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] Accordingly, under the above constraints, in order to reduce the operating cost of the integrated energy system, improve the utilization rate of wind power, and ensure system stability, the objective function for system scheduling is constructed as follows:

[0050]

[0051]

[0052]

[0053]

[0054] F t =F I +F p +F c +F h (16)

[0055] Where F(t) is the objective function of the system, F I The output cost of the thermal power unit is T, where T is the time scale considered when formulating the dispatch plan, and a is the time scale. i b i c i Let e ​​be the cost parameter for thermal power unit i. i h i For the valve point effect parameter of thermal power unit i; F p For the additional electricity costs of the wind-water-fire system, Cp For the price of additional electricity; F c As punishment for abandoning wind, C c For wind curtailment penalty factor; F h For capacity deviation penalty, EV h,t Let h be the expected capacity of the hydropower station during the time period t. C is the allowable deviation coefficient of hydropower station h during time period t. h1 C h2 It is a constant.

[0056] N represents the output of thermal power unit i during time period t. i This represents the number of thermal power units. For the output of the hydropower station h during time period t, N h This refers to the number of wind turbine units; N represents the output of the wth wind turbine during time period t. w Δt represents the number of wind turbine units; Δt represents the number of hours contained in time period t. For time period t, the amount of electricity the system purchases additionally from external sources. Let t be the amount of wind curtailed by the system in time period t. V represents the system's power demand during time period t. h,t I represents the capacity of the hydropower station h after scheduling during time period t. h,t Let Q be the natural inflow rate of the hydropower station h during time period t. h,t Let S be the power generation flow rate of the hydropower station h during time period t. h,t Let h be the water discharge rate of the hydropower station during time period t. Let t be the number of hydropower stations directly upstream of hydropower station h, k be the number of its directly upstream hydropower station, and t be the number of its upstream hydropower station. kh The time delay of water flow from upstream power station k to power station h; ξ h1 ~ξ h6 V is the power generation coefficient of the hydropower station h; h max With V h min These represent the upper and lower limits of the capacity h of the hydropower station, respectively. and These are the upper and lower limits of power generation flow, respectively. and These are the upper and lower limits of the outbound flow rate, respectively. and These represent the upper and lower limits of the power output h of the hydropower station. and These represent the upper and lower limits of the output of thermal power unit i, respectively; UR i With DR i These are the upper and lower limits of the ramp rate for thermal power unit i, respectively. P represents the output of the wind turbine w during time period t. ra,wThe rated power of the wind turbine generator w, v w,t Let v be the wind speed at point w during time period t. in,w v out,w v ra,w These refer to the cut-in, cut-out, and rated wind speed of the wind turbine unit w.

[0057] In step S2, based on the above optimization model, considering the spatiotemporal coupling relationship between wind power, cascaded hydropower, and thermal power, as well as the impact of wind power output uncertainty on the system, this invention applies a rolling optimization strategy to the wind-hydro-thermal integrated energy system. The specific process is as follows:

[0058] To facilitate understanding, a specific case is provided here, along with an appendix. Figure 3 Let's explain. In this case, the optimization scheduling cycle is set to 4 hours, the execution cycle to 1 hour, and rolling optimization begins at time t. When formulating the corresponding scheduling plan at time t, it's necessary to comprehensively consider information from time t to (t+3), a total of 4 hours, and formulate a corresponding scheduling plan to achieve overall optimization within those 4 hours. However, when actually executing the scheduling plan, since the execution cycle is 1 hour, only the scheduling plan for that 1 hour is executed, i.e., only the scheduling plan for time t is executed. Afterwards, as time progresses, the system enters time t+1. Based on the predicted information from time (t+1) to (t+4), a new round of optimization scheduling cycle decision information is constructed, and a new scheduling plan is formulated, executing only the scheduling plan for the t+1 time period. This process is repeated to achieve rolling optimization.

