Electricity-hydrogen coupling intelligent control method and system considering wind-solar power prediction errors

Abstract

This invention provides an intelligent control method and system for electric-hydrogen coupling that takes into account wind-solar power generation prediction errors. This method involves collecting historical wind-solar power generation data and deriving a prediction error probability distribution based on adaptive combined kernel density estimation. This distribution is used as an uncertainty constraint to perform deep reinforcement learning using a dual neural network, in which an action-value network dynamically adjusts the decision space, and a target network assesses risk and constrains decisions based on conditional risk and state value. Finally, a distributed collaborative network is constructed to control hydrogen production and decomposition power in real time based on dynamic consistency theory and synchronized potential functions. This invention improves the operational stability and economic efficiency of the electric-hydrogen coupling system, effectively addressing the uncertainties of wind-solar power generation.

Description

Technical Field

[0001] The present invention relates to electric power technology, and in particular to an electric-hydrogen coupling intelligent control method and system taking into account wind-solar power prediction errors. Background Art

[0002] As a clean and efficient energy carrier, hydrogen energy can achieve long-term storage of renewable energy through water electrolysis. At present, the control method of the electric-hydrogen coupling system is mainly based on deterministic forecast data for optimization and scheduling. However, the forecast errors of wind power and photovoltaic power generation have significant randomness and uncertainty, which poses a challenge to the safe and economic operation of the system. At the same time, as distributed devices, the coordinated control of the hydrogen production system and the hydrogen storage system has a significant impact on the overall performance of the system. The main problems of the existing technology are:

[0003] Existing control methods for electric-hydrogen coupled systems often use a single probability distribution model to describe wind and solar power forecast errors, which cannot accurately characterize the complex distribution characteristics of the forecast errors, resulting in insufficient robustness of system control.

[0004] Traditional scheduling optimization methods lack the ability to adaptively adjust to the short-term fluctuations and long-term trends of wind and solar power forecast errors, and do not fully consider system operation risks in the decision-making process, which can easily lead to unsatisfactory actual implementation results of scheduling plans.

[0005] The existing collaborative control methods for distributed hydrogen production systems have not established a mechanism to associate with system operation risks, and lack the ability to respond quickly based on real-time prediction deviations, making it difficult to achieve efficient collaborative operation of the system. Summary of the Invention

[0006] The embodiments of the present invention provide an electric-hydrogen coupled intelligent control method and system that takes into account wind-solar prediction errors, which can solve the problems in the prior art.

[0007] According to a first aspect of the embodiments of the present invention,

[0008] Provides an electric-hydrogen coupling intelligent control method that takes into account wind and solar power prediction errors, including:

[0009] Historical power generation data from wind farms and photovoltaic power stations were collected, and the spatiotemporal correlation characteristics of their forecast errors were calculated. The probability distribution of wind and solar power forecast errors was obtained using an adaptive combined kernel density estimation method using the Gaussian kernel function and the Epanechnikov kernel function.

[0010] Using the probability distribution as an uncertainty constraint, combined with grid load demand data and hydrogen storage system operating data, a dual neural network is used to perform deep reinforcement learning. The action-value network dynamically adjusts the decision space based on an adaptive boundary function of the short-term fluctuations and long-term trends of wind and solar forecast errors. The target network assesses risk through a dual branch of conditional risk and state value, and constructs a risk constraint coefficient based on Lyapunov stability theory to constrain decision-making risk and obtain dispatch instructions.

[0011] The electrolysis hydrogen production unit and the hydrogen storage system are constructed as a distributed collaborative network. Based on the dynamic consistency theory, the coupling strength of the distributed collaborative network is inversely mapped to the risk constraint coefficient. The deviation between the node state of the distributed collaborative network and the scheduling instruction and the state difference of adjacent nodes are used to construct a synchronization potential function. Based on the deviation between the actual power generation power of the wind farm and the predicted power generation power of the photovoltaic power station, the hydrogen production power and the hydrogen release power are controlled in real time along the gradient direction of the synchronization potential function.

[0012] In an optional embodiment,

[0013] The spatiotemporal correlation characteristics of wind farm and photovoltaic power station forecast errors are calculated, and the probability distribution of wind and solar power forecast errors is obtained by using the adaptive combined kernel density estimation method of Gaussian kernel function and Epanechnikov kernel function, including:

[0014] Calculate the temporal and spatial correlation coefficients of the forecast error data of wind farms and photovoltaic power plants to obtain the spatiotemporal coupling characteristics;

[0015] A Gaussian kernel function and an Epanechnikov kernel function are selected as basic kernel functions, the Gaussian kernel function is used for probability density estimation in the central area, and the Epanechnikov kernel function is used for probability density estimation in the boundary area; adaptive weights of the Gaussian kernel function and the Epanechnikov kernel function are calculated based on the local sample density, and the product of the Gaussian kernel function and the adaptive weight and the sum of the product of the Epanechnikov kernel function and the complement of the adaptive weight are used as the combined kernel function;

[0016] The initial bandwidth is calculated based on the standard deviation and interquartile range of historical power generation data, and the initial bandwidth is adjusted with quantile constraints according to the empirical distribution function to obtain the optimized bandwidth;

[0017] The spatiotemporal coupling characteristics, combined kernel function and optimized bandwidth are substituted into the kernel density estimation formula to obtain the probability density function of the wind and solar power generation prediction error; the prediction error samples are updated based on the sliding time window, the sample weights are calculated according to the time attenuation factor, and the sample weights and the probability density function are recursively updated to obtain a dynamically updated wind and solar power prediction error probability distribution.

[0018] In an optional embodiment,

[0019] A dual neural network is used to perform deep reinforcement learning. The action-value network dynamically adjusts the decision space based on an adaptive boundary function of the short-term fluctuations and long-term trends of wind and solar forecast errors. The target network assesses risk through a dual branch of conditional risk and state value, and constructs a risk constraint coefficient based on Lyapunov stability theory for decision-making. Risk constraints include:

[0020] Obtaining a system state vector, wherein the system state vector includes grid load demand data, hydrogen production power data, hydrogen storage capacity data, and hydrogen refueling demand data;

[0021] Inputting the system state vector into an action value network, the action value network calculating an adaptive boundary function based on a probability distribution of wind and solar power prediction errors, and dynamically adjusting a basic decision boundary according to the adaptive boundary function to obtain an adjusted decision space;

[0022] Inputting the system state vector and the adjusted decision space into a target network, the target network calculating a conditional risk value and a state value, and weighting the conditional risk value and the state value to obtain a risk metric value;

[0023] Constructing an optimization objective function including a system operation cost term, a risk metric term, and a renewable energy consumption term, and updating the action value network parameters based on the optimization objective function using a stochastic gradient descent method. The target network evaluates the decision risk based on the risk metric value, and periodically updates the network parameters using a preset soft update coefficient.

[0024] Perform iterative calculation of the value function based on the updated action value network parameters and risk metric values, and use the product of the iterative calculation result and the risk constraint coefficient as the risk constraint term. The risk constraint coefficient is obtained by constructing a Lyapunov function based on the quadratic term of the hydrogen storage deviation from the safe interval and the high-order term of the wind and solar power prediction error, and calculating the inverse of the stability margin through its time derivative;

[0025] Based on the adjusted decision space, a decision evaluation value considering the risk constraint item is calculated, and a decision corresponding to the maximum decision evaluation value is selected as a scheduling instruction.

[0026] In an optional embodiment,

[0027] The adaptive boundary function includes:

[0028] The wind and solar forecast error sequence is decomposed into short-term fluctuation error and long-term trend error by discrete wavelet transform.

[0029] Calculating the boundary adjustment characteristics of the short-term fluctuation error based on the local probability density characteristics of the forecast error samples;

[0030] Construct a cumulative influence function, which is obtained by weighted accumulation of long-term trend errors in the historical period according to an exponential decay coefficient, and calculate a trend strength index based on the cumulative influence function. The trend strength index is the ratio of the absolute value of the cumulative influence function to the maximum absolute value of the cumulative influence function within the evaluation time window;

[0031] A probability confidence interval is constructed based on the boundary adjustment feature, and the boundary value of the probability confidence interval is mapped to a fluctuation boundary adjustment amount, and the fluctuation boundary adjustment amount is dynamically corrected based on the frequency of fluctuation direction changes; a piecewise linear mapping function is constructed based on the trend strength indicator to calculate the trend boundary adjustment amount, and the trend boundary adjustment amount is corrected according to the duration of the trend;

[0032] An adaptive combination weight is calculated based on the trend strength index, and the adaptive combination weight is exponentially related to the trend strength index; the fluctuation boundary adjustment amount and the trend boundary adjustment amount are weightedly combined using the adaptive combination weight to obtain the adaptive boundary function.

