Adaptive energy scheduling method for off-grid energy storage system based on wind-solar complementation

Abstract

The application discloses a self-adaptive energy scheduling method of off-grid energy storage system based on wind-solar complementation, relates to the technical field of energy storage management, and has the following scheduling steps: S1, collecting data, including future weather data, energy storage state data, load demand and environmental parameters, and uploading the data after preprocessing; S2, predicting short-term wind and light energy based on the future weather data to generate a prediction interval; S3, constructing a multi-objective optimization function based on the obtained data, setting a constraint condition, solving the multi-objective optimization function by using an improved particle swarm optimization algorithm, and outputting an energy scheduling strategy.The application integrates future weather data, energy storage state data, load demand and environmental parameters in the data collection stage; in the prediction stage, a prediction interval of wind-solar energy is generated based on multi-source data, rather than a single prediction value, the prediction uncertainty can be quantified, and the abandoned wind and light or power shortage caused by inaccurate prediction can be effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of energy storage management technology, specifically to an adaptive energy dispatching method for off-grid energy storage systems based on wind-solar hybridization. Background Technology

[0002] With the rapid development of renewable energy technologies, wind and solar energy, due to their clean and renewable characteristics, have become the core sources of energy supply in off-grid areas (such as remote mountainous areas, islands, and field operation sites). Off-grid wind-solar hybrid energy storage systems integrate wind power, photovoltaic power generation devices, and energy storage equipment, and can independently meet local load demands without the main power grid. They have irreplaceable application value in scenarios with no grid coverage or low grid reliability, and have been widely used in fields such as power supply for communication base stations, electricity for remote residential areas, and temporary power supply for field projects. However, existing energy dispatch methods for off-grid wind-solar hybrid energy storage systems still have significant technical shortcomings. They are difficult to adapt to the strong volatility of wind and solar energy, the randomness of load demand, and the dynamic characteristics of energy storage systems, resulting in problems such as low system operating efficiency, poor power supply reliability, and shortened energy storage life. Specifically, these problems are manifested in the following aspects: Existing scheduling methods mostly rely on historical statistical data or simple empirical models to predict wind and solar power output, failing to fully integrate future weather data (such as short-term forecasts of wind speed, solar intensity, and temperature) with real-time environmental parameters (such as the impact of on-site humidity and air pressure on photovoltaic efficiency). Furthermore, they lack interval-based characterization of prediction results. Wind and solar energy are significantly affected by weather, and single-point predictions are prone to large deviations, leading to a mismatch between subsequent scheduling strategies and actual power output, which in turn causes wind and solar curtailment or load shortages. Traditional algorithms suffer from fixed inertial weights and a single learning factor, making them prone to getting trapped in local optima. Traditional algorithms also have slow convergence speeds and cannot output feasible scheduling strategies in a timely manner, affecting the real-time operation of the system. To address these issues, we propose an adaptive energy scheduling method for off-grid energy storage systems based on wind-solar complementarity. Summary of the Invention

[0003] To address the aforementioned technical issues, this technical solution provides an adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization, thus resolving the problems mentioned above.

[0004] To achieve the above objectives, the technical solution adopted in this invention is: an adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization, wherein the dispatch steps are as follows: S1. Collect data, including future weather data, energy storage status data, load demand and environmental parameters, and upload the data after preprocessing; S2. Based on future weather data, predict short-term wind and solar energy, and generate prediction intervals; S3. Based on the acquired data, construct a multi-objective optimization function, set constraints, and use an improved particle swarm optimization algorithm to solve the multi-objective optimization function. Output the energy dispatch strategy, including the wind and solar power generation allocation ratio, energy storage status data, and power supply start-up and shutdown times. S4. Real-time acquisition of the deviation between the actual and predicted values ​​of the system's energy storage status data. Based on the magnitude of the deviation, an adaptive correction mechanism is adopted. The optimization function and correction mechanism are updated based on the reinforcement learning algorithm to improve and correct the energy scheduling strategy. S5. The finalized scheduling strategy is then sent to the energy storage system for control and scheduling.

