Optical storage system and control method and device thereof

Abstract

The invention relates to the field of power system control, and discloses an optical storage system and a control method and device thereof, and the method comprises the steps: inputting the user load data of a first time period and related influence factor data into a load prediction model, and obtaining the user load data of a second time period in the future; inputting the photovoltaic power generation power of the photovoltaic power generation system in the first time period and the related influence factor data into the photovoltaic power generation power prediction model to obtain the photovoltaic power generation power in the second time period; constructing an optimization model based on a preset constraint, the user load data in the second time period and the photovoltaic power generation power; the first state of charge balance constraint in the preset constraint comprises that the absolute value of the deviation between the state of charge of the battery cluster and the average state of charge is smaller than or equal to a first threshold value; and performing iteration solution on the optimization model to obtain the charging and discharging power of the plurality of battery clusters, and adjusting the charging and discharging power of at least part of the battery clusters if the battery clusters which do not meet the first state-of-charge equilibrium constraint exist in each iteration. And the control precision and efficiency are improved.

Description

Technical Field

[0001] This application relates to the field of power system control technology, and in particular to a photovoltaic energy storage system and its control method and apparatus. Background Technology

[0002] Currently, photovoltaic power generation systems and energy storage systems, which are configured in synergy, have been widely used. On the one hand, photovoltaic power generation technology can effectively reduce electricity costs. On the other hand, energy storage systems can reduce demand-based electricity costs by implementing peak-valley electricity price arbitrage strategies, while also playing a role in smoothing load fluctuations and keeping the power change rate within a stable range. This can effectively address the dual challenges of emission reduction pressure and rising energy costs.

[0003] However, the control strategies of most photovoltaic-storage systems still have poor control accuracy and efficiency, making it difficult to fully leverage the advantages of photovoltaic-storage systems. Summary of the Invention

[0004] The purpose of this application is to provide at least one photovoltaic energy storage system and its control method and device, which can at least solve the problem that the control accuracy and efficiency of the control strategy of the photovoltaic energy storage system are still poor, and can at least achieve the effect of improving the control accuracy and efficiency of the control strategy of the photovoltaic energy storage system and reducing the cost of energy storage.

[0005] In a first aspect, this application provides a control method for a photovoltaic-energy storage system, the photovoltaic-energy storage system including a photovoltaic power generation system and an energy storage system having multiple battery clusters, the method comprising: Input the user load data and related influencing factor data of the first period into the load forecasting model to obtain the user load data of the second period in the future; The photovoltaic power generation power and related influencing factor data of the photovoltaic power generation system in the first time period are input into the photovoltaic power generation power prediction model to obtain the photovoltaic power generation power in the second time period; Based on preset constraints and user load data and photovoltaic power generation in the second time period, an optimization model is constructed with the charging and discharging power of the multiple battery clusters in the second time period as variables. The preset constraints include a first state of charge balance constraint, which includes the absolute value of the first deviation of the battery cluster being less than or equal to a first threshold. The first deviation is the deviation between the state of charge of the battery cluster and the average state of charge of the multiple battery clusters. The charging and discharging power of the plurality of battery clusters in the second time period is obtained by iteratively solving the optimization model. In each iteration, if there is a battery cluster that does not meet the first state of charge balance constraint, the charging and discharging power of at least some of the battery clusters is adjusted.

[0006] Optionally, adjusting the charging and discharging power of at least a portion of the plurality of battery clusters includes: The absolute values ​​of the first deviations of the plurality of battery clusters are sorted. Based on the sorting result, the priority order of the plurality of battery clusters is determined, wherein the larger the absolute value of the first deviation, the higher the priority of the battery cluster; Select at least a portion of the battery clusters from the plurality of battery clusters in descending order of priority; The charging and discharging power of at least a portion of the selected battery clusters is adjusted so that the plurality of battery clusters respectively satisfy the first state-of-charge balance constraint.

[0007] Optionally, the adjustment amount of the charging and discharging power of at least a portion of the battery clusters. for:

[0008] in, This indicates the state of charge of the k-th battery cluster at time t. This represents the average state of charge of the multiple battery clusters at time t; This is the preset adjustment coefficient; The rated capacity of a single battery cluster; This refers to the charging and discharging time between two adjacent moments.

[0009] Optionally, the optimization objective of the optimization model includes multiple objectives, including the lowest total electricity cost over a preset period, the longest energy storage life, and the lowest amount of curtailed solar power. The iterative solution of the optimization model to obtain the charging and discharging power of the multiple battery clusters in the second time period includes: The charging and discharging power of the multiple battery clusters in the second time period is obtained by iteratively solving the optimization model using the NSGA-III algorithm.

[0010] Optionally, the optimization model includes multiple objective functions corresponding to the multi-objective; the step of using the NSGA-III algorithm to iteratively solve the optimization model to obtain the charging and discharging power of the multiple battery clusters in the second time period includes: Generate reference points for the target space, the dimension of which is equal to the number of targets in the multi-target system; initialize a population, where each individual in the population includes the charging and discharging power of the multiple battery clusters; based on the population, perform the following iterative steps: Fitness evaluation is performed on each individual in the current population to obtain the fitness evaluation result for each individual; the fitness evaluation includes checking whether the individual meets the preset constraints; if there are battery clusters that do not meet the first state-of-charge equilibrium constraint, the charging and discharging power of at least some of the battery clusters is adjusted so that the battery clusters respectively meet the first state-of-charge equilibrium constraint; the values ​​of the multiple objective functions corresponding to the individual are calculated based on the individual; Based on the Pareto non-dominated relation, the population after fitness evaluation is sorted by non-dominated order to stratify the population and obtain different non-dominated levels. The values ​​of the multiple objective functions corresponding to the individuals in the stratified population are normalized to obtain normalization results; and based on the normalization results, the vertical distance from the individuals in the population to each of the reference points is calculated, and the individuals are associated with the reference points with the closest vertical distance. If the iteration termination condition is met, output the uniformly distributed Pareto optimal solution; If the iteration termination condition is not met, select some individuals from the population as parents according to the order of the non-dominated layer hierarchy from low to high and the order of the number of individuals associated with the reference point from small to large; select individuals from the parents for crossover and mutation to obtain offspring; based on the parents and the offspring, obtain the population for the next iteration, and re-execute the iteration steps.

[0011] Optionally, the objective function include:

[0012]

[0013] in, The possible values ​​are 1, 2, and 3. This indicates the total electricity cost. This represents the objective function corresponding to the minimum total electricity cost. Indicates energy storage lifespan loss. This represents the objective function corresponding to the longest energy storage lifetime. Indicates the amount of light discarded. This represents the objective function corresponding to minimizing light wastage. As a penalty item, Included by preset constraints The first constraint The amount of a constraint violation, Represents an individual. Indicates the dynamic penalty coefficient. Indicates the number of iterations. This represents the base penalty coefficient.

[0014] The preset constraints also include power balance constraints, peak-valley constraints, photovoltaic power generation constraints, battery cluster state of charge constraints, battery cluster charge-discharge state constraints, battery cluster power constraints, battery cluster state of charge continuity constraints, and second state of charge balance constraints; the second state of charge balance constraint includes the absolute value of the second deviation of the plurality of battery clusters being less than or equal to a second threshold; wherein, the second deviation is... and deviation, This indicates that at time t, the first of the multiple battery clusters... The state of charge of the aforementioned battery clusters, This represents the state of charge of the j-th battery cluster among the plurality of battery clusters at time t.

