Multi-time-scale optimization scheduling method and system for wind-light-water storage system

Abstract

The invention belongs to the related technical field of power system scheduling, and discloses a multi-time scale optimization scheduling method and system for a wind-light-water storage system, and the scheduling method comprises the steps: inputting the wind-light day-ahead prediction of a new scheduling period into a day-ahead optimization scheduling model before entering the new scheduling period, and enabling the optimization target to enable the overall cost of the scheduling period to be minimum, the overall cost comprises operation cost and power supply reliability punishment to obtain a day-ahead scheduling plan; and after entering a new scheduling period, inputting the system real-time state, the day-ahead scheduling plan and the wind and light intra-day prediction at the scheduling moment into an intra-day rolling optimization model, and obtaining an intra-day scheduling strategy corresponding to the scheduling moment by enabling the local cost at the scheduling moment to be minimum and the local cost to comprise loss load penalty and equipment adjustment penalty. According to the invention, a multi-scale collaborative framework of'day-ahead global planning + intra-day rolling correction 'is provided, and the problems of economy, robustness and sustainability in the wind-light-water storage multi-energy complementary system can be considered.

Description

Technical Field

[0001] This invention belongs to the technical field of power system dispatching, and more specifically, relates to a multi-timescale optimization dispatching method and system for wind, solar, hydro, and storage systems. Background Technology

[0002] The large-scale integration of new energy sources such as wind and solar power into the power system poses a severe challenge to the safe and stable operation of the power grid due to their randomness, volatility, and intermittency. Multi-energy complementary systems combining wind, solar, hydro, and energy storage, leveraging the large-capacity regulation capabilities of hydropower and the rapid response characteristics of energy storage, have become a core technological approach to alleviate the pressure of new energy consumption and enhance system flexibility.

[0003] Traditional scheduling methods typically employ a single time scale. However, this method has significant limitations: relying solely on day-ahead deterministic optimization makes it unable to cope with intraday wind and solar forecast errors and load fluctuations, which can easily lead to wind and solar curtailment or power shortages; and relying solely on intraday real-time scheduling lacks global economic planning, which can easily result in resource overdraft.

[0004] Therefore, there is a need to provide a technical solution that can balance the overall economic efficiency and real-time robustness of a multi-energy complementary system of wind, solar, hydro, and storage. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a multi-timescale optimization scheduling method and system for wind-solar-hydro-storage systems, aiming to balance the global economy and real-time robustness of multi-energy complementary wind-solar-hydro-storage systems.

[0006] To achieve the above objectives, the present invention is proposed.

[0007] According to a first aspect of the present invention, a multi-timescale optimization scheduling method for a wind-solar-hydro-storage system is provided, comprising: Before entering a new scheduling cycle, the predicted day-ahead output of wind power and photovoltaic power for the new scheduling cycle are input into the day-ahead optimization scheduling model to obtain the day-ahead scheduling plan for different time intervals within the new scheduling cycle. The day-ahead scheduling plan includes the planned day-ahead output of hydropower and the planned day-ahead output of energy storage. The optimization objective of the day-ahead optimization scheduling model is to minimize the overall cost of the scheduling cycle, which includes operating costs and power supply reliability penalties. After entering a new scheduling cycle, the real-time system status, day-ahead scheduling plan, and intraday predicted output of wind power and photovoltaic power at the scheduling time are input into the intraday rolling optimization model to obtain the intraday scheduling strategy for the corresponding scheduling time. The intraday scheduling strategy includes the intraday planned output of hydropower and the intraday planned output of energy storage. The optimization objective of the intraday rolling optimization model is to minimize the local cost at the scheduling time. The local cost includes load shedding penalty and equipment adjustment penalty. The equipment adjustment penalty includes the deviation penalty between the intraday scheduling strategy and the day-ahead scheduling plan at the scheduling time.

[0008] According to a second aspect of the present invention, a multi-timescale optimization scheduling system for a wind-solar-hydro-storage system is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a multi-timescale optimization scheduling method for the wind-solar-hydro-storage system.

[0009] In summary, compared with the prior art, the technical solutions conceived in this invention have the following main advantages: 1. This invention proposes a multi-scale collaborative framework of "day-ahead global planning + intraday rolling correction". In the day-ahead stage, a day-ahead optimized scheduling model is constructed, with the optimization objective of minimizing the operating cost and power supply reliability penalty of the scheduling cycle, generating a global day-ahead scheduling plan. Because day-ahead scheduling considers overall operating costs and power supply reliability, the generated day-ahead scheduling plan takes into account the overall economy and reliability of scheduling. In the intraday stage, an intraday rolling optimization model is constructed, with the optimization objective of minimizing the load shedding penalty and equipment adjustment penalty at the scheduling time. In the rolling time domain, real-time status and forecast information are used to follow and dynamically correct the day-ahead plan, generating an intraday scheduling strategy. Because intraday scheduling considers intraday forecasts and real-time information, it can achieve intelligent suppression of short-term power deviations and operational risks, improving the real-time robustness of scheduling. This method combines global optimization, dynamic adaptation, and physical feasibility, effectively solving the problem of unifying economy, robustness, and sustainability in wind-solar-hydro-storage multi-energy complementary systems, providing an efficient and intelligent optimization path for the safe and economical operation of high-proportion new energy power systems.

