Wind and light energy storage calculation and prediction cooperative control method and system

Abstract

The invention provides a wind and light energy storage calculation and prediction cooperative control method and system, and relates to the technical field of energy storage, and the method comprises the steps: obtaining the real-time operation data of a wind power plant, a photovoltaic power station and an energy storage power station, constructing a multi-head attention structure, and carrying out the feature extraction to obtain a power prediction value; constructing a generated power prediction curve and a confidence interval thereof based on the prediction value; an energy storage charging and discharging strategy is determined by adopting a hierarchical control structure, upper-layer control calculates a power regulation target value, lower-layer control generates a charging and discharging instruction, and the strategy is optimized by applying a time sequence difference method. According to the invention, accurate prediction and optimal regulation and control of the wind and light energy storage system can be realized, and the power generation smoothness and economy are improved.

Description

Technical Field

[0001] This invention relates to the field of energy storage technology, and in particular to a method and system for the coordinated control of wind and solar energy storage calculation and prediction. Background Technology

[0002] As the penetration rate of renewable energy in the power system continues to increase, wind and solar power generation, due to their intermittent, volatile, and random characteristics, pose significant challenges to the safe and stable operation of the power system. Energy storage technology, as a key technology capable of smoothing out fluctuations in renewable energy output and improving system stability, plays a crucial role in hybrid wind-solar-storage systems. By effectively coordinating the operation of wind power, solar power, and energy storage devices, hybrid wind-solar-storage systems can not only improve the utilization efficiency of renewable energy but also enhance the system's power supply reliability and economy.

[0003] Most existing energy storage control strategies lack a hierarchical collaborative control structure, failing to simultaneously address the system's short-term smooth control needs and long-term economic operation objectives. This makes it difficult to balance the contradiction between power generation smoothness and economic efficiency during energy storage resource scheduling. Traditional control methods often neglect the lifespan degradation and charging / discharging efficiency constraints of energy storage devices, leading to overuse or inefficiency of these devices, which is detrimental to the long-term stable operation of the energy storage system. Existing wind-solar-storage collaborative control methods generally lack adaptive optimization capabilities, failing to dynamically adjust control strategies based on system operating conditions and forecast uncertainties. When faced with forecast deviations and system fluctuations, real-time optimization control is difficult, impacting the overall system performance and economic benefits. Summary of the Invention

[0004] The embodiments of the present invention provide a method and system for coordinated control of wind, solar and energy storage calculation and prediction, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a method for coordinated control of wind-solar-storage energy calculation and prediction, comprising: The real-time operation data of the wind-solar-energy storage system is obtained. The real-time operation data includes the first power generation data of the wind farm, the second power generation data of the photovoltaic power station, and the real-time charging and discharging data of the energy storage power station. A multi-head attention structure is constructed to extract features from the real-time operation data of the wind-solar-energy storage system to obtain the power prediction value of the wind-solar-energy storage system. Based on the power prediction values ​​of the wind-solar-storage energy system, a power generation prediction curve for the wind-solar-storage energy system is constructed, and the upper and lower confidence limits of the power generation prediction curve are calculated based on historical power generation data; Based on the power generation prediction curve and its confidence upper and lower limits, a hierarchical control structure is constructed to determine the charging and discharging strategy of the energy storage power station. The upper-level control calculates the target value of energy storage power adjustment through the system power generation smoothness evaluation function and the economic evaluation function. The lower-level control generates real-time charging and discharging control commands by combining the energy storage equipment lifespan constraint function and the charging and discharging efficiency constraint function, and uses the time-series differential calculation method to dynamically optimize the charging and discharging strategy. The real-time charging and discharging control commands are sent to the energy management system of the energy storage power station for execution, thereby adjusting the charging and discharging power of the energy storage power station in real time.

[0006] A multi-head attention structure is constructed to extract features from the real-time operating data of the wind-solar-storage system, resulting in the following power prediction values ​​for the system: Calculate the correlation score between the real-time operation data and historical operation data of the wind-solar-storage system, and input the correlation score into the Softmax function to obtain the attention weight; The real-time running data is input into multiple parallel multi-head attention structures, each of which contains an independent query matrix, key matrix, and value matrix. For each multi-head attention structure, the real-time running data is multiplied by the query matrix to obtain the query vector, the historical running data is multiplied by the key matrix to obtain the key vector, the weighted features are calculated based on the dot product of the query vector and the key vector and combined with the attention weights, and the weighted features are multiplied by the value matrix to obtain the feature output. Multiple feature outputs are concatenated, and the power prediction value of the wind-solar energy storage system is obtained through linear transformation and nonlinear activation function.

[0007] Based on the power prediction values ​​of the wind-solar-storage energy system, a power generation prediction curve for the wind-solar-storage energy system is constructed, and the upper and lower confidence limits of the power generation prediction curve are calculated based on historical power generation data, including: The power prediction values ​​of the wind-solar-storage energy system are arranged in a time series, and the discrete prediction values ​​are processed into a continuous value using a cubic spline interpolation method to generate a smooth power generation prediction curve. Based on the power generation prediction curve, a kernel density estimation function is constructed to fit the probability density distribution of historical power generation data. The optimal bandwidth parameter of the kernel function is determined by an adaptive bandwidth selection algorithm to obtain the probability distribution characteristics of power generation. Based on the probability distribution characteristics, the conditional quantiles at different confidence levels are calculated using the quantile regression method. The upper quantile corresponding to the 95% confidence level is taken as the upper confidence limit, and the lower quantile corresponding to the 5% confidence level is taken as the lower confidence limit. The upper and lower confidence limits are smoothed to generate continuous confidence interval boundary curves, which are used to characterize the uncertainty range of the power generation prediction value.

[0008] The upper-level control calculates the target value for energy storage power regulation through the system power generation smoothness evaluation function and the economic evaluation function, including: Calculate the difference between the power generation of the wind-solar-storage system at the current moment and the power generation at the previous moment to obtain the power generation change; divide the power generation change by the time interval to obtain the power generation volatility; construct a smoothness evaluation function based on the power generation volatility. Obtain the unit price of charging and discharging electricity for the energy storage power station under peak and off-peak electricity prices, and calculate the electricity cost based on the charging and discharging power; calculate the equipment loss cost based on the charging and discharging depth and cycle number of the energy storage equipment; add the electricity cost and the equipment loss cost to obtain the total cost, and construct an economic evaluation function based on the total cost; The smoothness evaluation function and the economic evaluation function are weighted and combined using weighting coefficients to obtain the objective function. The objective function is then solved using a dynamic programming algorithm to output the target value for energy storage power regulation.