[0059] To facilitate subsequent solutions using deep reinforcement learning, the rolling optimization needs to be adapted to the data mechanisms of deep reinforcement learning. Therefore, the rolling optimization needs to be constructed as a Markov decision process. The specific process is attached. Figure 4 As shown. From an overall perspective, firstly, a new state variable s1 is created based on the decision information throughout the entire optimization scheduling cycle of the system, and this variable is input into the agent. The agent will output the corresponding action a1 according to its own control strategy, and the probability p1 of outputting a1 in state s1. Then, the environment provides a reward value r1 based on state s1 and action a1, and transitions the state to the next time step, state s2. Similarly, the agent provides a2 and p2 based on s2, repeating the above process until a reward value r4 is given. At this point, a1 to a4 constitute the scheduling scheme for the optimization scheduling cycle t1 to t4.

[0060] The Agent's state variables are represented as follows:

[0061]

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069] Among them, s t Let t be the state of the agent at time t.

[0070] The above construction method enables the agent to extract and utilize information on upstream water inflow, current reservoir capacity, thermal power unit output, and net load demand of the cascade reservoir. This facilitates the agent's analysis of differences between states and improves training efficiency.

[0071] The wind-water-thermal hybrid enhanced intelligent scheduling method based on deep reinforcement learning and evolutionary computation provided in this invention constructs data information such as upstream water inflow of cascade hydropower, natural water inflow between cascade hydropower stations, current output of thermal power units, and net system load as state variables for a deep reinforcement learning agent. This method maximizes the capture of key influencing factors of the system, provides sufficient system information for agent decision-making, and improves the training effect of the agent.

[0072] The agent's actions are represented as follows:

[0073] a t =[Q t ,Δp t (25)

[0074]

[0075]

[0076]

[0077] Among them, a t Let t be the action of the agent at time t.

[0078] Based on agent action a t It can determine the reservoir's power generation flow and the thermal power unit's ramp-up power, thereby obtaining the scheduling plan for the integrated wind-water-thermal energy system and realizing intelligent control of the system.

[0079] The wind-water-thermal hybrid enhanced intelligent scheduling method based on deep reinforcement learning and evolutionary computation provided in this invention achieves consistency between the actual physical system control variables and the agent's action variables by using the tiered hydropower generation flow and the thermal power climbing power as agent action variables in deep reinforcement learning. This allows the agent's action output to be applied to the wind-water-thermal joint scheduling system without further processing, increasing the convenience of the control system.

[0080] The reward mapping rules are established based on the system state after scheduling as follows:

[0081]

[0082] in, Let t be the reward value of the agent at time t.

[0083] Specifically, the reward value is defined as the negative of the objective function (16), that is:

[0084] r t =-F t (29)

[0085] In this invention, the environment is defined as a wind-water-fire integrated energy system. Based on state s t The environment can determine the current state information of the integrated energy system. Meanwhile, based on action a... t The environment can determine the current scheduling scheme of the integrated energy system. At this time, the objective function F can be calculated based on equations (1)-(16). t Furthermore, the reward value r that the environment should give to the Agent is calculated using equation (29). t .

[0086] The wind-water-fire hybrid enhanced intelligent scheduling method based on deep reinforcement learning and evolutionary computation provided in this invention constructs a deep reinforcement learning reward function according to the scheduling objective function, ensuring the consistency between the Markov decision process and the overall energy system scheduling objective. This enables the agent to learn and optimize its control strategy driven by the goal of optimizing the overall energy system scheduling.

[0087] In this invention, the Agent is the analysis-decision unit in the integrated energy system. Detailed construction is shown in the appendix. Figure 5 As shown, the decision network is used to give the mean μ and variance σ of the action based on the state s, and then obtains the action a through Gaussian distribution sampling. The evaluation network is used to evaluate the value V(s) of the state s and assists the policy network in updating in subsequent processes.

[0088] Furthermore, the method for training the agent's control policy is as follows:

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097] Where θ represents the policy network parameters and α represents the learning rate; Let π be the gradient of the objective function. θ (a t |s t ) for the intelligent agent in the control policy π θ Below, when the state is s t At that time, output action a t The probability of; For state s t At that time, action a t Advantages; For state s t At that time, action a t value, For state s t Value; For state s t At that time, action a t The reward value is τ, which is the trajectory, recording the states and actions the agent takes during its interaction with the environment. It is a set of states and actions.