[0010] 2. In a preferred embodiment, the day-to-day optimization scheduling model is solved using an improved sparrow search algorithm, which can more efficiently address the more complex wind-solar-hydro-storage complementary optimization problem and improve its convergence ability.

[0011] 3. In a preferred embodiment, the intraday rolling optimization model is constructed as a Markov decision process and solved using the SAC algorithm, which can quickly output the optimal action and generate an intraday scheduling strategy.

[0012] 4. In a preferred embodiment, the optimization objective in the intraday rolling optimization model considers M consecutive scheduling moments simultaneously, which can overcome the short-sightedness of the strategy caused by only considering the immediate state of the next scheduling moment and ensure the sustainability of scheduling. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the steps of a multi-timescale optimization scheduling method for a wind-solar-hydro-storage system according to an embodiment of the present invention.

[0014] Figure 2 This is a schematic diagram of a day-ahead and intraday collaborative scheduling optimization framework in one embodiment of the present invention.

[0015] Figure 3 This is a flowchart of the steps of an improved sparrow search algorithm in one embodiment of the present invention. Detailed Implementation

[0016] 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.

[0017] Firstly, this invention proposes a multi-timescale optimization scheduling method for wind, solar, hydro, and storage systems.

[0018] like Figure 1 The diagram shown is a flowchart of the multi-timescale optimization scheduling method for a wind-solar-hydro-storage system according to an embodiment of the present invention. Figure 2 The diagram shown is a schematic representation of a day-ahead and intraday collaborative scheduling optimization framework according to an embodiment of the present invention. The following is in conjunction with... Figure 1 and Figure 2 The method is described in detail.

[0019] S1. Before entering a new scheduling cycle, the predicted day-ahead output of wind power and photovoltaic power for the new scheduling cycle is input into the day-ahead optimization scheduling model to obtain the day-ahead scheduling plan for different time intervals within the new scheduling cycle. The day-ahead scheduling plan includes the planned day-ahead output of hydropower and the planned day-ahead output of energy storage. The optimization objective of the day-ahead optimization scheduling model is to minimize the overall cost of the scheduling cycle, which includes operating costs and power supply reliability penalties.

[0020] This step involves outputting a coarse-grained scheduling plan before entering the scheduling cycle. Assuming the scheduling cycle is one day, the day-ahead scheduling plan for the next 24 hours will be output based on the wind, solar and load forecast data for the next day.

[0021] Specifically, the day-ahead optimization scheduling model has day-ahead scheduling objectives and constraints.

[0022] The day-ahead scheduling objective in the day-ahead optimization scheduling model.

[0023] The current scheduling objective is to minimize the operating costs and power supply reliability penalties within the scheduling cycle. By adopting this scheduling objective, the scheduling problem is transformed into a comprehensive risk and cost optimization problem. By adjusting the weight coefficients between the two, a trade-off can be struck between pursuing economy and ensuring reliability to adapt to different operating requirements and risk preferences.

[0024] Therefore, the day-ahead scheduling objective is to minimize the overall cost of the scheduling cycle. The specific function of the overall cost of the scheduling cycle can be in the following form: ; In the formula, The cost function value represents the overall cost. and The weights for operating costs and power supply reliability penalties are respectively. The operating costs within the scheduling cycle, It is a power supply reliability penalty used to quantify the reliability risk cost caused by load loss or lack of backup resources throughout the entire dispatch cycle.

[0025] Operating costs The calculation formula is: ; In the formula, , , , They are respectively represented as The costs of purchasing electricity from the main power grid, operating costs of hydropower, operating costs of energy storage, and revenue from electricity sales are all considered at all times, where T is the total number of time intervals within the dispatch cycle. and The reservoir capacity status at the start and end times of the scheduling interval. and The energy storage status at the start and end times of the scheduling interval. and It is the penalty coefficient for the imbalance between the initial and final storage capacity and the initial and final energy storage state.