[0005] Preferably, in step S2, the prediction is made by constructing a training sample set from the preprocessed data, extracting wind speed statistics and illumination parameters from the weather data, performing dual-model fusion prediction, combining the physical model and the LSTM data-driven model, dynamically weighting according to weather stability to obtain the fused prediction value, and generating the prediction interval.

[0006] The LSTM data-driven model structure includes: The input layer contains six preprocessed feature parameters: historical wind speed, historical light intensity, ambient temperature, air density, photovoltaic panel temperature, and wind turbine operating status. The hidden layer consists of 3 LSTM units, with 64, 32, and 16 units per layer, respectively. The activation function is tanh. Dropout layers are added between each layer to prevent overfitting. The output layer is a fully connected layer that outputs the predicted wind power generation and photovoltaic power generation for the next 1-24 hours. The activation function is Linear. The training parameters used the Adam optimizer, with the initial learning rate set to 0.001 and adjusted using a learning rate decay strategy. The loss function used was the root mean square error (RMSE), and the number of training iterations was set to 500 epochs. Training was stopped when the loss on the validation set did not decrease for 20 consecutive epochs.

[0007] Preferably, the physical model output calculation includes the theoretical output of the wind turbine and the theoretical output of the photovoltaic system, wherein the formula for calculating the theoretical output of the wind turbine is:

[0008] in air density, The swept area of ​​the fan. For wind speed, The wind energy utilization coefficient ranges from 0.4 to 0.593. This is the theoretical output value; The theoretical output calculation formula for photovoltaic power is:

[0009] in Light intensity, For the area of ​​the photovoltaic panel, For standard efficiency, For temperature coefficient, For the temperature of the solar panel, The theoretical output value of photovoltaic power; The fusion prediction value is calculated using the following formula:

[0010] Among them, weight The dynamic adjustment method is as follows: Stable weather, , The standard deviation of wind speed; Complex weather, , The standard deviation of light intensity; The formula for calculating the prediction interval is: Maximum: ; Lower limit: ; in , , These represent the standard deviations of the model's historical prediction errors.

[0011] The weather stability judgment criteria in this application define stable weather as wind speed standard deviation ≤ 1.5 m / s and light intensity standard deviation ≤ 50 W / m² for 6 consecutive hours; complex weather is defined as wind speed standard deviation > 1.5 m / s or light intensity standard deviation > 50 W / m² for 6 consecutive hours. The prediction interval parameters are determined, and the standard deviations of the model's historical prediction errors σ1 and σ2 are calculated using the predicted and actual values ​​over the past 30 days. These are updated weekly to ensure the accuracy of the prediction interval.

[0012] Preferably, the step of constructing the multi-objective optimization function in step S3 is based on the actual application scenario, and the objectives to be balanced are determined, including economic objectives, environmental objectives, reliability objectives and energy storage life objectives; after determination, decision variables are defined, including wind and solar power generation allocation variables, energy storage system operation variables, power supply start-up and shutdown variables and grid interaction variables; the objectives are converted into mathematical expressions, and finally, the constraints are associated to obtain the multi-objective optimization function. Preferably, the step S3, which uses an improved particle swarm optimization algorithm to solve the multi-objective optimization function, is as follows: The ion swarm and algorithm parameters are initialized. For each particle, the particle position is substituted into the multi-objective optimization function to calculate the economic, environmental and reliability function values. Compare the fitness of the particle's current position with the historical individual optimal solution. If the current position is a dominant solution, update the individual optimal solution to the current position. If the two are not dominant, retain one of them. If the current position is a dominated solution, maintain the original individual optimal solution. Add the individual optimal solutions of all particles to a temporary set and merge it with the existing external archive. The particle velocity is adjusted based on adaptive inertia weight, hierarchical learning factor, individual optimal solution and global optimal solution, the particle position is adjusted based on the updated velocity, and the constraints are checked again after the update. If the number of iterations reaches the preset maximum value and there is no improvement after 10-20 consecutive rounds with the non-dominated solution in the external archive, then the iteration stops. From the externally archived set of non-dominated solutions, the optimal solution is selected, and the decision variables corresponding to the solution are transformed into specific scheduling strategies, including the wind and solar power generation allocation ratio at each time, the energy storage charging and discharging status and remaining capacity, and the start and stop times of interruptible power sources, thus forming an executable energy scheduling scheme.