[0098] Specifically, the Agent is trained using Proximal Policy Optimization (PPO), and the training flowchart is attached. Figure 6 As shown in the diagram, this process mainly consists of three stages: the sample generation stage through interaction with the environment, the policy network training stage, and the evaluation network training stage. The specific steps of the sample generation stage through interaction with the environment are as follows:

[0099] 1. The environment outputs state variables s based on the Agent's actions at the previous moment and historical data. t .

[0100] 2. The policy network is based on s tOutput μ and σ, and construct a Gaussian distribution based on them. Simultaneously through π θ (·|s t Action a is obtained by sampling t .

[0101] 3. Subsequently, the environment is based on s t With a t The state is obtained by transitioning from state s to state s. t+1 At the same time, the reward value r is output according to formula (29). t .

[0102] 4. Repeat the above process until the termination condition is met, and record the variable <s> generated during this process. t ,a t ,π(a t |s t ),r,s t+1 > Stored in the memory bank for subsequent network training.

[0103] A good strategy enables the agent to output the optimal action a* based on the current state s, thereby maximizing the reward value. The specific process of the policy network training phase is as follows:

[0104] 1. Obtain training data by sampling from memory.

[0105] 2. The policy network is based on s t Construct a Gaussian distribution π θ And according to a t π was calculated θ (a t |s t ).

[0106] 3. Based on π θ (a t |s t ), And evaluate the advantages output during the network training phase. Gradient updates are performed on the network. The calculation method is as follows:

[0107]

[0108]

[0109]

[0110] Evaluation networks are primarily used to assist policy networks in updating their functions. Their main principle is to fit a state-value function. This provides an advantage function for training policy networks. Therefore, we also need to update the evaluation network to better fit the state-value function. The process is as follows:

[0111] 1. To evaluate the network, first extract the states s from the training sample data. t and s t+1 Then its value was estimated, and the result was obtained. and Where θ c These are the policy network parameters.

[0112] 2. Based on the reward value r t and Computational advantages This advantage function is then passed on during the policy network training phase. The advantage function is calculated as follows:

[0113]

[0114]

[0115] 3. Based on the reward value r t and The parameters of the evaluation network are updated. The calculation method is as follows:

[0116]

[0117]

[0118]

[0119] Among them, a c To evaluate the learning rate of network updates.

[0120] Furthermore, in step S3, an evolutionary computation method, Particle Swarm Optimization (PSO), is used to further refine the scheduling scheme x. DRL Adjustments are made to enhance the algorithm's adaptability and robustness.

[0121] In this invention, the PSO algorithm is used to further refine the scheduling scheme x. DRL The adjustment process can be further divided into four phases: initialization, early iteration, middle iteration, and late iteration. Taking a scheduling cycle of 4Δt as an example, the flowchart is attached. Figure 7 As shown.

[0122] During the initialization phase, the integrated wind-water-fire energy system is first constructed as s0 by the environment. Subsequently, the agent interacts with the environment to obtain actions a1–a4 for four periods, which are then directly used as the scheduling scheme x. DRL (i.e., the preliminary scheduling plan).

[0123] In the early stages of the iteration, x DRL This will be used to guide the movement of particles in PSO, thereby enabling the particle population to move rapidly to a better position. This process can be represented by the following equation:

[0124]

[0125]

[0126] In the middle of the iteration, the particle population will continue to be affected by x. DRL The impact of x, in order to ensure that the population can make full use of x DRL At the same time, to improve the population's exploration ability and achieve a balance between algorithm exploration and utilization, this invention defines the magnitude of the guiding factor c3 and x. DRL Gbest k The difference between function values ​​is related, and its specific calculation method is shown below.

[0127]

[0128] in, Let N represent the velocity and position of the j-th particle in the k-th iteration, respectively, where j = 1, 2, ..., N. j k = 1, 2, ..., N k N j With N k These represent the total number of particles and the total number of iterations in the algorithm, respectively. Let Gbest be the optimal position of particle j in the first k iterations. k x represents the optimal position of the particle swarm in the first k iterations; DRL This is the initial scheduling scheme; w is the inertia weight, rand is a random number between (0,1], c1 and c2 are learning factors, and c3 is the guidance factor. The larger c3 is, the greater the x DRL The stronger the ability to guide particles; and Gbest k With x DRL The function value, C pso It is a constant.

