Energy management system data analysis method and device, equipment and storage medium

Abstract

The invention relates to the technical field of battery charging and discharging optimization, and discloses an energy management system data analysis method, device and equipment and a storage medium, and the method comprises the steps: collecting a working condition feature data set of a robot energy storage battery through an energy management system; performing state cooperative calculation on the working condition characteristic data set to obtain battery operation state data; establishing a mapping relation model of load power and charge-discharge efficiency, and determining a battery working interval boundary value to obtain charge-discharge characteristic data; inputting the charge and discharge characteristic data into a deep gating cycle unit network, and obtaining a battery performance prediction model through dynamic modeling; based on a battery performance prediction model, a gain factor self-tuning technology is adopted to construct a closed-loop controller, online optimization calculation is carried out on control parameters through a gradient descent algorithm, a control instruction containing a charging and discharging current limit value and a voltage threshold value is generated, accurate prediction of battery performance is further achieved, and stable operation of an energy management system is ensured.

Description

Technical Field

[0001] The present application relates to the technical field of battery charging and discharging optimization, and in particular to an energy management system data analysis method, device, equipment and storage medium. Background Art

[0002] When existing robot energy storage batteries operate under complex working conditions, the battery capacity can easily decay rapidly due to drastic fluctuations in load power and frequent changes in ambient temperature, seriously affecting the battery life and operating safety.

[0003] Traditional energy management systems mainly use fixed threshold control strategies, which cannot accurately perceive the dynamic characteristics of batteries under different working conditions, resulting in low charging and discharging control accuracy, and it is difficult to adaptively adjust control parameters according to the battery performance degradation. In addition, most of the existing battery status monitoring methods are based on a single algorithm and fail to effectively integrate multi-source information, resulting in large errors in the evaluation results of battery charge and health. The prior art lacks a systematic solution that can take into account both accurate battery status evaluation and optimal control of the charging and discharging process. Especially under high-load conditions of robots, the mapping relationship between battery performance and charging and discharging efficiency is complex and changeable. It is difficult for traditional methods to establish an accurate prediction model and achieve precise control of the charging and discharging process, which can easily cause overcharging and over-discharging, affecting the battery life. Summary of the invention

[0004] The present application provides an energy management system data analysis method, device, equipment and storage medium, thereby achieving accurate prediction of battery performance and ensuring stable operation of the energy management system.

[0005] A first aspect of the present application provides an energy management system data analysis method, the energy management system data analysis method comprising: Collect the operating characteristic data set of the robot's energy storage battery through the energy management system; Performing state collaborative calculation on the operating condition characteristic data set to obtain battery operating state data; According to the operating condition characteristic data set and the battery operating state data, a mapping relationship model between load power and charge and discharge efficiency is established, and a battery operating range boundary value is determined to obtain charge and discharge characteristic data; Inputting the charge and discharge characteristic data into a deep gated recurrent unit network, and obtaining a battery performance prediction model through dynamic modeling; Based on the battery performance prediction model, a closed-loop controller is constructed by adopting the gain factor self-tuning technology, and the control parameters are optimized and calculated online by the gradient descent algorithm to generate control instructions including charge and discharge current limits and voltage thresholds.

[0006] A second aspect of the present application provides an energy management system data analysis device, the energy management system data analysis device comprising: A collection module is used to collect the working condition characteristic data set of the robot energy storage battery through the energy management system; A calculation module, used for performing state collaborative calculation on the operating condition characteristic data set to obtain battery operating state data; An establishment module is used to establish a mapping relationship model between load power and charge and discharge efficiency according to the operating condition characteristic data set and the battery operating state data, and determine the boundary value of the battery working range to obtain charge and discharge characteristic data; A dynamic modeling module, used to input the charge and discharge characteristic data into a deep gated recurrent unit network, and obtain a battery performance prediction model through dynamic modeling; The generation module is used to construct a closed-loop controller based on the battery performance prediction model by using the gain factor self-tuning technology, and to perform online optimization calculation on the control parameters by using the gradient descent algorithm to generate control instructions including the charge and discharge current limit and the voltage threshold.

[0007] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned energy management system data analysis method.

[0008] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned energy management system data analysis method.

[0009] Compared with the prior art, the present application has the following beneficial effects: by screening the working condition data through the cross-validation-Voronoi adaptive sampling method, the representative characteristic data is extracted, the data redundancy is effectively reduced, and the collaborative calculation method of the particle filter, the recursive least squares method and the extended Kalman filter is adopted to realize the accurate evaluation of the battery charge and health, and provide a reliable basis for the formulation of the charge and discharge control strategy. Based on the support vector regression algorithm, a mapping relationship model between load power and charge and discharge efficiency is established. Combined with the electrochemical reaction kinetic characteristics, the battery working range is accurately defined to avoid overcharging and over-discharging. The battery charge and discharge process is dynamically modeled through a deep gated recurrent unit network to achieve accurate prediction of battery performance. The closed-loop controller is constructed by using the gain factor self-tuning technology. The online optimization of the control parameters is realized through the gradient descent algorithm, which improves the accuracy and adaptability of the charge and discharge control. Based on the variable step size recursive least squares method and the pole configuration method, a complete stability constraint system is established to ensure the stable operation of the energy management system. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0011] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0012] Figure 1 is a flow chart of a data analysis method for an energy management system provided by an embodiment of the present invention; Figure 2 is a schematic block diagram of the structure of a data analysis device for an energy management system provided by an embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0015] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0016] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 , an embodiment of the energy management system data analysis method in the embodiment of the present application includes: Step 100: Collecting a working condition characteristic data set of the robot energy storage battery through an energy management system; It is understandable that the execution subject of the present application may be an energy management system data analysis device, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0017] Specifically, the energy management system collects data on the terminal voltage, charge and discharge current, surface temperature, ambient temperature, and load power of the robot's energy storage battery to obtain the original data set, which reflects the battery's charge and discharge performance and working environment. In the data collection process, high-precision sensors and high-speed data acquisition modules are used to ensure the real-time and accuracy of the data, while avoiding data distortion caused by signal noise. The original data set is normalized to eliminate the dimensional differences between different physical quantities and improve the effectiveness of subsequent analysis. The normalization calculation uses the maximum-minimum normalization method or the Z-Score normalization method to compress the data values ​​into a unified range, such as between 0 and 1, through linear mapping or standardized formulas. Normalization processing can not only reduce the interference of extreme points, but also improve the model's ability to identify feature data, and obtain a normalized data set. The data space of the normalized data set is divided into k subspaces. A clustering algorithm (such as K-Means or DBSCAN) is used to ensure that the distribution of data points in each subspace has a certain similarity. After completing the subspace division, the data density of each subspace is calculated. Based on the density estimation method, the density value reflecting the data distribution characteristics of the subspace is obtained by counting the distribution frequency of the data points in each subspace. Based on the data space distribution characteristics, the Voronoi diagram partitioning algorithm is used to calculate the distance matrix between nodes. The Voronoi diagram is a space partitioning method. By taking each data point as the center and using the distance metric to construct the polygonal partitioning boundary, multiple non-overlapping Voronoi regions are formed. When calculating the distance matrix between nodes, the Euclidean distance, Manhattan distance or other distance measurement methods suitable for the data distribution characteristics are selected. Based on the distance matrix, the constructed Voronoi region can intuitively reflect the geometric distribution of data points and their neighboring relationships. The Voronoi region partitioning results are subjected to a 5-fold cross validation. By randomly dividing the data into five parts, four of them are selected as training sets and one as a validation set in turn, and the data representativeness of each region is evaluated. Through cross validation, the analysis result error caused by data deviation can be effectively reduced, while ensuring the scientificity and reliability of regional division. According to the verification results, the weight coefficient of each region is calculated. Based on the regional weight coefficient, the sampling probability of the data points in each Voronoi region is calculated. The calculation of sampling probability is based on regional weight. By combining the data density and distribution characteristics in the region, the sampling probability of data points in each region is adaptively adjusted to construct an adaptive sampling probability distribution. When using adaptive sampling probability distribution for data resampling, it ensures that data points in high-weight areas get more sampling opportunities, while data points in low-weight areas have fewer sampling opportunities. This sampling method can effectively balance the distribution of data sets, improve data representativeness, and generate target sampling data sets. The target sampling data sets are classified and labeled according to the working condition type to generate working condition feature data sets.When classifying and labeling, different operating condition categories are divided according to the characteristics such as charge and discharge status, temperature range, load power level, etc., combined with the actual working condition requirements. The labeling process requires a combination of manual experience and automated algorithms to ensure the accuracy and applicability of the working condition characteristic data set.

