Dynamic management and control method and system for energy storage battery

Abstract

The invention relates to the technical field of energy storage battery health state monitoring and dynamic management and control, and discloses an energy storage battery dynamic management and control method and system.The energy storage battery dynamic management and control method comprises the steps that energy storage battery state data and environment interference data are collected through a multi-modal sensor network, and a multi-modal feature set is obtained; processing the preprocessed multi-modal feature set by using a multi-layer time sequence convolutional network, extracting time sequence features of the health state of the battery, and meanwhile, realizing abnormal noise detection and suppression; establishing a mapping relation between environmental interference and prediction reliability, representing a prediction result as a confidence interval, and calculating a reliability index; dynamically adjusting parameters and a structure of the prediction model, and generating a health state prediction result and reliability evaluation thereof; generating an energy storage battery dynamic management and control strategy with safety margin; through a multi-mode signal acquisition and preprocessing technology, the battery state and environment interference information are effectively captured, and a high-quality data basis is provided for health state prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage battery health status monitoring and dynamic control, and more specifically, to a method and system for dynamic control of energy storage batteries. Background Art

[0002] As a key device for energy storage and conversion, energy storage batteries are widely used in large-scale energy storage power stations on the grid side, industrial microgrids, new energy vehicle battery management systems, and distributed energy storage systems. The health status of energy storage batteries directly affects the safety, reliability, and economic benefits of the system. Therefore, accurate prediction and effective control of the health status of energy storage batteries are of great significance.

[0003] At present, the methods for predicting and controlling the health status of energy storage batteries mainly include methods based on physical models, data-driven machine learning methods, and hybrid methods that combine the two. The physical model-based method predicts the health status by establishing a mathematical model of the electrochemical reactions and physical processes inside the battery. It has good interpretability, but it is difficult to accurately describe the nonlinear degradation process of the battery under complex working conditions. The data-driven machine learning method uses historical data to train the model and can capture complex nonlinear relationships, but it has a high dependence on data quality and has limited generalization capabilities when facing unseen working conditions. The hybrid method combines the advantages of both, but still faces many challenges under complex environmental interference conditions.

[0004] As the application environment of energy storage batteries becomes increasingly complex, environmental interference factors such as electromagnetic interference, temperature fluctuations, mechanical vibrations, etc. have an increasing impact on health status prediction. The existing technology mainly has the following problems: In a multi-source noise interference environment, the prediction model finds it difficult to effectively distinguish and suppress different types of noise interference, resulting in insufficient prediction accuracy; the ability to identify subtle characteristic changes such as micro-circulation discharge is limited, and it is unable to accurately capture small changes in the battery health status; it is unable to effectively quantify the reliability of the prediction results, and lacks an assessment of prediction uncertainty; the management and control strategy is fixed and cannot be adaptively adjusted according to interference conditions and prediction reliability; the safety margin design at different reliability levels is not considered, and there are potential safety risks.

[0005] Therefore, there is an urgent need for a technical solution that can cope with complex environmental interference, improve prediction accuracy and reliability, and achieve dynamic adaptive control. Summary of the invention

[0006] The present invention provides a method and system for dynamic control of energy storage batteries, which solve the technical problems in the related art that the prediction accuracy of the health status of energy storage batteries under complex environmental interference conditions is insufficient, the reliability is difficult to evaluate, and the control strategy is not flexible enough.

[0007] The present invention provides a method for dynamically managing and controlling an energy storage battery, including: Collecting energy storage battery state data and environmental interference data through a multi-modal sensor network, and performing noise reduction processing on the collected signals to obtain a multi-modal feature set; Using a multi-layer temporal convolutional network to process the preprocessed multi-modal feature set, extracting the temporal features of the battery health state, and simultaneously realizing abnormal noise detection and suppression; Based on the environmental interference data and the temporal features of the battery health state, establishing a mapping relationship between environmental interference and prediction reliability, applying the evidence theory to construct an uncertainty quantification framework, representing the prediction result as a confidence interval and calculating a reliability index; According to the interference condition and prediction reliability, dynamically adjusting the parameters and structure of the prediction model to generate a health state prediction result and its reliability evaluation; Based on the health state prediction result and reliability evaluation, generating a dynamic management and control strategy for the energy storage battery with a safety margin.

[0008] Furthermore, the multi-modal sensor network includes a battery state sensor group and an environmental interference sensor group; the battery state sensor group is used to collect parameters related to the battery health state, including voltage, current, temperature, and impedance data; the environmental interference sensor group is used to collect interference signals affecting the health state prediction, including electromagnetic field intensity, environmental noise, vibration, and humidity data.

[0009] Furthermore, the noise reduction processing adopts a multi-scale wavelet decomposition technique, including: Applying multi-scale wavelet decomposition to the time series data collected by each sensor; Applying an adaptive threshold function to the decomposed wavelet coefficients for noise suppression; Reconstructing the noise-reduced signal through inverse wavelet transform; Decomposing the signal into three frequency domain feature sets of high frequency, medium frequency, and low frequency, respectively corresponding to different physical meanings.