[0015] Optionally, both the load forecasting model and the photovoltaic power generation forecasting model include an LSTM model based on an attention mechanism; The attention-based LSTM model includes at least an input gate, a cell gate, a forget gate, a state layer, an output gate, a hidden layer, an attention mechanism, and an output layer. The input gate is used to obtain the input features to be written based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The unit gate is used to obtain the candidate memory to be written based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The forget gate is used to obtain the historical memory to be retained based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The state layer is used to obtain the current state based on the input features at the current time, the candidate memory, the historical memory, and the historical state at the previous time. The output gate is used to obtain the memory to be output based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The hidden layer is used to obtain the current hidden state based on the memory to be output at the current moment and the current state; The attention mechanism is used to calculate the current hidden state based on the attention mechanism, and obtain the calculation result of the attention mechanism; The output layer is used to obtain the prediction result based on the calculation result of the attention mechanism; the input feature is obtained based on the user load data and related influencing factor data of the first time period, and the prediction result includes the user load data of the second time period; or, the input feature is obtained based on the photovoltaic power generation power of the photovoltaic power generation system in the first time period and related influencing factor data, and the prediction result includes the photovoltaic power generation power of the second time period. The attention mechanism includes: calculating the temporal correlation between the current hidden state and the hidden state at each historical moment to obtain a correlation score; determining the attention weight corresponding to the hidden state at each historical moment based on the correlation score; performing a weighted summation based on the hidden states at all historical moments and their corresponding attention weights to obtain an intermediate vector; and obtaining the calculation result of the attention mechanism based on the intermediate vector and the current hidden state.

[0016] Secondly, this application provides a control device for a photovoltaic-energy storage system, the photovoltaic-energy storage system including a photovoltaic power generation system and an energy storage system having multiple battery clusters, the device comprising: The load forecasting module is used to input the user load data and related influencing factor data of the first period into the load forecasting model to obtain the user load data of the second period in the future. The photovoltaic prediction module is used to input the photovoltaic power generation power and related influencing factor data of the photovoltaic power generation system in the first time period into the photovoltaic power generation power prediction model to obtain the photovoltaic power generation power in the second time period. The model building module is used to build an optimization model based on preset constraints and user load data and photovoltaic power generation in the second time period, with the charging and discharging power of the multiple battery clusters in the second time period as variables. The preset constraints include a first state of charge balance constraint, which includes the absolute value of the first deviation of the battery cluster being less than or equal to a first threshold. The first deviation is the deviation between the state of charge of the battery cluster and the average state of charge of the multiple battery clusters. The model solving module is used to iteratively solve the optimization model to obtain the charging and discharging power of the plurality of battery clusters in the second time period. In each iteration, if there are battery clusters that do not meet the first state of charge balance constraint, the charging and discharging power of at least some of the battery clusters is adjusted.

[0017] Thirdly, this application provides a photoelectric storage system, comprising: Photovoltaic power generation system; An energy storage system comprising multiple battery clusters; A control unit, the control unit being used to execute the photovoltaic storage system control method as described in any of the above.

[0018] The advantages of this application compared to the prior art are: In the control method of the photovoltaic-storage system of this application, user load data and related influencing factor data for a first time period are input into a load prediction model to obtain user load data for a future second time period. Similarly, the photovoltaic power generation of the photovoltaic power generation system in the first time period and related influencing factor data are input into a photovoltaic power generation prediction model to obtain the photovoltaic power generation for the second time period. Based on preset constraints and the user load data and photovoltaic power generation for the second time period, an optimization model is constructed using the charging and discharging power of the multiple battery clusters in the second time period as variables. The preset constraints include a first state-of-charge (POC) balance constraint, which includes the absolute value of a first deviation of the battery cluster being less than or equal to a first threshold. The first deviation is the deviation between the POC of the battery cluster and the average POC of the multiple battery clusters. Then, the optimization model is iteratively solved to obtain the charging and discharging power of the multiple battery clusters in the second time period. In each iteration, if there are battery clusters that do not meet the first POC balance constraint, the charging and discharging power of at least some of the battery clusters is adjusted. This not only improves the control accuracy and efficiency of the photovoltaic-storage system's control strategy but also reduces the cost of energy storage.

[0019] It is understood that the beneficial effects of the second and third aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0020] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.

[0021] Figure 1 This is a flowchart of a photovoltaic energy storage system control method provided in one embodiment of this application; Figure 2 This is a schematic diagram of a control device for a photovoltaic storage system provided in another embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been provided in the various embodiments of this application to help readers better understand this application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and modifications based on the following embodiments. The division of the various embodiments below is for the convenience of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.

[0023] To facilitate understanding of the embodiments of this application, relevant content about the optical storage system will be introduced first.

[0024] Currently, photovoltaic power generation systems and energy storage systems, which are configured in synergy, have been widely used. On the one hand, photovoltaic power generation technology can effectively reduce electricity costs. On the other hand, energy storage systems can reduce demand-based electricity costs by implementing peak-valley electricity price arbitrage strategies, while also playing a role in smoothing load fluctuations and keeping the power change rate within a stable range. This can effectively address the dual challenges of emission reduction pressure and rising energy costs.

[0025] However, the control strategies of most photovoltaic-storage systems still have poor control accuracy and efficiency, making it difficult to fully leverage the advantages of photovoltaic-storage systems.

[0026] In industrial parks employing photovoltaic (PV) and energy storage systems, the park's energy structure is transitioning from unidirectional grid dependence to a coordinated operation of "source-grid-load-storage." However, the two-part tariff mechanism poses a significant challenge to the park's energy consumption structure and dispatch methods. Demand-based tariffs constitute a high proportion of the park's total electricity costs, and the park's large peak-valley load difference further exacerbates peak demand, driving up energy expenditures. In some scenarios, distributed PV requires self-consumption and anti-reverse flow control, which directly leads to a significant decrease in the park's overall economic viability when the curtailment rate is high. If users can predict PV power generation and user load consumption, and fully utilize the flexibility of energy storage power regulation to discharge during peak power consumption, controlling the maximum load demand, they can reduce basic electricity costs and simultaneously achieve peak-valley arbitrage to increase revenue.

[0027] Based on this, this application proposes a control method for a photovoltaic energy storage system. The implementation details of the control method for the photovoltaic energy storage system in this embodiment are described in detail below. The following implementation details are provided for ease of understanding and are not necessary for implementing this solution.

[0028] This embodiment provides a control method for a photovoltaic energy storage system, such as... Figure 1As shown, the photovoltaic-storage system includes a photovoltaic power generation system and an energy storage system with multiple battery clusters. The method includes: Step S101: Input the user load data and related influencing factor data of the first time period into the load forecasting model to obtain the user load data of the second time period.

[0029] For example, user load data includes user load power.

[0030] The second time period is a future period of the first time period. For example, the second time period includes a day in which the charging and discharging power of multiple battery clusters to be optimized, i.e., the target day, while the first time period includes a portion of the day preceding the second time period.

