Optimization control method for parallel direct-current power supply system of high-capacity storage battery

Abstract

The invention discloses an optimization control method for a parallel DC power supply system of a large-capacity storage battery. The optimization control method comprises the following steps: S1, collecting real-time data of a battery unit and an external environment, and constructing a data set; s2, performing noise filtering, standardization processing and missing value filling on the data set, and identifying a data mode and periodic fluctuation; s3, using the adaptive cross-layer neural network model to carry out modeling on the parallel DC power supply system, and evaluating the health state of the battery in real time; s4, analyzing a causal relationship between the battery units by applying a causal inference technology, and constructing a causal inference model; s5, reversely deducing an optimal battery charging and discharging scheduling scheme by adopting a reverse modeling technology, and dynamically adjusting a battery scheduling strategy; and S6, applying a game theory and a coevolution optimization algorithm to establish a game model between the battery units to realize optimization of local charging and discharging resources. The adaptive neural network, causal inference and the like are adopted, and intelligent control over the large-capacity storage battery parallel direct-current power supply system is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of optimized control of large-capacity storage batteries, and particularly to an optimized control method for a parallel DC power supply system for large-capacity storage batteries. Background Art

[0002] With the continuous development of battery technology, especially the wide application of large-capacity storage batteries in fields such as energy storage, smart grid, electric vehicles, and renewable energy, how to optimize the charge and discharge control of battery systems has become a current research hotspot. As a common form of large-capacity storage batteries, the parallel DC power supply system has been widely used in these fields due to its superior power output ability and high energy density. However, with the increase in the number of battery units and the improvement of system complexity, how to efficiently control the charge and discharge process of battery units and ensure the high efficiency, stability, and long life of the system in different working environments has become an urgent problem to be solved.

[0003] Most traditional control methods for parallel DC power supply systems of large-capacity storage batteries rely on static model-based designs. These methods usually use a simple battery management system (BMS) to monitor basic parameters such as the voltage, current, and temperature of the battery, and perform simple charge and discharge control on the battery according to set rules. However, this control method based on empirical rules cannot consider the complex dynamic relationships between battery units and the influence of the external environment, and it is difficult to achieve collaborative optimization between battery units. Especially during the charging and discharging processes, due to differences in factors such as the health state, load demand, and temperature between battery units, problems of uneven energy distribution often occur, resulting in premature aging or damage of some battery units, thereby affecting the overall performance of the system.

[0004] In addition, existing control methods usually fail to effectively utilize real-time data for dynamic optimization, but instead perform control under fixed rules. This makes the system unable to make timely adjustments in the face of rapid changes in the state of battery units, uncertain environmental factors, load fluctuations, etc., resulting in low charge and discharge efficiency, increased energy consumption, and shortened battery life. Especially in complex application scenarios, traditional methods cannot flexibly adjust the charge and discharge strategies of the battery according to real-time data and actual requirements, and fail to fully exploit the potential of battery resources.

[0005] In recent years, with the continuous development of artificial intelligence, big data, and deep learning technologies, battery optimization control methods based on real-time data and intelligent algorithms have gradually attracted the attention of the academic and industrial communities. In particular, the introduction of emerging technologies such as adaptive cross-layer neural networks, causal inference, and game theory has provided new ideas for the intelligent optimization of battery systems. These new technologies can model and analyze the complex relationships between battery cells, timely identify the operating status and potential problems of the system, and then dynamically adjust the charging and discharging strategies to optimize battery performance and extend battery life.

[0006] However, there are still some obvious defects in the existing technologies. First, although deep learning and causal inference technologies have been applied to battery management systems, most methods still remain at the optimization of single batteries or simple parallel systems, lacking the comprehensive optimization ability for large-scale parallel DC power systems. These methods cannot fully model the complex interactions among multiple battery cells and fail to consider the cooperation and competition relationships between battery cells, resulting in local optimization strategies being difficult to achieve the global optimum of the system. In addition, the existing charging and discharging optimization methods usually have difficulty in dealing with sudden situations in the system in real time, such as a sharp decline in the battery health state and load demand fluctuations, and the robustness and adaptability of the system are poor.

[0007] In addition, most of the existing optimization methods based on game theory are limited to the analysis of static models and fail to fully consider the dynamic changes of battery cells during the charging and discharging process. Traditional game models usually assume that the cooperation and competition relationships between battery cells are fixed. However, in fact, under different working conditions, the behavior of battery cells will be affected by many factors, such as temperature, charging and discharging efficiency, remaining battery capacity, etc., and these factors may fluctuate greatly in the short term. Therefore, traditional game theory methods cannot accurately reflect the complex dynamic interactions between battery cells.

[0008] More importantly, the existing inverse modeling methods usually fail to fully consider the multi-level optimization requirements of battery systems. Most traditional optimization methods use single-objective optimization functions and ignore the mutual relationships among multiple optimization objectives, such as energy consumption, load balancing, and battery life. In this way, the optimization process often tends to a certain objective while ignoring the overall performance of the system. Therefore, it is difficult for existing technologies to perform comprehensive optimization under multiple constraints and achieve the global optimal control of battery systems.