[0026] Power supply reliability penalty The calculation formula is: ; In the formula, The weights for the impact of power supply shortage and resource reserve are defined. EENS represents the expected power supply shortage during the dispatch period, which refers to the expected amount of electricity that cannot meet load demand due to insufficient power supply capacity during the dispatch period. This is a penalty that occurs when the actual reserve capacity of the reservoir is lower than the preset minimum reserve threshold. This is a penalty that is incurred when the actual standby capacity of energy storage is lower than the preset minimum standby threshold.

[0027] The constraints of the recent optimization scheduling model have been determined.

[0028] To ensure the feasibility and rationality of the day-ahead scheduling model, constraints are imposed on the operating status and boundary conditions of various resources in the system, mainly including constraints on wind and solar power output, constraints on battery energy storage power stations, and constraints on hydropower stations.

[0029] (1) Wind and solar power output constraints: The output of wind power stations and photovoltaic power stations in the system must be within the agreed range.

[0030] Wind power constraints are: ; Photovoltaic power constraints are: ; and Let be the wind power output and photovoltaic power output at time t, respectively. and These are the predicted maximum values ​​for wind power output and solar power output, respectively.

[0031] (2) Constraints of battery energy storage power stations.

[0032] The state of charge constraints of the energy storage system are: ; ; The charging and discharging constraints of the energy storage system are: ; ; The power balance constraint at the beginning and end of the dispatch cycle ensures the sustainability of the energy storage power station's operation in subsequent dispatch cycles. Its expression is: ; The energy storage charge / discharge state constraints are as follows: ; In the formula, for State of charge at time t, and They are respectively The charging power and discharging power at any given time. and These are charging efficiency and discharging efficiency, respectively. This refers to the rated capacity of the battery. and These are the minimum state of charge and the maximum state of charge, respectively. Minimum charging power for the battery. This is the maximum charging power of the battery. and This refers to the charging and discharging state of the energy storage power station. This is the minimum discharge power of the battery. This is the battery's maximum discharge power. The duration of a single time interval within the scheduling period, for example, 1 hour. and These represent the remaining electricity at the beginning and end of the energy storage power station, respectively.

[0033] (3) Constraints of hydropower stations.

[0034] The water balance constraint is: ; In the formula, for Shike Reservoir Storage capacity, for Shike Reservoir Storage capacity, , and They are respectively Shike Reservoir The inflow, power generation flow, and ecological wastewater discharge. The duration of a single time interval.

[0035] Outbound flow constraints are: ; In the formula, and Reservoirs The minimum and maximum outbound flow rates.

[0036] The storage capacity constraint is: ; ; In the formula, and Reservoirs Allowed upper and lower limits for warehouse capacity; Reservoir Reservoir capacity and upper reservoir water level The functional relationship between them.

[0037] The head constraint is: ; ; ; In the formula, for Shike Reservoir The water purifier head; and They are respectively Time and Reservoir of Time The water level of the upper reservoir; for Shike Reservoir The water level of the lower reservoir; for Reservoir of Time Head loss; Reservoir The functional relationship between the reservoir water level and the discharge flow rate; This represents the functional relationship between head loss and power generation flow.

[0038] Hydropower output constraints are:

[0039]

[0040]

[0041] In the formula, Reservoir The output coefficient, express Shike Reservoir Output of the hydropower unit and Reservoirs Lower and upper limits of output of hydropower units For reservoir Power limitations for hydropower units climbing slopes.

[0042] The above day-ahead optimization scheduling model is constructed and solved to output the day-ahead scheduling plan for different time intervals within the new scheduling cycle. Taking a scheduling cycle of 24 hours as an example, the day-ahead optimization scheduling model can be used to obtain preliminary scheduling schemes for the next day [0:00, 1:00), [1:00, 2:00), [2:00, 3:00), ..., [23:00, 24:00), where one time interval corresponds to one preliminary scheduling scheme.

[0043] In practical operation, the day-ahead optimal scheduling model can be solved using existing optimization algorithms. Its optimization objective is to minimize the overall cost of the scheduling cycle. Its decision variables include the day-ahead planned output of hydropower and the day-ahead planned output of energy storage. The day-ahead scheduling plan is obtained by optimizing the decision variables through optimization algorithms. For example, the day-ahead optimal scheduling model can be optimized using the Sparrow Search Algorithm (SSA model).

[0044] Considering that the SSA model initializes the individual positions of the sparrow population by randomly generating positions, this approach leads to low population diversity and slow convergence speed. In order to more efficiently deal with the more complex wind-solar-water-storage complementary optimization problem and improve its convergence ability, this embodiment improves the SSA model and proposes an improved sparrow search algorithm.

[0045] First, the sparrow population needs to be initialized, where the position of each sparrow represents the value of all day-ahead decision variables.

[0046] For example, using the Tent chaotic mapping makes the initial position distribution more uniform, and the position initialization formula is: ; ; In the formula, This represents the initial position of the i-th sparrow in the v-th dimension of the sparrow population, with parameters... It is the control factor of the Tent mapping. When the value is close to 0.5, the Tent mapping will exhibit more obvious chaotic characteristics.