[0009] The lower-level control, combining the energy storage device's lifetime constraint function and charge / discharge efficiency constraint function, generates real-time charge / discharge control commands, including: Based on the energy storage power adjustment target value, the cycle life loss of the energy storage device under different charge and discharge depths is calculated, and a life loss constraint function is constructed; an efficiency loss constraint function is constructed according to the correspondence between charge and discharge current and battery temperature. Using lifetime loss constraint function, efficiency loss constraint function and energy storage capacity limit as constraints, a quadratic programming problem is constructed with minimizing power tracking error as the optimization objective. The optimal charging and discharging power that satisfies the constraints is calculated by a convex optimization solver. Real-time charging and discharging control commands for the energy storage device are generated based on the optimal charging and discharging power.

[0010] Dynamic optimization of the charge / discharge strategy using a time-difference computation method includes: The charging and discharging power, available capacity, and real-time electricity price of the wind-solar-storage system are obtained. The instantaneous operating cost is calculated by combining the charging and discharging power and the real-time electricity price. The instantaneous reward is constructed by combining the available capacity and the operating cost. Based on historical data, predict future electricity price trends and load demand changes, while considering the physical security constraints of the wind, solar and energy storage system, calculate the long-term benefits under the current decision, and combine the immediate reward with the long-term benefits to obtain the state value assessment index. Establish a mapping relationship between system state and action, calculate the time difference objective based on immediate reward and maximum value of future state, construct a value network in combination with the state value evaluation index, and use the experience replay method to randomly extract decision sequences from historical samples for value network training. During the control execution phase, the real-time collected system status is input into the trained value network. Based on the state value evaluation index, the charging and discharging action sequence with the maximum cumulative instantaneous reward is obtained and a real-time control command is generated. Based on the real-time control command, charging and discharging control is executed and system response data is collected. The value network is continuously optimized using the system response data to achieve online learning of the charging and discharging strategy.

[0011] A second aspect of the present invention provides a wind-solar-storage energy calculation and prediction coordinated control system, comprising: The first unit is used to acquire real-time operating data of the wind-solar-energy storage system. The real-time operating data includes the first power generation data of the wind farm, the second power generation data of the photovoltaic power station, and the real-time charging and discharging data of the energy storage power station. A multi-head attention structure is constructed to extract features from the real-time operating data of the wind-solar-energy storage system to obtain the power prediction value of the wind-solar-energy storage system. The second unit is used to construct a power generation prediction curve for the wind-solar-storage energy system based on the predicted power value of the wind-solar-storage energy system, and to calculate the upper and lower confidence limits of the power generation prediction curve based on historical power generation data; The third unit is used to construct a hierarchical control structure based on the power generation prediction curve and its confidence upper and lower limits to determine the charging and discharging strategy of the energy storage power station. The upper-level control calculates the target value of energy storage power adjustment through the system power generation smoothness evaluation function and the economic evaluation function. The lower-level control generates real-time charging and discharging control commands by combining the energy storage equipment lifespan constraint function and the charging and discharging efficiency constraint function, and uses the time-series differential calculation method to dynamically optimize the charging and discharging strategy. The real-time charging and discharging control commands are sent to the energy management system of the energy storage power station for execution to adjust the charging and discharging power of the energy storage power station in real time.

[0012] A third aspect of the present invention, An electronic device is provided, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0013] Fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0014] The beneficial effects of this application are as follows: The wind-solar-storage energy calculation and prediction collaborative control method described in this invention extracts features from the real-time operating data of the wind-solar-storage system by constructing a multi-head attention structure. This accurately captures the interrelationships between wind power, photovoltaics, and energy storage systems, improves the accuracy of power prediction, and provides reliable data support for the formulation of subsequent collaborative control strategies.

[0015] This invention constructs a power generation prediction curve based on the power prediction value and calculates the confidence interval, which can effectively assess the uncertainty of the prediction results, making the system more adaptable to the fluctuations of wind and photovoltaic power generation, and improving the stability and reliability of the wind and solar energy storage system.

[0016] This invention employs a hierarchical control structure to determine the charging and discharging strategy of an energy storage power station. The upper-level control focuses on the overall smoothness and economy of the system, while the lower-level control considers the lifespan of the energy storage equipment and the charging and discharging efficiency. Dynamic optimization is achieved through a time-series differential calculation method, which realizes the efficient utilization of energy storage resources. This ensures the stable operation of the system, extends the service life of the energy storage equipment, and significantly improves the economic benefits of wind and solar energy storage systems. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the wind-solar-storage energy calculation and prediction coordinated control method according to an embodiment of the present invention; Figure 2 A schematic diagram illustrating the process of constructing and calculating confidence intervals for power generation prediction curves. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0020] Figure 1 This is a flowchart illustrating the wind-solar-storage energy calculation and prediction coordinated control method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes: The real-time operation data of the wind-solar-energy storage system is obtained. The real-time operation data includes the first power generation data of the wind farm, the second power generation data of the photovoltaic power station, and the real-time charging and discharging data of the energy storage power station. A multi-head attention structure is constructed to extract features from the real-time operation data of the wind-solar-energy storage system to obtain the power prediction value of the wind-solar-energy storage system. Based on the power prediction values ​​of the wind-solar-storage energy system, a power generation prediction curve for the wind-solar-storage energy system is constructed, and the upper and lower confidence limits of the power generation prediction curve are calculated based on historical power generation data; Based on the power generation prediction curve and its confidence upper and lower limits, a hierarchical control structure is constructed to determine the charging and discharging strategy of the energy storage power station. The upper-level control calculates the target value of energy storage power adjustment through the system power generation smoothness evaluation function and the economic evaluation function. The lower-level control generates real-time charging and discharging control commands by combining the energy storage equipment lifespan constraint function and the charging and discharging efficiency constraint function, and uses the time-series differential calculation method to dynamically optimize the charging and discharging strategy. The real-time charging and discharging control commands are sent to the energy management system of the energy storage power station for execution, thereby adjusting the charging and discharging power of the energy storage power station in real time.

[0021] In one optional implementation, a multi-head attention structure is constructed to extract features from the real-time operating data of the wind-solar-storage energy system to obtain the power prediction value of the wind-solar-storage energy system, including: Calculate the correlation score between the real-time operation data and historical operation data of the wind-solar-storage system, and input the correlation score into the Softmax function to obtain the attention weight; The real-time running data is input into multiple parallel multi-head attention structures, each of which contains an independent query matrix, key matrix, and value matrix. For each multi-head attention structure, the real-time running data is multiplied by the query matrix to obtain the query vector, the historical running data is multiplied by the key matrix to obtain the key vector, the weighted features are calculated based on the dot product of the query vector and the key vector and combined with the attention weights, and the weighted features are multiplied by the value matrix to obtain the feature output. Multiple feature outputs are concatenated, and the power prediction value of the wind-solar energy storage system is obtained through linear transformation and nonlinear activation function.