[0010] Furthermore, the multi-layer temporal convolutional network includes: A causal convolutional layer to ensure that the model only uses information before the current moment; An dilated convolutional layer to expand the receptive field through dilated convolution, and the dilation rate increases exponentially with the number of layers; A residual connection structure to optimize the gradient flow and avoid the problem of gradient disappearance; A skip connection to connect features at different levels and enhance the multi-scale feature fusion ability.

[0011] Furthermore, the abnormal noise detection adopts the Mahalanobis distance method, including: Calculating the mean vector and covariance matrix of the features; Calculate the Mahalanobis distance for each sample point; Set an adaptive threshold, and when the Mahalanobis distance is greater than the threshold, mark it as an outlier; For the detected outliers, use interpolation or prediction replacement methods for noise suppression.

[0012] Furthermore, the establishment of the mapping relationship between environmental interference and prediction reliability includes: Extract the feature vectors of the interference data collected by the environmental interference sensor group to obtain interference feature vectors; Through training with historical data, establish the mapping relationship between interference features and prediction errors; Quantify the influence degree of different types of interference on the prediction result as an influence coefficient matrix; Based on the influence coefficient matrix, calculate the prediction compensation value.

[0013] Furthermore, the uncertainty quantification framework is constructed based on the evidence theory, including: Define the judgment framework, including possible states or categories; Calculate the basic probability assignment for each evidence source; Calculate the belief function based on the basic probability assignment; Calculate the likelihood function; Express the prediction result as an interval, and calculate the uncertainty as the width of the prediction interval.

[0014] Furthermore, the steps of dynamically adjusting the prediction model parameters and structure include: Calculate the sensitivity matrix of the model parameters with respect to the interference features; Based on the sensitivity matrix and the current interference situation, calculate the parameter adjustment amount; Select different complexity model structures according to the reliability level; Adopt residual connection and recursive connection structures to enhance the model's ability to capture long-term dependencies and the efficiency of gradient flow.

[0015] Furthermore, the steps of generating the dynamic control strategy include: According to the reliability index of the prediction result, adaptively calculate the safety margin; Based on the health state prediction result and the safety margin, optimize the battery charge and discharge control strategy; According to the health state prediction and reliability assessment, optimize the load distribution and thermal management strategy; Implement preventive maintenance and abnormal state adaptive correction measures.

[0016] The present invention provides an energy storage battery dynamic control system for implementing the above-mentioned energy storage battery dynamic control method, including: The data acquisition and preprocessing module is used to obtain battery status and environmental interference information through a multi-modal sensor network and perform signal denoising processing; The feature extraction and denoising module is used to analyze the battery health status features by applying a multi-layer temporal convolutional network and filter out abnormal noises; The reliability evaluation module is used to construct an association model between environmental interference and prediction accuracy and quantify prediction uncertainty based on the evidence theory; The adaptive inference module is used to dynamically optimize the prediction algorithm structure and parameters according to the real-time interference situation and reliability indicators; The control strategy generation module is used to calculate the safety margin based on the health status prediction result and its confidence level and generate a differentiated control scheme.

[0017] The beneficial effects of the present invention are as follows: Through multi-modal signal acquisition and preprocessing technology, battery status and environmental interference information can be effectively captured, providing a high-quality data basis for health status prediction. Compared with traditional methods, the signal-to-noise ratio is improved; By adopting a multi-layer temporal convolutional network and the Mahalanobis distance anomaly detection method, efficient feature extraction and noise suppression are realized, the resistance to noise and interference signals is improved, and the recognition accuracy of the microcirculation discharge mode is improved; By constructing an interference and reliability mapping model and an evidence theory uncertainty quantification framework, a quantitative evaluation of the reliability of the prediction result is realized, the prediction result is extended from point estimation to interval representation, providing a reliability evaluation basis for decision-making and early warning of sudden interference; Based on the dynamic model adjustment mechanism of the interference condition and prediction reliability, the system can adapt to different working conditions, improve adaptability, reduce the consumption of computing resources at the same time, and support millisecond-level response; Through adaptive safety margin design and differentiated control strategies, while ensuring system performance, risks such as overcharging and over-discharging are effectively avoided, the incidence rate of safety risk events is reduced, and the overall service life of the battery pack is extended. Description of the Drawings

[0018] Figure 1 is a flowchart of a dynamic control method for an energy storage battery of the present invention; Figure 2 is a flowchart of the multi-modal signal acquisition and preprocessing steps of the present invention; Figure 3 is a detailed flowchart of the temporal convolutional feature extraction and noise suppression steps of the present invention; Figure 4 is a detailed flowchart of the interference reliability mapping and uncertainty quantification steps of the present invention; Figure 5 is a detailed flowchart of the adaptive inference and decision optimization steps of the present invention; Figure 6 This is a detailed flowchart of the dynamic control strategy generation step of the present invention. Detailed implementation manners

[0019] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0020] In at least one embodiment of the present invention, a dynamic control method for energy storage batteries is disclosed. As Figures 1 to 5 shown, it includes: Step 1: Collect energy storage battery state data and environmental interference data through a multi-modal sensor network, and perform noise reduction processing on the collected signals to obtain a multi-modal feature set; In this step, energy storage battery-related data and environmental interference data are collected through a multi-modal sensor network, and noise reduction processing is performed on the collected signals to extract different frequency domain feature information.