[0031] The load forecasting model is used to predict user load data for the second time period.

[0032] User load data is influenced by factors including data on indicators of at least one influencing factor. These factors include at least one of the following: economic factors, power policy factors, meteorological factors, time factors, industry development, production plans, and other random disturbances.

[0033] In industrial parks with their own photovoltaic and energy storage systems, power load data exhibits strong instability due to various unforeseen circumstances and influencing factors. To achieve efficient and accurate prediction of power load data, it is necessary not only to fully study the development patterns of power load data itself, but also to comprehensively consider external influencing factors.

[0034] Step S102: Input the photovoltaic power generation power and related influencing factor data of the photovoltaic power generation system in the first time period into the photovoltaic power generation power prediction model to obtain the photovoltaic power generation power in the second time period.

[0035] The photovoltaic power generation prediction model is used to predict the photovoltaic power generation in the second time period.

[0036] Data on factors influencing photovoltaic (PV) power generation include data on indicators of at least one of these factors. These at least one influencing factor includes at least one of the following: solar irradiance, temperature, time, PV panel angle, and power generation efficiency.

[0037] Step S103: Based on preset constraints and user load data and photovoltaic power generation in the second time period, construct an optimization model with the charging and discharging power of multiple battery clusters in the second time period as variables; the preset constraints include a first state of charge balance constraint, which includes the absolute value of the first deviation of the battery cluster being less than or equal to a first threshold, and the first deviation being the deviation between the state of charge of the battery cluster and the average state of charge of multiple battery clusters.

[0038] The optimization model is specifically designed for energy storage systems, optimizing the charging and discharging power of multiple battery clusters during the second time period. For example, an optimization model for a day-ahead energy storage system can be constructed. The charging and discharging power of a battery cluster includes both charging and discharging power.

[0039] For example, the preset constraints include a first state-of-charge (POC) balance constraint, and also include power balance constraints, peak-valley constraints, photovoltaic power generation constraints, POC constraints of the battery cluster, charge-discharge state constraints of the battery cluster, power constraints of the battery cluster, POC continuity constraints of the battery cluster, and a second POC balance constraint. Each of the preset constraints will be described in detail below.

[0040] First, power balance constraints include: .in This represents the power removed from the grid at time t. This represents the photovoltaic power generation at time t. express Real-time user load power This represents the sum of the discharge power of each battery cluster. This represents the sum of the charging power of each battery cluster.

[0041] Second, photovoltaic power generation constraints include: .in, This represents the predicted maximum photovoltaic power generation at time t.

[0042] Third, the state of charge constraints of the battery cluster include: .in, For time t, the first The state of charge of each battery cluster, and These are the lower and upper limits of the state of charge, respectively.

[0043] Fourth, the state-of-charge constraints of the battery cluster include: .in, and For variables that take the value of 0 or 1, respectively represent The discharge and charge states of the battery clusters at all times. 1 indicates The battery cluster is always in a discharging state. 1 indicates The battery clusters are always in a charging state.

[0044] Fifth, the power constraints of the battery clusters include:

[0045] in, and These represent the maximum discharge power and maximum charging power of the battery cluster, respectively.

[0046] Sixth, the continuity constraints of the state of charge of the battery cluster include: .in, express +1 moment The state of charge of each battery cluster; They represent Time of the first The charging and discharging power of each battery cluster is the sum of the charging and discharging power of the energy storage system at the same time. The charging and discharging time between two adjacent moments is, for example, 15 minutes. This indicates the charging efficiency of the battery cluster; This indicates the discharge efficiency of the battery cluster; The rated capacity of a single battery cluster.

[0047] Seventh, the second state-of-charge equilibrium constraint includes: the absolute value of the second deviation of multiple battery clusters is less than or equal to the second threshold, i.e. The second deviation is... and deviation, This indicates the first cell cluster among multiple cell clusters at time t. The state of charge of each battery cluster, This represents the state of charge of the j-th battery cluster among multiple battery clusters at time t. This indicates the second threshold.

[0048] The second state-of-charge (SOC) equalization constraint is used for accuracy constraints in string PCS to ensure the coordinated control effect of string PCS.

[0049] Eighth, peak-valley constraints include: .in, and These represent the maximum and minimum values ​​of user load data within the user's preset time period, respectively.

[0050] To prevent energy storage from being influenced by peak-valley electricity pricing and creating new load spikes, the peak-valley constraints for user loads are as described above.

[0051] Ninth, the first state-of-charge equilibrium constraint includes the absolute value of the first deviation of the battery cluster being less than or equal to the first threshold, where the first deviation is the deviation between the state of charge of the battery cluster and the average state of charge of multiple battery clusters.

[0052] The average state of charge of multiple battery clusters is the average value of the state of charge of multiple battery clusters.

[0053] Step S104: Iteratively solve the optimization model to obtain the charging and discharging power of multiple battery clusters in the second time period. In each iteration, if there are battery clusters that do not meet the first state of charge balance constraint, adjust the charging and discharging power of at least some of the multiple battery clusters.

[0054] For example, by iteratively solving the optimization model, the daily charging and discharging plan can be obtained, enabling peak-valley arbitrage and real-time scheduling.

[0055] For example, adjusting the charge and discharge power of at least a portion of a plurality of battery clusters includes: Sort the absolute values ​​of the first deviation of multiple battery clusters; Based on the sorting results, the priority order of multiple battery clusters is determined, where the larger the absolute value of the first deviation, the higher the priority of the battery cluster. Select at least a portion of the battery clusters from multiple battery clusters, in descending order of priority. The charging and discharging power of at least a portion of the selected battery clusters is adjusted so that the multiple battery clusters respectively satisfy the first state of charge balance constraint.

[0056] Thus, by applying real-time correction to the first state-of-charge (POC) balance constraint, if the POC of a battery cluster exceeds the allowable range, the charging and discharging power of the battery cluster is dynamically adjusted according to priority, ensuring that the first deviation converges to the necessary range within a single iteration. Based on this, control accuracy and efficiency can be improved.

[0057] Adjustment amount of charging and discharging power for at least part of the battery cluster as follows.

[0058]

[0059] in, This indicates the state of charge of the k-th battery cluster at time t. This represents the average state of charge of the multiple battery clusters at time t; This is the preset adjustment coefficient; The rated capacity of a single battery cluster; This refers to the charging and discharging time between two adjacent moments.

[0060] In this way, the charging and discharging power of the battery cluster can be accurately adjusted.

[0061] In this embodiment, a two-layer constraint processing mechanism for charge state balance is achieved through the first charge state balance constraint and the second charge state balance constraint.

[0062] The imbalance of state of charge (SOC) among battery clusters in an energy storage system leads to insufficient depth of charge and discharge, resulting in rapid lifespan degradation. This embodiment provides a cluster-level control strategy adapted to string-type PCS (Polymer Capacitor Systems). By adjusting the charging and discharging power between different battery clusters, the SOC balance between clusters is adjusted in a timely manner. Thus, through a two-layer constraint processing mechanism, the lifespan of the battery energy storage system can be improved, achieving SOC balance between clusters and enabling long-life energy storage, thereby reducing the cost of energy storage operation.

