Lithium battery pack charging and discharging control and energy storage optimization method based on artificial intelligence

Abstract

The invention discloses a lithium battery pack charge and discharge control and energy storage optimization method based on artificial intelligence. The method comprises the following steps: S1, collecting and preprocessing operation data of a lithium battery pack; s2, performing multi-level subspace mapping by adopting a Grassmann manifold to extract different working condition features; s3, dynamic prior distribution is constructed based on a Dirichlet process, and adaptive modeling is carried out on battery health parameters and load fluctuation; s4, cross-space mapping is established, and a Grassmann manifold and a Dirichlet process are fused; s5, in combination with Dirichlet process sampling and Grassmann manifold deviation quantification, a multi-objective cost function is optimized; s6, after the charging and discharging strategy is corrected, the charging and discharging instruction is executed in stages, and subspace mapping and prior distribution are adjusted in real time; s7, storing a staged calibration result, monitoring operation data, and updating iterative computation; and S8, when the multi-objective cost function satisfies a preset index, outputting a final optimization scheme and terminating iteration. Charging and discharging are accurately controlled, the service life of the battery is prolonged, scheduling is optimized, cost is reduced, and energy storage benefits are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic energy storage systems, and in particular to an artificial intelligence-based lithium battery pack charge and discharge control and energy storage optimization method. Background Art

[0002] Lithium-ion battery packs are widely used in electric vehicles, renewable energy storage, and smart grids due to their high energy density, long cycle life, and rapid charge and discharge capabilities. However, lithium-ion battery charge and discharge control and energy storage optimization still face many challenges. Existing charge and discharge management systems typically use fixed rule-based control strategies, such as overcharge or over-discharge protection using preset thresholds, or predict the battery's health status based on simple linear models. However, due to the complexity of the battery's internal chemical reactions and the dynamic changes in the battery's state during charge and discharge, these traditional methods have significant limitations in practical applications and cannot meet the requirements of efficient energy storage and safe operation.

[0003] The main existing approaches to charge and discharge control include charge and discharge management strategies based on empirical formulas, model-based state estimation methods, and energy scheduling methods based on optimization algorithms. Empirical formula-based methods typically rely on historical data statistics and SOC (State of Charge) and SOH (State of Health) reference values ​​provided by battery manufacturers to manage batteries. While simple and easy to implement, these methods lack adaptability and cannot cope with variations in battery performance under different operating conditions. In particular, in complex power grid environments or scenarios with fluctuating power demand, they can easily lead to overcharging or overdischarging, shortening battery life. Model-based state estimation methods, such as Kalman filters and particle filters, attempt to describe the internal state of the battery by constructing mathematical models and using sensor data for estimation. However, due to battery aging characteristics and changes in the external environment, these methods are prone to deviations during long-term operation, resulting in inaccurate SOC or SOH estimates, which in turn affect charge and discharge control decisions. Energy scheduling methods based on optimization algorithms, such as dynamic programming and genetic algorithms, determine the optimal charge and discharge strategy by establishing an objective function and performing iterative optimization. While theoretically feasible, these methods are difficult to apply in real time to large-scale energy storage systems due to their high computational complexity. Furthermore, traditional optimization methods suffer from significant uncertainty in parameter selection and have limited adaptability to diverse scenarios.

[0004] In terms of energy storage optimization, existing methods mainly rely on strategies based on fixed scheduling rules or static optimization. For example, many microgrids and energy storage systems adopt a fixed scheduling method based on electricity price time series, charging during low-price periods and discharging during high-price periods to maximize economic benefits. However, this method fails to fully consider the health status of the battery, which may lead to high-frequency charge and discharge cycles and accelerate battery aging. In addition, traditional energy storage optimization methods generally assume that load demand is stable and fail to fully consider the volatility of renewable energy (such as photovoltaic and wind power) power generation and the dynamic changes in grid load, thereby affecting the reliability and economy of the overall energy storage system.

[0005] In response to these issues, the application of artificial intelligence technology in the field of lithium battery management has gradually attracted attention in recent years. Machine learning-based methods can use historical data for pattern recognition, thereby optimizing charging and discharging strategies and improving the intelligence level of battery management systems. However, existing artificial intelligence applications mainly focus on SOC prediction based on neural networks and charging and discharging optimization based on reinforcement learning, and there are still many shortcomings. For example, although traditional neural network methods can learn the evolution of battery status from data, they often lack physical constraints and are prone to the problem of accumulated prediction errors. In addition, although deep reinforcement learning methods can dynamically adjust charging and discharging strategies, they require a large amount of simulation data during the training process, making it difficult to directly apply them to actual energy storage systems.

[0006] Most current research lacks unified modeling and coordinated optimization of charge and discharge control and energy storage optimization, making it difficult to simultaneously take into account battery life, energy conversion efficiency, and economy. Therefore, the adaptability of existing methods in complex application scenarios is still poor. For example, in the face of highly uncertain load demand and fluctuations in renewable energy, how to optimize charge and discharge scheduling while ensuring battery health remains a technical challenge. In addition, existing charge and discharge management methods are usually based on static or semi-static optimization strategies, which fail to fully utilize the dynamic characteristics of lithium battery packs under different operating conditions, resulting in low energy utilization efficiency.

[0007] Therefore, how to provide a lithium battery pack charge and discharge control and energy storage optimization method based on the fusion of Grassmann manifold and Dirichlet process to improve the charge and discharge control accuracy, optimize energy storage scheduling, extend battery life, and reduce energy storage costs is a problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0008] One objective of the present invention is to propose an artificial intelligence-based method for lithium battery charge and discharge control and energy storage optimization. This method uses Grassmann manifold mapping to extract battery operating characteristics, combines it with a Dirichlet process to establish a dynamic prior distribution, adaptively adjusts the charge and discharge strategy, and optimizes scheduling through cross-space mapping. This method can precisely control the charge and discharge process, optimize energy storage scheduling, improve energy utilization, and extend battery life. It is suitable for smart grids, microgrids, and new energy storage systems.