[0009] Therefore, how to provide an optimization control method for a parallel DC power system of large-capacity storage batteries is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0010] An object of the present invention is to propose an optimized control method for a parallel DC power supply system for large-capacity storage batteries. The present invention adopts an adaptive cross-layer neural network model, causal inference technology, and game theory and co-evolution optimization algorithm to comprehensively optimize the charge and discharge control strategy of the parallel DC power supply system for large-capacity storage batteries. Through dynamic scheduling and real-time feedback optimization, it can effectively improve energy efficiency, extend battery life and reduce losses, and further optimize the resource allocation between battery units. This method overcomes the deficiencies of traditional battery management systems in terms of energy efficiency and scheduling strategies in large-scale parallel systems, has a high level of intelligence, and can effectively improve the stability, reliability and economy of the system.

[0011] The optimized control method for a parallel DC power supply system for large-capacity storage batteries according to an embodiment of the present invention includes the following steps:

[0012] S1. Collect real-time data of battery units and the external environment of the parallel DC power supply system, including battery voltage, current, temperature, charge and discharge state, battery health state and internal resistance, and construct a data set;

[0013] S2. Perform noise filtering, normalization processing and missing value filling on the data set, and identify data patterns and periodic fluctuations through time series analysis to generate a cleaned data set;

[0014] S3. Based on the cleaned data set, use an adaptive cross-layer neural network model to model the parallel DC power supply system, perform dynamic scheduling optimization through the battery layer, decision layer and global optimization layer, and evaluate the battery health state in real time and output a short-term charge and discharge control strategy;

[0015] S4. According to the short-term charge and discharge control strategy, apply causal inference technology to analyze the causal relationship between battery units, identify the influence of temperature and discharge mode on the battery health state, and construct a causal inference model;

[0016] S5. Combine the causal inference model, adopt reverse modeling technology to reverse deduce the optimal battery charge and discharge scheduling scheme, and dynamically adjust the battery scheduling strategy to output a globally optimal charge and discharge control scheme;

[0017] S6. According to the globally optimal charge and discharge control scheme, apply game theory and co-evolution optimization algorithm to establish a game model between battery units, and realize the optimization of local charge and discharge resources through the cooperation and competition relationship between battery units.

[0018] Optionally, the S2 specifically includes:

[0019] S21. Perform noise filtering on the data set, adopt a median filtering method to process each data item, set a filtering window, process each data point through a sliding window, calculate the median within the window and replace the original data point value;

[0020] S22. Standardize the dataset after noise filtering. Use the Z-Score standardization method to calculate the mean and standard deviation of the data, and standardize each data item.

[0021] S23. Fill in the missing values in the standardized dataset. For continuous data, use linear interpolation to fill in the missing values, and calculate the missing data through interpolation of the previous and subsequent data points. For discrete data, use mode interpolation to replace the missing values with the most frequently occurring value in the dataset.

[0022] S24. Conduct time series analysis. Analyze the periodic fluctuations of the dataset after filling in the missing values, use the autoregressive moving average method to model the data, and identify the trends and periodic fluctuations in the data.

[0023] S25. Based on the results of the time series analysis, calculate the residual sequence in the data. Determine the goodness of fit by analyzing the standard deviation of the residual sequence. If the residual sequence conforms to the normal distribution, it is considered that the time series analysis is correct and the data pattern has been effectively identified; otherwise, if the data pattern is not effectively identified, it is discarded to obtain the finally screened dataset.

[0024] S26. Clean the finally screened dataset to generate a cleaned dataset.

[0025] Optionally, the specific steps of S3 are as follows:

[0026] S31. Based on the cleaned dataset, construct an adaptive cross-layer neural network model, which includes a battery layer, a decision layer, and a global optimization layer.

[0027] S32. The input of the battery layer is the real-time data of the battery unit. Use a multi-layer perceptron to process the input data and calculate the battery health state evaluation value:

[0028]

[0029] where S battery represents the battery health state evaluation value, w i represents the weight of the input feature of the i-th battery unit, X i represents the input feature of the i-th battery unit, b represents the bias term, and n represents the total number of battery units.

[0030] S33. Input the battery health state evaluation value S battery of the battery layer into the decision layer, analyze the correlation and interaction between batteries, and calculate the correlation weight of each battery unit by the method of weighted average.

[0031] S34. At the global optimization layer, dynamic scheduling optimization is performed in combination with the overall optimization goal of the parallel DC power supply system. The optimization objective function is:

[0032]

[0033] Among them, L represents the objective function, λ 1 , λ 2 and λ 3 represent weight coefficients, P charge,i represents the charging power of the i-th battery cell, P discharge,i represents the discharging power of the i-th battery cell, C discharge,i represents the discharging capacity of the i-th battery cell, D cycle,i represents the number of charge-discharge cycles of the i-th battery cell, L i represents the load of the i-th battery cell, represents the total load of all battery cells;

[0034] S35. Use the adaptive optimization algorithm for global optimization. The weight update during the optimization process is carried out by the gradient descent method:

[0035]

[0036] Among them, Δθ represents the parameter update amount of the adaptive cross-layer neural network model, η represents the learning rate, represents the gradient of the loss function with respect to the parameters of the adaptive cross-layer neural network model;

[0037] S36. Perform real-time scheduling according to the optimized parameters of the adaptive cross-layer neural network model and output the short-term charge-discharge control strategy.