[0033] In an optional embodiment,

[0034] The target network calculates a conditional risk value and a state value, and weights the conditional risk value and the state value to obtain a risk metric value, including:

[0035] Constructing a multi-dimensional state-coupled system energy function based on the system state vector and the adjusted decision space, wherein the system energy function is a nonlinear weighted combination of grid load demand data, hydrogen production power data, hydrogen storage data, and hydrogenation demand data and their corresponding reference values;

[0036] Construct the conditional risk branch and state value branch of the target network;

[0037] Input the adjusted decision space and historical decision sequence into the conditional risk branch, calculate the attention weight matrix based on the multi-level correlation degree between the system state vector and the historical state, and use the spatiotemporal convolution result of the attention weight matrix and the historical state as the conditional risk value;

[0038] Input the system state vector and the adjusted decision space into the state value branch, construct the posterior distribution of the state value using the probabilistic variational inference method, and extract the conditional expectation and uncertainty of the posterior distribution to obtain the state value;

[0039] Determining an adaptive time window length based on the system energy function, wherein the adaptive time window length is not less than a response time of the electric-hydrogen coupling system and not greater than a preset multiple of a time constant of the hydrogen storage system;

[0040] Obtaining a maximum value and a minimum value of the energy of the electric-hydrogen coupling system within an adaptive time window, and calculating a multi-scale stability index based on a difference between a current system energy value and the maximum value and the minimum value;

[0041] An adaptive weighting coefficient is constructed according to the multi-scale stability index, and the conditional risk value and the state value are nonlinearly weighted according to the adaptive weighting coefficient to obtain a risk measurement value.

[0042] In an optional embodiment,

[0043] The risk constraint coefficients include:

[0044] The real-time value of hydrogen storage, the upper and lower limits of the hydrogen storage safety interval, and the wind and solar power prediction error sequence are obtained; the degree of deviation of the hydrogen storage from the safety interval is constructed as a quadratic term, and the wind and solar power prediction error value is constructed as a high-order term to form a Lyapunov function that characterizes the stability of the system; the stability margin is calculated based on the derivative of the Lyapunov function with respect to time, and the inverse of the stability margin is used as the risk constraint coefficient.

[0045] In an optional embodiment,

[0046] Based on the dynamic consistency theory, the coupling strength of the distributed collaborative network is inversely mapped to the risk constraint coefficient, and the synchronization potential function is constructed by the deviation between the node state of the distributed collaborative network and the scheduling instruction and the state difference of the adjacent nodes.

[0047] Constructing a dynamic coupling strength based on the risk constraint coefficient, wherein the dynamic coupling strength is a ratio of the basic coupling strength to the risk adjustment term, wherein the risk adjustment term increases as the risk constraint coefficient increases, and the dynamic coupling strength also exhibits an exponential decay relationship with the state difference of adjacent nodes;

[0048] Constructing a synchronization potential function, the synchronization potential function including a scheduling instruction tracking term and a node coordination term, the scheduling instruction tracking term being the product of the square difference between the node state and the scheduling instruction and a first weight coefficient, and the node coordination term being the product of the square of the difference between adjacent node states and the dynamic coupling strength and a second weight coefficient;

[0049] Constructing a distributed control law based on the gradient of the synchronization potential function, the distributed control law including a scheduling tracking term and a coordinated adjustment term, the scheduling tracking term being the product of the deviation between the node state and the scheduling instruction and the first weight coefficient, and the coordinated adjustment term being the product of the weighted deviation between the node state and the adjacent node state and the second weight coefficient;

[0050] When the actual power generation of the wind farm and photovoltaic power station exceeds the predicted power generation, the hydrogen production power is increased based on the distributed control law; when the actual power generation is less than the predicted power generation, the hydrogen release power of the hydrogen storage system is increased based on the distributed control law, and the electric-hydrogen system is coordinated and regulated in real time.

[0051] In an optional embodiment,

[0052] The synchronization potential function includes a fast response layer and a steady-state tracking layer, including:

[0053] The scheduling instruction tracking item of the rapid response layer adopts the product of the square difference between the node state and the scheduling instruction and the first weight coefficient, and the node coordination item of the rapid response layer adopts the product of the square of the difference between the adjacent node states and the dynamic coupling strength and the second weight coefficient; the rapid response layer triggers calculation updates based on the rate of change of the node state, and updates the first weight coefficient and the second weight coefficient when the rate of change of the node state exceeds a preset threshold;

[0054] The scheduling instruction tracking item of the steady-state tracking layer adopts the product of the cumulative square difference between the node state and the scheduling instruction and the first weight coefficient, wherein the cumulative square difference is the time integral of the square difference between the node state and the scheduling instruction within a fixed time window, and the length of the fixed time window is not less than 10 times the calculation period of the fast response layer and not more than one-third of the system characteristic time, wherein the system characteristic time is the dynamic response time of the electric hydrogen system; the node coordination item of the steady-state tracking layer adopts the product of the square of the adjacent node state difference, the dynamic coupling strength and the second weight coefficient;

[0055] An inter-layer coupling weight is constructed based on the calculation results of the fast response layer and the steady-state tracking layer. When the rate of change of the node state is greater than a first preset threshold, the weight of the fast response layer is increased. When the rate of change of the node state is less than a second preset threshold, the weight of the steady-state tracking layer is increased.

[0056] According to a second aspect of the embodiments of the present invention,

[0057] Provides an electric-hydrogen coupled intelligent control system that takes into account wind and solar power prediction errors, including:

[0058] The first unit is used to collect historical power generation data of wind farms and photovoltaic power stations, calculate the spatiotemporal correlation characteristics of the prediction errors of wind farms and photovoltaic power stations, and use the adaptive combined kernel density estimation method of Gaussian kernel function and Epanechnikov kernel function to obtain the probability distribution of wind and solar power prediction errors;

[0059] The second unit is used to use the probability distribution as an uncertainty constraint, combined with grid load demand data and hydrogen storage system operation data, and perform deep reinforcement learning using a dual neural network. The action-value network dynamically adjusts the decision space based on an adaptive boundary function of the short-term fluctuations and long-term trends of wind and solar forecast errors. The target network assesses risk through a dual branch of conditional risk and state value, and constructs a risk constraint coefficient based on Lyapunov stability theory to constrain decision-making risk and obtain dispatch instructions.

[0060] The third unit is used to construct the electrolytic hydrogen production unit and the hydrogen storage system into a distributed collaborative network, map the coupling strength of the distributed collaborative network inversely to the risk constraint coefficient based on the dynamic consistency theory, and construct a synchronous potential function based on the deviation between the node state of the distributed collaborative network and the scheduling instruction and the state difference of the adjacent nodes; based on the deviation between the actual power generation power of the wind farm and the photovoltaic power station and the predicted power generation power, the hydrogen production power and hydrogen release power are controlled in real time along the gradient direction of the synchronous potential function.

[0061] According to a third aspect of the embodiments of the present invention,

[0062] An electronic device is provided, comprising:

[0063] processor;

[0064] a memory for storing processor-executable instructions;

[0065] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0066] The present invention adopts an adaptive combined kernel density estimation method to obtain the probability distribution of wind and solar power prediction errors, fully considering the spatiotemporal correlation characteristics of wind and solar power generation, improving the accuracy of the prediction error probability distribution, and providing reliable uncertainty constraints for subsequent regulatory decisions.

[0067] The present invention adopts a dual neural network to perform deep reinforcement learning. Through the coordinated cooperation of the action value network and the target network, it realizes the dynamic adjustment and risk constraint of the decision space, which not only ensures the adaptability of the scheduling instructions, but also ensures the safety of the system operation and improves the regulation efficiency of the electric-hydrogen coupling system.