[0047] In each iteration, the sparrow's position needs to be updated. In this embodiment, the update strategy adopted is as follows: Calculate the fitness of each sparrow u in the current population. And find its median. Wherein, the fitness function is the overall cost function of the day-ahead optimization scheduling model; For fitness satisfies The sparrow's location is updated to: ; In the formula, Let v be the value of the sparrow u in the r-th iteration. Indicates the movement step size. This represents the dot product operation. Denotes a random search path that follows a Lévy distribution, denoted as: The Levy flight strategy maintains the fast convergence speed of the SSA algorithm while improving the diversity of sparrow populations in the later stages. For fitness satisfies For a sparrow, Brownian motion is incorporated to update its position. The position update formula is:

[0048] In the formula, Let represent the current position vector of the sparrow at the r-th iteration. This represents the new position after s Brownian perturbation steps starting from the r-th iteration, where the variance of the difference between the position after the perturbation and the position before the perturbation is... The distribution follows a normal distribution, where r is the number of iterations and c is the diffusion coefficient of Brownian motion. As the number of iterations increases, the variance increases, the motion distance increases, and it becomes more likely to escape local optima.

[0049] After multiple iterations, the position of the sparrow with the lowest fitness is output as the optimization result, which is the optimal value of the decision variable.

[0050] like Figure 3 The diagram shows a flowchart of the improved sparrow search algorithm according to an embodiment of the present invention, which includes the following steps: Step 1: Parameter Initialization. Set parameters such as population size, maximum number of iterations, proportion of discoverers, proportion of watchers, upper and lower bounds of variables, and step size / perturbation coefficient, and initialize the iteration counter.

[0051] Step 2: Chaotic Population Initialization. Initial position vectors for each sparrow are generated using chaotic mapping, and then mapped to the range of variable values ​​to obtain the initial population.

[0052] Step 3: Handle constraints. Apply boundary and constraint conditions to the initial population to ensure that the positions of each individual satisfy the variable range and constraint requirements.

[0053] Step 4: Calculate fitness. Calculate the fitness value of each sparrow in the population and record the current best individual and its fitness.

[0054] Step 5: Sort by fitness. Sort the individuals in the population by fitness to obtain the sorted population sequence, and determine the median fitness value. .

[0055] Step Six: Divide the population into discoverers and followers. Divide the population into discoverer individuals and follower individuals according to a preset ratio.

[0056] Step 7: Update the positions of discoverers and followers. Determine if an individual is a discoverer: if it is a discoverer, update its position according to the discoverer position update strategy; if it is a follower, update its position according to the follower position update strategy.

[0057] Step 8: Update the vigilant's position. Determine if the individual is a vigilant: if so, update its position according to the vigilant position update strategy; otherwise, leave this step unchanged and proceed to the next step.

[0058] Step Nine: Constraint Processing and Fitness Calculation. Constraints / boundaries are processed on the individuals updated in Steps Seven and Eight, and the fitness is recalculated.

[0059] Step 10: Select a perturbation update strategy. Based on individual fitness... With median fitness The comparison results determine the perturbation method: if If the position is positive, Brownian motion is used to update it; otherwise, Levy flight strategy is used to update it.

[0060] Step 11: Re-constraint and Fitness Calculation. Constrain / boundary processing is applied to the updated individual positions after perturbation, and the updated fitness is calculated.

[0061] Step 12: Update the optimal sparrow position. Compare the fitness of each individual in the population after the update, update and save the optimal sparrow position (global optimal solution) and its fitness value for the current iteration.

[0062] Step Thirteen: Termination Check. Determine if the termination condition is met (reaching the maximum number of iterations or meeting the accuracy threshold, etc.); if met, output the optimal solution and end the iteration; if not, return to Step Six to begin the next iteration.

[0063] S2. After entering a new scheduling cycle, the real-time system status at the scheduling time, the day-ahead scheduling plan, the predicted daily output of wind power, and the predicted daily output of photovoltaic power are input into the intraday rolling optimization model to obtain the intraday scheduling strategy for the corresponding scheduling time. The intraday scheduling strategy includes the intraday planned output of hydropower and the intraday planned output of energy storage. The optimization objective of the intraday rolling optimization model is to minimize the local cost at the scheduling time. The local cost includes the load shedding penalty and the equipment adjustment penalty. The equipment adjustment penalty includes the deviation penalty between the intraday scheduling strategy and the day-ahead scheduling plan at the scheduling time.