[0018] Step 200, performing state collaborative calculation on the operating condition characteristic data set to obtain battery operating state data; Specifically, a battery equivalent circuit model is constructed based on the terminal voltage and charge and discharge current of the operating condition characteristic data set. A second-order RC equivalent circuit is used for state space modeling. By describing the internal resistance characteristics and dynamic characteristics of the battery with ohmic internal resistance, polarization internal resistance and polarization capacitance, a circuit model containing these parameters is constructed to reflect the dynamic characteristics of the battery. The equivalent circuit can simulate the voltage-current response relationship of the battery under different charge and discharge states. Based on the battery equivalent circuit model, the recursive least squares method of the forgetting factor is used to iteratively calculate the circuit parameters, effectively track the dynamic changes of the battery parameters, and ensure that the model always maintains accuracy during long-term use. The capacity state transfer equation is established by the coulomb counting method, and the capacity change is calculated by the current integral method of the battery to improve the parameter update model. Combining the recursive least squares method and the coulomb counting method makes the real-time estimation of the battery capacity more accurate and reliable. In order to deal with the capacity decay state of the battery, the particle filter technology is introduced. The particle filter is initialized for the capacity decay state to set the initial distribution of the particle group. After initialization, the state of each particle is predicted one step based on the initial state of the particle swarm. The predicted result is combined with the observation equation to calculate the likelihood function value of the particle. These likelihood function values ​​are converted into particle weight update values ​​through normalization, thereby reflecting the credibility of each particle in the current state. In order to ensure the calculation stability of the particle filter, the particle weight update value is judged by the number of effective particles. When the number of effective particles is less than half of the total number of particles, it indicates that the particles are severely degraded. At this time, the particle swarm distribution is improved by resampling, and the calculation result of capacity decay is obtained. Based on the calculation result of capacity decay, the nonlinear state equation and observation equation of the extended Kalman filter are constructed. The battery charge (SOC) and health (SOH) are used as state variables, and the terminal voltage is used as the observation variable. The state observation modeling is carried out in combination with the calculation result of capacity decay. The extended Kalman filter predicts the state variables through the nonlinear state equation, and estimates the observation variables through the observation equation, forming a closed-loop state estimation framework. In order to adapt the nonlinear equation to the calculation requirements of the Kalman filter, the state equation and the observation equation are linearized by the Jacobian matrix respectively. The core of linearization processing is to calculate the state transfer matrix and the observation matrix, and set the process noise covariance matrix and the observation noise covariance matrix to describe the uncertainty of the system state and observation. After completing the construction of the above linearized state space equation, the battery state is optimally estimated by the extended Kalman filter algorithm. The filtering algorithm gradually adjusts the deviation between the state prediction value and the observation value through iterative calculation to obtain the optimal estimate of the battery charge and health. The calculation results of the battery charge and health are fused, and the weights and observation information of each state variable are comprehensively considered to ensure that the final output battery operation status data can accurately reflect the actual state of the battery.

[0019] Step 300: Establish a mapping relationship model between load power and charge and discharge efficiency based on the operating condition characteristic data set and the battery operating status data, and determine the battery operating range boundary value to obtain charge and discharge characteristic data; It should be noted that the terminal voltage, charge and discharge current, surface temperature and load power data are extracted from the operating condition feature data set and paired according to the time series to ensure the time series consistency between the data. On this basis, the charge and health status in the battery operating status data are combined to establish a multidimensional feature matrix. This matrix integrates the internal and external state characteristics and load characteristics of the battery, and can fully reflect the dynamic behavior of the battery under different operating conditions. The multidimensional feature matrix is ​​segmented into power intervals, and the load power is divided into several intervals, and the charge and discharge efficiency in each interval is calculated separately. The interval charge and discharge efficiency data can reveal the performance changes of the battery in different load power ranges, and provide the necessary input data for establishing the mapping relationship between load power and charge and discharge efficiency. In this process, it is ensured that the power range between the partitions covers the actual operating range, and at the same time, it is ensured that the amount of data in each interval is sufficient to support statistical analysis. Based on the interval charge and discharge efficiency data, the support vector regression algorithm is used to establish the mapping relationship model between load power and charge and discharge efficiency. When selecting the kernel function, the Gaussian radial basis function is used as the kernel function to capture complex nonlinear relationships. In order to avoid overfitting or underfitting problems, the penalty factor is set to 0.1 to strike a balance between model complexity and prediction accuracy. By training and optimizing the support vector regression model, an efficiency prediction model that can accurately predict the charge and discharge efficiency under different load powers is constructed. Using the output results of the efficiency prediction model, the trend of the charge and discharge efficiency changing with the load power is analyzed to obtain the efficiency mapping relationship model. In this process, the key points in the efficiency curve are focused on, such as the efficiency peak point and the point where the efficiency drops sharply. The load power values ​​of these key points can reflect the performance limit of the battery under different load conditions. After extracting the load power values ​​of the key points of the efficiency mapping relationship model, combined with the battery surface temperature data, the reaction rate constant under the corresponding working conditions is calculated by the electrochemical reaction equation. The electrochemical reaction equation describes the diffusion and reaction process of lithium ions inside the battery, and the rate constant is the core parameter to characterize this dynamic process. The calculated kinetic characteristic parameters reveal the performance change characteristics of the battery under different load power and temperature conditions. Based on the kinetic characteristic parameters, the lithium ion diffusion coefficient and charge migration coefficient of the positive and negative electrode materials of the battery are calculated. These two parameters reflect the diffusion ability and electron transfer ability of the electrode material during the reaction process, which can help evaluate the dynamic performance of the battery. By using the electrochemical impedance model, the reaction process of the electrode interface is analyzed to obtain the polarization characteristic data of the electrode. The electrochemical impedance model can decompose the main components of battery polarization, including concentration polarization and electrochemical polarization. According to the electrode polarization characteristic data, a mathematical model including concentration polarization and electrochemical polarization is established, and the upper and lower limit thresholds of the polarization voltage are determined by numerical solution. These thresholds define the safe operating range of the battery to prevent excessive charging and discharging from damaging the battery life and safety.The working range restriction conditions are combined with the calculation results of the efficiency mapping relationship model to form the charge and discharge characteristic data, which includes both the efficiency characteristics of the battery and the boundaries of its safe operation.