[0068] The present invention constructs the electrolysis hydrogen production unit and the hydrogen storage system into a distributed collaborative network, realizes real-time control based on dynamic consistency theory and synchronous potential function, effectively responds to the real-time fluctuations of wind and solar power generation, improves the system's dynamic response capability and operational stability, and realizes real-time intelligent collaborative regulation of the electric-hydrogen coupling system. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1This is a flow chart of an electric-hydrogen coupled intelligent control method taking into account wind-solar prediction errors according to an embodiment of the present invention;

[0070] Figure 2 This is a comparison chart of the probability distribution estimation accuracy of the present invention, the traditional Gaussian kernel function method, and the fixed weight method;

[0071] Figure 3 This is a performance comparison chart of the double-layer control of the present invention and other single-layer controls. DETAILED DESCRIPTION

[0072] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0073] The following combination Figures 1 to 3 The technical solution of the present invention is described in detail with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0074] Figure 1 This is a flow chart of the electric-hydrogen coupling intelligent control method considering wind-solar prediction errors according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0075] Historical power generation data from wind farms and photovoltaic power stations were collected, and the spatiotemporal correlation characteristics of their forecast errors were calculated. The probability distribution of wind and solar power forecast errors was obtained using an adaptive combined kernel density estimation method using the Gaussian kernel function and the Epanechnikov kernel function.

[0076] Using the probability distribution as an uncertainty constraint, combined with grid load demand data and hydrogen storage system operating data, a dual neural network is used to perform deep reinforcement learning. The action-value network dynamically adjusts the decision space based on an adaptive boundary function of the short-term fluctuations and long-term trends of wind and solar forecast errors. The target network assesses risk through a dual branch of conditional risk and state value, and constructs a risk constraint coefficient based on Lyapunov stability theory to constrain decision-making risk and obtain dispatch instructions.

[0077] The electrolysis hydrogen production unit and the hydrogen storage system are constructed as a distributed collaborative network. Based on the dynamic consistency theory, the coupling strength of the distributed collaborative network is inversely mapped to the risk constraint coefficient. The deviation between the node state of the distributed collaborative network and the scheduling instruction and the state difference of adjacent nodes are used to construct a synchronization potential function. Based on the deviation between the actual power generation power of the wind farm and the predicted power generation power of the photovoltaic power station, the hydrogen production power and the hydrogen release power are controlled in real time along the gradient direction of the synchronization potential function.

[0078] In an optional embodiment,

[0079] The spatiotemporal correlation characteristics of wind farm and photovoltaic power station forecast errors are calculated, and the probability distribution of wind and solar power forecast errors is obtained by using the adaptive combined kernel density estimation method of Gaussian kernel function and Epanechnikov kernel function, including:

[0080] Calculate the temporal and spatial correlation coefficients of the prediction error data of wind farms and photovoltaic power plants to obtain the spatiotemporal coupling characteristics;

[0081] A Gaussian kernel function and an Epanechnikov kernel function are selected as basic kernel functions, the Gaussian kernel function is used for probability density estimation in the central area, and the Epanechnikov kernel function is used for probability density estimation in the boundary area; adaptive weights of the Gaussian kernel function and the Epanechnikov kernel function are calculated based on the local sample density, and the product of the Gaussian kernel function and the adaptive weight and the sum of the product of the Epanechnikov kernel function and the complement of the adaptive weight are used as the combined kernel function;

[0082] The initial bandwidth is calculated based on the standard deviation and interquartile range of historical power generation data, and the initial bandwidth is adjusted with quantile constraints according to the empirical distribution function to obtain the optimized bandwidth;

[0083] The spatiotemporal coupling characteristics, combined kernel function and optimized bandwidth are substituted into the kernel density estimation formula to obtain the probability density function of the wind and solar power generation prediction error; the prediction error samples are updated based on the sliding time window, the sample weights are calculated according to the time attenuation factor, and the sample weights and the probability density function are recursively updated to obtain a dynamically updated wind and solar power prediction error probability distribution.

[0084] For example, the spatiotemporal coupling characteristics of wind farm and photovoltaic power station forecast errors were analyzed. The forecast error time series for these two power stations were obtained, and the autocorrelation function of the forecast error time series was calculated using an autoregressive moving average model to obtain the temporal dimension characteristics. A spatial correlation matrix between the wind farm and photovoltaic power station was established, and the spatial correlation coefficient between the wind farm and photovoltaic power station was calculated to obtain the spatial dimension characteristics. A Kronecker product was performed on the time weight matrix corresponding to the temporal dimension characteristics and the spatial weight matrix corresponding to the spatial dimension characteristics to obtain the spatiotemporal coupling characteristics. For a 100MW wind farm and a 50MW photovoltaic power station, 15-minute resolution power generation data and forecast data were collected. The temporal correlation of the forecast error series was calculated using a sliding time window with a 24-hour window length. For the wind farm data, the correlation coefficient was 0.82 when the time delay was 1 hour, and it dropped to 0.45 when the delay was 4 hours. For the photovoltaic power station data, the correlation coefficient was 0.75 when the time delay was 1 hour, and it dropped to 0.38 when the delay was 4 hours. The spatial correlation coefficient between wind farms and photovoltaic power plants is 0.42, indicating that the prediction errors have significant spatial correlation. Compared with traditional separate analysis methods, this spatiotemporal coupled analysis can improve the accuracy of prediction error feature extraction by 25%.

[0085] In the kernel function selection phase, an adaptive combination of the Gaussian and Epanechnikov kernel functions is used. The Gaussian kernel function has excellent smoothness in the central region and is used to process the main part of the prediction error distribution. Specifically, when the prediction error value falls within the range of plus or minus one standard deviation, the Gaussian kernel function is used for probability density estimation. The Epanechnikov kernel function has excellent convergence in the boundary region and is used to characterize the tail characteristics of the prediction error distribution. When the prediction error value exceeds two standard deviations, the Epanechnikov kernel function is used for estimation.

[0086] Adaptive weights are calculated based on local sample density characteristics. Within a range of ±1 standard deviation, when the local sample density is greater than 0.8, the Gaussian kernel weight is 0.9, and the Epanechnikov kernel weight is 0.1. When the local sample density is between 0.4 and 0.8, the Gaussian kernel weight is 0.7, and the Epanechnikov kernel weight is 0.3. In boundary regions, when the local sample density is less than 0.2, the Gaussian kernel weight is reduced to 0.1, and the Epanechnikov kernel weight is increased to 0.9. Practical applications have shown that this adaptive weighting strategy improves probability distribution estimation accuracy by 30% compared to fixed-weight methods.

[0087] The bandwidth optimization process first determines initial values ​​based on historical data characteristics. For wind farm data, the standard deviation is 8 MW, the interquartile range is 12 MW, and the initial bandwidth is 2.5 MW. For photovoltaic power plant data, the standard deviation is 4 MW, the interquartile range is 6 MW, and the initial bandwidth is 1.5 MW. The initial bandwidth is optimized using a quantile constraint method. Specifically, when the difference between the 90th and 10th percentiles is less than a desired threshold, the bandwidth is reduced by 15%; when the difference is greater than the desired threshold, the bandwidth is increased by 10%. The optimized bandwidths are 1.8 MW and 1.2 MW, respectively, improving estimation accuracy by 20% compared to the initial bandwidths.

[0088] During the dynamic update of the probability distribution, the spatiotemporal coupling characteristics, the combined kernel function, and the optimized bandwidth are substituted into the kernel density estimation formula to obtain the probability density function of the wind and solar power generation forecast error. A 3-hour sliding time window is used to update the forecast error samples. Newly entered samples are given a larger weight, and the time decay factor is set to 0.95, meaning that the sample weight decreases by 5% with each time interval. The updated sample weights are recursively calculated with the existing probability density function to achieve dynamic updates of the probability distribution. When the forecast error experiences a sudden change, such as a power fluctuation exceeding 20% ​​of the rated capacity, the sliding window length is shortened to 1 hour to speed up the distribution update.

[0089] Figure 2 The comparison chart of probability distribution estimation accuracy shows the estimation accuracy performance of the three methods in different local sample density ranges. The present invention adopts an adaptive combined kernel function strategy to achieve an estimation accuracy of 95% in high-density areas (0.8-1.0), maintains an accuracy of 90% in medium-density areas (0.4-0.8), and maintains an accuracy of 85% even in low-density areas (0-0.2). Although the traditional Gaussian kernel function method performs well in high-density areas (reaching 80%), its accuracy drops significantly in medium and low-density areas (75% and 65% respectively). Although the fixed weight method is slightly better than the single Gaussian kernel function method, its accuracy in each density area is generally lower than that of the present technical solution by about 15-20 percentage points. Especially when processing boundary areas (low-density areas), the present technical solution improves the estimation accuracy by about 20 percentage points compared with the traditional method by increasing the weight of the Epanechnikov kernel function, which fully demonstrates the superiority of the adaptive combined kernel function strategy.