[0063] In this embodiment, user load data and related influencing factor data for the first time period are input into a load prediction model to obtain user load data for the second time period. Similarly, photovoltaic power generation data for the first time period and related influencing factor data are input into a photovoltaic power generation prediction model to obtain photovoltaic power generation data for the second time period. Based on preset constraints and the user load data and photovoltaic power generation data for the second time period, an optimization model is constructed using the charging and discharging power of multiple battery clusters as variables. The preset constraints include a first state-of-charge (POC) balance constraint, which specifies that the absolute value of a first deviation of a battery cluster is less than or equal to a first threshold. The first deviation is the deviation between the POC of a battery cluster and the average POC of multiple battery clusters. The optimization model is then iteratively solved to obtain the charging and discharging power of multiple battery clusters for the second time period. In each iteration, if there are battery clusters that do not meet the first POC balance constraint, the charging and discharging power of at least some of the battery clusters is adjusted. This not only improves the control accuracy and efficiency of the photovoltaic-storage system's control strategy but also reduces the cost of energy storage.

[0064] In some embodiments, the optimization objective of the optimization model includes multiple objectives, such as minimizing the total electricity cost over a preset period, maximizing energy storage lifetime, and minimizing solar power curtailment. Correspondingly, iteratively solving the optimization model to obtain the charging and discharging power of multiple battery clusters in the second period includes: using the Non-dominated Sorting Genetic Algorithm III (NSGA-III) algorithm to iteratively solve the optimization model to obtain the charging and discharging power of multiple battery clusters in the second period.

[0065] For example, the NSGA-III algorithm is used to iteratively solve the optimization model to obtain the charging and discharging power of multiple battery clusters on the target day. This allows for the determination of the daily charging and discharging plan based on the Pareto front, enabling peak-valley arbitrage and real-time scheduling. The next day, steps 101 to 104 are repeated sequentially.

[0066] The preset time period includes a second time period. For example, the preset time period is one day.

[0067] Traditional methods focus on single-objective optimization (such as reducing electricity costs), failing to balance demand control, energy storage lifetime, and minimizing solar curtailment. Scheduling schemes fail when multiple objectives conflict. This embodiment establishes a multi-objective optimization model, using the minimum total electricity cost, maximum energy storage lifetime, and minimum solar curtailment over a preset time period as multiple objectives. It balances multiple objectives such as demand, battery life, and reverse current prevention to generate a charging and discharging plan, adjusting the energy storage system's charging and discharging strategy. This reduces peak demand, lowers electricity costs, minimizes solar curtailment, and achieves demand control, peak-valley arbitrage, and real-time scheduling.

[0068] In some embodiments, the optimization model includes multiple objective functions corresponding to multiple objectives.

[0069] For example, the objective function include:

[0070]

[0071] in, The possible values ​​are 1, 2, and 3. This indicates the total electricity cost. This represents the objective function corresponding to the minimum total electricity cost. Indicates energy storage lifespan loss. This represents the objective function corresponding to the longest energy storage lifetime. Indicates the amount of light discarded. This represents the objective function corresponding to minimizing light wastage. As a penalty item, Included by preset constraints The first constraint The amount of a constraint violation, Represents an individual. Indicates the dynamic penalty coefficient. Indicates the number of iterations. This represents the base penalty coefficient.

[0072] In this embodiment, the objective function introduces a constraint penalty term, and dynamic penalty coefficients are designed for soft constraints such as anti-backflow and ramp rate. The constraints are relaxed in the early stage of iteration to improve the exploration capability, and the penalties are strengthened in the later stage to guide the feasible region, thereby improving the feasible solution generation rate.

[0073] For example, the objective function is to minimize the total daily electricity bill paid by large industrial users. The electricity bill payable by users includes electricity consumption fees and a basic electricity fee, which can be specifically expressed as: .in, This is the total electricity cost for the day; This represents the electricity cost for the day. This represents the basic electricity cost for the day. For example, if the charging / discharging time between two adjacent moments is 15 minutes, the specific formula for calculating the user's electricity cost is as follows: .in, It is a time-of-use electricity price.

[0074] According to the demand-based billing rules, the daily basic electricity cost is as follows:

[0075] in, The unit price for maximum demand; This is the demand verification value, which is the maximum demand value reported by the user; This represents the actual demand.

[0076] In addition, energy storage life targets must be considered. For example, based on the cycle life characteristics of energy storage batteries such as lithium iron phosphate batteries, life loss can be quantified by the number of complete charge-discharge cycles. .in, For the equivalent number of iterations, The standard deviation of the state of charge among battery clusters. This is a coefficient related to energy storage lifetime.

[0077] Minimize photovoltaic target: While meeting the requirements for preventing backflow, minimize the amount of curtailed solar power. Correspondingly, .in, This represents the amount of light wasted. Here, we use a time period of 96 as an example.

[0078] Accordingly, the NSGA-III algorithm is used to iteratively solve the optimization model to obtain the charging and discharging power of multiple battery clusters in the second time period, including: Reference points are generated for the target space, the dimension of which is equal to the number of targets in the multi-objective system. For example, if there are 3 targets, the target space is partitioned in a 3D hyperplane. Layering to generate uniform reference points. In this step, to improve the uniformity of the solution set distribution in the high-dimensional target space, the Das & Dennis layering method is used to generate reference points.

[0079] The population is initialized, with each individual representing the charging and discharging power of multiple battery clusters. Specifically, the multi-objective optimization problem of energy storage systems involves high-dimensional decision variables and strong constraints (anti-reverse current, state-of-charge equilibrium, etc.). The NSGA-III algorithm effectively solves high-dimensional multi-objective optimization problems by introducing a reference point mechanism, uniformly distributing reference points in the objective space to guide population evolution. This embodiment proposes an improved NSGA-III algorithm, achieving a dual breakthrough in solution efficiency and solution set quality through innovative constraint handling mechanisms, optimized computational architecture, and upgraded reference point generation strategies. For example, using real-number encoding, each individual represents the charging and discharging power vector of a battery cluster over 96 time periods; here, 10 clusters are used as an example.

[0080] During the initialization phase: an initial population of 100 is randomly generated, and individual variables are uniformly sampled within the rated power range. The variable value range is randomly generated according to the rated power constraint of the energy storage system. Individuals that violate the hard constraints of the upper and lower limits of the state of charge are removed through feasibility screening.

[0081] Based on the population, the following iterative steps are performed: The first step is to evaluate the fitness of each individual in the current population and obtain the fitness evaluation result for each individual. The fitness evaluation includes checking whether the individual meets the preset constraints. If there are battery clusters that do not meet the first state of charge equilibrium constraints, the charging and discharging power of at least some of the battery clusters is adjusted so that the multiple battery clusters meet the first state of charge equilibrium constraints respectively. The values ​​of multiple objective functions corresponding to the individual are calculated based on the individual.

[0082] The second step is to perform non-dominated ranking on the population after fitness evaluation based on Pareto non-dominated relations, so as to stratify the population and obtain different non-dominated levels.

[0083] Among them, the lower the level of the non-dominated hierarchy, the fewer the dominant individuals, and the higher the priority of the solution in the non-dominated hierarchy.