[0020] Step 400: Input the charge and discharge characteristic data into a deep gated recurrent unit network, and obtain a battery performance prediction model through dynamic modeling; Specifically, the charge and discharge characteristic data are divided to construct a training data set and a validation data set to ensure the uniformity and representativeness of the data set distribution. Most of the data is used to train the network through random sampling and partitioning, and the rest is used to verify the performance of the network. After the data set is divided, it is necessary to ensure that the training data contains sufficient timing information and the validation data can effectively detect the generalization ability of the model. Construct the input layer of the deep gated recurrent unit network. The design of the input layer is consistent with the number of features of the charge and discharge characteristic data, including five key features such as charge and discharge current, terminal voltage, temperature, efficiency, and working range boundary value. In order to capture the timing dependence of battery performance, the timing length is set to 100 time steps, that is, the network can learn the dynamic changes of the past 100 time steps through the input sequence. Construct the hidden layer of the deep gated recurrent unit network. The hidden layer is designed to contain three stacked GRU layers to capture the high-order timing relationship between features layer by layer. The number of units in the first layer of GRU is set to 256 to extract rich features of the original input; the number of units in the second layer of GRU is reduced to 128, focusing on the extracted intermediate features; the third layer of GRU is further compressed to 64 units to extract the final time-related features. Each layer of GRU selectively memorizes and forgets the temporal features by updating the gate, resetting the gate and the candidate hidden state. This structure can effectively handle the long-term dependency problem in battery performance prediction. In order to avoid network overfitting, a dropout layer is added between each layer of GRU in the hidden layer, and the dropout rate is set to 0.3 to randomly discard the output of some neurons, thereby enhancing the robustness of the model. At the same time, a batch normalization layer is added to the output of each layer of GRU, and the normalization parameter is set to 0.9. By normalizing the data of each mini-batch, the training of the model is more stable and the convergence is accelerated. The generalization ability of the model is improved by regularizing the structure. After completing the hidden layer design, the output layer of the deep gated recurrent unit network is constructed. The output layer adopts a fully connected layer structure, and its output dimension is set to the number of predicted targets, including three key parameters: charging and discharging current, terminal voltage, and remaining capacity at future moments. In order to ensure the continuity of the output, the activation function of the output layer adopts a linear function, which directly reflects the actual numerical range of the predicted target. The parameters of the deep gated recurrent unit network are initialized. During the initialization process, the weights are initialized using the Xavier initialization method. By reasonably allocating the initial value range of the weights, the effective propagation of the gradient is ensured to prevent the occurrence of gradient explosion or disappearance problems. The bias term is initialized to zero to avoid bias in the initial prediction of the model. In terms of optimizer selection, the Adam optimization algorithm is selected. Its adaptive learning rate adjustment mechanism can improve the optimization efficiency. At the same time, the initial value of the learning rate is set to 0.001 to ensure a smooth training process. After initialization, the training data set is input into the initialized network, and the mean square error loss function (MSE) is used for forward propagation calculation.In the forward propagation, the network generates a predicted value based on the current parameters and optimizes the model by calculating the error between the predicted value and the true value. The gradient is calculated by the back-propagation algorithm and the network parameters are updated according to the gradient. To avoid over-training of the model, an early stopping mechanism is adopted. When the loss of the validation data set does not decrease in 10 consecutive epochs, the training process is stopped. Through the above steps, a battery performance prediction model is formed that can accurately predict the future charge and discharge current, terminal voltage and remaining capacity.

[0021] Step 500: Based on the battery performance prediction model, a closed-loop controller is constructed using a gain factor self-tuning technique, and control parameters are optimized online using a gradient descent algorithm to generate control instructions including charge and discharge current limits and voltage thresholds.

[0022] Specifically, based on the battery performance prediction model, the initial controller structure of the gain factor self-tuning closed-loop controller is constructed. The initial structure contains basic control loops, such as the proportional integral derivative (PID) control framework, and combines the output data of the battery performance prediction model (such as current, voltage and remaining capacity at future times) to provide a dynamic reference signal for the controller. At this stage, a gain self-tuning objective function based on model prediction is designed. The form of the objective function needs to be able to reflect the performance indicators of the control system, including response time, steady-state error and robustness to disturbances. By constructing this objective function, a clear direction is provided for subsequent gain optimization. In order to optimize the gain factor, the gradient descent method is used to iteratively solve the objective function. The gradient vector is calculated based on the partial derivative of the objective function with respect to each gain factor, and the gain factor is gradually updated using the gradient descent method. In each iteration, the gain factor is adjusted according to the update formula so that the objective function gradually converges to the optimal value. Through optimization, a set of optimized gain factors are obtained, which can dynamically adjust the parameters of the controller to adapt to changes in battery performance and ensure that the response performance of the controller is optimal. According to the optimized gain factor, the variable step recursive least squares method is used to calculate the dynamic response characteristics of the controller in order to further analyze the performance of the closed-loop system. The recursive least squares method can efficiently capture the dynamic changes of the controller under different working conditions by gradually updating the model parameters. Based on this step, the dynamic model of the closed-loop system is constructed, and the stability analysis of the model is performed. The core of the stability analysis is to calculate the characteristic root distribution of the system, that is, whether the dynamic characteristics of the closed-loop system under different states are in the stable region. By analyzing the distribution of the characteristic roots, potential unstable points are identified, and targeted adjustment measures are taken. In order to ensure the stability of the closed-loop control system, the pole configuration method is used to constrain the gain factor and obtain the stability constraint condition. The pole configuration method can effectively constrain the value range of the gain factor by setting the poles of the system in the expected stable region, so that it can meet the dynamic performance requirements and avoid introducing system oscillation or instability. Based on the above stability constraints, the control parameters of the charging and discharging process are optimized online. Combined with real-time battery performance prediction data, the control parameters are adjusted through the optimization algorithm. For example, optimize the upper limit of the charging current and the lower limit of the discharging current to meet the current power demand and health status of the battery; optimize the upper limit threshold of the charging voltage and the lower limit threshold of the discharging voltage to prevent the battery from being overcharged or over-discharged. Through online optimization of these parameters, precise control of the charging and discharging process can be achieved to maximize battery efficiency while extending battery life. On the basis of optimizing the control parameters, combined with the output data of the battery performance prediction model, a switching strategy for the charging and discharging working mode is established. The working mode switching strategy refers to selecting the most suitable charging and discharging mode under different working conditions, such as fast charging mode, constant current discharge mode or deep cycle mode.The strategy dynamically adjusts based on optimized control parameters and predicted data to adapt to different load requirements and environmental conditions, ensuring that the battery always operates in the best state. The operating mode parameters are converted into control instructions, including the upper limit of the charging current, the lower limit of the discharge current, the upper threshold of the charging voltage, and the lower threshold of the discharge voltage. These instructions are directly passed to the energy management system (BMS) or robot control system to achieve real-time control and optimized management of the battery.

[0023] According to the optimized gain factor, the initial model of recursive calculation is constructed. The model contains the control input matrix and the system output matrix, which is used to describe the input-output relationship of the controller in the closed-loop system. In the recursive calculation process, in order to reflect the dynamic characteristics of the system in real time, the initial model is updated to generate a weighted data sequence. The weighted data sequence gives higher weights to the latest data by introducing the time weight factor, which enhances the sensitivity of the model to real-time changes. Based on the generated weighted data sequence, the controller parameters are recursively updated by the variable step-size recursive least squares method. The adjustment of the update step size follows a specific rule, that is, the step size is proportional to the square of the prediction error and inversely proportional to the trace of the covariance matrix. This design can balance the convergence speed and accuracy of the model, avoiding divergence caused by too large a step size or slow convergence caused by too small a step size. Through multiple iterative calculations, the dynamic response parameters are gradually obtained to describe the dynamic response characteristics of the controller under different input conditions. After obtaining the dynamic response parameters, system identification is performed to establish a closed-loop system dynamic model. The dynamic model of the closed-loop system consists of a state matrix, an input matrix, and an output matrix, which can describe the dynamic behavior and response law of the system. In this stage, the relationship between the controller input and the system output is comprehensively analyzed to ensure that the model can reflect the dynamic characteristics of the actual system. The dynamic model of the closed-loop system is discretized into a z-domain transfer function. Through discretization, the dynamic characteristics of the continuous-time system are mapped to the discrete time domain, which is convenient for further stability analysis. Through the discretized z-domain transfer function, the roots of the characteristic polynomial are calculated, and the calculation results are mapped to the s-plane to obtain the characteristic root distribution of the system. The characteristic root distribution is an important basis for system stability analysis, and its position directly determines the dynamic performance and stability of the system. In the s-plane, the dominant poles and non-dominant poles are divided according to the distribution of the characteristic roots. The dominant poles mainly determine the response speed and oscillation characteristics of the system, while the non-dominant poles have less impact on the system but still need to be considered. After the desired pole positions are determined, the state feedback controller is designed based on these positions to adjust the dynamic behavior of the system. In order to calculate the state feedback gain matrix, the Ackerman formula is used. The Ackerman formula is a pole configuration method that can calculate the state feedback gain matrix through linear algebra so that the poles of the system are accurately configured to the expected positions. Through this method, the dominant poles of the system are configured to the position that meets the performance requirements, and the distribution of the non-dominant poles is constrained to obtain the gain constraint equation. The gain constraint equation is matched with the optimized gain factor to ensure the stability of the closed-loop system. The core of the matching calculation is to verify whether the optimized gain factor meets the stability constraint condition and adjust the value range of the gain factor when necessary. Through the iterative calculation process, a set of stability constraints is finally generated.