[0038] Optionally, the S4 specifically includes:

[0039] S41. According to the short-term charge-discharge control strategy, apply causal inference technology to analyze the causal relationship between battery cells and identify the mutual influence between temperature, discharge mode and battery health status;

[0040] S42. Based on the causal inference framework, construct a causal reasoning model between battery cells, and use the autoregressive conditional heteroskedasticity model for modeling to describe the dynamic relationship between battery health status, temperature and discharge mode. The formula of the autoregressive conditional heteroskedasticity model is:

[0041]

[0042] Among them, SOH t represents the battery health status at time t, β 0 represents the constant term, p represents the lag period of the health status, β iRepresents the state of health (SOH) coefficient of the battery t-i Represents the state of health of the battery at time t-i, q represents the number of lag periods of temperature, θ j Represents the coefficient of the influence of temperature on the state of health of the battery, T t-j Represents the temperature data at time t-j, r represents the number of lag periods of the discharge mode, λ k Represents the coefficient of the influence of the discharge mode on the state of health, P t-k Represents the discharge mode data at time t-k, ∈ t Represents the error term;

[0043] S43. Optimize the parameters of the autoregressive conditional heteroskedasticity model through the maximum likelihood estimation method to determine the causal relationship between battery cells. The maximum likelihood estimation optimization objective function is:

[0044]

[0045] Wherein, Represents the optimized model parameters, T represents the total number of samples, p(SOH t |T t ,P t ,θ) represents the conditional probability of the state of health of the battery at time t under the temperature T at time t and the discharge mode data P at time t under the model parameters θ before optimization t and the discharge mode data P at time t t Under the condition, p(∈ t |θ) represents the probability density function of the error term;

[0046] S44. Further quantify the dynamic causal relationship between the state of health of the battery and temperature and discharge mode, and use a vector autoregressive model for modeling:

[0047]

[0048] Wherein, A y Represents the coefficient matrix of each lag period k, n represents the maximum lag period, SOH t-y Represents the state of health of the battery at time t-y, T t-y Represents the temperature data at time t-y, P t-y Represents the discharge mode data at time t-y;

[0049] S45. Use directed transfer entropy to analyze the dynamic causal relationship:

[0050]

[0051] Wherein, TE(Y→X) represents the transfer entropy from variable Y to variable X, T 1 Represents the length of the time series, xt The variable X at time t, p(x t |y t-1 ) represents the conditional probability of the state x t-1 occurring at the previous moment y t , and p(x t ) represents the marginal probability of x t ;

[0052] S46. Identify the dominant effects of temperature changes and discharge patterns on the battery health state based on the results of directional transfer entropy analysis.

[0053] Optionally, the S5 specifically includes:

[0054] S51. Combine the causal inference model and adopt reverse modeling technology to reverse-derive the optimal battery charge and discharge scheduling scheme based on the current battery health state and the parallel DC power supply system target;

[0055] S52. Define the optimization objective function during the reverse modeling process:

[0056]

[0057] where O represents the optimization objective function, n represents the total number of battery cells, E i represents the energy consumption of the i-th battery cell, L i represents the life loss of the i-th battery cell, T i represents the temperature fluctuation of the i-th battery cell, α i , β i and γ i represent the adjustment coefficients;

[0058] S53. Use the simulated annealing algorithm for global optimization, reverse-derive the battery charge and discharge strategy based on the optimization objective function, gradually optimize the scheme through the temperature decreasing strategy, update the battery charge and discharge parameters, and perform optimization search according to the following state transition equation:

[0059] u t+1 = u t + Δu·P(ψ);

[0060] where u t+1 represents the system state at time t + 1, u t represents the system state at time t, Δu represents the system state change amount, P(ψ) represents the temperature-dependent probability distribution function, and ψ represents the temperature parameter in the simulated annealing process;

[0061] S54. According to the battery health state information feedback during the reverse modeling process, perform real-time dynamic adjustment on the charge and discharge strategy, and continuously optimize the battery scheduling using the new feedback information.

[0062] S55. Output a globally optimal charge-discharge control scheme through the reverse modeling technology and the dynamic adjustment mechanism.

[0063] Optionally, the S6 specifically includes:

[0064] S61. According to the globally optimal charge-discharge control scheme, apply the game theory and co-evolution optimization algorithm to establish a game model among battery cells, define each battery cell as a game participant, set the charge-discharge strategies of the battery cells as optional actions, and determine the profit function of each battery cell.

[0065] S62. In the game model, set the cooperation and competition relationships among battery cells, define the profit function of the battery cells, and construct the profit function with the state of health of the battery, remaining power, charge-discharge efficiency, and energy consumption as input variables:

[0066]

[0067] where U i represents the profit of the i-th battery cell, R i represents the remaining power of the i-th battery cell, C i represents the maximum charge-discharge capacity of the i-th battery cell, SOH i represents the state of health of the battery of the i-th battery cell, E i represents the charge-discharge energy consumption of the i-th battery cell, and ξ, ζ, and τ represent adjustment coefficients;

[0068] S63. Through the Nash equilibrium analysis of game theory, determine the optimal strategies among battery cells and solve the strategy combination of each battery cell in the game model.