[0129] In the later stages of iteration, according to equation (40), the guidance factor c3 will gradually decrease to 0. At this point, the scheduling scheme given by deep reinforcement learning will no longer have a guiding role, and Gbest... k The function value is now superior. After this, the particle swarm will be in a completely autonomous state, while the system continues to be optimized. Finally, It will serve as the scheduling scheme (i.e., the optimal scheduling scheme) for the integrated wind-water-fire energy system.

[0130] The wind-water-fire hybrid enhanced intelligent scheduling method based on deep reinforcement learning and evolutionary computation provided in this invention achieves a balanced exploration and utilization capability of evolutionary computation by constructing an adaptive guidance factor. In the early stages of iteration, a larger guidance factor accelerates the algorithm's utilization capability and rapidly improves the quality of the scheduling scheme. When the scheduling scheme is superior to the scheme proposed by the agent, the exploration capability is increased to further explore a better scheduling scheme. This enables the algorithm to automatically adjust its optimization direction based on the quality of the scheduling scheme.

[0131] In summary, the wind-water-thermal hybrid enhanced intelligent scheduling method based on deep reinforcement learning and evolutionary computation provided in this invention considers the advantages and disadvantages of wind power, cascaded hydropower, and traditional thermal power, establishes the physical characteristics of wind power, hydropower, and thermal power, constructs a wind-water-thermal integrated complementary energy system, and then proposes the wind-water-thermal integrated complementary energy scheduling problem. Compared with traditional single-energy scheduling, the wind-water-thermal collaborative model constructed in this invention can fully leverage the complementary advantages of wind, water, and thermal power, thereby improving the economy and stability of the integrated energy system. By introducing a rolling optimization strategy to address the impact of wind power uncertainty on the integrated energy system, this strategy is based on the objective fact that wind speed prediction errors decrease over time. It uses prediction data to formulate a global scheduling plan, but only implements a portion of the scheduling plan, achieving local optimization instead of global optimization, and using the latest prediction data for feedback correction, thereby improving the scheduling effect. Furthermore, to ensure both the time and quality of solution generation for scheduling schemes, this invention proposes a hybrid enhanced intelligent scheduling algorithm. This algorithm combines deep reinforcement learning and evolutionary computation. Deep reinforcement learning leverages the value of historical data and interacts with the integrated wind-water-thermal energy system to optimize its own control strategy, thereby enabling rapid generation of scheduling schemes. Based on this, evolutionary computation is used to further adjust the scheme generated by deep reinforcement learning, enhancing the algorithm's adjustment capability and further optimizing the scheduling scheme. This achieves rapid decision-making for joint wind-water-thermal scheduling, ensuring power supply and demand balance, efficient absorption of new energy sources, and the safety of tiered hydropower capacity. The method provided by this invention can offer valuable reference for decision-makers in integrated energy systems such as wind-water-thermal systems to formulate fast and stable scheduling schemes; simultaneously, it improves the absorption capacity of new energy sources, reduces system response time, and enhances system stability and economy.

[0132] This invention provides an intelligent scheduling system for a wind-water-fire integrated energy system, comprising:

[0133] Computer-readable storage media and processors;

[0134] The computer-readable storage medium is used to store executable instructions;

[0135] The processor is configured to read executable instructions stored in the computer-readable storage medium and execute the method as described in any of the above embodiments.