[0009] According to an embodiment of the present invention, a method for controlling charging and discharging of a lithium battery pack and optimizing energy storage based on artificial intelligence includes the following steps:

[0010] S1. Collecting lithium battery pack operating data and preprocessing the operating data;

[0011] S2, using Grassmann manifold to perform multi-level subspace mapping on the pre-processed operating data to obtain subspace feature representations under different working conditions;

[0012] S3. Construct a dynamic prior distribution based on the Dirichlet process to perform non-parametric modeling of battery health parameters and load fluctuations, forming a state description that can be adaptively updated;

[0013] S4. Construct a multi-objective cost function based on the energy storage scheduling objectives and safety thresholds, establish a cross-space mapping between the characteristic subspace and the dynamic prior distribution, and achieve the fusion of the Grassmann manifold and the Dirichlet process through the association mapping function to generate a charge and discharge scheduling strategy;

[0014] S5. A cyclic verification mechanism is used to locally modify the charge and discharge control strategy. The iterative sampling of the Dirichlet process and the subspace deviation quantization of the Grassmann manifold are combined to collaboratively optimize the safety threshold and energy storage scheduling target in the multi-objective cost function.

[0015] S6. After the charge and discharge control strategy is modified, the charge and discharge instructions are allocated in segments, and the subspace mapping and the dynamic prior distribution are calibrated in stages simultaneously according to the real-time monitored battery operating status and energy storage demand changes;

[0016] S7. Store the periodic calibration results during the operation of the lithium battery pack, continuously monitor the temperature, voltage, and current, and incorporate the monitoring information into the next round of iterative calculations;

[0017] S8. When the performance indicators of the multi-objective cost function meet the preset requirements, the final charge and discharge control and energy storage optimization plan is output and the iterative process is terminated.

[0018] Optionally, the operating data includes voltage, temperature, current and environmental parameters, and the preprocessing includes unifying the data format and eliminating abnormal data.

[0019] Optionally, the S2 specifically includes:

[0020] S21. Based on the pre-processed lithium battery pack operation data, set the operation data matrix Wherein, m is the number of data samples, n is the feature dimension, and the operating data matrix X consists of voltage, temperature, current and environmental parameters, and is standardized to satisfy the zero mean unit variance distribution;

[0021] S22. Perform singular value decomposition on the running data matrix X, and set the decomposition result to:

[0022] X=UΣV T ;

[0023] in, is the left singular vector matrix, is a diagonal matrix of singular values, is the right singular vector matrix, and the singular vectors of the first d dimensions are taken to construct the characteristic subspace matrix V T is the transposed matrix of V, T is the transposed symbol;

[0024] S23, based on Grassmann manifold mapping, the feature subspace matrix V d Projection to the Grassmann manifold Gr(d,n):

[0025]

[0026] Among them, P represents the subspace representation of lithium battery pack operation data, P T is the transposed matrix of P, T is the transposed symbol, I d is the d-dimensional identity matrix, so that constrained optimization can be performed on the Grassmann manifold;

[0027] S24, Map the Grassmann manifold subspace representation P to different time series and construct the manifold trajectory matrix based on the working condition change Where t is the time step, and the subspace mapping transformation is performed on different time series data according to the trajectory matrix to form the final subspace feature representation under different working conditions.

[0028] Optionally, the S3 specifically includes:

[0029] S31. Define the lithium battery pack operation data matrix Among them, x t is the battery health parameter at time step t, including temperature, voltage and current, yt is the load fluctuation data at time step t, specifically referring to the environmental parameters, and Γ is the total number of sampling time steps;

[0030] S32. Construct a Dirichlet process mixture model, define the base measure distribution G0 and the concentration parameter α, and establish a state prior distribution G~DP(α,G0), where G represents the state distribution of the battery health state and load fluctuation, DP(α,G0) indicates that the state distribution obeys the Dirichlet process composed of the concentration parameter and the base measure, G0 is an empirical distribution based on historical data, and α controls the change in the number of state categories. The larger the value, the more categories there are and the greater the model flexibility.

[0031] S33. Establish a dynamic update mechanism based on the Dirichlet process and define the state transition equation:

[0032] s t =F(s t-1 ,x t )+∈ t ;

[0033] Among them, s t is the health status of the lithium battery at time step t, which represents the aging trend and load fluctuation characteristics of the lithium battery, F(·) is the state transfer function, and s t-1 is the health status of the lithium battery at time t-1, x t is the battery health parameter at time step t, ∈ t is a disturbance term that satisfies a zero-mean distribution and is used to describe unmeasurable uncertainties;

[0034] S34. Use discrete Beta distribution to dynamically adjust the health status category weights, and define the probability weight of each category k as:

[0035]

[0036] Among them, π k is the probability weight for each category k, v k Controls the weight of category k in the entire health status distribution, v k is an independent sampling variable and obeys the Beta distribution Beta(1,γ), where γ is a parameter that controls the sparsity of health status categories. The larger the value, the fewer categories there are.

[0037] S35. Dynamically cluster the battery health status and load fluctuation data, and calculate the probability density of each status category based on the non-homogeneous Poisson process:

[0038]

[0039] Among them, p(s t |xt ) indicates the battery health status t Depends on the current health parameter x t The conditional probability density, λ k is the occurrence rate of the state of category k, reflecting the battery aging rate or the severity of load fluctuation;

[0040] S36, update the Dirichlet process a posteriori and introduce new data x at time step Γ+1 Γ+1 ,y Γ+1 Then, update the state distribution:

[0041]

[0042] Among them, G Γ is the posterior state distribution at time step Γ, G Γ+1 The state distribution after correction for time step Γ+1;

[0043] S37, output the adaptive state description of the lithium battery pack, perform Bayesian inference on the classified health status and load fluctuation categories, and generate the predicted distribution p(s Γ+1 |x Γ+1 ,G Γ ), where x Γ+1 is the battery health parameter at time step Γ+1, s Γ+1 The health status of the lithium battery at time step Γ+1 is further formed into a state description that can be adaptively updated.