[0064] This step involves performing optimization at each scheduling moment after entering the scheduling cycle, outputting a fine-grained scheduling scheme. Assuming a 15-minute time scale, scheduling optimization is performed and the scheduling strategy is executed every 15 minutes using an intraday rolling optimization model.

[0065] Specifically, the intraday optimization scheduling model has intraday scheduling objectives and constraints.

[0066] Intraday scheduling objectives in the intraday optimized scheduling model.

[0067] The specific goal of intraday scheduling is to minimize the load loss penalty and equipment adjustment penalty at the scheduling time, while taking into account the load loss risk caused by power deviation and the equipment adjustment cost, so as to achieve a balance between risk control and plan tracking.

[0068] Therefore, the intraday scheduling objective is to minimize the local cost at each scheduling moment. The specific function of the local cost at each scheduling moment can be in the following form: ; In the formula, The cost function value represents the local cost. and These are the weights for the underload penalty and the adjustment penalty, respectively. The load shedding penalty function at the scheduling time. This is a device adjustment penalty function for local scheduling moments.

[0069] Among them, the underload penalty function The calculation formula can be: ; In the formula, k represents the previous completed scheduling time, and M represents the number of consecutive scheduling times. The cost factor per unit of load loss. Let be the system load demand at time t. The planned daily charging power for energy storage at time t. For the predicted wind power output at time t, To contribute to the daily hydropower plan at time t, For the photovoltaic power output at time t, Let be the planned daily power output of energy storage at time t. The planned daily output of hydropower and the planned daily output of energy storage (including planned daily power output for energy storage charging and discharging) are the decision variables for intraday scheduling. The predicted output of wind power and photovoltaic power are obtained in advance based on meteorological forecasts. These forecasts are made during day-ahead scheduling to obtain day-ahead forecast results, and are further predicted during intraday scheduling to obtain intraday forecast results.

[0070] The inclusion of equipment adjustment penalties in intraday dispatch targets is to ensure close alignment between intraday rolling dispatch and day-ahead dispatch plans, improve overall system performance and efficiency, minimize real-time output deviations of hydropower and energy storage power stations, and maintain consistency with day-ahead plans as much as possible. Specifically, equipment adjustment penalties include hydropower output deviation penalties, energy storage output deviation penalties, and total system output deviation penalties.

[0071] Among them, the equipment adjustment penalty function The calculation formula can be: ; ; ; ; In the formula, k represents the previous completed scheduling time, and M represents the number of consecutive scheduling times. and These represent the changes in hydropower output and energy storage output at time t, respectively. , and The deviation coefficients for hydropower units, energy equipment plans, and overall output are not specified. This represents the planned daily output of hydropower at time t. This represents the planned daily power output of hydropower at time t. Indicates the first The daily planned output of energy storage at any time. This represents the planned output of the energy storage system at time t. This represents the deviation of the overall output at time t. This represents the predicted photovoltaic power output at time t. This represents the predicted daily wind power output at time t. This represents the predicted photovoltaic output at time t. This represents the predicted wind power output at time t.

[0072] After determining the optimization objective and decision variables, the above intraday rolling optimization model is solved to find the optimal values ​​of the decision variables and generate the intraday scheduling strategy for the current scheduling time.

[0073] In one embodiment, the intraday rolling optimization model is constructed as a Markov decision process (MDP) and solved using a reinforcement learning algorithm, wherein the Markov decision process includes a state space, an action space, and a reward function.

[0074] (1) State space The state space aims to provide an agent with a comprehensive snapshot of the system's future predictions, current physical constraints, and predetermined operational goals, and is the foundation upon which reinforcement learning models can learn effective policies.

[0075] Specifically, the state space S includes real-time system state parameters and predicted parameters and day-ahead planning parameters for M consecutive scheduling times starting from the next scheduling time. The predicted parameters include intraday predicted output of wind power, intraday predicted output of photovoltaic power and system load. The day-ahead planning parameters include the day-ahead planned output of hydropower and the day-ahead planned output of energy storage. The current physical state parameters include the current reservoir capacity of the hydropower station and the current remaining capacity of the energy storage system.

[0076] The state space S can be specifically represented as: ; In the formula, Let these be the predicted daily output vectors of wind power, photovoltaic power, and system load for the next M consecutive scheduling times, respectively. , These are the planned daily output vectors of hydropower and energy storage for the next M consecutive scheduling times, respectively, and V and E are the current reservoir capacity of the hydropower station and the current remaining capacity of the energy storage system, respectively.

[0077] For example, Let be the vector representing the planned hydropower output for the three scheduled times of 8:15, 8:30, and 8:45. Since the planned hydropower output for 8:00 to 9:00 is 100 in the daily scheduling plan, then the planned hydropower output vector is... The result is (100, 100, 100).