[0069] S64. Through the co-evolution optimization algorithm, perform multi-generation iteration on the game model, evaluate the strategy combinations among battery cells in each generation, adjust the strategies of the battery cells, and output a new charge-discharge control strategy.

[0070] S65. Finally, realize the optimization of local charge-discharge resources and output a locally optimized charge-discharge resource allocation scheme.

[0071] The beneficial effects of the present invention are as follows:

[0072] First of all, by adopting the adaptive cross-layer neural network model to model the parallel DC power supply system, the present invention can dynamically adjust the charge-discharge strategies of battery cells and perform comprehensive optimization according to factors such as the real-time state of health, temperature, and load of the battery. Compared with the traditional control method based on a static model, the present invention can more accurately reflect the complex interaction relationships among battery cells and achieve the global optimal scheduling of the battery system.

[0073] Secondly, the present invention analyzes the causal relationship between battery cells through causal inference technology, identifies the impact of factors such as temperature and discharge mode on the battery health state, and further optimizes the battery management system. Through reverse modeling technology, combined with the battery health state and system objectives, the optimal charge-discharge scheduling scheme is deduced backward, and the charge-discharge strategy of the battery is dynamically adjusted. This process can quickly respond when the battery state changes, thus ensuring the long-term stability and high efficiency of the battery. Compared with the prior art, the present invention can respond to the changes in the battery health state and load fluctuations in real time, significantly improving the robustness and adaptability of the system.

[0074] In addition, the present invention establishes a game model between battery cells and adopts game theory and co-evolution optimization algorithm to optimize the local charge-discharge resources through the cooperation and competition relationship between battery cells. The behavior of battery cells in the game model is dynamically adjusted to maximize the charge-discharge efficiency and battery life of the entire system. Through this dynamic optimization method based on game theory, the present invention can effectively avoid the problem of uneven energy distribution between battery cells, prevent some battery cells from being damaged prematurely due to overcharge and overdischarge, and thus extend the service life of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0076] Figure 1 is the overall flowchart of the optimization control method for the parallel DC power supply system for large-capacity storage batteries proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0077] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0078] Refer to Figure 1 , the optimization control method for the parallel DC power supply system for large-capacity storage batteries includes the following steps:

[0079] S1. Collect real-time data of the battery cells and the external environment of the parallel DC power supply system, including battery voltage, current, temperature, charge-discharge state, battery health state, and internal resistance, and construct a data set;

[0080] S2. Perform noise filtering, normalization processing, and missing value filling on the data set, and identify data patterns and periodic fluctuations through time series analysis to generate a cleaned data set;

[0081] S3. Based on the cleaned dataset, use an adaptive cross-layer neural network model to model the parallel DC power supply system, perform dynamic scheduling optimization through the battery layer, decision layer, and global optimization layer, evaluate the battery health state in real time, and output a short-term charge and discharge control strategy;

[0082] S4. According to the short-term charge and discharge control strategy, apply causal inference technology to analyze the causal relationship between battery cells, identify the impact of temperature and discharge mode on the battery health state, and construct a causal inference model;

[0083] S5. Combine the causal inference model, use reverse modeling technology to reverse-derive the optimal battery charge and discharge scheduling scheme, dynamically adjust the battery scheduling strategy, and output the global optimal charge and discharge control scheme;

[0084] S6. According to the global optimal charge and discharge control scheme, apply game theory and co-evolution optimization algorithm to establish a game model between battery cells, and realize the optimization of local charge and discharge resources through the cooperation and competition relationship between battery cells.

[0085] In this embodiment, the S2 specifically includes:

[0086] S21. Perform noise filtering on the dataset, use the median filtering method to process each data item, set the filtering window, process each data point through the sliding window, calculate the median within the window, and replace the original data point value;

[0087] S22. Perform standardization processing on the dataset after noise filtering, use the Z-Score standardization method, calculate the mean and standard deviation of the data, and standardize each item of data;

[0088] S23. Fill in the missing values in the standardized dataset. For continuous data, use the linear interpolation method to fill in the missing values, and calculate and fill in the missing data through the interpolation of the front and back data points; for discrete data, use the mode interpolation method to replace the missing values with the most frequently occurring values in the dataset;

[0089] S24. Perform time series analysis, analyze the periodic fluctuations of the dataset after filling in the missing values, use the autoregressive moving average method to model the data, and identify the trends and periodic fluctuations in the data;

[0090] S25. Based on the results of the time series analysis, calculate the residual sequence in the data, determine the goodness of fit by analyzing the standard deviation of the residual sequence. If the residual sequence conforms to the normal distribution, it is considered that the time series analysis is correct and the data pattern has been effectively identified; otherwise, if the data pattern has not been effectively identified, it is discarded to obtain the finally screened dataset;

[0091] S26. Clean the finally screened data set to generate a cleaned data set.