[0084] The third step is to normalize the values ​​of multiple objective functions corresponding to individuals in the stratified population to obtain normalization results; and based on the normalization results, calculate the vertical distance from each individual in the population to each reference point, and associate the individual with the reference point with the closest vertical distance.

[0085] In this step, the target is standardized through normalization to eliminate dimensional differences. The normalization result... as follows.

[0086]

[0087] in, This is used to prevent division by zero and is a default value. and for The minimum and maximum values.

[0088] In this step, the vertical distance from the individual to the reference point is calculated. This is correlated to the nearest reference point to ensure that solutions are covered in all target directions. This achieves target space reference point optimization. Here, z represents the reference point.

[0089] Fourth step: If the iteration termination condition is met, output the uniformly distributed Pareto optimal solution.

[0090] Step 5: If the iteration termination condition is not met, select some individuals from the population as parents according to the order of the non-dominated layer hierarchy from low to high and the order of the number of individuals associated with the reference point from small to large; select individuals from the parents for crossover and mutation to obtain offspring; based on the parents and offspring, obtain the population for the next iteration and repeat the iteration steps.

[0091] If the iteration termination condition is not met in this step, selection and evolution operations are performed to obtain the population for the next iteration.

[0092] When making selections, the non-dominated ordering and the association with reference points are considered, and individuals with low-level and sparsely associated reference points are given priority in retention.

[0093] Crossover includes: Simulated Binary Crossover (SBX), with crossover probability... Distribution index The formula for generating offspring is as follows.

[0094]

[0095]

[0096] in, and They represent two parent generations, Indicates offspring, This is an intermediate quantity.

[0097] Mutation: Polynomial mutation, mutation probability Distribution index variability From uniform distribution Decide.

[0098]

[0099] For example, the iteration is terminated when the rate of change of the hypervolume index (HV) is less than a third threshold (e.g., 1%) after the number of iterations reaches a first preset generation (e.g., 300 generations) or after a second preset generation (e.g., 50 generations), and 20 uniformly distributed Pareto optimal solutions are output.

[0100] For users experiencing peak loads and large peak-valley price differences, guided by an optimized energy storage charging and discharging plan, the energy storage system automatically selects charging during off-peak hours and discharging during peak hours based on the price difference. Simultaneously, by reducing peak loads, demand is controlled within a set value. This achieves peak-valley arbitrage and demand control, ensuring a smooth power curve and minimal random fluctuations during static charging and discharging. This demand control method based on energy storage systems offers the following advantages: by optimizing demand and scheduling strategies, electricity costs can be effectively reduced; real-time scheduling and rolling optimization ensure stable system operation and reduce risks associated with load fluctuations. This method can dynamically adjust strategies based on real-time data, adapting to different electricity environments and demands, and can be applied to users of varying sizes, from residential users to large industrial users.

[0101] In some embodiments, both the load forecasting model and the photovoltaic power generation forecasting model include a Long Short-Term Memory (LSTM) model based on an attention mechanism. The attention-based LSTM model includes at least an input gate, a cell gate, a forget gate, a state layer, an output gate, a hidden layer, an attention mechanism, and an output layer. The input gate is used to obtain the input features to be written based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The unit gate is used to obtain the candidate memory to be written based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The forget gate is used to obtain the historical memory to be retained based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The state layer is used to obtain the current state based on the input features at the current time, the candidate memory, the historical memory, and the historical state at the previous time. The output gate is used to obtain the memory to be output based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The hidden layer is used to obtain the current hidden state based on the memory to be output at the current moment and the current state; The attention mechanism is used to calculate the current hidden state based on the attention mechanism, and obtain the calculation result of the attention mechanism; The output layer is used to obtain the prediction result based on the calculation result of the attention mechanism; the input feature is obtained based on the user load data and related influencing factor data of the first time period, and the prediction result includes the user load data of the second time period; or, the input feature is obtained based on the photovoltaic power generation power of the photovoltaic power generation system in the first time period and related influencing factor data, and the prediction result includes the photovoltaic power generation power of the second time period. The attention mechanism includes: calculating the temporal correlation between the current hidden state and the hidden state at each historical moment to obtain a correlation score; determining the attention weight corresponding to the hidden state at each historical moment based on the correlation score; performing a weighted summation based on the hidden states at all historical moments and their corresponding attention weights to obtain an intermediate vector; and obtaining the calculation result of the attention mechanism based on the intermediate vector and the current hidden state.

[0102] The current time can be time t.

[0103] For example, by collecting various influencing factors related to user load data and user load data as the raw data sequence. ,for Line × A matrix of columns, where, The number of variables includes user load power and indicators of related influencing factors. The time series length for each variable, i.e., the value of the variable. indivual.

[0104] For example, based on user load data samples from the first time period and related influencing factor data samples, an attention-based LSTM model is trained on user load data samples from the second time period.

[0105] The LSTM model inherently possesses a gate mechanism: the input gate selects the current information, and the forget gate selects to forget past information. However, LSTM needs to gradually capture temporal information. As the time step increases, the memory for older data gradually decays, making it difficult to retain all historically useful information. Introducing an attention mechanism weights the hidden layer states across all time steps of the LSTM, focusing attention on the more important hidden layer state information within the historical data. The main steps of implementing the LSTM model algorithm include: After quantitative analysis of the original data sequence, the input feature terms are determined. Construct the LSTM algorithm architecture, including network node parameter settings, input gates, cell gates, forget gates, state layer, output gates, hidden layer, and output layer design.

[0106] First, network node parameter settings. Input node inputSize = K, where K is the input feature term. This includes the total number of features, for example, timestep = 60, hidden number of nodes = hiddenSize = 64, number of connection layer nodes = linearSize = 64, final output number of nodes = outSize = 1, and batch training size = batchSize = 50.

[0107] Second, the input features obtained from the input gate are denoted as: .in, Represents an input gate network, at the current time t... This indicates that at the next moment t+1... express; This represents the sigmoid activation function. This represents the input value at the current moment. Let the hidden state at the previous time step t-1 be a vector of size hiddenSize. The bias of the input gate is represented by a vector of size `hiddenSize`. This represents the weight matrix corresponding to the input values ​​in the input gate, with a size of [inputSize × hiddenSize]. This represents the weight matrix corresponding to the output state value of the input gate at the previous time step, with a size of [hiddenSize×hiddenSize].

[0108] Third, the candidate memories obtained from the unit gate are recorded as follows: .in, Represents a gate network, at the current time t. This indicates that at the next moment t+1... express. This represents the bias in the unit gate, and is a vector of size `hiddenSize`. This represents the weight matrix corresponding to the input values ​​in the unit gate, with a size of [inputSize × hiddenSize]. This represents the weight matrix corresponding to the hidden state value of the previous time step in the cell gate, with a size of [hiddenSize×hiddenSize].

[0109] Fourth, the historical memories obtained through the Gate of Oblivion are recorded as follows: .in, Represents a forget gate network, where at the current time t... This means that at the current time t+1, express. The bias of the forget gate is represented by a vector of size `hiddenSize`. This represents the weight matrix corresponding to the input values ​​in the forget gate, with a size of [inputSize × hiddenSize]. This represents the weight matrix corresponding to the hidden state value of the previous time step in the forget gate, with a size of [hiddenSize×hiddenSize]. This represents the historical memory at time t+1.