[0024] In the embodiment of the present application, the working condition data is screened by the cross-validation-Voronoi adaptive sampling method, representative feature data is extracted, data redundancy is effectively reduced, and the collaborative calculation method of particle filter, recursive least squares method and extended Kalman filter is adopted to realize the accurate evaluation of battery charge and health, and provide a reliable basis for the formulation of charge and discharge control strategy. Based on the support vector regression algorithm, a mapping relationship model between load power and charge and discharge efficiency is established. Combined with the electrochemical reaction kinetic characteristics, the battery working range is accurately defined to avoid overcharging and over-discharging. The battery charging and discharging process is dynamically modeled through a deep gated recurrent unit network to achieve accurate prediction of battery performance. A closed-loop controller is constructed using gain factor self-tuning technology. The online optimization of control parameters is realized through the gradient descent algorithm, which improves the accuracy and adaptability of charge and discharge control. Based on the variable step size recursive least squares method and the pole configuration method, a complete stability constraint system is established to ensure the stable operation of the energy management system.

[0025] In a specific embodiment, the process of executing step 100 may specifically include the following steps: The energy management system collects data on the terminal voltage, charge and discharge current, surface temperature, ambient temperature, and load power of the robot's energy storage battery to obtain an original data set, and performs normalization calculation on the original data set to obtain a normalized data set; The data space of the normalized data set is divided into k subspaces, and the data density of each subspace is calculated to obtain the data space distribution characteristics; According to the spatial distribution characteristics of the data, the Voronoi diagram partitioning algorithm is used to calculate the distance matrix between nodes in each subspace, and the polygonal partitioning boundary is constructed based on the distance matrix to obtain the Voronoi region partitioning result. The Voronoi region division results were cross-validated 5-fold, and the data was randomly divided into 5 parts, 4 of which were selected as training sets and 1 as validation set. The data representativeness of each region was calculated to obtain the regional weight coefficient; According to the regional weight coefficient, the sampling probability of the data points in each Voronoi region is calculated to obtain an adaptive sampling probability distribution, and based on the adaptive sampling probability distribution, each Voronoi region is resampled to obtain the target sampling data set; The target sampling data set is classified and labeled according to the working condition type to obtain the working condition characteristic data set.

[0026] Specifically, the operating parameters of the robot energy storage battery are collected in real time through sensors, including terminal voltage (Indicates the battery time voltage value at the time of charge and discharge current (indicates the instantaneous current of the battery), surface temperature (battery case temperature), ambient temperature (external environment temperature) and load power (Represents the real-time power output of the battery). These data are aggregated by the acquisition module to form the original data set ,in Indicates the sampling time. In order to eliminate the dimensional differences between different physical quantities and improve the convergence of the model, the original data set is normalized. The normalization adopts the maximum-minimum normalization formula: ; in, Indicates A characteristic value of a data point, and are the minimum and maximum values ​​of the feature, respectively. is the normalized eigenvalue. After normalization, the value range of all data is compressed to [0, 1], and the normalized data set is obtained. In order to explore the distribution characteristics of the data, the normalized data set is divided into subspaces. -Means clustering algorithm, whose goal is to minimize the squared distance between data points and the centroid in each subspace: ; in, It is A data set of subspaces, It is a subspace The center of mass, Represents data points To the centroid After clustering, we get subspaces, and the data distribution density in each subspace is calculated by the following formula: ; in, Representation subspace The number of data points in It is a subspace To further describe the geometric relationship of the data, based on the above subspace distribution characteristics, the Voronoi diagram partitioning algorithm is used to calculate the distance matrix between nodes, and the polygon partitioning boundary is constructed based on the matrix. Each area of ​​the Voronoi diagram is controlled by the nearest centroid, and its mathematical expression is: ; in, It belongs to the center of mass The Voronoi region of is the feature space. By constructing the Voronoi diagram, the polygonal boundary of each subspace is formed, and the Voronoi region division result is obtained. Based on the Voronoi region division result, the data representativeness of each region is verified. Using five-fold cross validation, the data is randomly divided into five parts, four of which are selected as training sets each time, and the remaining one is used as a validation set. For the data representativeness of each region, the regional weight coefficient is calculated by the following formula: ; Among them, Val Indicates area Contribution to the model prediction performance in cross validation. Based on the regional weight coefficient , calculate the sampling probability for each data point in the Voronoi region. The adaptive sampling probability distribution formula is as follows: ; in, It is a data point The sampling probability of Contains data points Based on the adaptive sampling probability distribution, each Voronoi region is resampled to obtain the target sampling data set The target sampling data set is classified and labeled according to the operating condition type. For example, operating condition type 1 (low current charging) and operating condition type 2 (high current discharging) are defined according to the range of charge and discharge currents. The label of each data point is assigned according to the range in which its characteristic value falls, forming a labeled operating condition characteristic data set.

[0027] In a specific embodiment, the process of executing step 200 may specifically include the following steps: The terminal voltage and charge / discharge current of the operating condition characteristic data set are modeled in state space based on the second-order RC equivalent circuit, and a battery equivalent circuit model including ohmic internal resistance, polarization internal resistance and polarization capacitance is constructed; Based on the battery equivalent circuit model, the recursive least square method of the forgetting factor is used to iteratively calculate the circuit parameters, and the capacity state transfer equation based on the Coulomb counting method is established to obtain the parameter update model; Initialize the particle filter for the capacity decay state in the parameter update model to obtain the initial distribution of the particle swarm. According to the initial distribution of the particle swarm, make a one-step prediction of the state of each particle, calculate the likelihood function value based on the observation equation, and obtain the particle weight update value through normalization. The effective particle number is judged for the particle weight update value. When the effective particle number is less than half of the total number of particles, the capacity attenuation calculation result is obtained. Based on the calculation results of capacity decay, the nonlinear state equation and observation equation of the extended Kalman filter are constructed, and the battery charge and health status are used as state variables, and the terminal voltage is used as the observation variable to obtain the state observation model; The Jacobian matrix of the state observation model is linearized, the state transfer matrix and the observation matrix are calculated respectively, and the process noise covariance matrix and the observation noise covariance matrix are set to obtain the linearized state space equation; According to the linearized state space equation, the optimal state estimation value is calculated by the extended Kalman filter algorithm, and the calculation results of the battery charge and health status are fused to obtain the battery operation status data.