[0022] Real-time operational data of the wind-solar-storage energy system is acquired, and historical operational data is retrieved from the database. This historical data is organized into a time series, containing records of system operation under similar past conditions. The correlation score between real-time operational data and historical operational data is obtained through vector inner product operation. For example, if the real-time wind speed is 5.2 m / s, the wind speed at a similar time in the historical record is 5.3 m / s, and the two are highly correlated; while the real-time solar irradiance is 850 watts / m², and the historical record is 500 watts / m², and the correlation is low.

[0023] These relevance scores are processed by the Softmax function, which transforms all scores into a probability distribution between 0 and 1, ensuring that the sum of all weights is 1. The Softmax function assigns greater weight to historical data with high relevance and less weight to historical data with low relevance. For example, for five sets of historical data with original relevance scores of [2.5, 1.8, 0.5, 3.2, 1.0], the attention weights obtained after processing with the Softmax function are [0.28, 0.14, 0.04, 0.47, 0.07]. This means that the fourth set of historical data, due to its high relevance, accounts for the largest weight in the prediction, 47%.

[0024] The real-time running data is then fed into multiple parallel processing layers. The multi-head attention architecture uses eight attention heads, each focusing on different aspects or patterns of the data. Each multi-head attention architecture contains three key matrices: a query matrix, a key matrix, and a value matrix. The dimensions of these matrices are determined based on the number of features in the input data. Assuming the real-time running data contains 20 features, the query and key matrices for each attention head would have a dimension of 20×32, and the value matrix would have a dimension of 20×64. The elements of these matrices are determined through the neural network training process to extract key information from the data.

[0025] Within each multi-head attention structure, real-time running data is multiplied by the query matrix to obtain a query vector, which reflects the pattern or feature the system hopes to find from historical data. Historical running data is multiplied by the key matrix to obtain a key vector, which represents a feature in historical data that matches the current query. Taking one attention head as an example, assuming the real-time running data is [5.2, 850, 25, 60, ...] representing wind speed, light intensity, temperature, humidity, etc., multiplying it by the query matrix yields the query vector [0.35, 0.42, 0.28, ...], while the historical data, multiplied by the key matrix, yields the key vector [0.38, 0.40, 0.25, ...].

[0026] The similarity between the query vector and the key vector is calculated by performing a dot product operation. The larger the dot product, the more similar the real-time state is to the historical state. These dot product results are multiplied by the previously calculated attention weights to form a weighted feature. The weighted feature highlights historical data patterns that are highly correlated with the current state and suppresses the influence of irrelevant data. Continuing the example above, the dot product of the query vector and the key vector is 0.95. Multiplying this by the attention weight of 0.47 yields a weighted value of 0.4465. This weighted value is used to adjust the contribution of historical data to the prediction.

[0027] The weighted features are then multiplied by the value matrix to obtain the feature output of each attention head. The value matrix transforms the weighted features into a new feature space, extracting more abstract and useful feature representations. In practical applications, each attention head produces a 64-dimensional feature output, which captures different patterns or relationships in the data. For example, the first attention head focuses on the relationship between wind speed and power generation, the second focuses on the relationship between light intensity and power generation, and the remaining attention heads focus on other different combinations of features.

[0028] The feature outputs of multiple attention heads are concatenated into a larger feature vector. Each of the eight attention heads generates 64-dimensional features, which are then concatenated to obtain a 512-dimensional joint feature representation. This high-dimensional feature vector contains the results of data analysis from multiple perspectives, providing rich information for subsequent prediction. The concatenated feature vector is then passed through a linear transformation layer to reduce the dimension from 512 to 128. A nonlinear transformation capability is then introduced through the ReLU nonlinear activation function. Finally, the output layer yields the predicted power values ​​for the wind-solar-storage energy system in the future.

[0029] In the daily operation of wind-solar-storage energy systems, this forecasting method is executed every 15 minutes, continuously updating the power forecast for the next 48 hours. Based on the forecast results, the charging and discharging plans of the energy storage equipment are adjusted in advance, optimizing energy dispatch strategies. For example, if good sunlight conditions are predicted for the next 6 hours, the use of photovoltaic power generation during the current period will be appropriately reduced, and excess electricity will be stored to cope with the predicted period of low power generation in the future.

[0030] In one optional implementation, based on the power prediction value of the wind-solar-storage energy system, a power generation prediction curve for the wind-solar-storage energy system is constructed, and the upper and lower confidence limits of the power generation prediction curve are calculated based on historical power generation data, including: The power prediction values ​​of the wind-solar-storage energy system are arranged in a time series, and the discrete prediction values ​​are processed into a continuous value using a cubic spline interpolation method to generate a smooth power generation prediction curve. Based on the power generation prediction curve, a kernel density estimation function is constructed to fit the probability density distribution of historical power generation data. The optimal bandwidth parameter of the kernel function is determined by an adaptive bandwidth selection algorithm to obtain the probability distribution characteristics of power generation. Based on the probability distribution characteristics, the conditional quantiles at different confidence levels are calculated using the quantile regression method. The upper quantile corresponding to the 95% confidence level is taken as the upper confidence limit, and the lower quantile corresponding to the 5% confidence level is taken as the lower confidence limit. The upper and lower confidence limits are smoothed to generate continuous confidence interval boundary curves, which are used to characterize the uncertainty range of the power generation prediction value.

[0031] like Figure 2 As shown, the method includes: Power predictions for wind, solar, and energy storage systems are typically provided in discrete time points, such as hourly or half-hourly forecasts. These power predictions are arranged in a time series; for example, a 24-hour forecast yields discrete values ​​such as [23.5MW, 24.2MW, 22.8MW, 20.5MW...]. Cubic spline interpolation is applied to these discrete data points, creating a cubic polynomial function between every two adjacent data points. This ensures the curve and its first and second derivatives are continuous at the connection points, forming a smooth power prediction curve. In practical applications, assuming the predicted power at 8:00, 9:00, and 10:00 is 40MW, 42MW, and 38MW respectively, cubic spline interpolation will generate a smooth curve, making the predicted value at 8:30 approximately 41.2MW and the predicted value at 9:30 approximately 39.5MW. This interpolation method avoids the jagged curves caused by straight-line connections, providing a smoother prediction curve that better reflects actual power generation characteristics.

[0032] A smooth power generation forecast curve reflects the system's expectation of future power generation. However, actual power generation fluctuates due to various uncertainties. To quantify this uncertainty, a kernel density estimation function is constructed based on the power generation forecast curve. This function fits the historical power generation data to a probability density distribution. Kernel density estimation is a non-parametric statistical method that does not pre-determine the data distribution form but directly learns the distribution characteristics from the data. During implementation, the deviation between actual power generation data and forecast data under similar past conditions is collected. For example, with a forecast power of 40MW, historical actual power distributions range from 36MW to 44MW. For each forecast power point, the system uses a Gaussian kernel function to map the historical deviation data into a continuous probability density distribution.