[0110] Fifth, the current state output by the state layer is denoted as: ,in, This represents the historical state at the previous time step, and the state at the next time step t+1 is... , This indicates that corresponding elements are multiplied.

[0111] Sixth, the memory to be output by the output gate is recorded as follows: .in, This represents the output gate network, at the current time t. This means that at the current time t+1, express. The bias of the output gate is represented by a vector of size `hiddenSize`. This represents the weight matrix corresponding to the input values ​​in the output gate, with a size of [inputSize × hiddenSize]. This represents the weight matrix corresponding to the hidden state value in the previous time step of the output gate, with a size of [hiddenSize×hiddenSize].

[0112] Seventh, the current hidden state output by the hidden layer is denoted as: The size is hiddenSize. The hidden state at the previous time step t-1 was... .

[0113] Eighth, the prediction result of the final output layer is denoted as: .in, This represents the output layer network, at the current time t. express. This is the output layer weight matrix, with a size of [hiddenSize × outSize]. This represents the bias of the output layer, with a size of outSize.

[0114] Ninth, the loss function is denoted as: ,in: This represents the label value of the training sample, such as a user load data sample. The loss function at time t is... The loss function at time t+1 is: .

[0115] Parameter updates are performed based on the loss function. Tenth, updating weights and biases: Output layer weight and bias update:

[0116]

[0117]

[0118]

[0119] in, For learning efficiency. for The update volume. for The update volume. Indicates to Find the derivative.

[0120] Output gate weights and biases update:

[0121]

[0122]

[0123]

[0124]

[0125]

[0126] in, for Update volume for Update volume for The update volume.

[0127] Input gate weights and biases updates:

[0128]

[0129]

[0130]

[0131]

[0132] in, for Update volume for Update volume for The update volume.

[0133] Forget gate weight and bias update:

[0134]

[0135]

[0136]

[0137]

[0138] in, for Update volume for Update volume for The update volume.

[0139] Cell gate weight and bias update:

[0140]

[0141]

[0142]

[0143]

[0144]

[0145] in, for Update volume for Update volume for The update volume. For function The derivative of .

[0146] The formula for calculating the attention mechanism is as follows:

[0147]

[0148]

[0149]

[0150] in, This represents the number of all historical moments. This represents the nth historical moment. This represents the hidden state at the nth historical moment; express and The correlation score; express The corresponding attention weights; This represents the intermediate vector. This represents the weight matrix in the attention mechanism; This represents the computational result of the attention mechanism. For example, historical moments include historical moments within the second time period. This indicates transpose.

[0151] Based on processed historical data, the load power of users in the next 24 hours is predicted. The daily load curve consists of the load power of 96 users. To maximize the use of historical data and ensure sufficient model training, an iterative rolling prediction method is adopted, using the historical data of the previous day as input and the historical data of the next day as output to train the model. Through model training, an optimal load prediction model is obtained, and the model with an accuracy of over 95% on both the training and test sets is selected as the optimal load prediction model.

[0152] A photovoltaic power generation prediction model was established using the same method as the load forecasting model. The model with an accuracy of over 95% in both the training and test sets was selected as the photovoltaic power generation prediction model.

[0153] Industrial park loads are significantly affected by production plans and weather fluctuations. General LSTM models suffer from memory decay when processing long-term historical data, leading to inaccurate predictions. An attention-based LSTM model, however, processes time-series data, and the attention mechanism dynamically adjusts the model's focus on different time steps. This improves prediction accuracy. Accurate photovoltaic power generation forecasts and user load data prediction curves enhance the reliability and feasibility of maximum demand and scheduling strategies.

[0154] This scheme first collects user photovoltaic power generation and load data, along with related influencing factors. This data collection provides a comprehensive understanding of user production patterns and influencing factors, laying the foundation for subsequent analysis and forecasting. Next, the collected data undergoes preprocessing, including noise and outlier removal, and missing data imputation. After preprocessing, these key variables are input into an attention-based LSTM model. LSTM models excel at handling time-series data and can capture temporal dependencies. The attention mechanism dynamically adjusts the model's focus on different time steps, improving prediction accuracy. This model outputs photovoltaic power generation and user load data curves for a future period. After obtaining the photovoltaic power generation and user load prediction results, a multi-objective optimization model is developed. The NSGA-III algorithm is used to solve the Pareto optimal solution for the multi-objective optimization problem. Based on the Pareto optimal solution, a suitable charging and discharging strategy for the energy storage system is selected to reduce peak demand, decrease demand costs, and extend the system's lifespan. According to this plan, industrial parks can dynamically adjust the operating status of energy storage systems based on real-time load data and photovoltaic power generation data, achieving functions such as demand control, peak-valley arbitrage, and real-time scheduling. The advantage of daily rolling optimization is that it can adjust strategies in real time based on the latest data and information to ensure optimal system operation.

[0155] This solution enables cluster-level optimization and multi-objective demand control for photovoltaic and energy storage systems in industrial parks. It is suitable for industrial users (especially manufacturing and heavy industries) with photovoltaic power generation and energy storage systems, and can achieve synergistic optimization of reduced demand costs, extended energy storage lifespan, and improved photovoltaic absorption rate.

[0156] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0157] Another embodiment of this application relates to a photovoltaic-storage system control device. The photovoltaic-storage system control device described below can be referred to in correspondence with the photovoltaic-storage system control method described above. The implementation details of the photovoltaic-storage system control device of this embodiment are described in detail below. The following content is only for ease of understanding and is not necessary for implementing this solution. The schematic diagram of the photovoltaic-storage system control device of this embodiment can be seen as follows: Figure 2 As shown.

[0158] This embodiment provides a control device for a photovoltaic-energy storage system. The photovoltaic-energy storage system includes a photovoltaic power generation system and an energy storage system with multiple battery clusters. See [link to relevant documentation]. Figure 2 The device includes: The load forecasting module 201 is used to input the user load data and related influencing factor data of the first period into the load forecasting model to obtain the user load data of the second period in the future. The photovoltaic prediction module 202 is used to input the photovoltaic power generation power and related influencing factor data of the photovoltaic power generation system in the first time period into the photovoltaic power generation power prediction model to obtain the photovoltaic power generation power in the second time period; The model building module 203 is used to build an optimization model based on preset constraints and user load data and photovoltaic power generation in the second time period, with the charging and discharging power of the multiple battery clusters in the second time period as variables; the preset constraints include a first state of charge balance constraint, which includes the absolute value of the first deviation of the battery cluster being less than or equal to a first threshold, and the first deviation being the deviation between the state of charge of the battery cluster and the average state of charge of the multiple battery clusters. The model solving module 204 is used to iteratively solve the optimization model to obtain the charging and discharging power of the plurality of battery clusters in the second time period. In each iteration, if there are battery clusters that do not meet the first state of charge balance constraint, the charging and discharging power of at least some of the battery clusters is adjusted.