[0136] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart scheduling method for a wind-water-fire integrated energy system, characterized in that, include: S1, Establish the objective function and constraints of the integrated wind-water-fire energy system; S2, the process of solving the objective function is transformed into a Markov decision process. Based on the historical data of the wind-water-fire integrated energy system, a rolling optimization strategy is adopted to enable the agent to interact with the system, and the interaction data is collected to update the control strategy of the agent in order to train the agent. S3. The trained agent is used to solve the actual scheduling problem to obtain a preliminary scheduling scheme. The particle swarm optimization algorithm is then used to optimize the preliminary scheduling scheme to obtain the optimal scheduling scheme. The objective function is: ； ； ； ； ； in, The system's objective function, Cost of output of thermal power units The time scale considered when developing a scheduling plan. , , For thermal power units Cost parameters, , For thermal power units Valve point effect parameters; Additional electricity costs for the wind-water-fire system This is the price for additional electricity purchases; As punishment for abandoning the wind, As a penalty factor for wind curtailment; Penalty for capacity deviation, For hydroelectric power station exist Expected capacity for the time period For hydroelectric power station exist Permissible deviation coefficient for the time period , It is a constant; For thermal power units During the period of efforts, For thermal power units The lower limit of output, This refers to the number of thermal power units. For the first During a given period, the amount of electricity the system purchases additionally from external sources. For the first The amount of wind curtailed in the time-period system For hydroelectric power station exist The capacity after scheduling for a given time period; The state variables of the intelligent agent are: ； ； ； ； ； ； ； ； The action variables of the intelligent agent are: ； ； ； ； The reward function for the agent is: ; in, For hydroelectric power station During the period Natural water flow For hydroelectric power station exist The capacity after time-sharing. This refers to the number of wind turbine units. For the first Each wind turbine unit Constant effort For hydroelectric power station During the period Power generation flow, For the first Taiwan thermal power units Efforts made at all times; The secondary optimization process is as follows: ； ； ； in, , They represent the first The particle in the first Velocity and position in the next iteration , , and These represent the total number of particles and the total number of iterations in the algorithm, respectively. For particles in front The optimal position in the next iteration. The representative particle group is in front. The optimal position in the next iteration; This is a preliminary scheduling plan; For inertial weights, for Random numbers between , As a learning factor, As a guiding factor, The larger the value, the more likely it is to indicate The stronger the ability to guide particles; and They are respectively and The function value, It is a constant.

2. The method as described in claim 1, characterized in that, The constraints include: ； ； ； ； ； ； ； ； ； ； in, For thermal power units During the period of efforts, This represents the number of thermal power units. For hydroelectric power station exist Efforts during a specific time period; For the first Typhoon turbines during the period of efforts, This refers to the number of wind turbine units; For time period The number of hours included; For the first During a given period, the amount of electricity the system purchases additionally from external sources. For the first The amount of wind curtailed in the time-period system For the system in time period The electricity demand; For hydroelectric power station exist The capacity after time-sharing. For hydroelectric power station During the period Natural water flow For hydroelectric power station During the period Power generation flow, For hydroelectric power station During the period The discharge flow rate, For hydroelectric power station The number of directly upstream power plants, Number its directly upstream hydropower station. For upstream power stations to the power station Water flow time lag; For hydroelectric power station The power generation coefficient; and Hydropower stations Capacity upper and lower limits, and These are the upper and lower limits of power generation flow, respectively; and Hydropower stations Upper and lower limits of output and thermal power units The upper and lower limits of output; and thermal power units The upper and lower limits of the slope; For wind turbines During the period of efforts, For wind turbines Rated power, For wind turbines Period wind speed, , , Wind turbine The cut-in, cut-out, and rated wind speed.

3. The method as described in claim 1, characterized in that, In the rolling optimization strategy, the optimization scheduling period is longer than the execution period.

4. The method as described in claim 1, characterized in that, The agent is trained using the following proximal policy optimization method: ； ； ； ； ； ； ； in, For policy network parameters, The learning rate; The gradient of the objective function. For intelligent agents in control strategies Below, when the state is At that time, output action The probability of; The state is At that time, the action Advantages; The state is At that time, the action value, For state Value; The state is At that time, the action The reward value, The trajectory records the states and actions of an intelligent agent during its interaction with the environment, and is a collection of states and actions.

5. An intelligent dispatching system for a wind-water-fire integrated energy system, characterized in that, include: Computer-readable storage media and processors; The computer-readable storage medium is used to store executable instructions; The processor is configured to read executable instructions stored in the computer-readable storage medium and execute the method as described in any one of claims 1-4.

Images