[0090] Existing wind and solar power generation prediction error analysis methods ignore the spatiotemporal coupling relationship, do not consider the dynamic changes in data distribution characteristics, and are unable to quickly respond to sudden changes in prediction errors. The present invention proposes a wind and solar power generation prediction error probability distribution estimation method based on adaptive combined kernel density estimation. This method innovatively introduces a spatiotemporal coupling analysis framework, and comprehensively analyzes the spatiotemporal characteristics of the prediction errors of wind farms and photovoltaic power stations through a sliding time window, thereby improving the accuracy of feature extraction. In terms of kernel function selection, the Gaussian kernel function and the Epanechnikov kernel function are innovatively adaptively combined. The Gaussian kernel function is used for probability density estimation in the central area, and the Epanechnikov kernel function is used to model the boundary area. The two kernel functions complement each other through adaptive weight allocation based on local sample density. In terms of dynamic updating of probability distribution, a sample weight update mechanism and mutation response strategy based on the time attenuation factor are designed, and the system's response speed to sudden changes in prediction errors is improved by adaptively adjusting the sliding window length. The prediction error distribution estimation accuracy of the present invention is significantly improved, the computational efficiency is significantly improved, the spatiotemporal coupling analysis improves the feature extraction accuracy, the adaptive weight strategy optimizes the probability distribution estimation effect, the bandwidth optimization scheme improves the overall performance, and the mutation response speed is significantly accelerated, providing new ideas for the probabilistic modeling of wind and solar power generation prediction errors, and laying the foundation for improving the prediction accuracy of renewable energy power generation and the coordinated operation efficiency of the electric-hydrogen system.

[0091] In an optional embodiment,

[0092] A dual neural network is used to perform deep reinforcement learning. The action-value network dynamically adjusts the decision space based on an adaptive boundary function of the short-term fluctuations and long-term trends of wind and solar forecast errors. The target network assesses risk through a dual branch of conditional risk and state value, and constructs a risk constraint coefficient based on Lyapunov stability theory for decision-making. Risk constraints include:

[0093] Obtaining a system state vector, wherein the system state vector includes grid load demand data, hydrogen production power data, hydrogen storage capacity data, and hydrogen refueling demand data;

[0094] Inputting the system state vector into an action value network, the action value network calculating an adaptive boundary function based on a probability distribution of wind and solar power prediction errors, and dynamically adjusting a basic decision boundary according to the adaptive boundary function to obtain an adjusted decision space;

[0095] Inputting the system state vector and the adjusted decision space into a target network, the target network calculating a conditional risk value and a state value, and weighting the conditional risk value and the state value to obtain a risk metric value;

[0096] Constructing an optimization objective function including a system operation cost term, a risk metric term, and a renewable energy consumption term, and updating the action value network parameters based on the optimization objective function using a stochastic gradient descent method. The target network evaluates the decision risk based on the risk metric value, and periodically updates the network parameters using a preset soft update coefficient.

[0097] Perform iterative calculation of the value function based on the updated action value network parameters and risk metric values, and use the product of the iterative calculation result and the risk constraint coefficient as the risk constraint term. The risk constraint coefficient is obtained by constructing a Lyapunov function based on the quadratic term of the hydrogen storage deviation from the safe interval and the high-order term of the wind and solar power forecast error, and calculating the inverse of the stability margin through its time derivative;

[0098] Based on the adjusted decision space, a decision evaluation value considering the risk constraint item is calculated, and a decision corresponding to the maximum decision evaluation value is selected as a scheduling instruction.

[0099] Exemplarily, a system state vector is obtained, including grid load demand data, hydrogen production power data, hydrogen storage data, and hydrogen refueling demand data. Grid load demand data is obtained through a load forecasting system to obtain a 24-hour load forecast value. Hydrogen production power data is obtained based on the real-time operating status of the hydrogen production equipment. Hydrogen storage data is measured by a pressure sensor on the hydrogen storage tank. Hydrogen refueling demand data is predicted based on historical operating data of the hydrogen refueling station.

[0100] The system state vector is input into the action value network. The action value network first statistically analyzes the probability distribution characteristics of wind and solar power generation forecast errors based on historical data and calculates the upper and lower limits of the 95% confidence interval as an adaptive boundary function. Taking hydrogen production power as an example, when the wind and solar power generation forecast error is large, the adaptive boundary function appropriately narrows the decision range for hydrogen production power to avoid decision risks caused by forecast error. Specifically, if the standard deviation of the forecast error is 10%, the basic decision boundary for hydrogen production power is dynamically adjusted from 0-100% to 10%-90%.

[0101] The system state vector and the adjusted decision space are input into the target network. The target network consists of a conditional risk assessment branch and a state value assessment branch. The conditional risk assessment branch calculates the system's risk level based on conditions such as hydrogen storage capacity and refueling demand. For example, when hydrogen storage falls below the safe reserve limit by 20%, the risk value increases significantly. The state value assessment branch calculates the long-term benefits of each state. The outputs of the two branches are weighted by 0.3 and 0.7 to obtain a comprehensive risk metric.

[0102] An optimization objective function was constructed, encompassing system operating costs, risk metrics, and renewable energy consumption. Operating costs included the start-up and shutdown costs of the hydrogen production equipment and the grid's electricity purchase costs. Renewable energy consumption was calculated based on the predicted wind and solar power. Stochastic gradient descent was used to update the parameters of the action-value network, with the target network parameters updated every 100 training steps using a soft update coefficient of 0.01.

[0103] The value function is iterated to obtain the Q value. A Lyapunov function is constructed. When the hydrogen storage capacity deviates from the safe range of 35%-85%, the quadratic term increases; when the wind and solar power forecast error exceeds 10%, the higher-order terms increase. The time derivative of the Lyapunov function is calculated to obtain the stability margin, and its reciprocal is used as the risk constraint coefficient. The product of the Q value and the risk constraint coefficient is used as the risk constraint term.

[0104] Within the adjusted decision space, the risk constraint term is subtracted from the Q value to obtain the decision evaluation value. The decision corresponding to the maximum evaluation value is selected as the final dispatch instruction, including specific values ​​such as hydrogen production power and hydrogen storage release power.

[0105] The present invention dynamically adjusts the decision space through the adaptive boundary function of the action value network, effectively dealing with the uncertainty brought by wind and solar power generation prediction errors, and improving the robustness and reliability of decision-making; the target network adopts a dual-branch structure of conditional risk and state value to comprehensively evaluate system risks, and constructs a risk constraint mechanism based on Lyapunov stability theory to ensure safe and stable operation of the system; the optimization objective function comprehensively considers economy, safety and environmental protection, achieves economic and efficient operation while ensuring system safety, and improves the level of renewable energy absorption, which has important practical application value.

[0106] In an optional embodiment,

[0107] The adaptive boundary function includes:

[0108] The wind and solar forecast error sequence is decomposed into short-term fluctuation error and long-term trend error by discrete wavelet transform.

[0109] Calculating the boundary adjustment characteristics of the short-term fluctuation error based on the local probability density characteristics of the forecast error samples;

[0110] Construct a cumulative influence function, which is obtained by weighted accumulation of long-term trend errors in the historical period according to an exponential decay coefficient, and calculate a trend strength index based on the cumulative influence function. The trend strength index is the ratio of the absolute value of the cumulative influence function to the maximum absolute value of the cumulative influence function within the evaluation time window;

[0111] A probability confidence interval is constructed based on the boundary adjustment feature, and the boundary value of the probability confidence interval is mapped to a fluctuation boundary adjustment amount, and the fluctuation boundary adjustment amount is dynamically corrected based on the frequency of fluctuation direction changes; a piecewise linear mapping function is constructed based on the trend strength indicator to calculate the trend boundary adjustment amount, and the trend boundary adjustment amount is corrected according to the duration of the trend;

[0112] An adaptive combination weight is calculated based on the trend strength index, and the adaptive combination weight is exponentially related to the trend strength index; the fluctuation boundary adjustment amount and the trend boundary adjustment amount are weightedly combined using the adaptive combination weight to obtain the adaptive boundary function.