[0028] Specifically, for the terminal voltage in the operating characteristic data set and charge and discharge current The state space modeling based on the second-order RC equivalent circuit is carried out. The second-order RC battery model consists of three main parts: ohmic internal resistance , Polarization internal resistance and polarization capacitance The model can accurately describe the transient response and dynamic behavior of the battery. The terminal voltage of the battery It is expressed as: ; in, is the open circuit voltage, and They are respectively and The dynamic polarization voltage of the composition is defined as: ; After discretizing the above equation, we get the state space form: ; ; Among them, the state vector ,enter , output , the state transfer matrix , input matrix , output matrix and pass-through matrix By circuit parameters After the model is established, the parameters are dynamically updated. , using the recursive least squares (RLS) method with a forgetting factor. The goal of the recursive least squares method is to minimize the sum of squared prediction errors, and its update formula is: ; in, is a parameter vector containing is the gain vector; is the observation vector; the forgetting factor Used to balance historical data and new data. After recursive update, dynamically adjusted circuit parameters are obtained, and the capacity state transfer equation is established using the Coulomb counting method: ; in, is the state of charge, is the nominal capacity of the battery, is the sampling time. Based on the parameter update model, the particle filter is initialized for the capacity decay state to obtain the initial distribution of the particle swarm. Particle filtering is a nonlinear state estimation method. First, initialize particles, each particle represents a possible capacity state. For each particle state Make a one-step forecast and calculate the likelihood function value based on the observation equation: ; After normalization, the particle weight update value is obtained When the effective number of particles Less than When the charge is reduced, the particles are resampled and the calculation results of capacity decay are obtained. Using the capacity decay results, the nonlinear state equation and observation equation of the extended Kalman filter are constructed to convert the charge and health status As a state variable, the terminal voltage As the observed variable. The state equation is: ; The observation equation is: ; in, and They are process noise and observation noise respectively. The nonlinear equation is linearized by Jacobian matrix and the state transfer matrix is ​​calculated. and the observation matrix , and set the process noise covariance matrix and the observation noise covariance matrix , and obtain the linearized state space equation. By extending the prediction and update steps of the Kalman filter, the optimal state estimate is calculated. and health status The estimation results are fused to obtain complete battery operation status data.

[0029] In a specific embodiment, the process of executing step 300 may specifically include the following steps: The terminal voltage, charge and discharge current, surface temperature and load power data in the operating condition characteristic data set are paired in time series, and a multi-dimensional feature matrix is ​​established by combining the battery charge and health status in the battery operating status data; The multi-dimensional feature matrix is ​​processed into power intervals to obtain the charging and discharging efficiency data between the intervals; Based on the charging and discharging efficiency data between partitions, the support vector regression algorithm is used to establish the mapping relationship between load power and charging and discharging efficiency. The kernel function selects Gaussian radial basis function, and the penalty factor is set to 0.1 to obtain the efficiency prediction model. According to the output results of the efficiency prediction model, the variation trend of the charging and discharging efficiency with the load power is analyzed to obtain the efficiency mapping relationship model; Extract the load power value of the key point in the efficiency mapping relationship model, combine it with the battery surface temperature data, calculate the reaction rate constant under the corresponding working condition through the electrochemical reaction equation, and obtain the kinetic characteristic parameters; Based on the kinetic characteristic parameters, the lithium ion diffusion coefficient and charge migration coefficient of the positive and negative electrode materials of the battery are calculated, and the electrochemical impedance model is used to analyze the interface reaction process to obtain the electrode polarization characteristic data; Based on the electrode polarization characteristic data, a mathematical model including concentration polarization and electrochemical polarization is established. The upper and lower thresholds of the polarization voltage are determined by numerical solution to obtain the working range restriction conditions. The working range restriction conditions are then combined with the calculation results of the efficiency mapping relationship model to generate the charge and discharge characteristic data.

[0030] Specifically, the terminal voltage is extracted from the operating condition characteristic data set , charge and discharge current , surface temperature and load power The load power is calculated by the formula Calculated, among which is the time step, is the instantaneous voltage of the battery, is the current. At the same time, the battery charge (SOC) and health status (SOH) are obtained from the battery operating status data. These features are aligned in time to form a multi-dimensional feature matrix : ; Among them, each row represents the state of a time step, is the total amount of data, and the matrix dimension is . For the load power in the multidimensional feature matrix Perform interval segmentation processing. Assume that the load power range is divided into intervals, each interval is defined as , calculate the charge and discharge efficiency of each interval respectively. The calculation formula of charge and discharge efficiency is: ; in, and Respectively represent The average output power and input power in the interval, efficiency is a dimensionless quantity. Based on the data of charge and discharge efficiency between partitions, the support vector regression (SVR) algorithm is used to establish the mapping relationship between load power and charge and discharge efficiency. SVR uses Gaussian radial basis function (RBF kernel) to capture nonlinear relationships, and its kernel function is defined as ; in, is the sample point, Control the distribution width of the kernel function. During training, the goal of SVR is to minimize the following loss function: ; in, is the weight vector, is the penalty factor (set to 0.1), is a slack variable, which indicates the allowable prediction error range. The efficiency prediction model obtained through training outputs the charging and discharging efficiency under any load power. Based on the output results of the efficiency prediction model, the changing trend of the charging and discharging efficiency with the load power is analyzed to obtain the efficiency mapping relationship model. By observing the key points in the model (such as the efficiency peak point, the point where the efficiency drops rapidly, etc.), the corresponding load power value is extracted. Combined with the battery surface temperature , using the Arrhenius equation to calculate the reaction rate constant under working conditions : ; in, is the frequency factor, is the activation energy, is the gas constant, is the surface temperature. The value is used to characterize the reaction rate under different temperature and power conditions. , calculate the lithium ion diffusion coefficient of the positive and negative electrode materials and the charge transfer coefficient The diffusion coefficient is calculated by the following formula: ; in, is the lithium ion concentration, It's time. is the spatial second-order derivative of lithium ion concentration. The migration coefficient is obtained by fitting the experimental data. The electrochemical impedance model is used to analyze the interface reaction process and calculate the polarization characteristic data of the electrode, including concentration polarization. and electrochemical polarization The mathematical model of polarization voltage is: ; Determine the upper and lower thresholds of polarization voltage by numerical solution and . Combined with the results of the efficiency mapping model, the charge and discharge characteristic data is generated.

[0031] In a specific embodiment, the process of executing step 400 may specifically include the following steps: Divide the charge and discharge characteristic data to construct training data sets and verification data sets; Construct the input layer of the deep gated recurrent unit network. The input dimension is the number of features of the charge and discharge characteristic data, including charge and discharge current, terminal voltage, temperature, efficiency, and working range boundary value. The time series length is set to 100 time steps to obtain the input layer structure. Construct the hidden layer of the deep gated recurrent unit network, which includes three stacked GRU layers. The number of GRU units in the first layer is 256, the number of GRU units in the second layer is 128, and the number of GRU units in the third layer is 64. Each GRU layer contains an update gate, a reset gate, and a candidate hidden state to obtain the hidden layer structure. A dropout layer is added between each GRU layer of the hidden layer, and the dropout rate is set to 0.3. A batch normalization layer is added at the output of each GRU layer, and the normalization parameter is set to 0.9 to obtain a regularized structure. Construct the output layer of the deep gated recurrent unit network, using a fully connected layer structure. The output dimension is the number of predicted targets, including the charge and discharge current, terminal voltage, and remaining capacity at future times. The activation function uses a linear function to obtain the output layer structure. The parameters of the deep gated recurrent unit network are initialized. The weights are initialized using the Xavier method, the bias term is initialized to zero, the optimizer uses the Adam algorithm, and the learning rate is set to 0.001 to obtain the initialized network. The training data set is input into the initialized network, the mean square error loss function is used for forward propagation calculation, and the gradient is calculated through the back propagation algorithm to update the network parameters. The training is stopped when the loss of the verification data set does not decrease for 10 consecutive epochs to obtain the battery performance prediction model.