[0044] Optionally, the S4 specifically includes:

[0045] S41. Construct a charge and discharge scheduling objective function and define the multi-objective cost function of the lithium battery pack as:

[0046]

[0047] Among them, J is the overall optimization goal, Γ ' is the total number of time steps in the optimization cycle, P t is the charge and discharge power at time step t, s t is the health status of the lithium battery, C t is the energy storage cost per time step, f eff (P t ) represents the energy conversion efficiency loss function, f life (s t ) represents the battery life attenuation function, f cost (C t ) represents the charge and discharge economic loss function, w1, w2, w3 are the corresponding weight parameters;

[0048] S42. Establish the cross-space mapping between Grassmann manifold and Dirichlet process, and define the characteristic subspace matrix:

[0049]

[0050] Among them, V d represents the characteristic subspace matrix obtained by Grassmann manifold mapping, v i is the eigenvector in the subspace, n is the feature dimension, and d is the subspace dimension;

[0051] S43. Construct an association mapping function and define the cross-space projection relationship:

[0052]

[0053] Among them, M is the optimization mapping matrix, argmin is the independent variable value when the function takes the minimum value, is the characteristic subspace matrix at time step t, is the expected value of the state of the Dirichlet process at time step t. The goal is to minimize the subspace deviation and optimize the feature mapping;

[0054] S44. Generate a charge and discharge scheduling strategy based on the fusion mapping results and define the optimal scheduling instructions:

[0055]

[0056] in, is the optimal charge and discharge power at time step t, P t is the charge and discharge power at time step t, λ is the adjustment coefficient of the cross-space mapping error, and by adjusting P t The energy loss, life decay and economic cost are optimized while ensuring that the mapping deviation between the Grassmann manifold and the Dirichlet process is minimized.

[0057] Optionally, the S5 specifically includes:

[0058] S51. A cyclic verification mechanism is used to locally correct the charge and discharge control strategy. When new data arrives, the iterative sampling of the Dirichlet process and the subspace deviation quantization of the Grassmann manifold are combined to collaboratively optimize the safety threshold and energy storage scheduling target in the multi-objective cost function, and dynamically correct the optimal scheduling instructions:

[0059]

[0060] in, is the optimal charge and discharge power at time step t+1, is the optimal charge and discharge power at time step t, Pt is the charge and discharge power at time step t, λ is the adjustment coefficient of the cross-space mapping error, J is the total optimization target of the multi-objective cost function, and β is the learning rate, which ensures that the scheduling adjustment step size adapts to load fluctuations and realizes adaptive adjustment of charge and discharge optimization;

[0061] S52. When the multi-objective cost function J converges and meets the preset performance requirements, the iteration is stopped, the revised charge and discharge control strategy is output, and the saved optimization parameters are updated in real time so that the lithium battery pack performs periodic operations according to the optimal instructions.

[0062] Optionally, the S6 specifically includes:

[0063] S61, construct a segmented allocation strategy for charge and discharge instructions, and on the basis of the revised charge and discharge control strategy, calculate the charge and discharge power P at time step t. t Perform segmented calculations to satisfy the battery health status constraints and load fluctuation constraints, and define the segmented scheduling function:

[0064]

[0065] in, is the segmented charge and discharge power at the current moment t, ω i To assign weights, P t-i is the charge and discharge power of the past i time steps, N is the length of the sliding time window, which determines the smoothness of the segmented charge and discharge. i Make the charge and discharge power evenly distributed over time to avoid power surges that may cause damage to the battery pack;

[0066] S62: Real-time monitoring of battery operating status and energy storage demand changes, and adaptive calculation of charge and discharge scheduling errors:

[0067] δ t =‖s t -d t ‖;

[0068] Among them, s t is the health status of the lithium battery at time step t, d t is the load fluctuation variable at time step t, indicating the energy storage power demand, δ t is the state error at time step t, if δ t >∈, then adjust the charge and discharge strategy to make it consistent with the dynamic changes of the current battery health status and energy storage demand, where ∈ is the set threshold;

[0069] S63. Perform subspace mapping adjustment, update the charge-discharge state characteristic subspace matrix, project it on the Grassmann manifold, and optimize it in combination with the dynamic adjustment coefficient:

[0070]

[0071] in, is the characteristic subspace matrix at time step t, is the characteristic subspace matrix at time step t+1, μ t is the adaptive step size, X t is the running data matrix at time step t, Proj(·) represents the Grassmann manifold projection function, which ensures that the new data can be correctly mapped to the low-dimensional subspace;

[0072] S64. Calibrate the dynamic prior distribution, reconstruct the Dirichlet process based on the updated lithium battery health status and energy storage power requirements, and optimize the probability density of the state category:

[0073]

[0074] Among them, p(s t |s t-1 ) represents the battery health state s at time step t t Depends on the battery health state s at time step t-1 t-1 The conditional probability density of k is the probability weight for each category k, K is the total number of categories, μ k and Σ k are the mean and covariance matrices, respectively, so that the state distribution can adaptively change over time to match the health state of the battery pack;

[0075] S65, store the periodic calibration results, and the monitoring data at each time step t Perform discrete sampling and store in the database:

[0076]

[0077] in, represents the historical data storage structure, and Γ is the total number of sampling time steps.

[0078] The beneficial effects of the present invention are:

[0079] First, the present invention uses the Grassmann manifold mapping method to extract low-dimensional features of the lithium battery pack's operating status, enabling the charge-discharge control strategy to accurately capture dynamic changes in the battery's state of health. Compared to traditional SOC and SOH estimation methods based on fixed thresholds or empirical formulas, the present invention can maintain a precise description of the battery state under different operating conditions, improving the adaptability and accuracy of charge-discharge control.

[0080] Secondly, this invention uses a Dirichlet process to construct a dynamic prior distribution, enabling adaptive optimization by continuously updating the battery health status and load fluctuation model based on real-time monitoring data. Unlike existing energy storage scheduling methods based on static optimization or simple linear models, this invention can automatically adjust the charge and discharge strategy based on environmental changes, avoiding energy loss or battery overcharging and over-discharging caused by prediction errors, thereby effectively improving energy utilization and optimizing energy storage economics.