[0033] The key parameter in kernel density estimation is bandwidth, which controls the smoothness of the estimation results. Too small a bandwidth leads to excessive fluctuations, while too large a bandwidth results in over-smoothing and loss of important data features. An adaptive bandwidth selection algorithm is used to determine the optimal bandwidth parameter. This algorithm evaluates the performance of different bandwidth values ​​based on cross-validation, trying a series of candidate bandwidth values, such as 0.5MW to 2.5MW, in 0.1MW intervals. For each candidate bandwidth, a leave-one cross-validation error is calculated, and the bandwidth with the smallest error is selected as the optimal value. In a wind farm application case, the optimal bandwidth value calculated by the adaptive bandwidth selection algorithm is 1.2MW. The kernel density estimation function constructed under this bandwidth can accurately capture the probability distribution characteristics of power generation.

[0034] Quantile regression is used to calculate conditional quantiles at different confidence levels. Quantile regression considers not only the distribution of the deviation between predicted and actual values ​​but also the relationship between this deviation and the predicted value itself, thus more accurately characterizing the prediction uncertainty at different power levels. The upper quantile corresponding to the 95% confidence level is calculated as the upper confidence limit, and the lower quantile corresponding to the 5% confidence level is calculated as the lower confidence limit. This means that there is a 90% probability that the actual power generation will fall within this range. For example, when the predicted power value at a certain moment is 35MW, the 95% upper quantile calculated through quantile regression is 38.7MW, and the 5% lower quantile is 31.6MW, i.e., the upper confidence limit is 38.7MW and the lower confidence limit is 31.6MW.

[0035] The calculated upper and lower confidence limits are smoothed to ensure the smoothness and continuity of the confidence interval boundaries. The smoothing process uses an exponentially weighted moving average method, which assigns higher weights to recent data points and lower weights to older data points. The smoothing window size is typically set to 3 to 5 time units. The smoothed confidence interval boundary curves eliminate the abnormal fluctuations present in the original quantile calculation results, providing more stable and reliable uncertainty assessment results. In the daily operation of wind-solar-storage energy systems, these smoothed confidence interval boundary curves are visually displayed on the system monitoring interface, helping operators understand the reliability of the prediction results and providing important references for scheduling decisions.

[0036] This method has shown significant effectiveness in practical wind-solar-energy storage systems. In a real-world example of a 100MW photovoltaic power plant, traditional single-point prediction methods failed to provide information on prediction uncertainties, leading to frequent energy storage adjustments and increased equipment wear and operating costs. By adopting the method of this invention, a confidence interval for the predicted power is obtained, making the scheduling strategy more flexible and reasonable. When the actual power falls within the predicted confidence interval, the original scheduling plan is maintained; only when the actual power exceeds the confidence interval is the energy storage adjustment strategy triggered.

[0037] In one optional implementation, the upper-level control calculates the target value for energy storage power regulation using the system power generation smoothness evaluation function and the economic evaluation function, including: Calculate the difference between the power generation of the wind-solar-storage system at the current moment and the power generation at the previous moment to obtain the power generation change; divide the power generation change by the time interval to obtain the power generation volatility; construct a smoothness evaluation function based on the power generation volatility. Obtain the unit price of charging and discharging electricity for the energy storage power station under peak and off-peak electricity prices, and calculate the electricity cost based on the charging and discharging power; calculate the equipment loss cost based on the charging and discharging depth and cycle number of the energy storage equipment; add the electricity cost and the equipment loss cost to obtain the total cost, and construct an economic evaluation function based on the total cost; The smoothness evaluation function and the economic evaluation function are weighted and combined using weighting coefficients to obtain the objective function. The objective function is then solved using a dynamic programming algorithm to output the target value for energy storage power regulation.

[0038] The upper-level control system acquires real-time power generation data from the wind-solar-storage energy system, including the power generation values ​​at the current and previous moments. Assuming a wind-solar-storage energy system generates 42.8 MW at 10:00 and 40.2 MW at 9:55, the power generation difference between the two moments is calculated to be 2.6 MW. Considering a 5-minute sampling interval, dividing the power generation change by the time interval (0.0833 hours) yields a power generation volatility of 31.2 MW / h. This volatility reflects the rate of change in power generation and is a key indicator for evaluating system smoothness.

[0039] A smoothness evaluation function is constructed, employing a quadratic form. The smoothness evaluation function reaches its minimum value when the power generation fluctuation rate is zero. The greater the fluctuation rate, the larger the function value, indicating a more unstable power generation system. The smoothness evaluation function is expressed as the square of the fluctuation rate multiplied by a penalty coefficient. For example, for the aforementioned fluctuation rate of 31.2 MW / h, with a penalty coefficient set to 0.05, the calculated smoothness evaluation function value is 48.7. This value represents the degree of instability in the current system's power generation and will serve as an objective variable for subsequent optimization processes.

[0040] Obtain the unit electricity price for charging and discharging at the energy storage power station under peak and off-peak electricity pricing. For example, the electricity price is 1.2 yuan / kWh during peak hours (10:00-15:00 and 18:00-21:00), 0.8 yuan / kWh during normal hours (7:00-10:00 and 15:00-18:00), and 0.4 yuan / kWh during off-peak hours (21:00-7:00 the next day). Calculate the electricity cost based on the current charging and discharging power. Assuming the current time is 10:15, which is during peak hours, and the energy storage system plans to discharge 5MW for 1 hour, the electricity revenue would be 5MWh × 1.2 yuan / kWh = 6000 yuan; if the planned charging is 5MW, the electricity expenditure would be 5MWh × 1.2 yuan / kWh = 6000 yuan.

[0041] An economic evaluation of the loss costs of energy storage equipment is conducted, calculating the equipment loss costs based on the charge / discharge depth and cycle count. Assuming a total investment cost of 20 million yuan and a design life of 6000 complete charge / discharge cycles, the equivalent cost per complete cycle is 3333 yuan. When the charge / discharge depth is 50%, it is equivalent to 0.5 complete cycles, with a loss cost of 1667 yuan. Furthermore, different charge / discharge depths have different impacts on equipment lifespan; a nonlinear model is used for correction. For example, the loss at 80% charge / discharge depth is approximately 1.8 times that at 50% depth, rather than a nonlinear proportional relationship.