[0032] Specifically, the data set is divided from the charge and discharge characteristic data. The charge and discharge characteristic data includes the charge and discharge current , terminal voltage ,temperature ,efficiency and working range boundary value , these features constitute a time series data set. Assume that the total amount of data is , the data is sorted according to The data is divided into a training data set and a validation data set in a ratio of . In order to preserve the dependency of the time series, the partitioning process adopts a sliding window method, and the length of each window is set to 100 time steps, which means that the input of the model needs to consider the feature values ​​of the past 100 moments. The input layer of the deep gated recurrent unit network is constructed. The design of the input layer is based on the number of features of the charge and discharge characteristic data, that is, the input dimension is ,include , , , , . Assume the time series length is , so the shape of the input tensor is The input layer receives this tensor and completes the time series structured processing of the features. The hidden layer of the deep GRU network is constructed, and a three-layer stacked GRU structure is used to extract time series features layer by layer. The number of units in the first layer of GRU is set to 256, the second layer is set to 128, and the third layer is set to 64. The GRU unit contains an update gate , Reset Gate and candidate hidden states , and their update rules are as follows: ; ; ; ; in, and is the weight matrix, is the bias vector, is the sigmoid function, represents the element-wise multiplication operation, is the hidden state vector. Each layer of GRU captures different levels of features in the time series and passes them to the next layer. In order to prevent overfitting, a dropout layer is added between each layer of GRU, and the dropout rate is set to 0.3 to randomly discard the output of some neurons. At the same time, a batch normalization layer is added at the output of each layer of GRU, and its normalization formula is: ; in, is the input, and is the mean and variance of the mini-batch, and is a learnable parameter, is a small value for numerical stability. Batch normalization accelerates the training convergence of the network. After completing the hidden layer design, the output layer of the network is constructed. The output layer is a fully connected layer, and its output dimension is the number of predicted targets, including the charging and discharging currents at future moments. , terminal voltage and remaining capacity The activation function uses a linear function , to ensure the continuity of the output. Initialize the network parameters. The weight matrix uses the Xavier initialization method, and its standard deviation is defined as: ; The bias term is initialized to zero to avoid additional offset in the initial stage. The optimizer uses the Adam algorithm, and its update rule is: ; ; ; ; in, is the gradient, and are the first and second order momentum, is the learning rate (set to 0.001), is a numerically stable parameter. The training data set is input into the network and the mean square error loss function is used. Perform forward propagation calculations. In each iteration, use the backpropagation algorithm to calculate the gradient and update the network parameters. When the loss function value of the validation dataset does not decrease in 10 consecutive epochs, the training stops.

[0033] In a specific embodiment, the process of executing step 500 may specifically include the following steps: Based on the battery performance prediction model, the initial controller structure of the gain factor self-tuning closed-loop controller is constructed, and for the initial controller structure, a gain self-tuning objective function based on model prediction is constructed; Based on the gain self-tuning objective function, the gradient vector of each gain factor is iteratively updated using the gradient descent method to obtain the optimized gain factor; According to the optimized gain factor, the dynamic response characteristics of the controller are calculated by using the variable step-size recursive least squares method to obtain the closed-loop system dynamic model. The stability analysis of the closed-loop system dynamic model is performed, the characteristic root distribution of the system is calculated, and the gain factor is constrained by the pole configuration method to obtain the stability constraint condition. Based on the stability constraint, the control parameters of the charge and discharge process are optimized online to obtain the optimized control parameters. Based on the optimized control parameters and the output performance prediction data of the battery performance prediction model, the charge and discharge working mode switching strategy is established to obtain the working mode parameters. The working mode parameters are converted into control instructions, which include a charging current upper limit value, a discharging current lower limit value, a charging voltage upper limit threshold value, and a discharging voltage lower limit threshold value.

[0034] Specifically, based on the battery performance prediction model, the initial structure of the gain factor self-tuning closed-loop controller is constructed. The closed-loop controller is based on the proportional-integral-derivative (PID) controller, which controls the output It indicates the adjustment of current or voltage during the charging and discharging process. The formula is: ; in, , and are the proportional, integral and derivative gain factors respectively, is the systematic error, is the desired output (such as target voltage or current), is the actual output (such as current voltage or current). The goal of the initial structure design is to ensure basic tracking and steady-state performance while providing flexibility for optimization of gain factors. Based on the initial controller structure, a gain self-tuning objective function based on model prediction is constructed. The objective function is in the form of: ; in, is the prediction time domain length, is the regularization coefficient, which is used to balance the control error and the gain factor to avoid excessive gain leading to system instability or actuator saturation. In order to optimize the gain factor, the objective function Find the partial derivative and calculate the gradient of each gain factor: ; Similarly, calculate and The gain factor is iteratively updated by the gradient descent method, and the update formula is: ; in, is the learning rate. After several iterations, the optimized gain factor is obtained. Based on the optimized gain factor, the dynamic response characteristics of the controller are calculated using the variable step size recursive least squares method (RLS). The recursive least squares method is used to adjust the system parameters online, and its update formula is: ; in, is the system parameter vector, is the gain matrix, is the system output, is the input eigenvector. The recursive process ensures the stability of the update by adjusting the step size to make it proportional to the square of the prediction error and inversely proportional to the trace of the covariance matrix. After obtaining the closed-loop system dynamic model, stability analysis is performed to calculate the characteristic root distribution of the system. The state space model is discretized into -Domain transfer function: ; The characteristic root is The solution of . Map the characteristic roots to -plane, analyze the positions of the dominant poles and non-dominant poles, and adjust the gain factor by the pole placement method according to the system dynamic performance requirements. The formula is: ; in, is the positive definite matrix for solving the discrete time algebraic Riccati equation. Based on the characteristic root distribution and pole configuration results, stability constraints are formed. Combined with these constraints, the control parameters of the charging and discharging process are optimized online. The optimized control parameters include the upper limit of the charging current , discharge current lower limit , Charging voltage upper threshold and discharge voltage lower threshold The optimization goal is to ensure efficiency while avoiding overcharging and over-discharging of the battery. According to the output of the optimized control parameters and the battery performance prediction model, a charging and discharging working mode switching strategy is established. For example, a fast charging mode is used under high load conditions, and a constant current discharge mode is used under low load conditions. By calculating the current state of the battery, the working mode parameters are generated. , convert it into specific control instructions: .

[0035] In a specific embodiment, the execution step is to calculate the dynamic response characteristics of the controller by using the variable step-size recursive least squares method according to the optimized gain factor, obtain the closed-loop system dynamic model, perform stability analysis on the closed-loop system dynamic model, calculate the characteristic root distribution of the system, constrain the gain factor by the pole configuration method, and obtain the stability constraint condition. The process can specifically include the following steps: According to the optimized gain factor, a recursive calculation initial model including a control input matrix and a system output matrix is ​​constructed, and data of the recursive calculation initial model is updated to obtain a weighted data sequence; Based on the weighted data sequence, the controller parameters are recursively updated, and the update step is proportional to the square of the prediction error and inversely proportional to the trace of the covariance matrix to obtain the dynamic response parameters. Perform system identification on dynamic response parameters and establish a closed-loop system dynamic model including state matrix, input matrix and output matrix; Discretize the closed-loop system dynamic model into a z-domain transfer function, calculate the roots of the characteristic polynomial, and map the calculation results to the s-plane to obtain the system characteristic root distribution; Based on the distribution of system characteristic roots, the dominant poles and non-dominant poles are divided to obtain the desired pole positions. According to the desired pole positions, the state feedback gain matrix is ​​calculated using the Ackerman formula to obtain the gain constraint equation. The gain constraint equation and the optimized gain factor are matched and calculated to generate stability constraint conditions.