[0092] In this embodiment, the S3 specifically includes:

[0093] S31. Based on the cleaned data set, construct an adaptive cross-layer neural network model, which includes a battery layer, a decision layer, and a global optimization layer;

[0094] S32. The input of the battery layer is the real-time data of battery cells. Use a multi-layer perceptron to process the input data and calculate the battery health state evaluation value:

[0095]

[0096] where S battery represents the battery health state evaluation value, w i represents the weight of the input feature of the i-th battery cell, X i represents the input feature of the i-th battery cell, b represents the bias term, and n represents the total number of battery cells;

[0097] S33. Input the battery health state evaluation value S battery of the battery layer into the decision layer, analyze the correlation and interaction between batteries, and calculate the correlation weight of each battery cell by the method of weighted average;

[0098] S34. In the global optimization layer, perform dynamic scheduling optimization in combination with the overall optimization goal of the parallel DC power supply system. The optimization objective function is:

[0099]

[0100] where L represents the objective function, λ 1 , λ 2 and λ 3 represent the weight coefficients, P charge,i represents the charging power of the i-th battery cell, P discharge,i represents the discharging power of the i-th battery cell, C discharge,i represents the discharging capacity of the i-th battery cell, D cycle,i represents the charge-discharge cycle number of the i-th battery cell, L i represents the load of the i-th battery cell, represents the total load of all battery cells;

[0101] S35. Use an adaptive optimization algorithm for global optimization. The weight update in the optimization process is performed by the gradient descent method:

[0102]

[0103] where, Δθ represents the parameter update amount of the adaptive cross-layer neural network model, and η represents the learning rate, represents the gradient of the loss function with respect to the parameters of the adaptive cross-layer neural network model;

[0104] S36. Perform real-time scheduling according to the optimized parameters of the adaptive cross-layer neural network model, and output a short-term charge and discharge control strategy.

[0105] In this embodiment, the S4 specifically includes:

[0106] S41. According to the short-term charge and discharge control strategy, apply causal inference technology to analyze the causal relationship between battery cells, and identify the mutual influence between temperature, discharge mode and battery health state;

[0107] S42. Based on the causal inference framework, construct a causal inference model between battery cells, and use an autoregressive conditional heteroskedasticity model for modeling to describe the dynamic relationship between battery health state, temperature and discharge mode. The formula of the autoregressive conditional heteroskedasticity model is:

[0108]

[0109] where, SOH t represents the battery health state at time t, β 0 represents the constant term, p represents the lag period of the health state, β i represents the battery health state coefficient, SOH t-i represents the battery health state at time t - i, q represents the lag period of temperature, θ j represents the coefficient of the influence of temperature on the battery health state, T t-j represents the temperature data at time t - j, r represents the lag period of the discharge mode, λ k represents the coefficient of the influence of the discharge mode on the health state, P t-k represents the discharge mode data at time t - k, ∈ t represents the error term;

[0110] S43. Optimize the parameters of the autoregressive conditional heteroskedasticity model by the maximum likelihood estimation method to determine the causal relationship between battery cells. The maximum likelihood estimation optimization objective function is:

[0111]

[0112] where, represents the optimized model parameters, T represents the total number of samples, p(SOH t |T t ,P t, θ) represents the temperature T of the battery health state at time t under the model parameters θ before optimization t and the discharge mode data P at time t t under the condition, the conditional probability, p(∈ t | θ) represents the probability density function of the error term;

[0113] S44. Further quantify the dynamic causal relationship between the battery health state, temperature, and discharge mode, and use a vector autoregressive model for modeling:

[0114]

[0115] where, A y represents the coefficient matrix for each lag period k, n represents the maximum lag period, SOH t-y represents the battery health state at time t - y, T t-y represents the temperature data at time t - y, P t-y represents the discharge mode data at time t - y;

[0116] S45. Use directed transfer entropy to analyze the dynamic causal relationship:

[0117]

[0118] where, TE(Y→X) represents the transfer entropy from variable Y to variable X, T 1 represents the length of the time series, x t represents variable X at time t, p(x t |y t-1 ) represents the conditional probability of the occurrence of state x t-1 at the previous moment y t occurring, p(x t ) represents the marginal probability of x t ;

[0119] S46. According to the analysis results of directed transfer entropy, identify the dominant roles of temperature changes and discharge mode on the battery health state.

[0120] In this embodiment, the S5 specifically includes:

[0121] S51. Combine the causal inference model and use reverse modeling technology to reverse-derive the optimal battery charge and discharge scheduling scheme based on the current battery health state and the parallel DC power supply system target;

[0122] S52. During the reverse modeling process, define the optimization objective function:

[0123]

[0124] Among them, O represents the optimization objective function, n represents the total number of battery cells, and E i represents the energy consumption of the i-th battery cell, and L i represents the life loss of the i-th battery cell, and T i represents the temperature fluctuation of the i-th battery cell, and α i , β i and γ i represent adjustment coefficients;

[0125] S53. Perform global optimization using the simulated annealing algorithm, reverse-derive the battery charge and discharge strategy based on the optimization objective function, gradually optimize the scheme through the temperature decreasing strategy, update the battery charge and discharge parameters, and perform optimization search according to the following state transition equation:

[0126] u t+1 = u t + Δu·P(ψ);

[0127] where, u t+1 represents the system state at time t + 1, u t represents the system state at time t, Δu represents the system state change amount, P(ψ) represents the temperature-dependent probability distribution function, and ψ represents the temperature parameter in the simulated annealing process;

[0128] S54. According to the battery health state information fed back during the reverse modeling process, perform real-time dynamic adjustment on the charge and discharge strategy, and continuously optimize the battery scheduling using the new feedback information;

[0129] S55. Output the globally optimal charge and discharge control scheme through the reverse modeling technology and the dynamic adjustment mechanism.