[0042] The total cost is obtained by adding the electricity cost and the equipment depreciation cost. An economic evaluation function is constructed based on this total cost. The design objective of the economic evaluation function is to minimize the total cost, expressed as the absolute value of the total cost. Continuing the example, if 5MW of discharge is chosen at the current moment, the electricity revenue is 6000 yuan, the equipment depreciation cost is 1667 yuan, and the net revenue is 4333 yuan. The economic evaluation function value is -4333 (a negative value indicates revenue). If 5MW of charging is chosen, the electricity cost is 6000 yuan, the equipment depreciation cost is 1667 yuan, and the total cost is 7667 yuan. The economic evaluation function value is 7667.

[0043] The smoothness evaluation function and the economic evaluation function reflect two different objectives. They are weighted and combined using weighting coefficients to form the final objective function. The weighting coefficients reflect different emphases on smoothness and economy. In practical applications, the smoothness weighting coefficient is usually set to 0.6-0.8, and the economic weighting coefficient is set to 0.2-0.4. The specific values ​​can be adjusted according to grid requirements and market conditions. For example, in scenarios with high grid stability requirements, the smoothness weight can be set to 0.8, and the economic weight to 0.2; while in a highly competitive market environment, the economic weight can be appropriately increased to 0.4. Assuming a smoothness weight of 0.7 and an economic weight of 0.3, for the above case, if 5MW of discharge is chosen, the objective function value is 0.7×48.7+0.3×(-4333)≈-1266; if 5MW of charging is chosen, the objective function value is 0.7×48.7+0.3×7667≈2334.

[0044] Dynamic programming decomposes the entire scheduling process into multiple stages, each corresponding to a decision at a specific time point. In each stage, multiple energy storage power adjustment values ​​are considered, the corresponding objective function value is calculated, and the adjustment value that optimizes the objective function is selected as the decision for that stage. Dynamic programming can consider the impact of the current decision on future states, avoiding global suboptimal problems caused by local optima.

[0045] The energy storage power adjustment range is set to -10MW to 10MW (negative values ​​indicate charging, positive values ​​indicate discharging), with a step size of 0.5MW, totaling 41 adjustment options. For each time point, the objective function value corresponding to the 41 adjustment options is calculated, and the option with the smallest objective function value is selected as the target value for energy storage power adjustment at that time point. Considering the capacity limitations of the energy storage device, during the calculation process, it is also necessary to check whether each adjustment option will cause the energy storage capacity to exceed the upper or lower limit; if it will exceed the limit, the option is marked as invalid.

[0046] The key to dynamic programming lies in the design of the state transition equation. State variables include the current time *t*, the remaining energy storage capacity (SOC), and the previous generation power (P_prev). For each state, the objective function value and the next state are calculated under different energy storage power adjustment values. For example, if the current state is {t=10:15, SOC=60%, P_prev=42.8MW}, and the chosen energy storage power adjustment value is -3MW (3MW charging), then the next state is {t=10:20, SOC=62%, P_prev=39.8MW}. By recursively solving the problem, the optimal decision sequence from the initial state to the target state is obtained.

[0047] To improve computational efficiency, a forward dynamic programming algorithm is employed, progressively calculating the optimal decision at each time point from the initial moment. Simultaneously, to handle the state-of-charge (SOC) constraints of the energy storage device, a penalty term is introduced. A larger penalty value is added when the SOC approaches the upper or lower limit, preventing decisions from causing the energy storage to exceed its limits. In actual operation, the solution time of the dynamic programming algorithm is controlled within 100 milliseconds, meeting the requirements of real-time control.

[0048] In one optional implementation, the lower-level control, in conjunction with the energy storage device lifetime constraint function and the charge / discharge efficiency constraint function, generates real-time charge / discharge control commands, including: Based on the energy storage power adjustment target value, the cycle life loss of the energy storage device under different charge and discharge depths is calculated, and a life loss constraint function is constructed; an efficiency loss constraint function is constructed according to the correspondence between charge and discharge current and battery temperature. Using lifetime loss constraint function, efficiency loss constraint function and energy storage capacity limit as constraints, a quadratic programming problem is constructed with minimizing power tracking error as the optimization objective. The optimal charging and discharging power that satisfies the constraints is calculated by a convex optimization solver. Real-time charging and discharging control commands for the energy storage device are generated based on the optimal charging and discharging power.

[0049] The lower-level control system receives the target value for energy storage power adjustment from the upper-level control system. For example, if the target value received at a certain moment is -3.5MW, it means that the energy storage device needs to be charged at a power of 3.5MW. The system then reads the current state information of the energy storage device, including key parameters such as state of charge (SOC), individual cell voltage, current, and temperature. Taking a lithium-ion battery energy storage system as an example, the current SOC is 45%, the average individual cell voltage is 3.65V, and the average temperature is 28℃. This state information is the basic data for constructing constraint functions and optimization problems.

[0050] Different charge / discharge depths have significantly different effects on battery life. Based on a battery cycle life model, the cycle life loss at different charge / discharge depths is calculated. This model establishes a non-linear relationship between charge / discharge depth and the number of battery cycles: shallow charge / discharge (depth less than 20%) can support more than 15,000 cycles; medium charge / discharge (depth 20%-50%) can support 8,000-15,000 cycles; deep charge / discharge (depth 50%-80%) can support 3,000-8,000 cycles; and ultra-deep charge / discharge (depth greater than 80%) can support less than 3,000 cycles. Considering the total investment cost of energy storage equipment, the life loss per cycle is quantified as an economic cost. Taking an energy storage system with a total capacity of 10MWh and an investment cost of 20 million yuan as an example, if calculated based on 6,000 complete cycles, the equivalent cost per complete cycle is approximately 3,333 yuan. When the charge / discharge depth is 30%, it is equivalent to 0.3 complete cycles, and the life loss cost is approximately 1,000 yuan.

[0051] The lifetime loss constraint function represents the relationship between charging / discharging power and lifetime loss. It is a non-linear function, approximated using a piecewise linear approach. In the current case, if the charging power is in the 0-2MW range, the lifetime loss factor is set to 1; if the charging power is in the 2-4MW range, the lifetime loss factor is set to 1.2; and if the charging power exceeds 4MW, the lifetime loss factor is set to 1.5. This means that high-power charging leads to more severe lifetime loss. For a target charging power of 3.5MW, the calculated lifetime loss cost is approximately 1200 yuan.

[0052] The charging and discharging efficiency of energy storage devices is closely related to the charging and discharging current and battery temperature. Based on experimental data, a model was established to correlate charging and discharging current with battery temperature, and an efficiency loss constraint function was constructed based on this model. Experimental data shows that lithium-ion batteries can achieve an efficiency of over 95% at a charging and discharging current of 0.2C-0.5C (where C represents the charge / discharge rate, and 1C represents the current required for a full charge / discharge cycle in one hour); however, at higher currents above 1C, the efficiency drops below 85%. Temperature also significantly affects efficiency: efficiency is higher within the suitable temperature range of 15-35℃; efficiency decreases significantly below 5℃ or above 45℃. For a 10MWh energy storage system, a charging power of 3.5MW corresponds to a charging rate of approximately 0.35C, and combined with a battery temperature of 28℃, the estimated charging efficiency is approximately 94%.