[0036] Specifically, according to the optimized gain factor Construct the initial model of recursive calculation. Assume that the input of the system is , the output is , the control error is ,in is the reference signal. The goal of the closed-loop controller is to minimize the error while optimizing the dynamic performance of the system. The initial model of the recursive calculation is expressed in state space form as: ; ; Among them, the state matrix Represents the state transition relationship of the system, the input matrix Represents the influence of control input on system state, and the output matrix Mapping states to outputs, through matrix Indicates the direct impact of input on output. During initialization, assume is an approximately known initial value, and the system parameters will be continuously adjusted during the recursive process. Update the data of the recursive calculation initial model and construct a weighted data sequence. Assume that the current input-output relationship is expressed as: ; in, is the regression vector, containing the historical input and output , is the parameter vector to be estimated, is noise. By introducing the recursive least squares (RLS) method, the update rule is: ; Among them, the gain matrix The calculation formula is: ; The update rule for the covariance matrix is: ; in, is the covariance matrix, is the forgetting factor, which is used to weight the influence of historical data and new data. In this way, the recursive least squares method can update the model parameters in real time and obtain the dynamic response parameters. Based on the updated dynamic response parameters, the system is identified and the state matrix is ​​reconstructed. , input matrix , output matrix The closed-loop system dynamic model is a complete description of the dynamic behavior of the system under different input conditions. The model is discretized into -domain transfer function, the transfer function is expressed as: ; The characteristic polynomial is given by Determine, its root represents the eigenvalue of the system. -Domains are mapped to - plane, by transforming ,in is the sampling time, and the characteristic root distribution of the system is obtained. According to the characteristic root distribution, the dominant poles and non-dominant poles are divided. The dominant poles mainly determine the dynamic characteristics of the system (such as response time and oscillation frequency), and the non-dominant poles have little effect on the system. Based on the expected dynamic performance, the expected pole position is set , the state feedback gain matrix is ​​calculated using the Ackerman formula: ; in, is a positive definite symmetric matrix, solved by the algebraic Riccati equation, is the control weight matrix. The gain matrix Used to adjust the system feedback to ensure that all poles are configured to the desired position. The gain constraint equation and the optimized gain factor are matched and calculated to generate stability constraints. These conditions define the dynamic performance and stability range of the system and are used to guide the online adjustment of the controller.

[0037] The above describes the energy management system data analysis method in the embodiment of the present application. The following describes the energy management system data analysis device 10 in the embodiment of the present application. Figure 2 In the embodiment of the present application, an energy management system data analysis device 10 includes: The acquisition module 11 is used to collect the working condition characteristic data set of the robot energy storage battery through the energy management system; A calculation module 12 is used to perform state collaborative calculation on the operating condition characteristic data set to obtain battery operating state data; Establishing module 13, used to establish a mapping relationship model between load power and charge and discharge efficiency according to the operating condition characteristic data set and the battery operating status data, and determine the boundary value of the battery working range to obtain charge and discharge characteristic data; A dynamic modeling module 14, used to input the charge and discharge characteristic data into a deep gated recurrent unit network, and obtain a battery performance prediction model through dynamic modeling; The generation module 15 is used to construct a closed-loop controller based on the battery performance prediction model by using the gain factor self-tuning technology, and to perform online optimization calculation on the control parameters by using the gradient descent algorithm to generate control instructions including the charge and discharge current limit and voltage threshold.

[0038] Through the collaboration of the above components, the operating data is screened by the cross-validation-Voronoi adaptive sampling method, representative feature data is extracted, and data redundancy is effectively reduced. The collaborative calculation method of particle filter, recursive least squares method and extended Kalman filter is adopted to achieve accurate evaluation of battery charge and health, providing a reliable basis for the formulation of charge and discharge control strategy. Based on the support vector regression algorithm, a mapping relationship model between load power and charge and discharge efficiency is established. Combined with the electrochemical reaction kinetic characteristics, the battery working range is accurately defined to avoid overcharge and overdischarge. The battery charge and discharge process is dynamically modeled through the deep gated recurrent unit network to achieve accurate prediction of battery performance. The closed-loop controller is constructed by the gain factor self-tuning technology. The online optimization of control parameters is achieved through the gradient descent algorithm, which improves the accuracy and adaptability of charge and discharge control. Based on the variable step size recursive least squares method and the pole configuration method, a complete stability constraint system is established to ensure the stable operation of the energy management system.

[0039] See also Figure 3 , Figure 3 This is a schematic block diagram of the structure of an electronic device 300 provided in an embodiment of the present application. The electronic device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0040] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above energy management system data analysis methods.

[0041] The processor 301 is used to provide computing and control capabilities to support the operation of the entire electronic device 300 .

[0042] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned energy management system data analysis methods.

[0043] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a partial structure related to the present application scheme, and does not constitute a limitation on the electronic device 300 involved in the present application scheme. The specific electronic device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0044] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0045] It should be noted that technicians in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working process of the electronic device 300 described above can refer to the corresponding process of the aforementioned energy management system data analysis method, and will not be repeated here.

[0046] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the energy management system data analysis method provided in the embodiment of the present application.

[0047] The computer-readable storage medium may be an internal storage unit of the electronic device 300 in the aforementioned embodiment, such as a hard disk or memory of the electronic device 300. The computer-readable storage medium may also be an external storage device of the electronic device 300, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped with the electronic device 300.

[0048] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0049] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0050] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A data analysis method for an energy management system, characterized in that: The method comprises: Collect the operating characteristic data set of the robot's energy storage battery through the energy management system; Performing state collaborative calculation on the operating condition characteristic data set to obtain battery operating state data; According to the operating condition characteristic data set and the battery operating state data, a mapping relationship model between load power and charge and discharge efficiency is established, and a battery operating range boundary value is determined to obtain charge and discharge characteristic data; Inputting the charge and discharge characteristic data into a deep gated recurrent unit network, and obtaining a battery performance prediction model through dynamic modeling; Based on the battery performance prediction model, a closed-loop controller is constructed by adopting the gain factor self-tuning technology, and the control parameters are optimized and calculated online by the gradient descent algorithm to generate control instructions including charge and discharge current limits and voltage thresholds.

2. The energy management system data analysis method according to claim 1, characterized in that: The collecting of the working condition characteristic data set of the robot energy storage battery through the energy management system includes: The energy management system collects data on the terminal voltage, charge and discharge current, surface temperature, ambient temperature, and load power of the robot's energy storage battery to obtain an original data set, and performs normalization calculation on the original data set to obtain a normalized data set; Dividing the data space of the normalized data set into k subspaces, and calculating the data density of each subspace to obtain data space distribution characteristics; According to the data space distribution characteristics, a Voronoi diagram partitioning algorithm is used to calculate the distance matrix between nodes in each subspace, and a polygon partitioning boundary is constructed based on the distance matrix to obtain a Voronoi region partitioning result; A 5-fold cross validation was performed on the Voronoi region division result, the data was randomly divided into 5 parts, 4 parts were selected as training sets and 1 part was selected as a validation set, the data representativeness of each region was calculated, and the regional weight coefficient was obtained; Calculating the sampling probability of the data points in each Voronoi region according to the regional weight coefficient to obtain an adaptive sampling probability distribution, and resampling each Voronoi region based on the adaptive sampling probability distribution to obtain a target sampling data set; The target sampling data set is classified and labeled according to the operating condition type to obtain an operating condition feature data set.

3. The energy management system data analysis method according to claim 2, characterized in that: The performing state collaborative calculation on the operating condition characteristic data set to obtain battery operating state data includes: The terminal voltage and charge / discharge current of the operating condition characteristic data set are modeled in state space based on a second-order RC equivalent circuit, and a battery equivalent circuit model including ohmic internal resistance, polarization internal resistance and polarization capacitance is constructed; Based on the battery equivalent circuit model, the circuit parameters are iteratively calculated using a recursive least squares method of a forgetting factor, and a capacity state transfer equation based on a coulomb counting method is established to obtain a parameter update model; Initializing the capacity decay state in the parameter update model through particle filtering to obtain the initial distribution of the particle swarm, and performing a one-step prediction on the state of each particle based on the initial distribution of the particle swarm, calculating the likelihood function value based on the observation equation, and obtaining the particle weight update value through normalization processing; The effective particle number is judged for the particle weight update value, and when the effective particle number is less than half of the total number of particles, a capacity attenuation calculation result is obtained; Based on the capacity decay calculation results, a nonlinear state equation and observation equation of an extended Kalman filter are constructed, and the battery charge and health status are used as state variables, and the terminal voltage is used as an observation variable to obtain a state observation model; Performing Jacobian matrix linearization processing on the state observation model, calculating the state transfer matrix and the observation matrix respectively, and setting the process noise covariance matrix and the observation noise covariance matrix to obtain a linearized state space equation; According to the linearized state space equation, the optimal state estimation value is calculated by using the extended Kalman filter algorithm, and the calculation results of the battery charge and health status are fused to obtain the battery operation status data.