[0130] In this embodiment, the S6 specifically includes:

[0131] S61. According to the globally optimal charge and discharge control scheme, apply the game theory and co-evolution optimization algorithm to establish a game model among the battery cells, define each battery cell as a game participant, set the charge and discharge strategy of the battery cell as an optional action, and determine the revenue function of each battery cell;

[0132] S62. In the game model, set the cooperation and competition relationships among the battery cells, define the revenue function of the battery cell, and construct the revenue function with the battery health state, remaining power, charge and discharge efficiency, and energy consumption as input variables:

[0133]

[0134] where, U i represents the revenue of the i-th battery cell, R i represents the remaining power of the i-th battery cell, Ci represents the maximum charge and discharge capacity of the ith battery cell, SOH i represents the battery health status of the ith battery cell, E i represents the charging and discharging energy consumption of the i-th battery cell, ξ, ζ and τ represent the regulation coefficients;

[0135] S63. Determine the optimal strategy between battery cells through Nash equilibrium analysis of game theory, and solve the strategy combination of each battery cell in the game model;

[0136] S64, iterating the game model for multiple generations through the co-evolutionary optimization algorithm, evaluating the strategy combination between each generation of battery cells, adjusting the strategy of the battery cells and outputting a new charge and discharge control strategy;

[0137] S65. Finally, local charging and discharging resource optimization is achieved, and a local optimized charging and discharging resource allocation plan is output.

[0138] Embodiment 1:

[0139] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the optimization control of a large-capacity battery parallel DC power supply system. The system is widely used in scenarios such as electric vehicle charging stations, solar energy storage systems, and large-scale factory backup power supplies. In these application scenarios, the battery charging and discharging strategy directly affects the stability, energy efficiency, and battery life of the system, so reasonable battery charging and discharging control is crucial.

[0140] In actual scenarios, due to the complex and changeable battery usage environment, the health status of the battery will be affected by multiple factors, such as temperature changes, battery charge and discharge frequency, discharge mode, etc. If the battery charge and discharge strategy cannot be monitored and adjusted in real time, it may cause the battery to over-discharge or over-charge, thereby shortening the battery life and even causing system failure. Therefore, optimizing the battery charge and discharge strategy, especially the dynamic scheduling in the parallel DC power supply system, is the key to improving system efficiency, extending battery life and reducing energy consumption.

[0141] In the actual application of an electric vehicle charging station, the site is equipped with dozens of large-capacity parallel battery cells, which are mainly used to store electricity from the power grid and release electricity during peak loads. The total capacity of the battery is 500kWh. Due to the performance differences of each cell in the battery pack, traditional charging and discharging control methods are difficult to achieve optimal results. Traditional methods mainly rely on fixed charging power and discharging power. This method ignores factors such as battery health status and temperature changes, causing some battery cells to age prematurely while other battery cells fail to perform at their best.

[0142] After applying the optimization control method in the present invention, we first collect the real-time data of the battery unit, including battery voltage, current, temperature, charge and discharge status, etc. These data are fed back to the control system in real time through sensors. In the data processing stage, techniques such as noise filtering, normalization processing, and missing value filling are used to ensure the accuracy and reliability of the data. Then, an adaptive cross-layer neural network model is used to model the battery system, and the charge and discharge strategy is optimized by evaluating the battery health state in real time.

[0143] In this embodiment, we monitored the performance of the system for one month, and the monitoring data is shown in the following table:

[0144] Table 1 Real-time Monitoring Data of Battery Unit and Optimization Effect of Charge and Discharge

[0145]

[0146] The above Table 1 shows the data such as the battery health state evaluation value, charge and discharge power, charge and discharge cycle number, and battery life loss before and after implementing the optimization control method of the present invention. Through this method, we can adjust the charge and discharge strategy of the battery in real time, and dynamically optimize the charge and discharge process according to factors such as the battery health state and temperature.

[0147] Before implementation, the average battery health state evaluation value was 0.93, and the life loss of most battery units was about 2.5 years. After implementing the optimization control of the present invention, the battery health state evaluation value of the battery unit generally increased, and the average value reached 0.96, and the battery life loss was significantly reduced. After 30 days of optimized scheduling, the energy efficiency of the overall system increased by about 12%, and the battery life loss of the battery unit decreased by 15%.

[0148] Through this dynamic optimization control, not only the overall performance of the battery system is improved, the service life of the battery is extended, but also unnecessary energy waste is effectively reduced, and the energy efficiency of the system is significantly improved. During the actual operation process, the load scheduling of the system is more flexible, and it can cope with the fluctuations of different load demands, making the battery charge and discharge process more balanced and efficient.