[0013] Preferably, the parameters include weight values, and the weight value adjustment formula is:

[0014] in, For the first The inertia weight in each iteration balances global search and local optimization capabilities. and These are the maximum and minimum inertia weights, with values ​​ranging from 0.9 to 0.4. This represents the maximum number of iterations. The formula for adjusting particle velocity is:

[0015] in For particle velocity, For the particle position, This is the optimal solution for the individual. This is the globally optimal solution. , As a learning factor, It is a random number in the range [0, 1].

[0016] Preferably, the particle position update formula is:

[0017] in For the particle position, This represents the number of iterations. Filter non-dominated solutions from external archives, for the first Given several objective functions, the crowding density of the intermediate solutions after sorting is:

[0018] in This represents the crowding degree of the j-th non-dominated solution under the m-th objective function after sorting. Let m be the objective function value of the (j+1)th non-dominated solution under the m-th objective function. Let be the objective function value of the (j-1)th non-dominated solution under the m-th objective function; The m-th objective function is the maximum value among all non-dominated solutions. Let m be the minimum value of the m-th objective function among all non-dominated solutions; set the congestion degree of the boundary solutions to infinity to prioritize their retention.

[0019] Preferably, the formula for calculating the deviation value in step S4 is:

[0020] Where P is the calculated deviation value, B1 is the actual energy storage state data value, and B2 is the predicted energy storage state data value; When the deviation is less than the preset threshold, the initial energy scheduling plan remains unchanged; when the deviation exceeds the preset threshold, the energy storage charging and discharging power and load priority are dynamically adjusted based on fuzzy control rules.

[0021] Preferably, the deviation information is fuzzified and transformed into fuzzy linguistic variables. Based on the preset fuzzy control rules, the corresponding control strategy is matched. By defuzzifying, the fuzzy control quantity is transformed into a specific execution value, and the adjusted energy storage charging and discharging power and load priority ranking are output.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention integrates future weather data, energy storage status data, load demand, and environmental parameters during the data acquisition phase. During the prediction phase, it generates prediction ranges for wind and solar energy based on multi-source data, rather than a single prediction value. This quantifies prediction uncertainty and effectively reduces wind and solar curtailment or power shortages caused by inaccurate predictions. By adaptively adjusting inertia weights and setting learning factors in a hierarchical manner, it solves the problems of traditional algorithms easily getting trapped in local optima and having slow convergence speed. It quickly outputs feasible scheduling strategies, meets the real-time requirements of off-grid scenarios, and avoids system operational imbalances caused by strategy generation delays. Attached Figure Description

[0023] Figure 1 This is a flowchart of the scheduling steps of the present invention. Detailed Implementation

[0024] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0025] Reference Figure 1 As shown, the adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization includes the following dispatch steps: S1. Collect data, including future weather data, energy storage status data, load demand and environmental parameters, and upload the data after preprocessing; S2. Based on future weather data, predict short-term wind and solar energy, and generate prediction intervals; S3. Based on the acquired data, construct a multi-objective optimization function, set constraints, and use an improved particle swarm optimization algorithm to solve the multi-objective optimization function. Output the energy dispatch strategy, including the wind and solar power generation allocation ratio, energy storage status data, and power supply start-up and shutdown times. S4. Real-time acquisition of the deviation between the actual and predicted values ​​of the system's energy storage status data. Based on the magnitude of the deviation, an adaptive correction mechanism is adopted. The optimization function and correction mechanism are updated based on the reinforcement learning algorithm to improve and correct the energy scheduling strategy. S5. The finalized scheduling strategy is then sent to the energy storage system for control and scheduling.