[0053] The efficiency loss constraint function correlates charging and discharging power with efficiency loss, approximating the nonlinear relationship between power and efficiency using a piecewise linear approach. For the current case, if the charging power is in the 0-3MW range, the efficiency is set to 95%; if the charging power is in the 3-5MW range, the efficiency is set to 94%; and if the charging power exceeds 5MW, the efficiency is set to 92%. Considering battery temperature, a temperature correction coefficient is introduced: when the temperature is in the 25-30℃ range, the correction coefficient is 1; when the temperature is below 25℃ or above 30℃, the correction coefficient is less than 1, indicating a decrease in efficiency. For a target charging power of 3.5MW and a battery temperature of 28℃, the calculated actual charging efficiency is 94%, meaning that approximately 3.72MW of grid power is required to achieve an effective charging power of 3.5MW.

[0054] In addition to constraints related to lifespan degradation and efficiency loss, energy storage capacity limitations must also be considered. Currently, the State of Charge (SOC) is 45%, and the upper limit for energy storage system capacity is 95%, with a lower limit of 10%. If charged at 3.5MW for 1 hour, the SOC will increase by approximately 35%, reaching 80%, which is within the upper limit. If discharged at 3.5MW for 1 hour, the SOC will decrease by approximately 35%, dropping to 10%, just reaching the lower limit. Considering the rated power limitation of the power converter, for example, if the rated power of the power converter is 8MW, the absolute value of the charging and discharging power must not exceed 8MW.

[0055] After establishing the constraint functions, a quadratic programming problem is constructed with the goal of minimizing the power tracking error. The optimization objective can be expressed as: minimizing the square of the difference between the actual charging / discharging power and the target power. The constraints include the aforementioned lifetime loss constraints, efficiency loss constraints, energy storage capacity limitations, and rated power limitations of the power converter. The decision variable of the quadratic programming problem is the actual charging / discharging power, which takes values ​​within the rated power range of the power converter (-8MW to 8MW).

[0056] This quadratic programming problem is solved using a convex optimization solver. Convex optimization is a class of optimization problems where the objective function is a convex function and the constraint set is a convex set. It has the characteristic of having a unique global optimum, making it suitable for real-time control scenarios. Using efficient algorithms such as the interior-point method or the principal-dual interior-point method, the solution time is typically controlled within 10 milliseconds, meeting the requirements of real-time control. For this case, considering all constraints, the optimal charging power obtained by optimization is 3.2MW, slightly less than the target value of 3.5MW. This is because under the current SOC and temperature conditions, a charging power of 3.5MW would lead to a significant efficiency loss; reducing it to 3.2MW can minimize efficiency loss while maintaining charging performance.

[0057] Based on the calculated optimal charging and discharging power, real-time charging and discharging control commands for the energy storage device are generated. These commands include parameters such as power setpoints, current limits, and voltage limits. For a charging power of 3.2MW, the generated control commands include: a power setpoint of 3.2MW, a current limit of 600A, and a voltage limit of 850V. These control commands are sent to the battery management system (BMS) and power converter (PCS) via a communication network, where they execute specific control actions, such as adjusting the charging current and monitoring battery temperature.

[0058] The control method described in this embodiment has good scalability and is applicable to different types of energy storage technologies, such as lithium-ion batteries, lead-acid batteries, and flow batteries. By configuring different lifetime and efficiency models, it can adapt to the characteristics of different energy storage technologies. Furthermore, this method supports coordinated control of multiple energy storage units. Through distributed optimization algorithms, power tasks are rationally allocated among multiple energy storage units, further improving the overall efficiency and lifetime of the system.

[0059] In one optional implementation, the dynamic optimization of the charging and discharging strategy using a time-differential calculation method includes: The charging and discharging power, available capacity, and real-time electricity price of the wind-solar-storage system are obtained. The instantaneous operating cost is calculated by combining the charging and discharging power and the real-time electricity price. The instantaneous reward is constructed by combining the available capacity and the operating cost. Based on historical data, predict future electricity price trends and load demand changes, while considering the physical security constraints of the wind, solar and energy storage system, calculate the long-term benefits under the current decision, and combine the immediate reward with the long-term benefits to obtain the state value assessment index. Establish a mapping relationship between system state and action, calculate the time difference objective based on immediate reward and maximum value of future state, construct a value network in combination with the state value evaluation index, and use the experience replay method to randomly extract decision sequences from historical samples for value network training. During the control execution phase, the real-time collected system status is input into the trained value network. Based on the state value evaluation index, the charging and discharging action sequence with the maximum cumulative instantaneous reward is obtained and a real-time control command is generated. Based on the real-time control command, charging and discharging control is executed and system response data is collected. The value network is continuously optimized using the system response data to achieve online learning of the charging and discharging strategy.

[0060] The system periodically collects key parameters of the energy storage device, such as charging and discharging power, remaining available capacity, and current grid electricity price. Taking a 100MWh energy storage power station as an example, at a certain moment, the charging power is 4.5MW, the available capacity is 62MWh, and the real-time electricity price is 0.4 yuan / kWh. Based on this data, the immediate operating cost is calculated. During charging, the immediate operating cost equals the charging power multiplied by the charging electricity price; during discharging, the immediate operating cost is negative, equal to the discharging power multiplied by the opposite of the discharging electricity price. For the above case, the immediate operating cost is 4.5MW × 0.4 yuan / kWh = 1800 yuan / hour. This cost data reflects the direct economic impact of the system's current operation.

[0061] The immediate reward is a direct indicator of the merits of the current charging and discharging strategy. It is constructed by combining available capacity and operating costs. Consideration of available capacity ensures the sustainable operation of the energy storage system and avoids system losses caused by overcharging and discharging. A non-linear function maps available capacity to a capacity reward value: when available capacity is within the suitable range of 40%-80%, the capacity reward value is higher; when available capacity approaches the upper limit (95%) or lower limit (10%), the capacity reward value decreases rapidly to prevent overcharging or over-discharging. For 62% available capacity, the capacity reward value is 0.85 (out of 1). The immediate reward is composed of the capacity reward and the operating cost reward, with the operating cost reward being inversely proportional to the operating cost. In the current example, the immediate reward value is calculated as the capacity reward of 0.85 minus the normalized operating cost of 0.3, resulting in 0.55. This comprehensive reward value considers both economic efficiency and the health status of the energy storage system.