[0021] Step 1.1: Multi-modal data collection; Construct a multi-modal sensor network, which includes the following two types of sensor groups; Battery state sensor group, used to collect parameters related to the battery health state, including but not limited to: Voltage sensor, monitoring the voltage of individual battery cells and the overall voltage; Current sensor, monitoring the charge and discharge current; Temperature sensor, monitoring the surface and ambient temperature of the battery; Impedance measurement unit, collecting the electrochemical impedance spectrum data of the battery.

[0022] Environmental interference sensor group, used to collect interference signals that may affect the prediction of the battery health state, including but not limited to: Electromagnetic field intensity sensor, detecting changes in the environmental electromagnetic field intensity; Environmental noise sensor, monitoring acoustic and electrical noise; Vibration sensor, monitoring mechanical vibration interference; Humidity sensor, monitoring changes in the environmental humidity.

[0023] All sensor data are synchronously collected by the data acquisition unit, and the sampling frequency is configured according to the variation characteristics of different parameters. For example, the voltage and current sampling frequencies are usually 1 - 10 kHz, the temperature sampling frequency is 1 - 10 Hz, and the impedance measurement frequency is 0.1 - 0.01 Hz.

[0024] Step 1.2, signal preprocessing and feature separation; Preprocess the collected original signal and use the multi-scale wavelet decomposition technique for signal denoising and feature separation. The specific steps are as follows: Signal denoising processing, for the time series data collected by each sensor Apply multi-scale wavelet decomposition to decompose the signal into wavelet coefficients of different scales: ; where is the wavelet basis function at time , represents the scale parameter, represents the translation parameter, is the corresponding wavelet coefficient, is the time series data at time , represents the integral operation.

[0025] Threshold processing, apply the threshold function to the wavelet coefficients to suppress noise: ; where is the threshold, adaptively determined according to the signal-to-noise ratio; is the processed wavelet coefficient; is the threshold function for suppressing noise.

[0026] Signal reconstruction, reconstruct the denoised signal through inverse wavelet transform: ; where is the reconstructed signal, represents the summation operation, and are the indices of the scale parameter and the translation parameter respectively, represents the processed wavelet coefficient, represents the time when the scale parameter is and the translation parameter is of the wavelet basis function.

[0027] Frequency domain feature separation, decompose the signal into three frequency domain feature sets: high frequency, medium frequency, and low frequency, corresponding to different physical meanings: Low frequency feature set , corresponding to the basic state changes of the battery (such as SOC, SOH, etc.), with a frequency range of 0 - 0.1 Hz; Intermediate-frequency feature set , corresponding to the dynamic charging and discharging characteristics of the battery, with a frequency range of 0.1 - 10 Hz; High-frequency feature set , corresponding to noise and interference information, with a frequency range above 10 Hz.

[0028] Through the above multi-modal signal acquisition and preprocessing, output the multi-modal feature set after noise reduction , providing a high-quality data basis for subsequent feature extraction and analysis.

[0029] Step 2: Use a multi-layer temporal convolutional network to process the preprocessed multi-modal feature set, extract the temporal features of the battery health state, and simultaneously achieve abnormal noise detection and suppression; In this step, taking the multi-modal feature set output in Step 1 as the input, use a multi-layer temporal convolutional network to process these preprocessed multi-modal feature data, extract the multi-scale temporal features of the battery health state, simultaneously achieve abnormal noise detection and suppression, and identify the characteristics of the microcirculation discharge mode.

[0030] Step 2.1: Construction of a multi-layer temporal convolutional network; Construct a multi-layer temporal convolutional network that combines causal convolution and dilated convolution. This network includes the following key components: Causal convolution layer, ensuring that the model only uses the information before the current moment to avoid information leakage. Its mathematical expression is: ; Where is the output feature, is the convolutional kernel parameter, is the convolutional kernel size, is the input at time is the time offset, represents the summation operation.

[0031] Dilated convolution layer, by inserting holes between the convolutional kernel elements, expands the receptive field and captures long-term dependencies, defined as: ; Where is the dilation rate, which exponentially increases as the number of network layers increases ( , is the layer index), is the output feature of the dilated convolution, is the convolutional kernel parameter, is the input at time

[0032] Residual connection structure, used to optimize gradient flow and avoid the problem of gradient disappearance: ; where is the output of the th layer, is the non-linear transformation function of the th layer, is the output of the th layer.

[0033] Skip connection, connecting features of different levels to enhance the ability of multi-scale feature fusion: ; where is the weight coefficient of the th layer, is the output of the th layer, is the feature after skip connection, represents the summation operation.

[0034] Step 2.2, Mahalanobis distance-based anomaly noise detection; Apply the Mahalanobis distance method to the extracted features to detect anomaly noise: Calculate the mean vector and covariance matrix of the features: ; ; where is the feature vector, is the number of samples, is the mean vector, is the covariance matrix, represents the summation operation, is the sample index, represents transpose of, is the feature vector minus the mean vector.

[0035] Calculate the Mahalanobis distance of each sample point: ; where is the Mahalanobis distance of the sample point , is the inverse matrix of the covariance matrix, represents the square root operation.

[0036] Set the adaptive threshold , when , mark it as an outlier.