[0159] In some embodiments, the model solving module is specifically used for: The absolute values ​​of the first deviations of the plurality of battery clusters are sorted. Based on the sorting result, the priority order of the plurality of battery clusters is determined, wherein the larger the absolute value of the first deviation, the higher the priority of the battery cluster; Select at least a portion of the battery clusters from the plurality of battery clusters in descending order of priority; The charging and discharging power of at least a portion of the selected battery clusters is adjusted so that the plurality of battery clusters respectively satisfy the first state-of-charge balance constraint.

[0160] In some embodiments, the adjustment amount of the charging and discharging power of at least a portion of the battery clusters for:

[0161] in, This indicates the state of charge of the k-th battery cluster at time t. This represents the average state of charge of the multiple battery clusters at time t; This is the preset adjustment coefficient; The rated capacity of a single battery cluster; This refers to the charging and discharging time between two adjacent moments.

[0162] In some embodiments, the optimization objectives of the optimization model include multiple objectives, including the lowest total electricity cost over a preset period, the longest energy storage life, and the lowest amount of curtailed solar power. The model solving module is specifically used for: The charging and discharging power of the multiple battery clusters in the second time period is obtained by iteratively solving the optimization model using the NSGA-III algorithm.

[0163] In some embodiments, the optimization model includes multiple objective functions corresponding to the multi-objective; the model solving module is specifically used for: Generate reference points for the target space, the dimension of which is equal to the number of targets in the multi-target system; initialize a population, where each individual in the population includes the charging and discharging power of the multiple battery clusters; based on the population, perform the following iterative steps: Fitness evaluation is performed on each individual in the current population to obtain the fitness evaluation result for each individual; the fitness evaluation includes checking whether the individual meets the preset constraints; if there are battery clusters that do not meet the first state-of-charge equilibrium constraint, the charging and discharging power of at least some of the battery clusters is adjusted so that the battery clusters respectively meet the first state-of-charge equilibrium constraint; the values ​​of the multiple objective functions corresponding to the individual are calculated based on the individual; Based on the Pareto non-dominated relation, the population after fitness evaluation is sorted by non-dominated order to stratify the population and obtain different non-dominated levels. The values ​​of the multiple objective functions corresponding to the individuals in the stratified population are normalized to obtain normalization results; and based on the normalization results, the vertical distance from the individuals in the population to each of the reference points is calculated, and the individuals are associated with the reference points with the closest vertical distance. If the iteration termination condition is met, output the uniformly distributed Pareto optimal solution; If the iteration termination condition is not met, select some individuals from the population as parents according to the order of the non-dominated layer hierarchy from low to high and the order of the number of individuals associated with the reference point from small to large; select individuals from the parents for crossover and mutation to obtain offspring; based on the parents and the offspring, obtain the population for the next iteration, and re-execute the iteration steps.

[0164] In some embodiments, the objective function include:

[0165]

[0166] in, The possible values ​​are 1, 2, and 3. This indicates the total electricity cost. This represents the objective function corresponding to the minimum total electricity cost. Indicates energy storage lifespan loss. This represents the objective function corresponding to the longest energy storage lifetime. Indicates the amount of light discarded. This represents the objective function corresponding to minimizing light wastage. As a penalty item, Included by preset constraints The first constraint The amount of a constraint violation, Represents an individual. Indicates the dynamic penalty coefficient. Indicates the number of iterations. This represents the base penalty coefficient.

[0167] In some embodiments, the preset constraints further include power balance constraints, peak-valley constraints, photovoltaic power generation constraints, battery cluster state of charge constraints, battery cluster charge-discharge state constraints, battery cluster power constraints, battery cluster state of charge continuity constraints, and a second state of charge equilibrium constraint; the second state of charge equilibrium constraint includes the absolute value of a second deviation of the plurality of battery clusters being less than or equal to a second threshold; wherein, the second deviation is... and deviation, This indicates that at time t, the first of the multiple battery clusters... The state of charge of the aforementioned battery clusters, This represents the state of charge of the j-th battery cluster among the plurality of battery clusters at time t.

[0168] In some embodiments, both the load forecasting model and the photovoltaic power generation forecasting model include an LSTM model based on an attention mechanism; The attention-based LSTM model includes at least an input gate, a cell gate, a forget gate, a state layer, an output gate, a hidden layer, an attention mechanism, and an output layer. The input gate is used to obtain the input features to be written based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The unit gate is used to obtain the candidate memory to be written based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The forget gate is used to obtain the historical memory to be retained based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The state layer is used to obtain the current state based on the input features at the current time, the candidate memory, the historical memory, and the historical state at the previous time. The output gate is used to obtain the memory to be output based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The hidden layer is used to obtain the current hidden state based on the memory to be output at the current moment and the current state; The attention mechanism is used to calculate the current hidden state based on the attention mechanism, and obtain the calculation result of the attention mechanism; The output layer is used to obtain the prediction result based on the calculation result of the attention mechanism; the input feature is obtained based on the user load data and related influencing factor data of the first time period, and the prediction result includes the user load data of the second time period; or, the input feature is obtained based on the photovoltaic power generation power of the photovoltaic power generation system in the first time period and related influencing factor data, and the prediction result includes the photovoltaic power generation power of the second time period. The attention mechanism includes: calculating the temporal correlation between the current hidden state and the hidden state at each historical moment to obtain a correlation score; determining the attention weight corresponding to the hidden state at each historical moment based on the correlation score; performing a weighted summation based on the hidden states at all historical moments and their corresponding attention weights to obtain an intermediate vector; and obtaining the calculation result of the attention mechanism based on the intermediate vector and the current hidden state.

[0169] This application also relates to a photovoltaic energy storage system, including: Photovoltaic power generation system; An energy storage system comprising multiple battery clusters; A control unit is used to execute the photovoltaic storage system control method as described in the above embodiments.

[0170] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.

Claims

1. A control method for a photovoltaic-energy storage system, characterized in that, The photovoltaic-storage system includes a photovoltaic power generation system and an energy storage system with multiple battery clusters, and the method includes: Input the user load data and related influencing factor data of the first period into the load forecasting model to obtain the user load data of the second period in the future; The photovoltaic power generation power and related influencing factor data of the photovoltaic power generation system in the first time period are input into the photovoltaic power generation power prediction model to obtain the photovoltaic power generation power in the second time period; Based on preset constraints and user load data and photovoltaic power generation in the second time period, an optimization model is constructed with the charging and discharging power of the multiple battery clusters in the second time period as variables. The preset constraints include a first state of charge balance constraint, which includes the absolute value of the first deviation of the battery cluster being less than or equal to a first threshold. The first deviation is the deviation between the state of charge of the battery cluster and the average state of charge of the multiple battery clusters. The charging and discharging power of the plurality of battery clusters in the second time period is obtained by iteratively solving the optimization model. In each iteration, if there is a battery cluster that does not meet the first state of charge balance constraint, the charging and discharging power of at least some of the battery clusters is adjusted.

2. The photovoltaic-storage system control method according to claim 1, characterized in that, Adjusting the charging and discharging power of at least a portion of the plurality of battery clusters includes: The absolute values ​​of the first deviations of the plurality of battery clusters are sorted. Based on the sorting result, the priority order of the plurality of battery clusters is determined, wherein the larger the absolute value of the first deviation, the higher the priority of the battery cluster; Select at least a portion of the battery clusters from the plurality of battery clusters in descending order of priority; The charging and discharging power of at least a portion of the selected battery clusters is adjusted so that the plurality of battery clusters respectively satisfy the first state-of-charge balance constraint.