[0026] This application improves upon the traditional particle swarm optimization algorithm by adaptively adjusting the inertia weight and setting the learning factor in stages. This solves the problems of traditional algorithms being prone to getting trapped in local optima and having slow convergence speed. Within a scheduling cycle of 15-30 minutes in an off-grid system, the algorithm's convergence speed is improved by 40%-60%, and feasible scheduling strategies can be output quickly to meet the real-time requirements of off-grid scenarios and avoid system imbalance caused by strategy generation delays.

[0027] Future weather data comes from real-time push notifications from regional meteorological stations and short-term forecasts from third-party meteorological platforms, including parameters such as wind speed, wind direction, light intensity, ambient temperature, and precipitation probability. Energy storage status data is collected in real time through the energy storage system BMS, including battery SOC, single cell voltage, charge and discharge current, battery temperature, and cycle count, with a collection frequency of 1 minute / time; Load demand data is collected through smart meters and sensors on the load side. It is divided into interruptible loads and non-interruptible loads according to load type. The power demand curve, minimum continuous operation time and number of allowed start-stop times are recorded for each load. The collection frequency is 5 minutes / time. In addition to weather data, environmental parameters include the air density at the installation area of ​​the wind and solar equipment, calculated using barometric pressure and temperature sensors, with the formula ρ = P / (R). T), where P is the gas pressure, R is the gas constant, T is the absolute temperature, and T is the surface cleanliness of the photovoltaic panel.

[0028] In step S2, the prediction is made by constructing a training sample set from the preprocessed data, extracting wind speed statistics and illumination parameters from the weather data, performing dual-model fusion prediction, combining the physical model and the LSTM data-driven model, dynamically weighting according to weather stability to obtain the fused prediction value, and generating the prediction interval.

[0029] Outlier handling uses the 3σ criterion to identify outlier data and replaces outliers using linear interpolation of data from adjacent time points. For missing data with short durations (≤30 minutes), the moving average method is used for imputation; for missing data with long durations (>30 minutes), the similarity matching method based on historical data of the same period under similar weather conditions is used for imputation. Data standardization transforms all data to the [0, 1] interval, eliminating the influence of units. For highly correlated parameters, principal component analysis (PCA) is used to extract principal components for data dimensionality reduction. Principal components with a cumulative contribution rate of ≥95% are retained to reduce the amount of subsequent calculations.

[0030] The physical model output calculation includes the theoretical output of the wind turbine and the theoretical output of the photovoltaic system. The formula for calculating the theoretical output of the wind turbine is as follows:

[0031] in air density, The swept area of ​​the fan. For wind speed, The wind energy utilization coefficient ranges from 0.4 to 0.593. This is the theoretical output value; The theoretical output calculation formula for photovoltaic power is:

[0032] in Light intensity, For the area of ​​the photovoltaic panel, For standard efficiency, For temperature coefficient, For the temperature of the solar panel, The theoretical output value of photovoltaic power; The fusion prediction value is calculated using the following formula:

[0033] Among them, weight The dynamic adjustment method is as follows: Stable weather, , The standard deviation of wind speed; Complex weather, , The standard deviation of light intensity; The formula for calculating the prediction interval is: Maximum: ; Lower limit: ; in , , These represent the standard deviations of the model's historical prediction errors.

[0034] The physical model in this application calculates power output based on the underlying principles of wind and solar power generation. The wind turbine formula is based on the wind energy conversion law, and the photovoltaic formula is combined with temperature correction to ensure that the prediction does not exceed the physical boundary and fits the actual operating conditions, avoiding deviations from objective laws. Dynamic weighted fusion adjusts the weights according to weather stability. Stable weather emphasizes the high stability of the physical model, while complex weather emphasizes the time-series prediction capability of LSTM, maximizing the model's advantages under different scenarios and significantly improving prediction accuracy. The prediction interval quantifies uncertainty through the standard deviation of historical errors, providing upper and lower limits for power output, providing a risk buffer for scheduling, avoiding the limitations of single-point prediction, and preventing problems such as power outages or overcharging of energy storage. The three elements work together to build a reasonable, accurate, and controllable prediction system, ensuring the reliable and economical operation of the system.