4. The energy management system data analysis method according to claim 3, characterized in that: The mapping relationship model between load power and charge-discharge efficiency is established according to the operating condition characteristic data set and the battery operating status data, and the boundary value of the battery operating range is determined to obtain the charge-discharge characteristic data, including: Performing time-series pairing on the terminal voltage, charge and discharge current, surface temperature and load power data in the operating condition characteristic data set, and establishing a multi-dimensional characteristic matrix in combination with the battery charge and health status in the battery operating status data; Performing power interval segmentation processing on the multidimensional characteristic matrix to obtain charging and discharging efficiency data between the segments; Based on the charge and discharge efficiency data between the partitions, a support vector regression algorithm is used to establish a mapping relationship between load power and charge and discharge efficiency, a Gaussian radial basis function is selected as the kernel function, and a penalty factor is set to 0.1 to obtain an efficiency prediction model; According to the output result of the efficiency prediction model, the variation trend of the charging and discharging efficiency with the load power is analyzed to obtain an efficiency mapping relationship model; Extracting the load power value of the key point in the efficiency mapping relationship model, combining the battery surface temperature data, and calculating the reaction rate constant under the corresponding working condition through the electrochemical reaction equation to obtain the kinetic characteristic parameters; Based on the kinetic characteristic parameters, the lithium ion diffusion coefficient and charge migration coefficient of the positive and negative electrode materials of the battery are calculated, and the interface reaction process is analyzed using an electrochemical impedance model to obtain electrode polarization characteristic data; According to the electrode polarization characteristic data, a mathematical model including concentration polarization and electrochemical polarization is established, and the upper and lower thresholds of the polarization voltage are determined by numerical solution to obtain the working range restriction conditions. The working range restriction conditions are combined with the calculation results of the efficiency mapping relationship model to generate charge and discharge characteristic data.

5. The energy management system data analysis method according to claim 4, characterized in that: The step of inputting the charge and discharge characteristic data into a deep gated recurrent unit network and obtaining a battery performance prediction model through dynamic modeling includes: Dividing the charge and discharge characteristic data to construct a training data set and a verification data set; Constructing an input layer of a deep gated recurrent unit network, the input dimension is the number of features of the charge and discharge characteristic data, including charge and discharge current, terminal voltage, temperature, efficiency, and working range boundary value, and the time series length is set to 100 time steps to obtain an input layer structure; Construct the hidden layer of the deep gated recurrent unit network, which includes three stacked GRU layers. The number of GRU units in the first layer is 256, the number of GRU units in the second layer is 128, and the number of GRU units in the third layer is 64. Each GRU layer contains an update gate, a reset gate, and a candidate hidden state to obtain the hidden layer structure. A dropout layer is added between each GRU layer of the hidden layer, the dropout rate is set to 0.3, and a batch normalization layer is added at the output end of each GRU layer, and the normalization parameter is set to 0.9 to obtain a regularized structure; Construct the output layer of the deep gated recurrent unit network, using a fully connected layer structure. The output dimension is the number of predicted targets, including the charge and discharge current, terminal voltage, and remaining capacity at future times. The activation function uses a linear function to obtain the output layer structure. Initialize the parameters of the deep gated recurrent unit network, use the Xavier initialization method for weights, initialize the bias term to zero, use the Adam algorithm as the optimizer, and set the learning rate to 0.001 to obtain an initialized network; The training data set is input into the initialized network, the mean square error loss function is used for forward propagation calculation, and the gradient is calculated by the back propagation algorithm to update the network parameters. The training is stopped when the loss of the verification data set does not decrease for 10 consecutive epochs to obtain a battery performance prediction model.

6. The energy management system data analysis method according to claim 5, characterized in that: Based on the battery performance prediction model, a closed-loop controller is constructed by using a gain factor self-tuning technology, and control parameters are optimized and calculated online by a gradient descent algorithm to generate control instructions including charge and discharge current limits and voltage thresholds, including: Based on the battery performance prediction model, constructing an initial controller structure of a gain factor self-tuning closed-loop controller, and constructing a gain self-tuning objective function based on model prediction for the initial controller structure; Based on the gain self-tuning objective function, the gradient vector of each gain factor is iteratively updated by using a gradient descent method to obtain an optimized gain factor; According to the optimized gain factor, the dynamic response characteristics of the controller are calculated by using the variable step-size recursive least squares method to obtain a closed-loop system dynamic model, stability analysis is performed on the closed-loop system dynamic model, characteristic root distribution of the system is calculated, and the gain factor is constrained by the pole placement method to obtain stability constraint conditions; Based on the stability constraint condition, the control parameters of the charge and discharge process are optimized online to obtain the optimized control parameters, and according to the optimized control parameters and the output performance prediction data of the battery performance prediction model, a charge and discharge working mode switching strategy is established to obtain the working mode parameters; The working mode parameters are converted into control instructions, wherein the control instructions include a charging current upper limit value, a discharging current lower limit value, a charging voltage upper limit threshold value, and a discharging voltage lower limit threshold value.

7. The energy management system data analysis method according to claim 6, characterized in that: The method of calculating the dynamic response characteristics of the controller by using the variable step-size recursive least squares method according to the optimized gain factor to obtain a closed-loop system dynamic model, performing stability analysis on the closed-loop system dynamic model, calculating the characteristic root distribution of the system, and constraining the gain factor by the pole placement method to obtain stability constraint conditions includes: According to the optimized gain factor, a recursive calculation initial model including a control input matrix and a system output matrix is ​​constructed, and data of the recursive calculation initial model is updated to obtain a weighted data sequence; Based on the weighted data sequence, recursively update the controller parameters, wherein the update step is proportional to the square of the prediction error and inversely proportional to the trace of the covariance matrix, to obtain dynamic response parameters; Performing system identification on the dynamic response parameters, and establishing a closed-loop system dynamic model including a state matrix, an input matrix, and an output matrix; Discretizing the closed-loop system dynamic model into a z-domain transfer function, calculating the roots of the characteristic polynomial, and mapping the calculation results to the s-plane to obtain the system characteristic root distribution; Based on the characteristic root distribution of the system, the dominant poles and non-dominant poles are divided to obtain the desired pole positions, and according to the desired pole positions, the state feedback gain matrix is ​​calculated using the Ackerman formula to obtain the gain constraint equation; A matching calculation is performed on the gain constraint equation and the optimized gain factor to generate a stability constraint condition.

8. An energy management system data analysis device, characterized in that: Used to execute the energy management system data analysis method according to any one of claims 1 to 7, the energy management system data analysis device comprises: A collection module is used to collect the working condition characteristic data set of the robot energy storage battery through the energy management system; A calculation module, used for performing state collaborative calculation on the operating condition characteristic data set to obtain battery operating state data; An establishment module is used to establish a mapping relationship model between load power and charge and discharge efficiency according to the operating condition characteristic data set and the battery operating state data, and determine the boundary value of the battery working range to obtain charge and discharge characteristic data; A dynamic modeling module, used to input the charge and discharge characteristic data into a deep gated recurrent unit network, and obtain a battery performance prediction model through dynamic modeling; The generation module is used to construct a closed-loop controller based on the battery performance prediction model by using the gain factor self-tuning technology, and to perform online optimization calculation on the control parameters by using the gradient descent algorithm to generate control instructions including the charge and discharge current limit and the voltage threshold.

9. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes the energy management system data analysis method according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the energy management system data analysis method according to any one of claims 1 to 7 is implemented.

Images