[0113] For example, a series of wind and solar power forecast errors is obtained and decomposed at multiple scales using a discrete wavelet transform. Using the db4 wavelet basis function, the error series is decomposed into four scales, yielding short-term fluctuation errors and long-term trend errors. Short-term fluctuation errors correspond to high-frequency components, while long-term trend errors correspond to low-frequency components. For example, decomposing the 15-minute forecast error series for a photovoltaic power station yields fluctuation components reflecting 5-minute, 15-minute, 30-minute, and 60-minute characteristics.

[0114] Next, we calculate the boundary adjustment characteristics of short-term fluctuation errors. Using kernel density estimation, we calculate the local probability density characteristics using a sliding window of 15 minutes. For each error sample within each time window, we estimate the probability density curve using a Gaussian kernel function and extract the 95% confidence interval as the boundary adjustment characteristic. For example, the boundary adjustment characteristic at a certain moment is ±10% of the rated capacity.

[0115] Then, a cumulative impact function is constructed. 24-hour historical data is selected, and long-term trend errors are weighted and accumulated using an exponential decay coefficient of 0.95. The cumulative impact function reflects the degree of impact of historical trend errors on the current boundary. A trend strength index is calculated based on the cumulative impact function, ranging from 0 to 1. For example, if the cumulative impact function is 8% of rated capacity at a certain moment and reaches a maximum of 10% within 24 hours, the trend strength index is 0.8.

[0116] Adaptive boundaries are further constructed. Fluctuation boundary adjustments are derived from probability confidence interval mappings, and compression corrections are applied when the frequency of fluctuation direction changes is high. Trend boundary adjustments are calculated using a piecewise linear mapping function, with larger adjustments occurring as the trend duration increases. The two boundary adjustments are weighted using an adaptive combination weight to obtain the final boundary. The combined weight is exponentially related to the trend strength indicator, with the trend boundary weight increasing as the trend becomes more significant. For example, if the trend strength at a certain moment is 0.8, the fluctuation boundary adjustment is 8%, and the trend boundary adjustment is 12%, with combined weights of 0.3 and 0.7, respectively, resulting in a final boundary adjustment of 10.8%.

[0117] Existing methods for adjusting the boundaries of wind and solar power prediction errors mainly use fixed boundaries or simple adaptive boundaries. The present invention proposes a method for constructing an adaptive boundary function, innovatively introduces a cumulative impact function to characterize the impact of historical trends, and designs a trend intensity index to quantitatively characterize trend characteristics. Fluctuation boundaries and trend boundaries are constructed respectively through probability confidence intervals and piecewise linear mappings, and adaptive weight combination is achieved based on the trend intensity index. The present invention achieves accurate characterization of the multi-scale characteristics of wind and solar power prediction errors, and significantly improves the adaptability and flexibility of boundary adjustment. Compared with existing technologies, the boundary prediction accuracy is improved, the dynamic tracking performance is enhanced, and the system responds more accurately to short-term fluctuations and long-term trends, providing strong support for the precise regulation of the electric hydrogen system.

[0118] In an optional embodiment,

[0119] The target network calculates a conditional risk value and a state value, and weights the conditional risk value and the state value to obtain a risk metric value, including:

[0120] Constructing a multi-dimensional state-coupled system energy function based on the system state vector and the adjusted decision space, wherein the system energy function is a nonlinear weighted combination of grid load demand data, hydrogen production power data, hydrogen storage data, and hydrogenation demand data and their corresponding reference values;

[0121] Construct the conditional risk branch and state value branch of the target network;

[0122] Input the adjusted decision space and historical decision sequence into the conditional risk branch, calculate the attention weight matrix based on the multi-level correlation degree between the system state vector and the historical state, and use the spatiotemporal convolution result of the attention weight matrix and the historical state as the conditional risk value;

[0123] Input the system state vector and the adjusted decision space into the state value branch, construct the posterior distribution of the state value using the probabilistic variational inference method, and extract the conditional expectation and uncertainty of the posterior distribution to obtain the state value;

[0124] Determining an adaptive time window length based on the system energy function, wherein the adaptive time window length is not less than a response time of the electric-hydrogen coupling system and not greater than a preset multiple of a time constant of the hydrogen storage system;

[0125] Obtaining a maximum value and a minimum value of the energy of the electric-hydrogen coupling system within an adaptive time window, and calculating a multi-scale stability index based on a difference between a current system energy value and the maximum value and the minimum value;

[0126] An adaptive weighting coefficient is constructed according to the multi-scale stability index, and the conditional risk value and the state value are nonlinearly weighted according to the adaptive weighting coefficient to obtain a risk measurement value.

[0127] For example, a multi-dimensional state-coupled system energy function is constructed based on the system state vector and the adjusted decision space. This energy function includes four dimensions: grid load demand data, hydrogen production power data, hydrogen storage data, and hydrogen refueling demand data. The difference between the data in each dimension and its corresponding reference value is calculated, and the system energy function is obtained through nonlinear weighted combination. The reference value can be obtained based on historical data statistics, such as the grid load demand reference value taking the historical average value for the same period, the hydrogen production power reference value taking 80% of the equipment rated power, the hydrogen storage reference value taking 50% of the hydrogen storage tank capacity, and the hydrogen refueling demand reference value taking 70% of the historical peak value.

[0128] A two-branch target network is constructed, consisting of a conditional risk branch and a state-value branch. The conditional risk branch uses a multi-head attention mechanism, taking the adjusted decision space and historical decision sequences as input. An attention weight matrix is ​​calculated based on the multi-level correlation between the system state vector and historical states. Specifically, the state vector is divided into three time scales: hourly, daily, and weekly. The similarity between the current state and historical states is calculated to obtain weight coefficients. A spatiotemporal convolution is performed on the historical states to extract spatiotemporal features, which are then multiplied by the attention weight matrix to obtain the conditional risk value.

[0129] The state value branch employs probabilistic variational inference to map the system state vector and the adjusted decision space into a latent variable space. A multi-layer neural network is used to construct the posterior distribution of the state value. The mean of the distribution is extracted as the conditional expectation, and the variance is used as the uncertainty measure. Together, these two constitute the state value.

[0130] The adaptive time window length is determined based on the system energy function. The start-up and shutdown time of water electrolysis hydrogen production equipment is typically on the order of minutes, while the charging and discharging process of the hydrogen storage system lasts for several hours. Therefore, the window length is set to a lower limit of 5 minutes and an upper limit of three times the time constant of the hydrogen storage system. Within the adaptive window, the maximum and minimum system energy values ​​are calculated, and a multi-scale stability index is calculated based on the normalized difference between the current energy value and the extreme value. This index reflects the stability of the system at different time scales.

[0131] An adaptive weighting coefficient is constructed based on the multi-scale stability index. When stability is good, the state value is weighted more heavily; when stability is poor, the conditional risk is weighted more heavily. A nonlinear function is used to map the stability index to the range of 0-1 to obtain the weighting coefficient. Finally, the conditional risk value and state value are nonlinearly combined according to the weighting coefficient to obtain the risk metric.

[0132] The present invention comprehensively characterizes the operating state of the electric-hydrogen coupling system by constructing a system energy function with multi-dimensional state coupling, thereby improving the accuracy and reliability of risk measurement. It adopts a dual-branch target network structure to evaluate system risk from two dimensions: conditional risk and state value, and adaptively adjusts weights based on system stability, making risk measurement more reasonable and scientific. It introduces an adaptive time window mechanism to automatically adjust the analysis scale according to the dynamic characteristics of the system, thereby improving the adaptability and robustness of risk measurement to dynamic changes in the system.

[0133] In an optional embodiment,

[0134] The risk constraint coefficients include:

[0135] The real-time value of hydrogen storage, the upper and lower limits of the hydrogen storage safety interval, and the wind and solar power prediction error sequence are obtained; the degree of deviation of the hydrogen storage from the safety interval is constructed as a quadratic term, and the wind and solar power prediction error value is constructed as a high-order term to form a Lyapunov function that characterizes the stability of the system; the stability margin is calculated based on the derivative of the Lyapunov function with respect to time, and the inverse of the stability margin is used as the risk constraint coefficient.