[0078] In one embodiment, to maintain consistency of input dimensions within the rolling optimization framework, all vectors can be padded to the same dimension in each direction using zero-padding.

[0079] Through this state representation method, the agent can embed constraint information directly into the input of the state space based on its vision of the future. When learning, the agent will naturally make decisions based on this information, thereby achieving compliance with constraints and making optimal real-time adjustment decisions.

[0080] (2) Action space The state space A includes the planned daily output of hydropower and the planned daily output of energy storage for M consecutive scheduling times in the future. A positive planned daily output of energy storage indicates discharging, and a negative planned output indicates charging.

[0081] Based on observations of state S, the agent selects an action from the action space A. The action space consists of two continuous variables: the power generation flow of the hydropower station and the operating power of the energy storage system, specifically defined as: ; In the formula, This is represented as the daily planned hydropower output vector for M consecutive scheduling times in the future. It is represented as the daily planned output vector of energy storage for M consecutive scheduling times in the future, where positive indicates discharge and negative indicates charging.

[0082] It should be noted that although the action space includes actions at M scheduling times, only the action at the next scheduling time is actually taken as the intraday scheduling plan for the next scheduling time. Since the current decision has a continuous impact on the running results of multiple future scheduling times, if only the immediate state of the next scheduling time is considered, it is easy to lead to short-sighted strategy. Therefore, multiple scheduling times are considered in the optimization.

[0083] (3) Reward function After an agent performs an action, the environment provides an immediate reward R to evaluate the quality of the decision. To align the reward maximization objective of reinforcement learning with the optimization objective of minimizing the local cost at intraday scheduling moments, the reward in MDP is defined as:

[0084] In the formula, , intraday objective function and Weighting factors.

[0085] After constructing the above MDP process, the optimal action is found through reinforcement learning to obtain the intraday scheduling strategy for the next scheduling time.

[0086] In one embodiment, the reinforcement learning algorithm used to solve the intraday rolling optimization model is the SAC (Flexible Motion Evaluation) algorithm.

[0087] The SAC algorithm constructs a "soft" value function by introducing policy entropy into the objective function. Its soft state-value function and optimal policy... The definition is as follows: ; ; In the formula, This is the entropy regularization coefficient, used to balance immediate rewards and policy entropy. This allows for the control of the randomness of the optimal strategy.

[0088] To implement the SAC algorithm above, a neural network structure can be constructed, including a policy network (parameters). Two value networks (parameters) ) and two corresponding target value networks (parameters) The dual-value network structure is used to alleviate the overestimation of Q-values ​​and improve training stability.

[0089] Before practical application, the SAC neural network needs to be trained so that it can quickly output the best action based on the input.

[0090] Specifically, the preprocessed dataset is divided into training, validation, and test sets proportionally. The training set data is first input into the day-ahead optimization scheduling model to obtain the day-ahead scheduling plan, which is then used as one of the input data to train the SAC neural network in the intraday rolling optimization model. During SAC training, "state (including day-ahead plan) - action - reward" data is stored in the experience replay pool, and the validation set is used to optimize parameters such as the network learning rate and entropy coefficient to ensure the model's dynamic response capability to real-time operating conditions.

[0091] After training, the historical operational data of wind, solar, hydro, and energy storage from the test set are input into the complete day-ahead and intraday scheduling model: first, the day-ahead optimization scheduling model generates an initial 24-hour hydropower / energy storage output plan, and then the intraday rolling optimization model outputs 15-minute rolling scheduling instructions based on real-time status. In the actual scheduling process, the scheduling results output by the intraday SAC model are ultimately executed.

[0092] In the first aspect, the present invention proposes a multi-timescale optimization scheduling system for wind, solar, hydro, and storage systems.

[0093] The system includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0094] This system can be installed on computing devices such as desktop computers, laptops, handheld computers, and cloud servers. The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The memory can be used to store computer programs and / or modules. The processor implements various system functions by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory.

[0095] The technical solution of this invention is compared with widely used methods below.

[0096] The experiment included three typical scenarios: Scenario 1 served as the baseline scenario, used to evaluate the model's economy and multi-energy coordination performance under normal forecasting conditions; Scenario 2 considered wind power, photovoltaic, and load forecasting errors, and analyzed the robustness of the strategy by superimposing Gaussian noise with a standard deviation of 30% onto the baseline curve; Scenario 3 simulated extreme conditions, evaluating the system's operational resilience under the combined disturbances of a sudden drop in photovoltaic output and an unexpected increase in peak load.