[0035] The step S3, which involves constructing a multi-objective optimization function, is based on the actual application scenario. It determines the balanced objectives, including economic, environmental, reliability, and energy storage lifespan objectives. After determination, it defines decision variables, including wind and solar power generation allocation variables, energy storage system operation variables, power supply start-up and shutdown variables, and grid interaction variables. The objectives are then converted into mathematical expressions, and finally, constraints are associated to obtain the multi-objective optimization function. This application avoids the pitfalls of optimizing a single objective by clearly defining multiple objectives such as economic efficiency and environmental protection; secondly, it defines decision variables such as wind and solar power allocation and energy storage operation to accurately map controllable scheduling elements of the system, ensuring a strong correlation between the optimization direction and actual operation; and thirdly, it transforms the objectives into mathematical expressions and associates them with constraints, making the abstract objectives concrete into calculable and optimizable quantitative indicators, providing clear boundaries for the algorithm to solve the problem.

[0036] The steps in step S3, which use an improved particle swarm optimization algorithm to solve the multi-objective optimization function, are as follows: The ion swarm and algorithm parameters are initialized. For each particle, the particle position is substituted into the multi-objective optimization function to calculate the economic, environmental and reliability function values. Compare the fitness of the particle's current position with the historical individual optimal solution. If the current position is a dominant solution, update the individual optimal solution to the current position. If the two are not dominant, retain one of them. If the current position is a dominated solution, maintain the original individual optimal solution. Add the individual optimal solutions of all particles to a temporary set and merge it with the existing external archive. The particle velocity is adjusted based on adaptive inertia weight, hierarchical learning factor, individual optimal solution and global optimal solution, the particle position is adjusted based on the updated velocity, and the constraints are checked again after the update. If the number of iterations reaches the preset maximum value and there is no improvement after 10-20 consecutive rounds with the non-dominated solution in the external archive, then the iteration stops. From the externally archived set of non-dominated solutions, the optimal solution is selected, and the decision variables corresponding to the solution are transformed into specific scheduling strategies, including the wind and solar power generation allocation ratio at each time, the energy storage charging and discharging status and remaining capacity, and the start and stop times of interruptible power sources, thus forming an executable energy scheduling scheme.

[0037] The optimal solution selection criteria use the Analytic Hierarchy Process (AHP) to determine the weights of each objective. If the application scenario is power supply to remote residents, the weights are: reliability 0.4, economy 0.3, energy storage lifespan 0.2, and environmental protection 0.1. If it is power supply to an ecological protection zone, the weights are: environmental protection 0.4, reliability 0.3, energy storage lifespan 0.2, and economy 0.1. Calculate the overall score of the non-dominated solution: S = ω1S1 + ω2S2 + ω3S3 + ω4S4, where ω1-ω4 are the weights of each objective, and S1-S4 are the standardized scores of each objective, with a maximum value of 1 and a minimum value of 0. Select the non-dominated solution with the highest overall score as the optimal solution.

[0038] The parameters include weight values, and the formula for adjusting the weight values ​​is:

[0039] in, For the first The inertia weight in each iteration balances global search and local optimization capabilities. and These are the maximum and minimum inertia weights, with values ​​ranging from 0.9 to 0.4. This represents the maximum number of iterations. The formula for adjusting particle velocity is:

[0040] in For particle velocity, For the particle position, This is the optimal solution for the individual. This is the globally optimal solution. , As a learning factor, It is a random number in the range [0, 1].

[0041] The particle position update formula is:

[0042] in For the particle position, This represents the number of iterations. Filter non-dominated solutions from external archives, for the first Given several objective functions, the crowding density of the intermediate solutions after sorting is:

[0043] in This represents the crowding degree of the j-th non-dominated solution under the m-th objective function after sorting. Let m be the objective function value of the (j+1)th non-dominated solution under the m-th objective function. Let be the objective function value of the (j-1)th non-dominated solution under the m-th objective function; The m-th objective function is the maximum value among all non-dominated solutions. Let m be the minimum value of the m-th objective function among all non-dominated solutions; set the congestion degree of the boundary solutions to infinity to prioritize their retention.