[0136] For example, a method for determining the risk constraint coefficient of a wind-solar hydrogen storage system first obtains the real-time value of the hydrogen storage tank through real-time monitoring, and at the same time determines the upper and lower thresholds of the hydrogen storage safety interval based on the system design parameters. The real-time value of the hydrogen storage amount is generally measured in standard cubic meters and is collected in real time by a hydrogen flow meter and pressure sensor. For example, the hydrogen storage tank capacity of a certain system is 1,000 standard cubic meters, the upper limit of the safety interval is 900 standard cubic meters, and the lower limit is 100 standard cubic meters. The wind-solar power prediction system obtains the predicted wind and solar output value for the future period, compares the predicted value with the actual value, and obtains a prediction error sequence. For example, the prediction error sequence of a wind-solar system for the next four hours is [-50kW, 30kW, -20kW, 40kW]. The deviation between the real-time value of the hydrogen storage amount and the upper and lower limits of the safety interval is squared to construct a quadratic term. When the hydrogen storage capacity is 600 standard cubic meters, the square of the deviation from the upper limit of 900 standard cubic meters is 90,000, and the square of the deviation from the lower limit of 100 standard cubic meters is 250,000. The wind and solar power forecast error sequence is processed to a higher order to construct higher-order terms. Each value in the forecast error sequence is raised to the fourth power to obtain a higher-order error sequence. The higher-order result of the above forecast error sequence is [6,250,000, 810,000, 160,000, 2,560,000]. A weighted combination of the quadratic term and the higher-order term forms a Lyapunov function that characterizes system stability. This function is differentiated to obtain the stability margin, and its inverse is then calculated as the risk constraint coefficient. For example, if the calculated stability margin at a certain moment is 0.8, the corresponding risk constraint coefficient is 1.25.

[0137] The present invention constructs a Lyapunov function through the quadratic term of the hydrogen storage deviation and the high-order term of the wind and solar power prediction error, comprehensively considering the key influencing factors of the system stability, making the risk constraint more accurate and reasonable; calculates the risk constraint coefficient based on the Lyapunov stability theory, organically combines the system dynamic characteristics and stability requirements, and ensures the theoretical basis and practicality of the constraint; adopts real-time monitoring of hydrogen storage and wind and solar power prediction error sequence as input, realizes real-time dynamic adjustment of risk constraints, and improves the safety and reliability of system operation.

[0138] In an optional embodiment,

[0139] Based on the dynamic consistency theory, the coupling strength of the distributed collaborative network is inversely mapped to the risk constraint coefficient, and the synchronization potential function is constructed by the deviation between the node state of the distributed collaborative network and the scheduling instruction and the state difference of the adjacent nodes.

[0140] Constructing a dynamic coupling strength based on the risk constraint coefficient, wherein the dynamic coupling strength is a ratio of the basic coupling strength to the risk adjustment term, wherein the risk adjustment term increases as the risk constraint coefficient increases, and the dynamic coupling strength also exhibits an exponential decay relationship with the state difference of adjacent nodes;

[0141] Constructing a synchronization potential function, the synchronization potential function including a scheduling instruction tracking term and a node coordination term, the scheduling instruction tracking term being the product of the square difference between the node state and the scheduling instruction and a first weight coefficient, and the node coordination term being the product of the square of the difference between adjacent node states and the dynamic coupling strength and a second weight coefficient;

[0142] Constructing a distributed control law based on the gradient of the synchronization potential function, the distributed control law including a scheduling tracking term and a coordinated adjustment term, the scheduling tracking term being the product of the deviation between the node state and the scheduling instruction and the first weight coefficient, and the coordinated adjustment term being the product of the weighted deviation between the node state and the adjacent node state and the second weight coefficient;

[0143] When the actual power generation of the wind farm and photovoltaic power station exceeds the predicted power generation, the hydrogen production power is increased based on the distributed control law; when the actual power generation is less than the predicted power generation, the hydrogen release power of the hydrogen storage system is increased based on the distributed control law, and the electric-hydrogen system is coordinated and regulated in real time.

[0144] For example, a distributed collaborative network control method based on dynamic consistency theory realizes real-time collaborative regulation of the electric-hydrogen system by constructing dynamic coupling strength and synchronization potential functions.

[0145] Construct dynamic coupling strength. The dynamic coupling strength is calculated by dividing the base coupling strength by the risk adjustment factor. The base coupling strength is set to a fixed value, such as 0.8. The risk adjustment factor is positively correlated with the risk constraint coefficient. For every 0.1 increase in the risk constraint coefficient, the risk adjustment factor increases by 0.2. As the difference in adjacent node states increases, the dynamic coupling strength decays exponentially. For example, when the difference in adjacent node states is 10%, the dynamic coupling strength decays to 0.9 times its original value.

[0146] Construct a synchronization potential function. The synchronization potential function consists of two terms: a scheduling instruction tracking term and a node coordination term. The scheduling instruction tracking term is the square of the deviation between the node state and the scheduling instruction multiplied by the first weight coefficient, which is set to 0.6. The node coordination term is the square of the difference between the states of adjacent nodes multiplied by the dynamic coupling strength and the second weight coefficient, which is set to 0.4.

[0147] A distributed control law is constructed based on the gradient of the synchronized potential function. The distributed control law includes a scheduling tracking term and a coordinated adjustment term. The scheduling tracking term is the deviation between the node state and the scheduling instruction multiplied by a first weight coefficient. The coordinated adjustment term is the weighted deviation between the node state and the states of adjacent nodes multiplied by a second weight coefficient.

[0148] Real-time coordinated control of the electric-hydrogen system is achieved. When the wind farm's actual power generation exceeds the predicted value by 10%, hydrogen production capacity increases by 8%. When the photovoltaic power plant's actual power generation falls below the predicted value by 15%, the hydrogen storage system's hydrogen discharge capacity increases by 12%. Distributed control laws enable coordinated regulation of each node, ensuring overall system synchronization.

[0149] The present invention realizes adaptive adjustment of system risk and control intensity through inverse mapping of dynamic coupling strength and risk constraint coefficient, thereby improving the safety and reliability of system operation; constructs distributed control law based on synchronization potential function, realizes accurate tracking of node status and scheduling instructions, and coordinated consistency between adjacent nodes, ensuring the synchronization stability of the system; applies distributed control to real-time coordinated regulation of electric hydrogen system, effectively smoothes fluctuations in renewable energy power generation, and improves the operating economy and flexibility of the system.

[0150] In an optional embodiment,

[0151] The synchronization potential function includes a fast response layer and a steady-state tracking layer, including:

[0152] The scheduling instruction tracking item of the rapid response layer adopts the product of the square difference between the node state and the scheduling instruction and the first weight coefficient, and the node coordination item of the rapid response layer adopts the product of the square of the difference between the adjacent node states and the dynamic coupling strength and the second weight coefficient; the rapid response layer triggers calculation updates based on the rate of change of the node state, and updates the first weight coefficient and the second weight coefficient when the rate of change of the node state exceeds a preset threshold;

[0153] The scheduling instruction tracking item of the steady-state tracking layer adopts the product of the cumulative square difference between the node state and the scheduling instruction and the first weight coefficient, wherein the cumulative square difference is the time integral of the square difference between the node state and the scheduling instruction within a fixed time window, and the length of the fixed time window is not less than 10 times the calculation period of the fast response layer and not more than one-third of the system characteristic time, wherein the system characteristic time is the dynamic response time of the electric hydrogen system; the node coordination item of the steady-state tracking layer adopts the product of the square of the adjacent node state difference, the dynamic coupling strength and the second weight coefficient;

[0154] An inter-layer coupling weight is constructed based on the calculation results of the fast response layer and the steady-state tracking layer. When the rate of change of the node state is greater than a first preset threshold, the weight of the fast response layer is increased. When the rate of change of the node state is less than a second preset threshold, the weight of the steady-state tracking layer is increased.

[0155] Exemplarily, the synchronization potential function includes a fast response layer and a steady-state tracking layer, and the coordinated control of the electric hydrogen system is achieved through a two-layer structure.