[0081] Finally, the present invention optimizes the charge-discharge scheduling strategy through cross-space mapping, constructing a joint optimization mechanism between charge-discharge status and energy storage demand, and achieving multi-objective scheduling control. This method not only extends battery life but also provides a stable and efficient energy storage optimization solution under complex load fluctuations and renewable energy grid integration, improving the reliability and economic benefits of the energy storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0083] Figure 1 This is a flow chart of an artificial intelligence-based lithium battery pack charge and discharge control and energy storage optimization method proposed by the present invention. DETAILED DESCRIPTION

[0084] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0085] refer to Figure 1 , a lithium battery pack charge and discharge control and energy storage optimization method based on artificial intelligence, comprising the following steps:

[0086] S1. Collecting lithium battery pack operating data and preprocessing the operating data;

[0087] S2, using Grassmann manifold to perform multi-level subspace mapping on the pre-processed operating data to obtain subspace feature representations under different working conditions;

[0088] S3. Construct a dynamic prior distribution based on the Dirichlet process to perform non-parametric modeling of battery health parameters and load fluctuations, forming a state description that can be adaptively updated;

[0089] S4. Construct a multi-objective cost function based on the energy storage scheduling objectives and safety thresholds, establish a cross-space mapping between the characteristic subspace and the dynamic prior distribution, and achieve the fusion of the Grassmann manifold and the Dirichlet process through the association mapping function to generate a charge and discharge scheduling strategy;

[0090] S5. A cyclic verification mechanism is used to locally modify the charge and discharge control strategy. The iterative sampling of the Dirichlet process and the subspace deviation quantization of the Grassmann manifold are combined to collaboratively optimize the safety threshold and energy storage scheduling target in the multi-objective cost function.

[0091] S6. After the charge and discharge control strategy is modified, the charge and discharge instructions are allocated in segments, and the subspace mapping and the dynamic prior distribution are calibrated in stages simultaneously according to the real-time monitored battery operating status and energy storage demand changes;

[0092] S7. Store the periodic calibration results during the operation of the lithium battery pack, continuously monitor the temperature, voltage, and current, and incorporate the monitoring information into the next round of iterative calculations;

[0093] S8. When the performance indicators of the multi-objective cost function meet the preset requirements, the final charge and discharge control and energy storage optimization plan is output and the iterative process is terminated.

[0094] The present invention provides an artificial intelligence-based lithium battery pack charge and discharge control and energy storage optimization method, which accurately models the battery health status and load fluctuations through the fusion of Grassmann manifold mapping and Dirichlet process, and realizes adaptive charge and discharge optimization. At the same time, cross-space mapping and cyclic verification mechanisms are adopted to optimize the multi-objective cost function, improve battery energy utilization and charge and discharge efficiency, reduce energy loss, and extend battery life. In addition, through real-time monitoring and dynamic adjustment, energy storage scheduling is optimized, economic efficiency is improved, and stable system operation is ensured. The present invention improves charge and discharge accuracy, optimizes energy storage scheduling, and reduces operating costs. It is suitable for smart grids, microgrids and new energy storage systems.

[0095] In this embodiment, the operating data includes voltage, temperature, current and environmental parameters, and the preprocessing includes unifying the data format and eliminating abnormal data.

[0096] The present invention improves data quality by unifying data formats and eliminating anomalies, ensuring accurate and reliable charge and discharge control.

[0097] In this embodiment, S2 specifically includes:

[0098] S21. Based on the pre-processed lithium battery pack operation data, set the operation data matrix Wherein, m is the number of data samples, n is the feature dimension, and the operating data matrix X consists of voltage, temperature, current and environmental parameters, and is standardized to satisfy the zero mean unit variance distribution;

[0099] S22. Perform singular value decomposition on the running data matrix X, and set the decomposition result to:

[0100] X=UΣV T ;

[0101] in, is the left singular vector matrix, is a diagonal matrix of singular values, is the right singular vector matrix, and the singular vectors of the first d dimensions are taken to construct the characteristic subspace matrix V T is the transposed matrix of V, T is the transposed symbol;

[0102] S23, based on Grassmann manifold mapping, the feature subspace matrix V d Projection to the Grassmann manifold Gr(d,n):

[0103]

[0104] Among them, P represents the subspace representation of lithium battery pack operation data, P T is the transposed matrix of P, T is the transposed symbol, I d is the d-dimensional identity matrix, so that constrained optimization can be performed on the Grassmann manifold;

[0105] S24, Map the Grassmann manifold subspace representation P to different time series and construct the manifold trajectory matrix based on the working condition change Where t is the time step, and the subspace mapping transformation is performed on different time series data according to the trajectory matrix to form the final subspace feature representation under different working conditions.

[0106] This method uses Grassmann manifolds to perform multi-level subspace mapping on preprocessed operating data, effectively extracting key features of lithium battery packs under different operating conditions and improving the adaptability and accuracy of charge and discharge control strategies. By using the low-dimensional feature representation of the Grassmann manifold, this method maintains the integrity of battery status information while reducing computational complexity, enabling rapid convergence of charge and discharge optimization in complex dynamic environments.

[0107] In this embodiment, S3 specifically includes:

[0108] S31. Define the lithium battery pack operation data matrix Among them, x tis the battery health parameter at time step t, including temperature, voltage and current, y t is the load fluctuation data at time step t, specifically referring to the environmental parameters, and Γ is the total number of sampling time steps;

[0109] S32. Construct a Dirichlet process mixture model, define the base measure distribution G0 and the concentration parameter α, and establish a state prior distribution G~DP(α,G0), where G represents the state distribution of the battery health state and load fluctuation, DP(α,G0) indicates that the state distribution obeys the Dirichlet process composed of the concentration parameter and the base measure, G0 is an empirical distribution based on historical data, and α controls the change in the number of state categories. The larger the value, the more categories there are and the greater the model flexibility.