[0037] For the detected abnormal points, interpolation or prediction replacement methods are used for noise suppression: ; where is the interpolation function, is the time window size, denotes to the data at time, denotes to the data at time, is the corrected data, is the original data at time, is the Mahalanobis distance of, is the threshold.

[0038] Step 2.3, multi-head self-attention mechanism and microcirculation discharge pattern recognition; Introduce the multi-head self-attention mechanism to recognize the characteristics of microcirculation discharge patterns; Construct query, key, and value matrices: ; ; ; where , , respectively represent the weight matrices of the query, key, and value, is the input feature, is the query matrix, is the key matrix, is the value matrix.

[0039] Calculate the attention weights: ; where is the dimension of the key vector, is the transpose of, is the normalized exponential function, is the attention output.

[0040] Implement the multi-head attention structure: ; ; where is the number of attention heads, is the output weight matrix, is the output of multi-head attention, represents the concatenation operation, is the output of the -th and -th attention heads respectively represent the weight matrices of the query, key, and value of the -th attention head,

[0041] Feature adaptive weighting is achieved through the channel attention mechanism: ; ; where is the weight coefficient for channel , is the feature of channel , is the activation function, is the multi-layer perceptron, is the global average pooling operation, is the weighted feature.

[0042] Through the above temporal convolutional feature extraction and noise suppression steps, a multi-scale temporal feature representation of the battery health state and the identified microcirculation discharge pattern features are output, providing high-quality feature inputs for subsequent steps.

[0043] Step 3: Based on the environmental interference data and the temporal features of the battery health state, establish a mapping relationship between environmental interference and prediction reliability, apply the evidence theory to construct an uncertainty quantification framework, represent the prediction result as a confidence interval, and calculate the reliability index; This step takes the environmental interference data collected in Step 1 and the health state features extracted in Step 2 as inputs, establishes a mapping relationship between environmental interference and prediction reliability, applies the evidence theory to construct an uncertainty quantification framework, expands the prediction result from point estimation to confidence interval representation, and provides a reliability evaluation basis for decision-making.

[0044] Step 3.1: Construction of the interference and reliability mapping model; Construct a mapping relationship between interference patterns and prediction reliability to quantify the influence degree of different interferences on the prediction result: Interference feature extraction: Extract features from the interference data collected by the environmental interference sensor group to obtain interference feature vectors; ; where represents the set of interference feature vectors, , , represent the 1st, 2nd, and th interference features respectively, and

[0045] Construct an interference impact model, train it with historical data, and establish a mapping relationship between interference features and prediction errors: ; where represents the reliability evaluation index, is the mapping function, which can be implemented by a multi-layer perceptron: ; where and are the weights and biases of the th layer respectively, is the activation function, is the number of network layers, is the input of the interference feature vector, is the reliability evaluation index output by the mapping function, represents the index of the network layer, is the weight matrix of the output layer, is the bias of the output layer, is the weight matrix of the first layer, is the bias of the first layer, is the weight matrix of the second-to-last layer, is the bias of the second-to-last layer.

[0046] Quantify the influence degree, and quantify the influence degree of different types of interference on the prediction result as the influence coefficient matrix : ; where represents the influence degree of the th type of interference on the th prediction index, which is calculated through the Jacobian matrix, represents 's partial derivative with respect to .

[0047] Interference compensation calculation, based on the influence coefficient matrix, calculate the prediction compensation value: ; where is the prediction compensation value, which is used to adjust the original prediction result, is the influence coefficient vector of the th type of interference, is the Eigenvalues of class interference Denotes the summation operation Is the total number of interference types

[0048] Step 3.2, Uncertainty quantification framework based on evidence theory Apply evidence theory to construct an uncertainty quantification framework to achieve interval representation of prediction results Construct an identification framework and define a judgment framework ; Where Represents the set of judgment frameworks , , Respectively represent the 1st, 2nd, th possible states or categories Is the total number of states or categories

[0049] Basic probability assignment calculation. For each evidence source (Such as different sensors, different feature sets), calculate its basic probability assignment : ; Where Is A subset of Is the evidence source The credibility weight of, provided by the interference and reliability mapping model Is the evidence source For the subset Support degree of Denotes the summation operation Represents All subsets of (i.e., subsets of the judgment framework, representing possible combinations of states or categories) Is the evidence source For the subset Basic probability assignment of

[0050] Belief function calculation. Calculate the belief function based on the basic probability assignment ; Where Is the combined basic probability assignment, obtained through the Dempster combination rule Is the belief degree for the subset Of Denotes the summation operation Represents All subsets of

[0051] Plausibility function calculation. Calculate the plausibility function ; where is the complement set of, is the likelihood of the subset , is the belief degree of the complement set, represents the summation operation, represents all subsets of.

[0052] Interval representation generation, representing the prediction result as an interval , rather than a single point estimate. The uncertainty is calculated as: ; where is the uncertainty of the subset , is the likelihood, is the belief degree.

[0053] Step 3.3, real-time interference impact assessment and reliability index calculation; Based on the currently detected interference situation and historical model performance, real-time evaluate the reliability of the prediction result: Interference state detection, based on the environmental interference sensor data, detect the current interference state: ; where is the interference state vector, is the currently collected interference data, is the interference detection function.