[0156] The implementation process of the rapid response layer is as follows: first, the node status and scheduling instructions are obtained. The squared difference between the node status and the scheduling instruction is calculated and multiplied by the first weight coefficient to obtain the scheduling instruction tracking term. Simultaneously, the squared difference between adjacent node states is calculated and multiplied by the dynamic coupling strength and the second weight coefficient to obtain the node coordination term. When the node status change rate exceeds 0.1, the first weight coefficient is updated to 1.2 times the original value, and the second weight coefficient is updated to 0.8 times the original value, achieving rapid response.

[0157] The steady-state tracking layer is implemented as follows: within a fixed time window (set to 100 seconds, which is 10 times longer than the 10-second calculation period of the rapid response layer and less than one-third of the 300-second system characteristic time), the squared difference between the node state and the scheduling instruction is accumulated. The accumulated value is multiplied by the first weight coefficient to obtain the scheduling instruction tracking term. The node coordination term is calculated in the same way as the rapid response layer.

[0158] The inter-layer coupling weights are constructed as follows: The node state change rate is monitored in real time. When the change rate is greater than 0.2, the weight of the fast response layer is increased to 0.8, while the weight of the steady-state tracking layer is reduced to 0.2. When the change rate is less than 0.05, the weight of the fast response layer is reduced to 0.3, while the weight of the steady-state tracking layer is increased to 0.7. Finally, the calculation results of the two layers are weighted to obtain the control output.

[0159] Figure 3 This is a performance comparison chart of the double-layer control of the present invention and other single-layer control. Figure 3 As shown, traditional electric hydrogen system control methods usually adopt a single control hierarchy structure, which makes it difficult to take into account both the system's rapid response capability and steady-state tracking accuracy. The existing technologies either have a fast response speed but poor steady-state accuracy, or have high steady-state accuracy but slow dynamic response, which cannot meet the dual requirements of the electric hydrogen system for rapid adjustment and stable operation. The present invention proposes a two-layer synchronous potential function structure, which innovatively divides the control into a rapid response layer and a steady-state tracking layer. The rapid response layer adopts a weight update mechanism based on the rate of change trigger to ensure the system's rapid response to disturbances; the steady-state tracking layer ensures control accuracy through the cumulative square difference and fixed time window design. Seamless switching of the two layers of control is achieved through dynamic adjustment of the inter-layer coupling weights. The present invention significantly improves the control performance of the electric hydrogen system and achieves the unity of rapid response and steady-state accuracy. The system response time is shortened, the steady-state control accuracy is improved, and the dynamic disturbance suppression capability is enhanced, providing effective guarantees for the efficient and coordinated operation of the electric hydrogen system. The response speed of the present invention is 70% higher than that of traditional single-layer steady-state control, the control accuracy is 87% higher than that of single-layer rapid control, the system stability is improved by 45%, and adaptive adjustment of the change rate within the range of 0.05-0.25 is supported.

[0160] According to a second aspect of the embodiments of the present invention,

[0161] Provides an electric-hydrogen coupled intelligent control system that takes into account wind and solar power prediction errors, including:

[0162] The first unit is used to collect historical power generation data of wind farms and photovoltaic power stations, calculate the spatiotemporal correlation characteristics of the prediction errors of wind farms and photovoltaic power stations, and use the adaptive combined kernel density estimation method of Gaussian kernel function and Epanechnikov kernel function to obtain the probability distribution of wind and solar power prediction errors;

[0163] The second unit is used to use the probability distribution as an uncertainty constraint, combined with grid load demand data and hydrogen storage system operation data, and perform deep reinforcement learning using a dual neural network. The action-value network dynamically adjusts the decision space based on an adaptive boundary function of the short-term fluctuations and long-term trends of wind and solar forecast errors. The target network assesses risk through a dual branch of conditional risk and state value, and constructs a risk constraint coefficient based on Lyapunov stability theory to constrain decision-making risk and obtain dispatch instructions.

[0164] The third unit is used to construct the electrolytic hydrogen production unit and the hydrogen storage system into a distributed collaborative network, map the coupling strength of the distributed collaborative network inversely to the risk constraint coefficient based on the dynamic consistency theory, and construct a synchronous potential function based on the deviation between the node state of the distributed collaborative network and the scheduling instruction and the state difference of the adjacent nodes; based on the deviation between the actual power generation power of the wind farm and the photovoltaic power station and the predicted power generation power, the hydrogen production power and hydrogen release power are controlled in real time along the gradient direction of the synchronous potential function.

[0165] According to a third aspect of the embodiments of the present invention,

[0166] An electronic device is provided, comprising:

[0167] processor;

[0168] a memory for storing processor-executable instructions;

[0169] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0170] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements 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. An electric-hydrogen coupled intelligent control method considering wind-solar forecast errors is characterized by: include: Historical power generation data from wind farms and photovoltaic power stations were collected, and the spatiotemporal correlation characteristics of their forecast errors were calculated. The probability distribution of wind and solar power forecast errors was obtained using an adaptive combined kernel density estimation method using the Gaussian kernel function and the Epanechnikov kernel function. Using the probability distribution as an uncertainty constraint, combined with grid load demand data and hydrogen storage system operation data, a dual neural network is used to perform deep reinforcement learning, where the action value network dynamically adjusts the decision space based on an adaptive boundary function of the short-term fluctuations and long-term trends of wind and solar forecast errors; The target network evaluates risk through dual branches of conditional risk and state value, and constructs risk constraint coefficients based on Lyapunov stability theory to perform decision-making risk constraints and obtain scheduling instructions, which specifically include: using dual neural networks to perform deep reinforcement learning, wherein the action value network dynamically adjusts the decision space based on the adaptive boundary function of the short-term fluctuation and long-term trend of wind and solar forecast errors; the target network evaluates risk through dual branches of conditional risk and state value, and constructs risk constraint coefficients based on Lyapunov stability theory to perform decision-making risk constraints; obtains the system state vector, which includes grid load demand data, hydrogen production power data, hydrogen storage data and hydrogen refueling demand data; inputs the system state vector into the action value network, and the action value network calculates the adaptive boundary function based on the probability distribution of wind and solar forecast errors, and dynamically adjusts the basic decision boundary according to the adaptive boundary function to obtain the adjusted decision space; the system state vector and the adjusted decision space are combined. Input the target network, the target network calculates the conditional risk value and the state value, and weights the conditional risk value and the state value to obtain the risk measurement value; construct an optimization objective function including the system operation cost item, the risk measurement item and the renewable energy consumption item, and adopt the stochastic gradient descent method to update the action value network parameters based on the optimization objective function, the target network evaluates the decision risk based on the risk measurement value, and periodically updates the network parameters through the preset soft update coefficient; perform iterative calculation of the value function according to the updated action value network parameters and the risk measurement value, and take the product of the iterative calculation result and the risk constraint coefficient as the risk constraint item, the risk constraint coefficient is obtained by constructing the Lyapunov function based on the quadratic term of the hydrogen storage deviation from the safe interval and the high-order term of the wind and solar prediction error and calculating the inverse of the stability margin through its time derivative; based on the adjusted decision space, calculate the decision evaluation value considering the risk constraint item, and select the decision corresponding to the maximum decision evaluation value as the scheduling instruction; The electrolytic hydrogen production unit and the hydrogen storage system are constructed into a distributed collaborative network. Based on the dynamic consistency theory, the coupling strength of the distributed collaborative network is inversely mapped to the risk constraint coefficient. The deviation between the node state of the distributed collaborative network and the scheduling instruction and the state difference of the adjacent nodes is used to construct a synchronization potential function. Based on the deviation between the actual power generation power of the wind farm and the predicted power generation power of the photovoltaic power station, the hydrogen production power and the hydrogen release power are controlled in real time along the gradient direction of the synchronization potential function. Specifically, the following steps are performed: based on the risk constraint coefficient, a dynamic coupling strength is constructed. The dynamic coupling strength is the ratio of the basic coupling strength to the risk adjustment item. The risk adjustment item increases with the increase of the risk constraint coefficient. The dynamic coupling strength also has an exponential decay relationship with the state difference of the adjacent nodes. A synchronization potential function is constructed. The synchronization potential function includes a scheduling instruction tracking item and a node collaboration item. The dispatch instruction tracking item is the product of the square difference between the node state and the dispatch instruction and the first weight coefficient, and the node coordination item is the product of the square of the difference between the adjacent node states and the dynamic coupling strength and the second weight coefficient; a distributed control law is constructed based on the gradient of the synchronization potential function, and the distributed control law includes a dispatch tracking item and a coordinated adjustment item, the dispatch tracking item is the product of the deviation between the node state and the dispatch instruction and the first weight coefficient, and the coordinated adjustment item is the product of the weighted deviation between the node state and the adjacent node state and the second weight coefficient; when the actual power generation power of the wind farm and the photovoltaic power station exceeds the predicted power generation power, the hydrogen production power is increased based on the distributed control law; when the actual power generation power is less than the predicted power generation power, the hydrogen release power of the hydrogen storage system is increased based on the distributed control law, and the electric hydrogen system is coordinated and regulated in real time.