[0097] To comprehensively compare performance, three representative scheduling strategies were set up as references: ① Day-ahead optimization only strategy, which generates plans only during the day-ahead phase without intraday adjustments; ② Intraday optimization only strategy, which omits the day-ahead phase and optimizes in real time based on short-term forecasts in each 15-minute period; ③ Conventional rule-based scheduling strategy, which adopts the fixed logic of energy storage "valley charging and peak releasing" and hydropower output according to the net load ratio, and makes adjustments according to the rule of "hydropower priority, energy storage second, and grid backup". By comparing the operating costs, power shortages, and system stability of each strategy under different scenarios, the comprehensive advantages of the proposed method in terms of economy, robustness, and resilience are verified.

[0098] Table 1 below shows the scheduling operation results for the baseline scenario, Table 2 shows the scheduling operation results for the 30% uncertainty scenario, and Table 3 shows the scheduling operation results for the extreme scenario.

[0099]

[0100]

[0101]

[0102] As shown in Tables 1, 2, and 3, through multiple simulation comparisons under benchmark, uncertainty, and extreme combined scenarios, the results strongly demonstrate that the method proposed in this invention is comprehensively superior to comparative strategies such as day-ahead optimization only, intraday optimization only, and conventional rules in terms of overall operating cost, uncertainty robustness, and system resilience. Under the premise of ensuring the safety and stability of the power grid, it provides an efficient, reliable, and highly intelligent technical path for solving the optimization scheduling problem of complex multi-energy complementary systems.

[0103] Overall, the multi-timescale optimization scheduling method for wind-solar-hydro-storage systems proposed in this invention, through a multi-scale collaborative framework of "day-ahead global planning + intraday rolling correction," combines global optimization, dynamic adaptation, and physical feasibility. This method effectively solves the problem of unifying economy, robustness, and sustainability in multi-energy complementary wind-solar-hydro-storage systems, providing an efficient and intelligent optimization path for the safe and economical operation of high-proportion renewable energy power systems. Furthermore, in solving the model, the global optimization advantage of the improved sparrow search algorithm is combined with the online adaptive capability of the SAC algorithm. During the intraday phase, a Markov decision process is constructed based on the SAC algorithm, and real-time forecast information is used to dynamically correct the day-ahead plan in the rolling time domain, achieving intelligent suppression of short-term power deviations and operational risks.

[0104] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. It should be noted that the terms "in one embodiment," "for example," and "again" are intended to illustrate the present invention and are not intended to limit the present invention.

[0105] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A multi-timescale optimization scheduling method for a wind-solar-hydro-storage system, characterized in that, include: Before entering a new scheduling cycle, the predicted day-ahead output of wind power and photovoltaic power for the new scheduling cycle are input into the day-ahead optimization scheduling model to obtain the day-ahead scheduling plan for different time intervals within the new scheduling cycle. The day-ahead scheduling plan includes the planned day-ahead output of hydropower and the planned day-ahead output of energy storage. The optimization objective of the day-ahead optimization scheduling model is to minimize the overall cost of the scheduling cycle, which includes operating costs and power supply reliability penalties. After entering a new scheduling cycle, the real-time system status, day-ahead scheduling plan, and intraday predicted output of wind power and photovoltaic power at the scheduling time are input into the intraday rolling optimization model to obtain the intraday scheduling strategy for the corresponding scheduling time. The intraday scheduling strategy includes the intraday planned output of hydropower and the intraday planned output of energy storage. The optimization objective of the intraday rolling optimization model is to minimize the local cost at the scheduling time. The local cost includes load shedding penalty and equipment adjustment penalty. The equipment adjustment penalty includes the deviation penalty between the intraday scheduling strategy and the day-ahead scheduling plan at the scheduling time.

2. The multi-timescale optimization scheduling method for wind-solar-hydro-storage systems as described in claim 1, characterized in that, In the aforementioned day-ahead optimization scheduling model, the cost function for calculating the overall cost of the scheduling cycle is: ； In the formula, The cost function value represents the overall cost. and The weights for operating costs and power supply reliability penalties are respectively. The operating costs within the scheduling cycle, Penalty for power supply reliability.

3. The multi-timescale optimization scheduling method for wind-solar-hydro-storage systems as described in claim 2, characterized in that, Operating costs The calculation formula is: ； In the formula, , , , They are respectively represented as The costs of purchasing electricity from the main power grid, operating costs of hydropower, operating costs of energy storage, and revenue from electricity sales are all considered at all times, where T is the total number of time intervals within the dispatch cycle. and The reservoir capacity status at the start and end times of the scheduling interval. and The energy storage status at the start and end times of the scheduling interval. and This is the penalty coefficient for the imbalance between the initial and final storage capacity and the initial and final energy storage state. Power supply reliability penalty The calculation formula is: ； In the formula, The weights for the impact of power supply shortage and resource reserve are defined. EENS represents the expected power supply shortage during the dispatch period, which refers to the expected amount of electricity that cannot meet load demand due to insufficient power supply capacity during the dispatch period. This is a penalty that occurs when the actual reserve capacity of the reservoir is lower than the preset minimum reserve threshold. This is a penalty that is incurred when the actual standby capacity of energy storage is lower than the preset minimum standby threshold.