[0054] Reliability index calculation: Considering the interference situation, historical model performance and prediction uncertainty comprehensively, calculate the reliability index: ; where is the reliability index, and the value range is , , , represent the interference state impact weight, historical accuracy weight and uncertainty weight respectively, is the interference sensitivity parameter, is the norm of the interference state vector, is the historical accuracy of the model under similar conditions, is the prediction uncertainty.

[0055] Reliability level division, dividing the reliability index into different levels for subsequent decision-making adjustment: ; where is the reliability level, is the reliability index; Through the interference and reliability mapping and uncertainty quantification steps, the confidence interval representation of the prediction result is output and the reliability index , providing a basis for reliability assessment for subsequent adaptive decision-making.

[0056] Step 4, Dynamically adjust the parameters and structure of the prediction model according to the interference condition and prediction reliability, and generate the health state prediction result and its reliability assessment; This step integrates the output results of the previous three steps. According to the interference condition information (from the original interference data in Step 1) and prediction reliability assessment obtained in Step 3, dynamically adjust the parameters and structure of the prediction model, and combine the multi-modal feature information extracted in Step 2 to generate an optimized health state prediction result and reliability assessment.

[0057] Step 4.1, Adaptive model parameter adjustment; Based on the interference condition and reliability index, dynamically adjust the parameters and structure of the prediction model: Model parameter sensitivity analysis, calculate the model parameters: ; where represents the set of model parameters, , , respectively represent the 1st, 2nd, th model parameter, is the total number of model parameters.

[0058] The sensitivity matrix with respect to the interference feature : ; where, represents the index of the model parameter, represents the index of the interference feature, represents the th model parameter's sensitivity to the th interference feature; is the model loss function, is the th model parameter, is the interference feature, is the sensitivity matrix element, represents the loss function with respect to the parameter and the interference feature The second-order partial derivative.

[0059] Parameter adaptive adjustment. Based on the sensitivity matrix and the current interference situation, calculate the parameter adjustment amount: ; where is the adjustment step size, is the current interference eigenvalue, is the reliability index, is the parameter adjustment amount, represents the summation operation, is the interference feature dimension, is the sensitivity matrix element, indicating the sensitivity of the th model parameter to the th interference feature.

[0060] Adaptive model structure switching: Select different complexity model structures according to the reliability level: ; where , and represent high, medium, and low complexity model structures respectively, is the reliability level, is the selected model.

[0061] Step 4.2, Long-term dependence enhancement and gradient flow optimization; Adopt residual connection and recursive connection structures to enhance the model's long-term dependence capture ability and gradient flow efficiency: Recursive connection structure: Introduce recursive connection to capture long-term dependencies: ; where is the hidden state at the current moment, is the input at the current moment, is the hidden state at the previous moment, is the non-linear transfer function, which can be implemented by a gated recurrent unit (GRU): ; ; ; ; where is the update gate, controlling the degree of retention of the previous moment's hidden state information; is the reset gate, controlling the influence degree of the previous moment's hidden state on the current candidate hidden state; is the candidate hidden state, representing the new information at the current moment; is the hidden state at the current moment, which is a weighted combination of the hidden state at the previous moment and the candidate hidden state; is the hidden state at the previous moment, containing historical information; is the input feature at the current moment; represents the Hadamard product (element-wise multiplication), which multiplies the corresponding elements in the vector or matrix; is the activation function, mapping the input to the range between 0 and 1; is the hyperbolic tangent activation function; is the weight matrix of the update gate, used to learn the parameters for controlling information update; is the weight matrix of the reset gate, used to learn the parameters for controlling historical information reset; is the weight matrix of the candidate hidden state, used to learn the parameters for new information; represents concatenating the hidden state at the previous moment and the current input along the feature dimension; represents the forgetting factor, controlling the degree of discarding the information of the hidden state at the previous moment.

[0062] Residual connection optimization, applying residual connection in the multi-layer structure to optimize the gradient flow: ; where represents the layer normalization operation, improving the training stability; is the non-linear transformation function of the th layer, is the output of the th layer, is the output of the th layer.

[0063] Gradient clipping, preventing the problem of gradient explosion: ; where is the parameter gradient, is the clipping threshold, is the clipping function, restricting the gradient value within the range of .

[0064] Step 4.3, Multimodal Feature Fusion and Health State Prediction; Adopt a non-linear feature fusion algorithm to integrate multimodal information and form the final health state prediction result: Weighted Feature Fusion: Weighted fusion of different modal features: ; where is the fused feature, denotes the th feature mode, is the corresponding weight coefficient, automatically determined by the attention mechanism, denotes the summation operation, is the number of feature modes: ; ; where and respectively represent the query vector and the feature transformation matrix in the attention mechanism, both of which are learnable parameters, denotes the exponential function, is the hyperbolic tangent activation function, denotes transpose of, is the feature attention score. Through this attention mechanism, the model can automatically learn the importance weights of different feature modes, making the fusion process more intelligent and adaptive.

[0065] Cross-modal feature complementarity: Establish a cross-modal feature correlation matrix : ; where is the element of the correlation matrix, representing the degree of correlation between the th feature mode and the th feature mode, denotes the transpose of the feature vector , is the th feature mode and the th feature mode mapping matrix.