2. The method according to claim 1, characterized in that The spatiotemporal correlation characteristics of wind farm and photovoltaic power station forecast errors are calculated, and the probability distribution of wind and solar power forecast errors is obtained by using the adaptive combined kernel density estimation method of Gaussian kernel function and Epanechnikov kernel function, including: Calculate the temporal and spatial correlation coefficients of the prediction error data of wind farms and photovoltaic power plants to obtain the spatiotemporal coupling characteristics; A Gaussian kernel function and an Epanechnikov kernel function are selected as basic kernel functions, the Gaussian kernel function is used for probability density estimation in the central area, and the Epanechnikov kernel function is used for probability density estimation in the boundary area; adaptive weights of the Gaussian kernel function and the Epanechnikov kernel function are calculated based on the local sample density, and the product of the Gaussian kernel function and the adaptive weight and the sum of the product of the Epanechnikov kernel function and the complement of the adaptive weight are used as the combined kernel function; The initial bandwidth is calculated based on the standard deviation and interquartile range of historical power generation data, and the initial bandwidth is adjusted with quantile constraints according to the empirical distribution function to obtain the optimized bandwidth; The spatiotemporal coupling characteristics, combined kernel function and optimized bandwidth are substituted into the kernel density estimation formula to obtain the probability density function of the wind and solar power generation prediction error; the prediction error samples are updated based on the sliding time window, the sample weights are calculated according to the time attenuation factor, and the sample weights and the probability density function are recursively updated to obtain a dynamically updated wind and solar power prediction error probability distribution.

3. The method according to claim 1, characterized in that The adaptive boundary function includes: The wind and solar forecast error sequence is decomposed into short-term fluctuation error and long-term trend error by discrete wavelet transform. Calculating the boundary adjustment characteristics of the short-term fluctuation error based on the local probability density characteristics of the forecast error samples; Construct a cumulative influence function, which is obtained by weighted accumulation of long-term trend errors in the historical period according to an exponential decay coefficient, and calculate a trend strength index based on the cumulative influence function. The trend strength index is the ratio of the absolute value of the cumulative influence function to the maximum absolute value of the cumulative influence function within the evaluation time window; A probability confidence interval is constructed based on the boundary adjustment feature, and the boundary value of the probability confidence interval is mapped to a fluctuation boundary adjustment amount, and the fluctuation boundary adjustment amount is dynamically corrected based on the frequency of fluctuation direction changes; a piecewise linear mapping function is constructed based on the trend strength indicator to calculate the trend boundary adjustment amount, and the trend boundary adjustment amount is corrected according to the duration of the trend; An adaptive combination weight is calculated based on the trend strength index, and the adaptive combination weight is exponentially related to the trend strength index; the fluctuation boundary adjustment amount and the trend boundary adjustment amount are weightedly combined using the adaptive combination weight to obtain the adaptive boundary function.

4. The method according to claim 1, wherein The target network calculates a conditional risk value and a state value, and weights the conditional risk value and the state value to obtain a risk metric value, including: Constructing a multi-dimensional state-coupled system energy function based on the system state vector and the adjusted decision space, wherein the system energy function is a nonlinear weighted combination of grid load demand data, hydrogen production power data, hydrogen storage data, and hydrogenation demand data and their corresponding reference values; Construct the conditional risk branch and state value branch of the target network; Input the adjusted decision space and historical decision sequence into the conditional risk branch, calculate the attention weight matrix based on the multi-level correlation degree between the system state vector and the historical state, and use the spatiotemporal convolution result of the attention weight matrix and the historical state as the conditional risk value; Input the system state vector and the adjusted decision space into the state value branch, construct the posterior distribution of the state value using the probabilistic variational inference method, and extract the conditional expectation and uncertainty of the posterior distribution to obtain the state value; Determining an adaptive time window length based on the system energy function, wherein the adaptive time window length is not less than a response time of the electric-hydrogen coupling system and not greater than a preset multiple of a time constant of the hydrogen storage system; Obtaining a maximum value and a minimum value of the energy of the electric-hydrogen coupling system within an adaptive time window, and calculating a multi-scale stability index based on a difference between a current system energy value and the maximum value and the minimum value; An adaptive weighting coefficient is constructed according to the multi-scale stability index, and the conditional risk value and the state value are nonlinearly weighted according to the adaptive weighting coefficient to obtain a risk measurement value.

5. The method according to claim 1, wherein The risk constraint coefficients include: The real-time value of hydrogen storage, the upper and lower limits of the hydrogen storage safety interval, and the wind and solar power prediction error sequence are obtained; the degree of deviation of the hydrogen storage from the safety interval is constructed as a quadratic term, and the wind and solar power prediction error value is constructed as a high-order term to form a Lyapunov function that characterizes the stability of the system; the stability margin is calculated based on the derivative of the Lyapunov function with respect to time, and the inverse of the stability margin is used as the risk constraint coefficient.

6. The method according to claim 1, characterized in that The synchronization potential function includes a fast response layer and a steady-state tracking layer, specifically including: The scheduling instruction tracking item of the rapid response layer adopts the product of the square difference between the node state and the scheduling instruction and the first weight coefficient, and the node coordination item of the rapid response layer adopts the product of the square of the difference between the adjacent node states and the dynamic coupling strength and the second weight coefficient; the rapid response layer triggers calculation updates based on the rate of change of the node state, and updates the first weight coefficient and the second weight coefficient when the rate of change of the node state exceeds a preset threshold; The scheduling instruction tracking item of the steady-state tracking layer adopts the product of the cumulative square difference between the node state and the scheduling instruction and the first weight coefficient, wherein the cumulative square difference is the time integral of the square difference between the node state and the scheduling instruction within a fixed time window, and the length of the fixed time window is not less than 10 times the calculation period of the fast response layer and not more than one-third of the system characteristic time, wherein the system characteristic time is the dynamic response time of the electric hydrogen system; the node coordination item of the steady-state tracking layer adopts the product of the square of the adjacent node state difference, the dynamic coupling strength and the second weight coefficient; An inter-layer coupling weight is constructed based on the calculation results of the fast response layer and the steady-state tracking layer. When the rate of change of the node state is greater than a first preset threshold, the weight of the fast response layer is increased. When the rate of change of the node state is less than a second preset threshold, the weight of the steady-state tracking layer is increased.

7. An electric-hydrogen coupled intelligent control system considering wind-solar forecast errors, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to collect historical power generation data of wind farms and photovoltaic power stations, calculate the spatiotemporal correlation characteristics of the prediction errors of wind farms and photovoltaic power stations, and use the adaptive combined kernel density estimation method of Gaussian kernel function and Epanechnikov kernel function to obtain the probability distribution of wind and solar power prediction errors; The second unit is configured to use the probability distribution as an uncertainty constraint, combine the grid load demand data and the hydrogen storage system operation data, and use a dual neural network to perform deep reinforcement learning, wherein the action value network dynamically adjusts the decision space based on an adaptive boundary function of the short-term fluctuation and long-term trend of the wind and solar forecast errors; The target network evaluates risk through the dual branches of conditional risk and state value, and constructs risk constraint coefficients based on Lyapunov stability theory to constrain decision-making risks and obtain scheduling instructions; The third unit is used to construct the electrolytic hydrogen production unit and the hydrogen storage system into a distributed collaborative network, map the coupling strength of the distributed collaborative network inversely to the risk constraint coefficient based on the dynamic consistency theory, and construct a synchronous potential function based on the deviation between the node state of the distributed collaborative network and the scheduling instruction and the state difference of the adjacent nodes; based on the deviation between the actual power generation power of the wind farm and the photovoltaic power station and the predicted power generation power, the hydrogen production power and hydrogen release power are controlled in real time along the gradient direction of the synchronous potential function.

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

Images