4. The multi-timescale optimization scheduling method for wind-solar-hydro-storage systems as described in claim 1, characterized in that, The day-ahead optimization scheduling model has constraints, including: wind and solar power output constraints, battery energy storage power station constraints, and hydropower station constraints. The wind and solar power output constraints include: wind power constraints and photovoltaic power constraints; The constraints of the battery energy storage power station include: energy storage system state of charge constraints, energy storage system charge and discharge constraints, energy balance constraints at the beginning and end of the scheduling cycle, and energy storage charge and discharge state constraints. The constraints of the hydropower station include: water balance constraints, outflow constraints, head constraints, and hydropower output constraints.

5. The multi-timescale optimization scheduling method for wind-solar-hydro-storage systems as described in claim 1, characterized in that, In the intraday rolling optimization model: Its underload penalty function for: ； In the formula, k represents the previous completed scheduling time, and M represents the number of consecutive scheduling times. Cost per unit of load loss Let be the system load demand at time t. The planned daily charging power for energy storage at time t. For the predicted wind power output at time t, To contribute to the daily hydropower plan at time t, For the photovoltaic power output at time t, Let be the planned daily power of energy storage discharge at time t; Its equipment adjustment penalty function for: ； ； ； ； In the formula, , , These represent the changes in hydropower output, energy storage output, and overall output deviation at time t, respectively. , and The deviation coefficient for hydropower units, energy equipment plans, and overall output is not specified. , , , Let represent the planned daily output of hydropower, the predicted daily output of photovoltaic power, the predicted daily output of wind power, and the planned daily output of energy storage at time t, respectively. It contributes to the energy storage plan at time t within the day.

6. The multi-timescale optimization scheduling method for wind-solar-hydro-storage systems as described in claim 1, characterized in that, The day-ahead optimization scheduling model is solved using an improved sparrow search algorithm, where the position of each sparrow represents the value of all day-ahead decision variables, and the fitness function used by the algorithm is the cost function for calculating the overall cost. In each iteration of the improved sparrow search algorithm, the update strategy used to update the sparrow's position is as follows: Calculate the fitness of each sparrow u in the current population. And find its median. ; For fitness satisfies The sparrow's location is updated to: ； In the formula, Let v be the value of the sparrow u in the r-th iteration. Indicates the movement step size. This represents the dot product operation. Indicates a random search path. Let be the parameters of the Lévy distribution, which follows a Lévy distribution, denoted as . ; For fitness satisfies For a sparrow, Brownian motion is incorporated to update its position. The position update formula is: In the formula, Let represent the current position vector of the sparrow at the r-th iteration. This represents the new position after s Brownian perturbation steps starting from the r-th iteration, where the variance of the difference between the position after the perturbation and the position before the perturbation is... The normal distribution is given by r, where r is the number of iterations and c is the diffusion coefficient of Brownian motion.

7. The multi-timescale optimization scheduling method for wind-solar-hydro-storage systems as described in claim 1, characterized in that, The intraday rolling optimization model is constructed as a Markov decision process and solved using a reinforcement learning algorithm. The Markov decision process has a state space, an action space, and a reward function. The state space S includes real-time system state parameters and predicted parameters and day-ahead planning parameters for M consecutive scheduling times starting from the next scheduling time. The predicted parameters include predicted daily output of wind power, predicted daily output of photovoltaic power and system load. The day-ahead planning parameters include planned daily output of hydropower and planned daily output of energy storage. The current physical state parameters include the current reservoir capacity of the hydropower station and the current remaining capacity of the energy storage system. The state space A includes the planned daily output of hydropower and the planned daily output of energy storage for the next M consecutive scheduling times; The reward function R is the inverse of the local cost in the intraday rolling optimization model.

8. The multi-timescale optimization scheduling method for wind-solar-hydro-storage systems as described in claim 7, characterized in that, The reinforcement learning algorithm used to solve the intraday rolling optimization model is the SAC algorithm.

9. The multi-timescale optimization scheduling method for wind-solar-hydro-storage systems as described in any one of claims 1 to 8, characterized in that, The day-ahead optimization scheduling model uses a 1-hour time interval and outputs a day-ahead scheduling plan that includes the planned hydropower output and energy storage output for each hour within the next 24 hours. The intraday rolling optimization model uses a 15-minute scheduling period and outputs an intraday scheduling strategy that includes the planned hydropower output and energy storage output for the next scheduling period.

10. A multi-timescale optimization scheduling system for a wind-solar-hydro-storage system, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 9.

Images