[0044] The formula for calculating the deviation value in step S4 is:

[0045] Where P is the calculated deviation value, B1 is the actual energy storage state data value, and B2 is the predicted energy storage state data value; When the deviation is less than the preset threshold, the initial energy scheduling plan remains unchanged; when the deviation exceeds the preset threshold, the energy storage charging and discharging power and load priority are dynamically adjusted based on fuzzy control rules.

[0046] This application's design, which makes decisions based on the comparison between the deviation and a preset threshold, achieves "layered control." When the deviation is less than the threshold, the original plan remains unchanged, avoiding frequent adjustments to energy storage charging / discharging or load status due to minor fluctuations, thus ensuring system operational stability. Adjustments are only initiated when the deviation exceeds the threshold, ensuring that resources are not wasted on unnecessary interventions and balancing system stability with responsiveness.

[0047] The deviation information is fuzzified and transformed into fuzzy linguistic variables. Based on the preset fuzzy control rules, the corresponding control strategy is matched. By defuzzifying, the fuzzy control quantity is transformed into a specific execution value, and the adjusted energy storage charging and discharging power and load priority ranking are output.

[0048] This application transforms deviation information into fuzzy linguistic variables, which can accurately capture the fuzziness and uncertainty of deviations in real-world scenarios. This avoids the problems of traditional precision control being overly sensitive to subtle deviations or rigid in judging fuzzy deviations, making the control strategy more aligned with the complex and ever-changing reality of energy storage and load operation. The preset fuzzy control rules are formulated based on actual operating experience and scenario requirements, and can directly match different deviation scenarios without the need for complex mathematical modeling and derivation, reducing the decision-making difficulty of dynamic adjustments. At the same time, it ensures that the strategy can quickly respond to deviation changes and reduce scheduling lag.

[0049] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. An adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization, characterized in that, The scheduling steps are as follows: S1. Collect data, including future weather data, energy storage status data, load demand and environmental parameters, and upload the data after preprocessing; S2. Based on future weather data, predict short-term wind and solar energy, and generate prediction intervals; S3. Based on the acquired data, construct a multi-objective optimization function, set constraints, and use an improved particle swarm optimization algorithm to solve the multi-objective optimization function. Output the energy dispatch strategy, including the wind and solar power generation allocation ratio, energy storage status data, and power supply start-up and shutdown times. S4. Real-time acquisition of the deviation between the actual and predicted values ​​of the system's energy storage status data. Based on the magnitude of the deviation, an adaptive correction mechanism is adopted. The optimization function and correction mechanism are updated based on the reinforcement learning algorithm to improve and correct the energy scheduling strategy. S5. The finalized scheduling strategy is then sent to the energy storage system for control and scheduling.

2. The adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization according to claim 1, characterized in that: In step S2, the prediction is made by constructing a training sample set from the preprocessed data, extracting wind speed statistics and illumination parameters from the weather data, performing dual-model fusion prediction, combining the physical model and the LSTM data-driven model, dynamically weighting according to weather stability to obtain the fused prediction value, and generating the prediction interval.

3. The adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization according to claim 2, characterized in that, The physical model output calculation includes the theoretical output of the wind turbine and the theoretical output of the photovoltaic system. The formula for calculating the theoretical output of the wind turbine is as follows: in air density, The swept area of ​​the fan. For wind speed, The wind energy utilization coefficient ranges from 0.4 to 0.

593. This is the theoretical output value; The theoretical output calculation formula for photovoltaic power is: in Light intensity, For the area of ​​the photovoltaic panel, For standard efficiency, For temperature coefficient, For the temperature of the solar panel, The theoretical output value of photovoltaic power; The fusion prediction value is calculated using the following formula: Among them, weight The dynamic adjustment method is as follows: Stable weather, , The standard deviation of wind speed; Complex weather, , The standard deviation of light intensity; The formula for calculating the prediction interval is: Maximum: ; Lower limit: ; in , , These represent the standard deviations of the model's historical prediction errors.