[0062] The forecasting of future periods based on historical data employs time series analysis, incorporating factors such as weather forecasts, historical load patterns, and market supply and demand. For example, system analysis of electricity price data from the past three months reveals that the average electricity price during the weekday period from 2:00 PM to 4:00 PM is 1.2 yuan / kWh, while the average price during the period from 10:00 PM to 6:00 AM the next day is only 0.3 yuan / kWh. Based on these historical patterns, the forecast predicts that the electricity price for the next 24 hours will fluctuate between 0.3 and 1.2 yuan / kWh, and a detailed forecast curve is plotted. Simultaneously, the physical safety constraints of wind, solar, and energy storage systems are considered, including charging and discharging power limits (e.g., ±10MW), capacity upper and lower limits (e.g., 10%-95%), and power change rate limits (e.g., no more than 1MW per minute).

[0063] Long-term returns are calculated using a discounted cash flow (DCF) method. A discount factor (typically 0.9-0.99) is preset, and the expected returns at different future points in time are discounted and summed to obtain a long-term return estimate. The discount factor reflects the impact of future uncertainty on decision-making; returns further in the future are discounted more. In the current case, the system predicts that if charging continues at 4.5MW until the off-peak electricity price ends late at night, and then discharging at 8MW during the peak period the next day, the total expected return within 24 hours is approximately 5000 yuan, with a discounted long-term return estimate of 4500 yuan.

[0064] The combination of immediate reward and long-term benefit constitutes the state value assessment index. This index comprehensively reflects the overall value of a specific charging / discharging action in the current state. The combination method usually adopts a weighted sum, with an immediate reward weight of 0.3 and a long-term benefit weight of 0.7, highlighting the importance of long-term planning. For the above case, the state value assessment index is calculated as 0.3 × 0.55 + 0.7 × (4500 / 10000) = 0.48. This index value will be used as the target value for training the value network, guiding the system to learn the optimal decision-making strategy.

[0065] A mapping relationship between states and actions is established, which is implemented by a value network. The value network adopts a deep neural network structure. The input layer receives the system state vector, including features such as current time, available capacity, charging and discharging power, and grid electricity price. The hidden layer contains 3-5 fully connected layers, each with 64-128 neurons, using the ReLU activation function. The output layer corresponds to the value assessment of different charging and discharging actions. For example, if the system discretizes the charging and discharging power into 21 levels (-10MW to +10MW, with a step size of 1MW), the output layer of the value network will contain 21 nodes, each representing the value assessment of each power level.

[0066] The training of the value network is based on a temporal difference objective, which combines immediate reward with the maximum value of the future state. Formally, it equals the immediate reward plus the discounted maximum future value, minus the value estimate of the current state. During training, an experience replay method is used to randomly sample decision sequences from historical samples. The experience replay buffer stores state transition samples of the system over a period of time. Each sample contains information such as the current state, the action performed, the reward obtained, and the next state. Random sampling breaks the temporal correlation between samples, improving training efficiency and stability. Assuming the experience replay buffer has a capacity of 10,000 samples, 256 samples are randomly selected in each training session to form a batch. The Adam optimizer is used for gradient descent updates, with a learning rate of 0.0005. Training continues until the loss function converges.

[0067] During the control execution phase, real-time collected state data is input into the trained value network. For example, if the current time is 15:30, the available capacity is 78MWh, and the grid electricity price is 1.1 yuan / kWh, these data form a state vector and are input into the value network. The value network calculates the value assessment of each charging and discharging power option and selects the action with the highest value to execute. In this case, the value network outputs the highest value of 8MW discharge power, which is 0.82. It generates an 8MW discharge control command and sends it to the energy storage device for execution.

[0068] While executing charge and discharge control, response data is continuously collected, including actual executed power, system state changes, and economic benefits. This real-time data is used to evaluate the control effect and added as new samples to the experience playback buffer. For example, after discharging 8MW for one hour, if the available capacity drops to 70MWh and the economic benefit is 8MW × 1.1 yuan / kWh = 8800 yuan, this information constitutes a new experience sample. The value network is periodically updated using the accumulated new samples, enabling online learning and continuous optimization of the strategy. This closed-loop optimization mechanism allows the system to adapt to constantly changing environmental conditions and operational requirements.

[0069] The online learning of value networks adopts an incremental update approach, updating the network parameters every 4 hours using the most recently accumulated samples. During the update process, a small learning rate (such as 0.0001) is used to maintain the stability of the policy while absorbing new information. This incremental learning approach balances the relationship between policy stability and adaptability, enabling the system to make appropriate adjustments to environmental changes while maintaining the basic decision-making framework.

[0070] This invention relates to a wind-solar-storage energy calculation and prediction collaborative control system, the system comprising: The first unit is used to acquire real-time operating data of the wind-solar-energy storage system. The real-time operating data includes the first power generation data of the wind farm, the second power generation data of the photovoltaic power station, and the real-time charging and discharging data of the energy storage power station. A multi-head attention structure is constructed to extract features from the real-time operating data of the wind-solar-energy storage system to obtain the power prediction value of the wind-solar-energy storage system. The second unit is used to construct a power generation prediction curve for the wind-solar-storage energy system based on the predicted power value of the wind-solar-storage energy system, and to calculate the upper and lower confidence limits of the power generation prediction curve based on historical power generation data; The third unit is used to construct a hierarchical control structure based on the power generation prediction curve and its confidence upper and lower limits to determine the charging and discharging strategy of the energy storage power station. The upper-level control calculates the target value of energy storage power adjustment through the system power generation smoothness evaluation function and the economic evaluation function. The lower-level control generates real-time charging and discharging control commands by combining the energy storage equipment lifespan constraint function and the charging and discharging efficiency constraint function, and uses the time-series differential calculation method to dynamically optimize the charging and discharging strategy. The real-time charging and discharging control commands are sent to the energy management system of the energy storage power station for execution to adjust the charging and discharging power of the energy storage power station in real time.

[0071] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0072] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0073] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for coordinated control of wind-solar-storage energy calculation and prediction, characterized in that, include: The real-time operation data of the wind-solar-energy storage system is obtained. The real-time operation data includes the first power generation data of the wind farm, the second power generation data of the photovoltaic power station, and the real-time charging and discharging data of the energy storage power station. A multi-head attention structure is constructed to extract features from the real-time operation data of the wind-solar-energy storage system to obtain the power prediction value of the wind-solar-energy storage system. Based on the power prediction values ​​of the wind-solar-storage energy system, a power generation prediction curve for the wind-solar-storage energy system is constructed, and the upper and lower confidence limits of the power generation prediction curve are calculated based on historical power generation data; Based on the power generation prediction curve and its confidence upper and lower limits, a hierarchical control structure is constructed to determine the charging and discharging strategy of the energy storage power station. The upper-level control calculates the target value of energy storage power adjustment through the system power generation smoothness evaluation function and the economic evaluation function. The lower-level control generates real-time charging and discharging control commands by combining the energy storage equipment lifespan constraint function and the charging and discharging efficiency constraint function, and uses the time-series differential calculation method to dynamically optimize the charging and discharging strategy. The real-time charging and discharging control commands are sent to the energy management system of the energy storage power station for execution, thereby adjusting the charging and discharging power of the energy storage power station in real time.