[0066] Final health status prediction. Fusing the above features to generate a health status prediction result: ; where is the health status prediction value, is the prediction function, considering the fused features, cross-modal correlations, and reliability index, is the fused feature, is the cross-modal feature correlation matrix, is the reliability index.

[0067] Reliability-aware interval prediction. Adjust the prediction interval width according to the reliability index: ; where is the prediction interval, is the lower prediction bound, is the upper prediction bound, is the predicted value of the health state, is the basic interval width parameter, is the reliability index. The lower the reliability, the wider the interval, and represent subtraction and addition operations respectively.

[0068] Through the adaptive reasoning and decision optimization steps, the predicted results of the health state considering the interference effect and reliability assessment are output, including the predicted value and the prediction interval , providing a decision-making basis for generating the final dynamic control strategy.

[0069] Step 5: Based on the predicted results of the health state and reliability assessment, generate a dynamic control strategy for the energy storage battery with a safety margin; This step takes the predicted results of the health state generated in Step 4 and their reliability assessment ( of Step 3) as the core input, and at the same time refers to the real-time state data of the battery and environmental condition data obtained in Step 1 to generate a dynamic control strategy for the energy storage battery with a safety margin, avoiding safety risks such as overcharge and over-discharge while ensuring system performance.

[0070] Step 5.1: Calculation of the safety margin based on reliability; According to the reliability index of the prediction result, the safety margin is adaptively calculated: The mapping relationship between the safety margin and reliability: ; where is the safety margin, is the basic safety margin parameter, is the reliability index, and represent the amplification factor and the non-linear adjustment parameter respectively, controlling the influence degree of reliability on the safety margin. Usually , , represents to the power of.

[0071] Fusion of the prediction interval and the safety margin: ; where is the prediction interval considering the safety margin, is the lower safety prediction bound, is the upper safety prediction bound, is the original lower prediction bound, is the original upper prediction bound, is the safety margin.

[0072] Step 5.2, Optimization of charge and discharge strategies; Based on the health state prediction results and the safety margin, optimize the battery charge and discharge control strategies: Adjustment of the charge cut-off condition: ; where is the adjusted charge cut-off state of charge, is the maximum allowable state of charge, is the nominal charge cut-off state of charge, is the safety margin for SOC, represents the minimum value function.

[0073] Adjustment of the discharge cut-off condition: ; where is the adjusted discharge cut-off state of charge, is the minimum allowable state of charge, is the nominal discharge cut-off state of charge, is the safety margin for SOC, represents the maximum value function.

[0074] Charge and discharge power limit: ; ; where is the adjusted maximum charging power, is the adjusted maximum discharging power, is the nominal maximum power, is the power adjustment coefficient, is the reliability index, represents the power adjustment factor.

[0075] Step 5.3, Load distribution and thermal management strategies; According to the health state prediction and reliability assessment, optimize the load distribution and thermal management strategies: Load distribution strategy, for a multi-battery system, distribute the load according to the health state and reliability: ; where is the load ratio assigned to the th battery, is the th battery's health state, is the predicted reliability index of the th battery, is the sum of the products of the state of health and the reliability index of all batteries, is the total load demand, is the number of batteries, is a summation variable representing the battery number, used to calculate the sum of the products of the state of health and the reliability index of all batteries. This formula ensures that batteries with better state of health and more reliable prediction results bear more load, thus optimizing the overall performance of the system and extending the service life of the battery pack.

[0076] Thermal management strategy adjustment: ; where is the cooling intensity, is the basic cooling intensity, is the adjustment amount, is the reliability index of the temperature prediction, represents the degree of unreliability of the temperature prediction.

[0077] Step 5.4, Fault prevention measures and adaptive correction; Implement preventive maintenance and adaptive correction measures for abnormal states: Abnormal state detection, based on the deviation between the prediction interval and the actual measurement value, detect abnormal states: ; where is the anomaly score, is the actual measurement value, is the predicted value, is the upper bound of the safety prediction interval, is the lower bound of the safety prediction interval, represents the absolute value of the difference between the actual value and the predicted value, represents the half-width of the safety prediction interval.

[0078] Preventive maintenance trigger: ; where is the maintenance flag, with a value of 1 indicating maintenance is required and a value of 0 indicating no maintenance is required; is the anomaly score threshold, is the reliability index, is the minimum acceptable reliability index, represents the logical OR operation.

[0079] Policy adaptive correction: According to historical data and current status, adaptively correct the control policy: ; wherein is the control strategy for the next moment, is the current strategy, is the historical optimal strategy, is the learning rate, which controls the strategy update speed, represents the gap between the current strategy and the optimal strategy.

[0080] Through the dynamic control strategy generation step, a comprehensive control strategy including charge and discharge strategies, load distribution, thermal management, and fault prevention measures is output. This strategy has self - adaptability and safety, can dynamically adjust the safety margin according to the reliability of the prediction results, effectively avoid safety risks such as over - charge and over - discharge, and improve the overall safety and service life of the battery system. Through the dynamic control strategy generation step, a comprehensive control strategy including charge and discharge strategies, load distribution, thermal management, and fault prevention measures is output. This strategy has self - adaptability and safety, can dynamically adjust the safety margin according to the reliability of the prediction results, effectively avoid safety risks such as over - charge and over - discharge, and improve the overall safety and service life of the battery system.