[0110] S33. Establish a dynamic update mechanism based on the Dirichlet process and define the state transition equation:

[0111] s t =F(s t-1 ,x t )+∈ t ;

[0112] Among them, s t is the health status of the lithium battery at time step t, which represents the aging trend and load fluctuation characteristics of the lithium battery, F(·) is the state transfer function, and s t-1 is the health status of the lithium battery at time t-1, x t is the battery health parameter at time step t, ∈ t is a disturbance term that satisfies a zero-mean distribution and is used to describe unmeasurable uncertainties;

[0113] S34. Use discrete Beta distribution to dynamically adjust the health status category weights, and define the probability weight of each category k as:

[0114]

[0115] Among them, π k is the probability weight for each category k, v k Controls the weight of category k in the entire health status distribution, v k is an independent sampling variable and obeys the Beta distribution Beta(1,γ), where γ is a parameter that controls the sparsity of health status categories. The larger the value, the fewer categories there are.

[0116] S35. Dynamically cluster the battery health status and load fluctuation data, and calculate the probability density of each status category based on the non-homogeneous Poisson process:

[0117]

[0118] Among them, p(s t |x t ) indicates the battery health status t Depends on the current health parameter x t The conditional probability density, λ k is the occurrence rate of the state of category k, reflecting the battery aging rate or the severity of load fluctuation;

[0119] S36, update the Dirichlet process a posteriori and introduce new data x at time step Γ+1 Γ+1 ,y Γ+1 Then, update the state distribution:

[0120]

[0121] Among them, G Γ is the posterior state distribution at time step Γ, G Γ+1 The state distribution after correction for time step Γ+1;

[0122] S37, output the adaptive state description of the lithium battery pack, perform Bayesian inference on the classified health status and load fluctuation categories, and generate the predicted distribution p(s Γ+1 |x Γ+1 ,G Γ ), where x Γ+1 is the battery health parameter at time step Γ+1, s Γ+1 The health status of the lithium battery at time step Γ+1 is further formed into a state description that can be adaptively updated.

[0123] This method uses the Dirichlet process to construct a dynamic prior distribution, enabling nonparametric adaptive modeling of battery health status and load fluctuations, continuously optimizing the state description based on real-time monitoring data. Compared to traditional fixed threshold or static models, this method can accurately predict battery degradation trends, dynamically adjust charge and discharge strategies, reduce the impact of overcharge and overdischarge on battery life, improve energy utilization, and ensure the stability and economy of the energy storage system.

[0124] In this embodiment, the S4 specifically includes:

[0125] S41. Construct a charge and discharge scheduling objective function and define the multi-objective cost function of the lithium battery pack as:

[0126]

[0127] Among them, J is the overall optimization goal, Γ ' is the total number of time steps in the optimization cycle, P t is the charge and discharge power at time step t, s t is the health status of the lithium battery, C tis the energy storage cost per time step, f eff (P t ) represents the energy conversion efficiency loss function, f life (s t ) represents the battery life attenuation function, f cost (C t ) represents the charge and discharge economic loss function, w1, w2, w3 are the corresponding weight parameters;

[0128] S42. Establish the cross-space mapping between Grassmann manifold and Dirichlet process, and define the characteristic subspace matrix:

[0129]

[0130] Among them, V d represents the characteristic subspace matrix obtained by Grassmann manifold mapping, v i is the eigenvector in the subspace, n is the feature dimension, and d is the subspace dimension;

[0131] S43. Construct an association mapping function and define the cross-space projection relationship:

[0132]

[0133] Among them, M is the optimization mapping matrix, argmin is the independent variable value when the function takes the minimum value, is the characteristic subspace matrix at time step t, is the expected value of the state of the Dirichlet process at time step t. The goal is to minimize the subspace deviation and optimize the feature mapping;

[0134] S44. Generate a charge and discharge scheduling strategy based on the fusion mapping results and define the optimal scheduling instructions:

[0135]

[0136] in, is the optimal charge and discharge power at time step t, P t is the charge and discharge power at time step t, λ is the adjustment coefficient of the cross-space mapping error, and by adjusting P t The energy loss, life decay and economic cost are optimized while ensuring that the mapping deviation between the Grassmann manifold and the Dirichlet process is minimized.

[0137] This method achieves adaptive optimization of charge and discharge scheduling by constructing a multi-objective cost function, establishing a cross-space mapping between characteristic subspaces and dynamic prior distributions, and integrating Grassmann manifolds with Dirichlet processes. This method can accurately match battery health status with load demand, improve charge and discharge efficiency, reduce overcharge and overdischarge, optimize energy storage scheduling strategies, reduce energy loss, extend battery life, and improve energy storage economics. It is suitable for intelligent energy storage management in complex energy environments.

[0138] In this embodiment, the S5 specifically includes:

[0139] S51. A cyclic verification mechanism is used to locally correct the charge and discharge control strategy. When new data arrives, the iterative sampling of the Dirichlet process and the subspace deviation quantization of the Grassmann manifold are combined to collaboratively optimize the safety threshold and energy storage scheduling target in the multi-objective cost function, and dynamically correct the optimal scheduling instructions:

[0140]

[0141] in, is the optimal charge and discharge power at time step t+1, is the optimal charge and discharge power at time step t, P t is the charge and discharge power at time step t, λ is the adjustment coefficient of the cross-space mapping error, J is the total optimization target of the multi-objective cost function, and β is the learning rate, which ensures that the scheduling adjustment step size adapts to load fluctuations and realizes adaptive adjustment of charge and discharge optimization;

[0142] S52. When the multi-objective cost function J converges and meets the preset performance requirements, the iteration is stopped, the revised charge and discharge control strategy is output, and the saved optimization parameters are updated in real time so that the lithium battery pack performs periodic operations according to the optimal instructions.