2. The method according to claim 1, characterized in that, A multi-head attention structure is constructed to extract features from the real-time operating data of the wind-solar-storage system, resulting in the following power prediction values ​​for the system: Calculate the correlation score between the real-time operation data and historical operation data of the wind-solar-storage system, and input the correlation score into the Softmax function to obtain the attention weight; The real-time running data is input into multiple parallel multi-head attention structures, each of which contains an independent query matrix, key matrix, and value matrix. For each multi-head attention structure, the real-time running data is multiplied by the query matrix to obtain the query vector, the historical running data is multiplied by the key matrix to obtain the key vector, the weighted features are calculated based on the dot product of the query vector and the key vector and combined with the attention weights, and the weighted features are multiplied by the value matrix to obtain the feature output. Multiple feature outputs are concatenated, and the power prediction value of the wind-solar energy storage system is obtained through linear transformation and nonlinear activation function.

3. The method according to claim 1, characterized in that, Based on the power prediction values ​​of the wind-solar-storage energy system, a power generation prediction curve for the wind-solar-storage energy system is constructed, and the upper and lower confidence limits of the power generation prediction curve are calculated based on historical power generation data, including: The power prediction values ​​of the wind-solar-storage energy system are arranged in a time series, and the discrete prediction values ​​are processed into a continuous value using a cubic spline interpolation method to generate a smooth power generation prediction curve. Based on the power generation prediction curve, a kernel density estimation function is constructed to fit the probability density distribution of historical power generation data. The optimal bandwidth parameter of the kernel function is determined by an adaptive bandwidth selection algorithm to obtain the probability distribution characteristics of power generation. Based on the probability distribution characteristics, the conditional quantiles at different confidence levels are calculated using the quantile regression method. The upper quantile corresponding to the 95% confidence level is taken as the upper confidence limit, and the lower quantile corresponding to the 5% confidence level is taken as the lower confidence limit. The upper and lower confidence limits are smoothed to generate continuous confidence interval boundary curves, which are used to characterize the uncertainty range of the power generation prediction value.

4. The method according to claim 1, characterized in that, The upper-level control calculates the target value for energy storage power regulation through the system power generation smoothness evaluation function and the economic evaluation function, including: Calculate the difference between the power generation of the wind-solar-storage system at the current moment and the power generation at the previous moment to obtain the power generation change; divide the power generation change by the time interval to obtain the power generation volatility; construct a smoothness evaluation function based on the power generation volatility. Obtain the unit price of charging and discharging electricity for the energy storage power station under peak and off-peak electricity prices, and calculate the electricity cost based on the charging and discharging power; calculate the equipment loss cost based on the charging and discharging depth and cycle number of the energy storage equipment; add the electricity cost and the equipment loss cost to obtain the total cost, and construct an economic evaluation function based on the total cost; The smoothness evaluation function and the economic evaluation function are weighted and combined using weighting coefficients to obtain the objective function. The objective function is then solved using a dynamic programming algorithm to output the target value for energy storage power regulation.

5. The method according to claim 1, characterized in that, The lower-level control, combining the energy storage device's lifetime constraint function and charge / discharge efficiency constraint function, generates real-time charge / discharge control commands, including: Based on the energy storage power adjustment target value, the cycle life loss of the energy storage device under different charge and discharge depths is calculated, and a life loss constraint function is constructed; an efficiency loss constraint function is constructed according to the correspondence between charge and discharge current and battery temperature. Using lifetime loss constraint function, efficiency loss constraint function and energy storage capacity limit as constraints, a quadratic programming problem is constructed with minimizing power tracking error as the optimization objective. The optimal charging and discharging power that satisfies the constraints is calculated by a convex optimization solver. Real-time charging and discharging control commands for the energy storage device are generated based on the optimal charging and discharging power.

6. The method according to claim 1, characterized in that, Dynamic optimization of the charge / discharge strategy using a time-difference computation method includes: The charging and discharging power, available capacity, and real-time electricity price of the wind-solar-storage system are obtained. The instantaneous operating cost is calculated by combining the charging and discharging power and the real-time electricity price. The instantaneous reward is constructed by combining the available capacity and the operating cost. Based on historical data, predict future electricity price trends and load demand changes, while considering the physical security constraints of the wind, solar and energy storage system, calculate the long-term benefits under the current decision, and combine the immediate reward with the long-term benefits to obtain the state value assessment index. Establish a mapping relationship between system state and action, calculate the time difference objective based on immediate reward and maximum value of future state, construct a value network in combination with the state value evaluation index, and use the experience replay method to randomly extract decision sequences from historical samples for value network training. During the control execution phase, the real-time collected system status is input into the trained value network. Based on the state value evaluation index, the charging and discharging action sequence with the maximum cumulative instantaneous reward is obtained and a real-time control command is generated. Based on the real-time control command, charging and discharging control is executed and system response data is collected. The value network is continuously optimized using the system response data to achieve online learning of the charging and discharging strategy.

7. A wind-solar-storage energy calculation and prediction coordinated control system, used to implement the method as described in any one of claims 1-6, characterized in that, include: The first unit is used to acquire real-time operating data of the wind-solar-energy storage system. The real-time operating data includes the first power generation data of the wind farm, the second power generation data of the photovoltaic power station, and the real-time charging and discharging data of the energy storage power station. A multi-head attention structure is constructed to extract features from the real-time operating data of the wind-solar-energy storage system to obtain the power prediction value of the wind-solar-energy storage system. The second unit is used to construct a power generation prediction curve for the wind-solar-storage energy system based on the predicted power value of the wind-solar-storage energy system, and to calculate the upper and lower confidence limits of the power generation prediction curve based on historical power generation data; The third unit is used to construct a hierarchical control structure based on the power generation prediction curve and its confidence upper and lower limits to determine the charging and discharging strategy of the energy storage power station. The upper-level control calculates the target value of energy storage power adjustment through the system power generation smoothness evaluation function and the economic evaluation function. The lower-level control generates real-time charging and discharging control commands by combining the energy storage equipment lifespan constraint function and the charging and discharging efficiency constraint function, and uses the time-series differential calculation method to dynamically optimize the charging and discharging strategy. The real-time charging and discharging control commands are sent to the energy management system of the energy storage power station for execution to adjust the charging and discharging power of the energy storage power station in real time.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images