[0081] An energy storage battery dynamic control system for implementing the above - mentioned energy storage battery dynamic control method, comprising: A data acquisition and pre - processing module for obtaining battery state and environmental interference information through a multi - modal sensor network and performing signal denoising processing; A feature extraction and denoising module for analyzing battery health state features using a multi - layer temporal convolutional network and filtering abnormal noise; A reliability evaluation module for constructing a correlation model between environmental interference and prediction accuracy and quantifying prediction uncertainty based on the evidence theory; An adaptive inference module for dynamically optimizing the prediction algorithm structure and parameters according to real - time interference conditions and reliability indicators; A control strategy generation module for calculating the safety margin based on the health state prediction results and their confidence levels and generating a differentiated control scheme.

[0082] Here, the present invention provides an implementation example: Actual application tests were carried out in a large - scale industrial micro - grid energy storage power station. The energy storage power station contains 100 groups of lithium - ion battery packs with a total capacity of 10 MWh, mainly used for load peak shaving and valley filling and power grid frequency regulation. The power station is located in an industrial park, and there are large - scale motor equipment and substations around, with complex and variable environmental interference.

[0083] The test environment has the following characteristics: Electromagnetic interference sources: multiple high - power inverters, transformers, and motor equipment; Temperature change range: - 10°C to 45°C, and the daily temperature difference can reach 20°C; Load characteristics: Intermittent high-power discharge and alternating microcirculation charge and discharge appear alternately; Sudden interference: Unpredictable interferences such as voltage fluctuations on the grid side and sudden electromagnetic pulses.

[0084] Test system configuration: Battery state sensors: Each group of batteries is equipped with 8 voltage, current, and temperature sensors, and 1 set of impedance measurement units; Environmental interference sensors: 10 electromagnetic field intensity sensors, 8 environmental noise sensors, 12 vibration sensors, and 6 humidity sensors; Computing platform: Edge computing server, configured with an Intel i7 processor and 16GB of memory, for model inference and generation of control strategies.

[0085] The original signal collected is denoised through multi-scale wavelet decomposition technology, effectively improving the signal quality.

[0086] The system continuously collects data during a 6-month operation cycle, and the data sampling frequency is configured as shown in Table 1: Table 1: Data acquisition parameters and preprocessing effects

[0087] The denoised signal is decomposed into three feature sets: low frequency, medium frequency, and high frequency, providing a high-quality data basis for subsequent analysis. The system can automatically identify and mark the abnormal points in the data, effectively avoiding prediction deviations caused by abnormal data.

[0088] The Mahalanobis distance method is applied for abnormal noise detection, and the characteristics of the microcirculation discharge mode are identified through the multi-head self-attention mechanism. The performance under different working conditions is shown in Table 2: Table 2: Results of abnormal noise detection and microcirculation discharge mode recognition

[0089] Compared with the traditional method, this implementation method can maintain a high abnormal detection accuracy rate in a complex interference environment, and at the same time, the recognition accuracy rate of the microcirculation discharge mode is increased from 60% of the traditional method to an average of 89.9%.

[0090] By constructing an interference and reliability mapping model, the influence of different types of interference on the prediction results is quantified, and a reliability index is generated. The influence coefficients and reliability indices under different interference types are shown in Table 3: Table 3: Relationship between interference types and reliability indices

[0091] The system can evaluate the environmental interference status in real time, dynamically adjust the width of the prediction interval according to the interference degree, ensure that the prediction interval can contain the actual state value, and provide a basis for reliability evaluation for decision-making.

[0092] Under different interference conditions, the system dynamically adjusts the parameters and structure of the prediction model through adaptive reasoning and decision optimization technology to generate optimized health state prediction results and control strategies. The comparison of the root mean square error (RMSE) of SOH prediction under different interference intensities and the system safety and life extension effects are shown in Table 4: Table 4: Analysis of Adaptive Reasoning Effect and System Safety Benefits

[0093] In an extremely strong interference environment, the prediction accuracy of this solution is improved by 53.6% compared with the traditional method, the incidence rate of safety events is reduced by 65.8%, the expected battery life is extended by 20.2%, and at the same time, the consumption of computing resources is reduced by 46.8%, verifying the effectiveness of the adaptive reasoning mechanism.

[0094] This embodiment has been continuously operated in the energy storage power station for 6 months, and a large amount of operation data has been collected for long-term performance verification.

[0095] The health state prediction performances of this embodiment and the traditional method under different working conditions are compared, and the data are from the statistical results under different interference environments during 6 months of continuous operation. As shown in Table 5: Table 5: Comparison of Prediction Accuracy and Reliability during Long-Term Operation

[0096] This embodiment is superior to the traditional method in all prediction indexes. Especially in a strong interference environment, the prediction performance improvement is more obvious, verifying the robustness and adaptability of the system.

[0097] After applying this embodiment, the incidence rate of system safety events is reduced, and the average reduction amplitude reaches 58.9%. As the interference intensity increases, the advantages of this method become more obvious, and the safety event reduction rate reaches 65.8% under extremely strong interference conditions. At the same time, the battery capacity attenuation rate slows down, the expected life is extended by 18.5% on average, and the maintenance cost is reduced by about 32.7%, fully verifying the effectiveness of the reliability-based dynamic control strategy.