[0143] This invention dynamically modifies the charge-discharge control strategy through a cyclic verification mechanism, combines iterative sampling of the Dirichlet process to improve the accuracy of battery health state prediction, and utilizes Grassmann manifold subspace deviation quantization to optimize battery state mapping, making the charge-discharge strategy more precise. This method effectively avoids the misjudgments caused by traditional fixed threshold control, ensuring that the charge and discharge process meets battery health requirements. It also synergistically optimizes safety thresholds and energy storage scheduling targets, improving energy utilization, reducing the risk of overcharge and overdischarge, extending battery life, and enhancing the economic efficiency and stability of the energy storage system.

[0144] In this embodiment, S6 specifically includes:

[0145] S61, construct a segmented allocation strategy for charge and discharge instructions, and on the basis of the revised charge and discharge control strategy, calculate the charge and discharge power P at time step t.t Perform segmented calculations to satisfy the battery health status constraints and load fluctuation constraints, and define the segmented scheduling function:

[0146]

[0147] in, is the segmented charge and discharge power at the current moment t, ω i To assign weights, P t-i is the charge and discharge power of the past i time steps, N is the length of the sliding time window, which determines the smoothness of the segmented charge and discharge. i Make the charge and discharge power evenly distributed over time to avoid power surges that may cause damage to the battery pack;

[0148] S62: Real-time monitoring of battery operating status and energy storage demand changes, and adaptive calculation of charge and discharge scheduling errors:

[0149] δ t =‖s t -d t ‖;

[0150] Among them, s t is the health status of the lithium battery at time step t, d t is the load fluctuation variable at time step t, indicating the energy storage power demand, δ t is the state error at time step t, if δ t >∈, then adjust the charge and discharge strategy to make it consistent with the dynamic changes of the current battery health status and energy storage demand, where ∈ is the set threshold;

[0151] S63. Perform subspace mapping adjustment, update the charge-discharge state characteristic subspace matrix, project it on the Grassmann manifold, and optimize it in combination with the dynamic adjustment coefficient:

[0152]

[0153] in, is the characteristic subspace matrix at time step t, is the characteristic subspace matrix at time step t+1, μ t is the adaptive step size, X t is the running data matrix at time step t, Proj(·) represents the Grassmann manifold projection function, which ensures that the new data can be correctly mapped to the low-dimensional subspace;

[0154] S64. Calibrate the dynamic prior distribution, reconstruct the Dirichlet process based on the updated lithium battery health status and energy storage power requirements, and optimize the probability density of the state category:

[0155]

[0156] Among them, p(s t |s t-1 ) represents the battery health state s at time step t t Depends on the battery health state s at time step t-1 t-1 The conditional probability density of k is the probability weight for each category k, K is the total number of categories, μ k and Σ k are the mean and covariance matrices, respectively, so that the state distribution can adaptively change over time to match the health state of the battery pack;

[0157] S65, store the periodic calibration results, and the monitoring data at each time step t Perform discrete sampling and store in the database:

[0158]

[0159] in, represents the historical data storage structure, and Γ is the total number of sampling time steps.

[0160] This invention utilizes a segmented allocation and dynamic calibration mechanism for charge and discharge instructions, combined with real-time monitoring of battery operating status and energy storage demand changes, to adaptively adjust subspace mapping and dynamic prior distributions to ensure the accuracy and flexibility of charge and discharge strategies. This method effectively reduces power fluctuations, optimizes energy scheduling, and reduces battery loss, thereby improving the efficiency of energy storage systems, extending battery life, and enhancing the overall economic efficiency of energy storage.

[0161] Example 1:

[0162] To verify the feasibility of this invention, we applied it to a new energy microgrid energy storage system. This system includes a 5MWh lithium-ion battery energy storage station, primarily used to regulate grid load, optimize the energy storage utilization of wind-solar hybrid power generation, and improve the capacity to absorb new energy. This energy storage station, equipped with multiple battery packs, employs a traditional fixed-threshold charge and discharge control strategy. This strategy suffers from rigid charging and discharging, inaccurate battery health management, and inflexible energy storage scheduling. This results in low system efficiency, accelerated battery degradation, and an inability to fully exploit the economic advantages of charging at low electricity prices and discharging at high electricity prices.

[0163] In this scenario, the present invention uses the Grassmann manifold mapping method to perform a low-dimensional subspace representation of the battery operation data, obtains the evolution law of the battery pack health state, and combines the Dirichlet process to construct a dynamic prior distribution to achieve adaptive optimization of the charging and discharging strategy. At the same time, the present invention optimizes the charging and discharging scheduling strategy through cross-space mapping, and improves the overall economy of the energy storage system while ensuring the battery life. During the test, the system was empirically verified for 6 months from October 2024 to March 2025. The method of the present invention was used to optimize the energy storage station, and compared with the unoptimized traditional scheduling strategy to evaluate its improvement effect in charging and discharging efficiency, battery health, energy storage economy, etc.

[0164] In practical applications, the present invention first collects historical operating data of the energy storage station, including parameters such as charging and discharging voltage, current, temperature, load demand, grid electricity price, and renewable energy power generation, and extracts features from these data through Grassmann manifold. Then, a dynamic model of battery health status is established using the Dirichlet process, and iterative updates are performed based on real-time monitoring data. The charge and discharge scheduling of the energy storage system is dynamically adjusted through the cross-space mapping optimization strategy of the present invention, so that the battery operates within the optimal SOC range (20% to 80%), while giving priority to charging during the low electricity price period (22:00-06:00) and discharging during the peak electricity price period (08:00-12:00, 18:00-22:00), thereby maximizing energy storage benefits.

[0165] During the six-month experiment, key performance indicators before and after optimization were recorded, including battery health degradation, battery energy utilization, charge and discharge efficiency, energy storage economic benefits, etc. The application effect data of the present invention are shown in the following table:

[0166] Table 1 Comparison of energy storage system performance before and after optimization of the present invention

[0167]

[0168]

[0169] Battery health decay is reduced: Before optimization, the number of abnormal SOC fluctuations (below 20% or above 90%) per month was about 87 times, which was reduced to 12 times after optimization. The average battery health (SOH) decay rate was reduced from 1.42% per month to 0.86%, and the battery life is expected to be extended by 1.5-2 years.