[0149] Therefore, the method of the present invention shows great advantages in solving the performance degradation problems caused by factors such as temperature, discharge mode, and health state during the battery charge and discharge process. Through optimization control, the service life of the battery is extended, the overall efficiency of the system is improved, and the balance of the battery pack is also improved, which can effectively reduce the risks caused by battery imbalance.

[0150] Through the application of this solution, not only the performance of the battery unit is improved, but also important technical support is provided for the operation and maintenance management of the entire battery pack, providing a more intelligent and sustainable solution for the battery management system.

[0151] As described above, it is only the preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered within the protection scope of the present invention.

Claims

1. An optimization control method for a parallel DC power supply system of large-capacity batteries, characterized in that: The steps include: S1. Collect real-time data of the battery cells and external environment of the parallel DC power system, including battery voltage, current, temperature, charge and discharge status, battery health status and internal resistance, and build a data set; S2, filter the data set for noise, standardize it, and fill in missing values, and identify data patterns and periodic fluctuations through time series analysis to generate a cleaned data set; S3. Based on the cleaned data set, an adaptive cross-layer neural network model is used to model the parallel DC power supply system, and dynamic scheduling optimization is performed through the battery layer, decision layer and global optimization layer to evaluate the battery health status in real time and output a short-term charge and discharge control strategy; S4. According to the short-term charge and discharge control strategy, causal inference technology is applied to analyze the causal relationship between battery cells, identify the impact of temperature and discharge mode on battery health status, and build a causal inference model; S5. Combining the causal reasoning model, using the inverse modeling technology to reversely derive the optimal battery charging and discharging scheduling plan, and dynamically adjust the battery scheduling strategy to output the global optimal charging and discharging control plan; S6. According to the global optimal charge and discharge control scheme, game theory and collaborative evolution optimization algorithm are applied to establish a game model between battery cells, and local charge and discharge resource optimization is achieved through the cooperation and competition relationship between battery cells.

2. The optimization control method for a parallel DC power supply system for large-capacity batteries according to claim 1, characterized in that: The S2 specifically includes: S21, perform noise filtering on the data set, use the median filter method to process each data item, set the filter window, process each data point through the sliding window, calculate the median value in the window and replace the original data point value; S22, standardizing the noise-filtered data set, using the Z-Score standardization method, calculating the mean and standard deviation of the data, and standardizing each item of data; S23, fill missing values ​​in the standardized data set. For continuous data, linear interpolation is used to fill missing values, and missing data are filled by interpolation calculation of previous and next data points; for discrete data, mode interpolation is used to replace missing values ​​with the most frequently occurring values ​​in the data set; S24. Perform time series analysis to analyze the periodic fluctuations of the data set after filling in missing values, use the autoregressive moving average method to model the data, and identify trends and periodic fluctuations in the data; S25, based on the result of the time series analysis, the residual sequence in the data is calculated, and the goodness of fit is determined by analyzing the standard deviation of the residual sequence. If the residual sequence conforms to the normal distribution, it is considered that the time series analysis is correct and the data pattern has been effectively identified; otherwise, the data pattern is discarded if it is not effectively identified, and the final screened data set is obtained; S26. Clean the final screened data set to generate a cleaned data set.

3. The optimization control method for a parallel DC power supply system for large-capacity batteries according to claim 1, characterized in that: The S3 specifically includes: S31, constructing an adaptive cross-layer neural network model based on the cleaned data set, including a battery layer, a decision layer and a global optimization layer; S32, the battery layer input is the real-time data of the battery unit, and the input data is processed using a multi-layer perceptron to calculate the battery health status assessment value: Among them, S battery Represents the battery health status assessment value, w i represents the weight of the input feature of the ith battery cell, X i represents the input feature of the i-th battery cell, b represents the bias term, and n represents the total number of battery cells; S33, the battery health status evaluation value S of the battery layer battery Input decision layer, analyze the correlation and interaction between batteries, and calculate the correlation weight of each battery unit by weighted average method; S34. In the global optimization layer, dynamic scheduling optimization is performed in combination with the overall optimization goal of the parallel DC power supply system. The optimization objective function is: Among them, L represents the objective function, λ1, λ2 and λ3 represent weight coefficients, and P charge,i represents the charging power of the ith battery cell, P discharge,i represents the discharge power of the ith battery cell, C discharge,i represents the discharge capacity of the ith battery cell, D cycle,i represents the number of charge and discharge cycles of the ith battery cell, L i represents the load of the ith battery cell, Indicates the total load of all battery cells; S35. Use the adaptive optimization algorithm for global optimization. The weight update in the optimization process is performed by the gradient descent method: Among them, Δθ represents the parameter update amount of the adaptive cross-layer neural network model, η represents the learning rate, Represents the gradient of the loss function to the parameters of the adaptive cross-layer neural network model; S36. Perform real-time scheduling based on the optimized adaptive cross-layer neural network model parameters and output a short-term charging and discharging control strategy.