3. The control method for a photovoltaic energy storage system according to claim 2, characterized in that, Adjustment amount of charging and discharging power of at least a portion of the battery clusters for: in, This indicates the state of charge of the k-th battery cluster at time t. This represents the average state of charge of the multiple battery clusters at time t; This is the preset adjustment coefficient; The rated capacity of a single battery cluster; This refers to the charging and discharging time between two adjacent moments.

4. The control method for a photovoltaic energy storage system according to any one of claims 1 to 3, characterized in that, The optimization objectives of the optimization model include multiple objectives, including the lowest total electricity cost over a preset period, the longest energy storage life, and the minimum amount of curtailed solar power. The iterative solution of the optimization model to obtain the charging and discharging power of the multiple battery clusters in the second time period includes: The charging and discharging power of the multiple battery clusters in the second time period is obtained by iteratively solving the optimization model using the NSGA-III algorithm.

5. The photovoltaic-storage system control method according to claim 4, characterized in that, The optimization model includes multiple objective functions corresponding to the multi-objective; the step of using the NSGA-III algorithm to iteratively solve the optimization model to obtain the charging and discharging power of the multiple battery clusters in the second time period includes: Generate reference points for the target space, the dimension of which is equal to the number of targets in the multi-target system; initialize a population, where each individual in the population includes the charging and discharging power of the multiple battery clusters; based on the population, perform the following iterative steps: Fitness evaluation is performed on each individual in the current population to obtain the fitness evaluation result for each individual; the fitness evaluation includes checking whether the individual meets the preset constraints; if there are battery clusters that do not meet the first state-of-charge equilibrium constraint, the charging and discharging power of at least some of the battery clusters is adjusted so that the battery clusters respectively meet the first state-of-charge equilibrium constraint; the values ​​of the multiple objective functions corresponding to the individual are calculated based on the individual; Based on the Pareto non-dominated relation, the population after fitness evaluation is sorted by non-dominated order to stratify the population and obtain different non-dominated levels. The values ​​of the multiple objective functions corresponding to the individuals in the stratified population are normalized to obtain normalization results; and based on the normalization results, the vertical distance from the individuals in the population to each of the reference points is calculated, and the individuals are associated with the reference points with the closest vertical distance. If the iteration termination condition is met, output the uniformly distributed Pareto optimal solution; If the iteration termination condition is not met, select some individuals from the population as parents according to the order of the non-dominated layer hierarchy from low to high and the order of the number of individuals associated with the reference point from small to large; select individuals from the parents for crossover and mutation to obtain offspring; based on the parents and the offspring, obtain the population for the next iteration, and re-execute the iteration steps.

6. The control method for a photovoltaic energy storage system according to claim 5, characterized in that, The objective function include: in, The possible values ​​are 1, 2, and 3. This indicates the total electricity cost. This represents the objective function corresponding to minimizing total electricity costs. Indicates energy storage lifespan loss. This represents the objective function corresponding to the longest energy storage lifetime. Indicates the amount of light discarded. This represents the objective function that minimizes light wastage. As a penalty item, Included by preset constraints The first constraint The amount of a constraint violation, Represents an individual. Indicates the dynamic penalty coefficient. Indicates the number of iterations. This represents the base penalty coefficient.

7. The control method for a photovoltaic energy storage system according to claim 1, characterized in that, The preset constraints also include power balance constraints, peak-valley constraints, photovoltaic power generation constraints, battery cluster state of charge constraints, battery cluster charge-discharge state constraints, battery cluster power constraints, battery cluster state of charge continuity constraints, and second state of charge balance constraints; the second state of charge balance constraint includes the absolute value of the second deviation of the plurality of battery clusters being less than or equal to a second threshold; wherein, the second deviation is... and deviation, This indicates that at time t, the first of the multiple battery clusters... The state of charge of the aforementioned battery clusters, This represents the state of charge of the j-th battery cluster among the plurality of battery clusters at time t.

8. The control method for a photovoltaic energy storage system according to claim 1, characterized in that, Both the load forecasting model and the photovoltaic power generation forecasting model include an LSTM model based on an attention mechanism; The attention-based LSTM model includes at least an input gate, a cell gate, a forget gate, a state layer, an output gate, a hidden layer, an attention mechanism, and an output layer. The input gate is used to obtain the input features to be written based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The unit gate is used to obtain the candidate memory to be written based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The forget gate is used to obtain the historical memory to be retained based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The state layer is used to obtain the current state based on the input features at the current time, the candidate memory, the historical memory, and the historical state at the previous time. The output gate is used to obtain the memory to be output based on the input feature terms and the hidden state output by the hidden layer at the previous time step. The hidden layer is used to obtain the current hidden state based on the memory to be output at the current moment and the current state; The attention mechanism is used to calculate the current hidden state based on the attention mechanism, and obtain the calculation result of the attention mechanism; The output layer is used to obtain the prediction result based on the calculation result of the attention mechanism; the input feature is obtained based on the user load data and related influencing factor data of the first time period, and the prediction result includes the user load data of the second time period; or, the input feature is obtained based on the photovoltaic power generation power of the photovoltaic power generation system in the first time period and related influencing factor data, and the prediction result includes the photovoltaic power generation power of the second time period. The attention mechanism includes: calculating the temporal correlation between the current hidden state and the hidden state at each historical moment to obtain a correlation score; determining the attention weight corresponding to the hidden state at each historical moment based on the correlation score; performing a weighted summation based on the hidden states at all historical moments and their corresponding attention weights to obtain an intermediate vector; and obtaining the calculation result of the attention mechanism based on the intermediate vector and the current hidden state.

9. A control device for a photovoltaic energy storage system, characterized in that, A photovoltaic-storage system includes a photovoltaic power generation system and an energy storage system with multiple battery clusters. The device includes: The load forecasting module is used to input the user load data and related influencing factor data of the first period into the load forecasting model to obtain the user load data of the second period in the future. The photovoltaic prediction module is used to input the photovoltaic power generation power and related influencing factor data of the photovoltaic power generation system in the first time period into the photovoltaic power generation power prediction model to obtain the photovoltaic power generation power in the second time period. The model building module is used to build an optimization model based on preset constraints and user load data and photovoltaic power generation in the second time period, with the charging and discharging power of the multiple battery clusters in the second time period as variables. The preset constraints include a first state of charge balance constraint, which includes the absolute value of the first deviation of the battery cluster being less than or equal to a first threshold. The first deviation is the deviation between the state of charge of the battery cluster and the average state of charge of the multiple battery clusters. The model solving module is used to iteratively solve the optimization model to obtain the charging and discharging power of the plurality of battery clusters in the second time period. In each iteration, if there are battery clusters that do not meet the first state of charge balance constraint, the charging and discharging power of at least some of the battery clusters is adjusted.

10. A photovoltaic energy storage system, characterized in that, include: Photovoltaic power generation system; An energy storage system comprising multiple battery clusters; A control unit, the control unit being configured to perform the photovoltaic storage system control method as described in any one of claims 1 to 8.

Images