[0170] Improved battery energy utilization: Under traditional methods, the average discharge depth of the energy storage system is 68%, which is increased to 82% after optimization, and the system's effective energy utilization rate is increased by 20.5%.

[0171] Improved charging and discharging efficiency: Before and after optimization, the charging efficiency increased from 89.2% to 94.8%, the discharging efficiency increased from 88.4% to 93.1%, and the overall system energy conversion loss was reduced by 7.4%.

[0172] The economic benefits of energy storage have been enhanced: Before optimization, the daily economic benefits gained through peak-valley electricity price arbitrage were 1,245 yuan. After optimization, this increased to 1,878 yuan. The cumulative benefits increased by 114,000 yuan in six months, and the economic efficiency improved by 50.8%.

[0173] The experimental results demonstrate that the method proposed in this paper significantly improves battery charge and discharge efficiency, optimizes energy storage economics, and extends battery life, demonstrating greater practical value compared to traditional methods. This method is particularly suitable for scenarios such as new energy microgrids, electric vehicle charging stations, and large-scale energy storage power stations, effectively enhancing the operational reliability and economic benefits of energy storage systems.

[0174] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A lithium battery pack charge and discharge control and energy storage optimization method based on artificial intelligence, characterized in that: The steps include: S1. Collecting lithium battery pack operating data and preprocessing the operating data; S2, using Grassmann manifold to perform multi-level subspace mapping on the pre-processed operating data to obtain subspace feature representations under different working conditions; S3. Construct a dynamic prior distribution based on the Dirichlet process to perform non-parametric modeling of battery health parameters and load fluctuations, forming a state description that can be adaptively updated; S4. Construct a multi-objective cost function based on the energy storage scheduling objectives and safety thresholds, establish a cross-space mapping between the characteristic subspace and the dynamic prior distribution, and achieve the fusion of the Grassmann manifold and the Dirichlet process through the association mapping function to generate a charge and discharge scheduling strategy; S5. A cyclic verification mechanism is used to locally modify the charge and discharge control strategy. The iterative sampling of the Dirichlet process and the subspace deviation quantization of the Grassmann manifold are combined to collaboratively optimize the safety threshold and energy storage scheduling target in the multi-objective cost function. S6. After the charge and discharge control strategy is modified, the charge and discharge instructions are allocated in segments, and the subspace mapping and the dynamic prior distribution are calibrated in stages simultaneously according to the real-time monitored battery operating status and energy storage demand changes; S7. Store the periodic calibration results during the operation of the lithium battery pack, continuously monitor the temperature, voltage, and current, and incorporate the monitoring information into the next round of iterative calculations; S8. When the performance indicators of the multi-objective cost function meet the preset requirements, the final charge and discharge control and energy storage optimization plan is output and the iterative process is terminated.

2. The artificial intelligence-based lithium battery charge and discharge control and energy storage optimization method according to claim 1, characterized in that: The operating data includes voltage, temperature, current and environmental parameters, and the preprocessing includes unifying the data format and eliminating abnormal data.

3. The artificial intelligence-based lithium battery charge and discharge control and energy storage optimization method according to claim 1, characterized in that: The S2 specifically includes: S21. Based on the pre-processed lithium battery pack operation data, set the operation data matrix Wherein, m is the number of data samples, n is the feature dimension, and the operating data matrix X consists of voltage, temperature, current and environmental parameters, and is standardized to satisfy the zero mean unit variance distribution; S22. Perform singular value decomposition on the running data matrix X, and set the decomposition result to: X=UΣV T ; in, is the left singular vector matrix, is a diagonal matrix of singular values, is the right singular vector matrix, and the singular vectors of the first d dimensions are taken to construct the characteristic subspace matrix V T is the transposed matrix of V, T is the transposed symbol; S23, based on Grassmann manifold mapping, the feature subspace matrix V d Projection to the Grassmann manifold Gr(d,n): Among them, P represents the subspace representation of lithium battery pack operation data, P T is the transposed matrix of P, T is the transposed symbol, I d is the d-dimensional identity matrix, so that constrained optimization can be performed on the Grassmann manifold; S24, Map the Grassmann manifold subspace representation P to different time series and construct the manifold trajectory matrix based on the working condition change Where t is the time step, and the subspace mapping transformation is performed on different time series data according to the trajectory matrix to form the final subspace feature representation under different working conditions.

4. The artificial intelligence-based lithium battery charge and discharge control and energy storage optimization method according to claim 1, characterized in that: The S3 specifically includes: S31. Define the lithium battery pack operation data matrix Among them, x t is the battery health parameter at time step t, including temperature, voltage and current, y t is the load fluctuation data at time step t, specifically referring to the environmental parameters, and Γ is the total number of sampling time steps; S32. Construct a Dirichlet process mixture model, define the base measure distribution G0 and the concentration parameter α, and establish a state prior distribution G~DP(α,G0), where G represents the state distribution of the battery health state and load fluctuation, DP(α,G0) indicates that the state distribution obeys the Dirichlet process composed of the concentration parameter and the base measure, G0 is an empirical distribution based on historical data, and α controls the change in the number of state categories. The larger the value, the more categories there are and the greater the model flexibility. S33. Establish a dynamic update mechanism based on the Dirichlet process and define the state transition equation: s t =F(s t-1 ,x t )+∈ t ; Among them, s t is the health status of the lithium battery at time step t, which represents the aging trend and load fluctuation characteristics of the lithium battery, F(·) is the state transfer function, and s t-1 is the health status of the lithium battery at time t-1, x t is the battery health parameter at time step t, ∈ t is a disturbance term that satisfies a zero-mean distribution and is used to describe unmeasurable uncertainties; S34. Use discrete Beta distribution to dynamically adjust the health status category weights, and define the probability weight of each category k as: Among them, π k is the probability weight for each category k, v k Controls the weight of category k in the entire health status distribution, v k is an independent sampling variable and obeys the Beta distribution Beta(1,γ), where γ is a parameter that controls the sparsity of health status categories. The larger the value, the fewer categories there are. S35. Dynamically cluster the battery health status and load fluctuation data, and calculate the probability density of each status category based on the non-homogeneous Poisson process: Among them, p(s t |x t ) indicates the battery health status t Depends on the current health parameter x t The conditional probability density, λ k is the occurrence rate of the state of category k, reflecting the battery aging rate or the severity of load fluctuation; S36, update the Dirichlet process a posteriori and introduce new data x at time step Γ+1 Γ+1 ,y Γ+1 Then, update the state distribution: Among them, G Γ is the posterior state distribution at time step Γ, G Γ+1 The state distribution after correction for time step Γ+1; S37, output the adaptive state description of the lithium battery pack, perform Bayesian inference on the classified health status and load fluctuation categories, and generate the predicted distribution p(s Γ+1 |x Γ+1 ,G Γ ), where x Γ+1 is the battery health parameter at time step Γ+1, s Γ+1 The health status of the lithium battery at time step Γ+1 is further formed into a state description that can be adaptively updated.