[0098] The above describes the embodiments of the present invention, but these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative rather than restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. A method for dynamic control of energy storage batteries, characterized in that: include: The energy storage battery status data and environmental interference data are collected through a multimodal sensor network, and the collected signals are subjected to noise reduction processing to obtain a multimodal feature set; Use a multi-layer temporal convolutional network to process the preprocessed multimodal feature set, extract the temporal features of the battery health status, and realize abnormal noise detection and suppression; Based on the time series characteristics of environmental interference data and battery health status, a mapping relationship between environmental interference and prediction reliability is established, and an uncertainty quantification framework is constructed using evidence theory. The prediction results are expressed as confidence intervals and the reliability index is calculated. According to the interference conditions and prediction reliability, the prediction model parameters and structure are dynamically adjusted to generate health status prediction results and their reliability evaluation; Based on the health status prediction results and reliability assessment, a dynamic management and control strategy for energy storage batteries with safety margin is generated.

2. A method for dynamic control of energy storage batteries according to claim 1, characterized in that: The multimodal sensor network includes a battery status sensor group and an environmental interference sensor group; the battery status sensor group is used to collect battery health status related parameters, including voltage, current, temperature and impedance data; the environmental interference sensor group is used to collect interference signals that affect health status prediction, including electromagnetic field strength, environmental noise, vibration and humidity data.

3. A method for dynamic control of energy storage batteries according to claim 1, characterized in that: The noise reduction process adopts multi-scale wavelet decomposition technology, including: Apply multi-scale wavelet decomposition to the time series data collected by each sensor; Applying adaptive threshold function to the decomposed wavelet coefficients to suppress noise; Reconstruct the denoised signal through inverse wavelet transform; The signal is decomposed into three frequency domain feature sets: high frequency, medium frequency and low frequency, which correspond to different physical meanings respectively.

4. A method for dynamic control of energy storage batteries according to claim 1, characterized in that: The multi-layer temporal convolutional network includes: Causal convolutional layers ensure that the model only uses information before the current moment; Dilated convolutional layers expand the receptive field through dilated convolutions, and the dilation rate increases exponentially with the number of layers; Residual connection structure, optimizes gradient flow and avoids the gradient vanishing problem; Skip connections connect features at different levels and enhance the ability to fusion multi-scale features.

5. The method for dynamic control of energy storage batteries according to claim 1, characterized in that: The abnormal noise detection adopts the Mahalanobis distance method, including: Calculate the mean vector and covariance matrix of the features; Calculate the Mahalanobis distance of each sample point; Set an adaptive threshold. When the Mahalanobis distance is greater than the threshold, it is marked as an outlier. For the detected outliers, interpolation or prediction replacement methods are used to suppress noise.

6. A method for dynamic control of energy storage batteries according to claim 1, characterized in that: The establishing of a mapping relationship between environmental interference and prediction reliability includes: Extracting features from the interference data collected by the environmental interference sensor group to obtain an interference feature vector; Through historical data training, a mapping relationship between interference features and prediction errors is established; The influence of different types of interference on the prediction results is quantified as an influence coefficient matrix; Based on the influence coefficient matrix, the predicted compensation value is calculated.

7. A method for dynamic control of energy storage batteries according to claim 1, characterized in that: The uncertainty quantification framework is based on the theory of evidence and includes: Define the decision framework, including possible states or categories; Calculate the basic probability distribution for each source of evidence; Calculate belief functions based on basic probability assignments; Calculate the likelihood function; Express the prediction results as an interval, and calculate the uncertainty as the width of the prediction interval.

8. A method for dynamic control of energy storage batteries according to claim 1, characterized in that: The step of dynamically adjusting the prediction model parameters and structure includes: Calculate the sensitivity matrix of model parameters with respect to disturbance signatures; Calculate parameter adjustments based on the sensitivity matrix and current interference conditions; Select model structures of different complexity according to the reliability level; Residual connection and recursive connection structures are used to enhance the model's ability to capture long-term dependencies and the efficiency of gradient flow.

9. A method for dynamic control of energy storage batteries according to claim 1, characterized in that: The steps of generating the dynamic control strategy include: Adaptively calculate the safety margin based on the reliability index of the prediction results; Optimize battery charge and discharge control strategy based on health status prediction results and safety margin; Optimize load distribution and thermal management strategies based on health status prediction and reliability assessment; Implement preventive maintenance and adaptive corrective measures for abnormal conditions.

10. A dynamic management and control system for energy storage batteries, characterized in that: A method for dynamically controlling an energy storage battery according to any one of claims 1 to 9, comprising: Data acquisition and preprocessing module, used to obtain battery status and environmental interference information through a multimodal sensor network and perform signal noise reduction processing; Feature extraction and noise reduction module, which is used to apply multi-layer temporal convolutional networks to analyze battery health status characteristics and filter abnormal noise; Reliability assessment module, which is used to build a correlation model between environmental disturbance and prediction accuracy and quantify prediction uncertainty based on evidence theory; Adaptive reasoning module, used to dynamically optimize the prediction algorithm structure and parameters according to real-time interference conditions and reliability indicators; The control strategy generation module is used to calculate the safety margin based on the health status prediction results and their confidence and generate differentiated control plans.

Images