4. The adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization according to claim 1, characterized in that: The S3 step of constructing a multi-objective optimization function is based on the actual application scenario and determines the balanced objectives, including economic objectives, environmental objectives, reliability objectives, and energy storage lifetime objectives. Once determined, decision variables are defined, including wind and solar power generation allocation variables, energy storage system operation variables, power supply start-up and shutdown variables, and grid interaction variables; the objective is converted into a mathematical expression, and finally, constraints are associated to obtain a multi-objective optimization function.

5. The adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization according to claim 1, characterized in that, The steps in step S3, which use an improved particle swarm optimization algorithm to solve the multi-objective optimization function, are as follows: The particle swarm and algorithm parameters are initialized. For each particle, the particle position is substituted into the multi-objective optimization function to calculate the economic, environmental and reliability function values. Compare the fitness of the particle's current position with the historical individual optimal solution. If the current position is a dominant solution, update the individual optimal solution to the current position. If the two are not dominant, retain one of them. If the current position is a dominated solution, maintain the original individual optimal solution. Add the individual optimal solutions of all particles to a temporary set and merge it with the existing external archive. The particle velocity is adjusted based on adaptive inertia weight, hierarchical learning factor, individual optimal solution and global optimal solution, the particle position is adjusted based on the updated velocity, and the constraints are checked again after the update. If the number of iterations reaches the preset maximum value and there is no improvement after 10-20 consecutive rounds with the non-dominated solution in the external archive, then the iteration stops. From the externally archived set of non-dominated solutions, the optimal solution is selected, and the decision variables corresponding to the solution are transformed into specific scheduling strategies, including the wind and solar power generation allocation ratio at each time, the energy storage charging and discharging status and remaining capacity, and the start and stop times of interruptible power sources, thus forming an executable energy scheduling scheme.

6. The adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization according to claim 5, characterized in that: The parameters include weight values, and the formula for adjusting the weight values ​​is: in, For the first The inertia weight in each iteration balances global search and local optimization capabilities. and These are the maximum and minimum inertia weights, with values ​​ranging from 0.9 to 0.

4. This represents the maximum number of iterations. The formula for adjusting particle velocity is: in For particle velocity, For the particle position, This is the optimal solution for the individual. This is the globally optimal solution. , As a learning factor, It is a random number in the range [0, 1].

7. The adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization according to claim 5, characterized in that, The particle position update formula is: in For the particle position, This represents the number of iterations. Filter non-dominated solutions from external archives, for the first Given several objective functions, the crowding density of the intermediate solutions after sorting is: in This represents the crowding degree of the j-th non-dominated solution under the m-th objective function after sorting. Let m be the objective function value of the (j+1)th non-dominated solution under the m-th objective function. Let be the objective function value of the (j-1)th non-dominated solution under the m-th objective function; The m-th objective function is the maximum value among all non-dominated solutions. Let m be the minimum value of the m-th objective function among all non-dominated solutions; set the congestion degree of the boundary solutions to infinity to prioritize their retention.

8. The adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization according to claim 1, characterized in that, The formula for calculating the deviation value in step S4 is: Where P is the calculated deviation value, B1 is the actual energy storage state data value, and B2 is the predicted energy storage state data value; When the deviation is less than the preset threshold, the initial energy scheduling plan remains unchanged; When the deviation exceeds the preset threshold, the energy storage charging and discharging power and load priority are dynamically adjusted based on fuzzy control rules.

9. The adaptive energy dispatch method for off-grid energy storage systems based on wind-solar hybridization according to claim 8, characterized in that: The deviation information is fuzzified and transformed into fuzzy linguistic variables. Based on the preset fuzzy control rules, the corresponding control strategy is matched. By defuzzifying, the fuzzy control quantity is transformed into a specific execution value, and the adjusted energy storage charging and discharging power and load priority ranking are output.

Images