5. The artificial intelligence-based lithium battery charge and discharge control and energy storage optimization method according to claim 1, characterized in that: The S4 specifically includes: S41. Construct a charge and discharge scheduling objective function and define the multi-objective cost function of the lithium battery pack as: Among them, J is the overall optimization goal, Γ ' is the total number of time steps in the optimization cycle, P t is the charge and discharge power at time step t, s t is the health status of the lithium battery, C t is the energy storage cost per time step, f eff (P t ) represents the energy conversion efficiency loss function, f life (s t ) represents the battery life attenuation function, f cost (C t ) represents the charge and discharge economic loss function, w1, w2, w3 are the corresponding weight parameters; S42. Establish the cross-space mapping between Grassmann manifold and Dirichlet process, and define the characteristic subspace matrix: Among them, V d represents the characteristic subspace matrix obtained by Grassmann manifold mapping, v i is the eigenvector in the subspace, n is the feature dimension, and d is the subspace dimension; S43. Construct an association mapping function and define the cross-space projection relationship: Among them, M is the optimization mapping matrix, argmin is the independent variable value when the function takes the minimum value, is the characteristic subspace matrix at time step t, is the expected value of the state of the Dirichlet process at time step t. The goal is to minimize the subspace deviation and optimize the feature mapping; S44. Generate a charge and discharge scheduling strategy based on the fusion mapping results and define the optimal scheduling instructions: in, is the optimal charge and discharge power at time step t, P t is the charge and discharge power at time step t, λ is the adjustment coefficient of the cross-space mapping error, and by adjusting P t The energy loss, life decay and economic cost are optimized while ensuring that the mapping deviation between the Grassmann manifold and the Dirichlet process is minimized.

6. The artificial intelligence-based lithium battery charge and discharge control and energy storage optimization method according to claim 1, characterized in that: The S5 specifically includes: S51. A cyclic verification mechanism is used to locally correct the charge and discharge control strategy. When new data arrives, the iterative sampling of the Dirichlet process and the subspace deviation quantization of the Grassmann manifold are combined to collaboratively optimize the safety threshold and energy storage scheduling target in the multi-objective cost function, and dynamically correct the optimal scheduling instructions: in, is the optimal charge and discharge power at time step t+1, is the optimal charge and discharge power at time step t, P t is the charge and discharge power at time step t, λ is the adjustment coefficient of the cross-space mapping error, J is the total optimization target of the multi-objective cost function, and β is the learning rate, which ensures that the scheduling adjustment step size adapts to load fluctuations and realizes adaptive adjustment of charge and discharge optimization; S52. When the multi-objective cost function J converges and meets the preset performance requirements, the iteration is stopped, the revised charge and discharge control strategy is output, and the saved optimization parameters are updated in real time so that the lithium battery pack performs periodic operations according to the optimal instructions.

7. The artificial intelligence-based lithium battery charge and discharge control and energy storage optimization method according to claim 1, characterized in that: The S6 specifically includes: S61, construct a segmented allocation strategy for charge and discharge instructions, and on the basis of the revised charge and discharge control strategy, calculate the charge and discharge power P at time step t. t Perform segmented calculations to satisfy the battery health status constraints and load fluctuation constraints, and define the segmented scheduling function: in, is the segmented charge and discharge power at the current moment t, ω i To assign weights, P t-i is the charge and discharge power of the past i time steps, N is the length of the sliding time window, which determines the smoothness of the segmented charge and discharge. i Make the charge and discharge power evenly distributed over time to avoid power surges that may cause damage to the battery pack; S62: Real-time monitoring of battery operating status and energy storage demand changes, and adaptive calculation of charge and discharge scheduling errors: d t =‖s t -d t ‖; Among them, s t is the health status of the lithium battery at time step t, d t is the load fluctuation variable at time step t, indicating the energy storage power demand, δ t is the state error at time step t, if δ t >∈, then adjust the charge and discharge strategy to make it consistent with the dynamic changes of the current battery health status and energy storage demand, where ∈ is the set threshold; S63. Perform subspace mapping adjustment, update the charge-discharge state characteristic subspace matrix, project it on the Grassmann manifold, and optimize it in combination with the dynamic adjustment coefficient: in, is the characteristic subspace matrix at time step t, is the characteristic subspace matrix at time step t+1, μ t is the adaptive step size, X t is the running data matrix at time step t, Proj(·) represents the Grassmann manifold projection function, which ensures that the new data can be correctly mapped to the low-dimensional subspace; S64. Calibrate the dynamic prior distribution, reconstruct the Dirichlet process based on the updated lithium battery health status and energy storage power requirements, and optimize the probability density of the state category: Among them, p(s t |s t-1 ) represents the battery health state s at time step t t Depends on the battery health state s at time step t-1 t-1 The conditional probability density of k is the probability weight for each category k, K is the total number of categories, μ k and Σ k are the mean and covariance matrices, respectively, so that the state distribution can adaptively change over time to match the health state of the battery pack; S65, store the periodic calibration results, and the monitoring data at each time step t Perform discrete sampling and store in the database: in, represents the historical data storage structure, and Γ is the total number of sampling time steps.

Images