4. The optimization control method for a parallel DC power supply system for large-capacity batteries according to claim 1, characterized in that: The S4 specifically includes: S41, according to the short-term charge and discharge control strategy, applying causal inference technology to analyze the causal relationship between battery cells, and identifying the mutual influence between temperature, discharge mode and battery health status; S42. Based on the causal inference framework, a causal inference model between battery cells is constructed, and an autoregressive conditional heteroscedastic model is used for modeling to describe the dynamic relationship between the battery health state and the temperature and discharge mode. The autoregressive conditional heteroscedastic model formula is: Among them, SOH t represents the battery health status at time t, β0 represents the constant term, p represents the hysteresis period of the health status, β i Indicates the battery health status factor, SOH t-i represents the battery health status at time ti, q represents the hysteresis period of temperature, θ j The coefficient that represents the effect of temperature on the battery health status, T t-j represents the temperature data at time tj, r represents the hysteresis period of the discharge mode, λ k The coefficient representing the effect of discharge mode on health status, P t-k Represents the discharge mode data at time tk, ∈ t represents the error term; S43, optimizing the parameters of the autoregressive conditional heteroskedasticity model by a maximum likelihood estimation method to determine the causal relationship between battery cells, wherein the maximum likelihood estimation optimization objective function is: in, represents the optimized model parameters, T represents the total number of samples, p(SOH t |T t ,P t ,θ) represents the temperature T of the battery health state at time t under the model parameter θ before optimization t and the discharge mode data P at time t t The conditional probability under the condition, p(∈ t |θ) represents the probability density function of the error term; S44, further quantify the dynamic causal relationship between battery health status, temperature and discharge mode, and use vector autoregression model to model: Among them, A y represents the coefficient matrix for each lag period k, n represents the maximum lag period, SOH t-y Indicates the battery health status at time ty, T t-y Represents the temperature data at time ty, P t-y Represents the discharge mode data at time ty; S45. Use directed transfer entropy to analyze dynamic causal relationships: Among them, TE(Y→X) represents the transfer entropy from variable Y to variable X, T1 represents the length of the time series, and x t represents the variable X at time t, p(x t |y t-1 ) indicates that given the previous moment y t-1 The state x at the moment t The conditional probability of occurrence, p(x t ) represents x t The marginal probability of S46. Based on the results of the directional transfer entropy analysis, identify the dominant effects of temperature changes and discharge patterns on the battery health status.

5. The optimization control method for a parallel DC power supply system for large-capacity batteries according to claim 1, characterized in that: The S5 specifically includes: S51, combining the causal reasoning model, using the reverse modeling technology, based on the current battery health status and the parallel DC power supply system target, reversely deriving the optimal battery charging and discharging scheduling plan; S52. In the process of reverse modeling, define the optimization objective function: Where O represents the optimization objective function, n represents the total number of battery cells, and E i represents the energy consumption of the ith battery cell, L i represents the life loss of the ith battery cell, T i represents the temperature fluctuation of the i-th battery cell, α i , β i and γ i represents the adjustment coefficient; S53, using a simulated annealing algorithm for global optimization, reversely deriving the battery charging and discharging strategy based on the optimization objective function, gradually optimizing the solution through a temperature reduction strategy, updating the battery charging and discharging parameters, and performing optimization search according to the following state transfer equation: you t+1 =u t +Δu·P(ψ); Among them, u t+1 represents the system state at time t+1, u t represents the system state at time t, Δu represents the change in the system state, P(ψ) represents the temperature-dependent probability distribution function, and ψ represents the temperature parameter in the simulated annealing process; S54, according to the battery health status information fed back during the reverse modeling process, the charging and discharging strategy is adjusted in real time and dynamically, and the battery scheduling is continuously optimized using the new feedback information; S55. Output a global optimal charge and discharge control solution through the inverse modeling technology and dynamic adjustment mechanism.

6. The optimization control method for a parallel DC power supply system for large-capacity batteries according to claim 1, characterized in that: The S6 specifically includes: S61. According to the global optimal charge and discharge control scheme, game theory and co-evolutionary optimization algorithm are applied to establish a game model between battery cells, define each battery cell as a game participant, set the charge and discharge strategy of the battery cell as an optional action, and determine the benefit function of each battery cell; S62. In the game model, the cooperation and competition relationship between the battery cells is set, the benefit function of the battery cells is defined, and the battery health status, remaining power, charge and discharge efficiency and energy consumption are used as input variables to construct the benefit function: Among them, U i represents the revenue of the ith battery cell, R i represents the remaining capacity of the ith battery cell, C i represents the maximum charge and discharge capacity of the ith battery cell, SOH i represents the battery health status of the ith battery cell, E i represents the charging and discharging energy consumption of the i-th battery cell, ξ, ζ and τ represent the regulation coefficients; S63. Determine the optimal strategy between battery cells through Nash equilibrium analysis of game theory, and solve the strategy combination of each battery cell in the game model; S64, iterating the game model for multiple generations through the co-evolutionary optimization algorithm, evaluating the strategy combination between each generation of battery cells, adjusting the strategy of the battery cells and outputting a new charge and discharge control strategy; S65. Finally, local charging and discharging resource optimization is achieved, and a local optimized charging and discharging resource allocation plan is output.

Images