Electric appliance battery management system integrated with charger control

Abstract

The invention discloses an electric appliance battery management system integrated with charger control, and the system comprises a main control unit which is responsible for the global management, receiving parameters monitored by a slave control unit, estimating the SOC and SOH of a battery pack, introducing an adaptive multi-model fusion SOC estimation technology, carrying out the parallel operation of EKF, UKF and particle filtering, dynamically selecting the optimal output through the confidence coefficient weight, and carrying out the calculation of the optimal output. The main control unit interacts control parameters and state information of a charging process with the charger control module, sends a battery equalization control command based on an inductance / capacitance topological structure, exchanges data with other modules through a communication network, and controls the charger control module to control the charger. The main control unit adopts dynamic mixed critical scheduling to optimize charger control tasks, the tasks are divided into a time-critical task and a calculation-intensive task, it is ensured that the key tasks are processed in time, and meanwhile the execution efficiency of the calculation-intensive task is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of electric energy storage, and particularly to a battery management system for electrical appliances integrated with a charger control. Background Art

[0002] With the transformation of the global energy structure and the improvement of environmental awareness, new energy technologies have developed rapidly. The popularization of new energy devices such as electric vehicles has put forward higher requirements for charging devices and management systems. With the rapid development of new energy technologies such as electric vehicles, integrating charger control and battery management for electrical appliances has become a key link to support the wide application of these technologies. They help improve the performance and reliability of new energy devices and promote the continuous progress of new energy technologies.

[0003] Different types of batteries (such as ternary lithium batteries and lithium iron phosphate batteries) have significant differences in chemical composition, charge and discharge characteristics, etc. Fixed parameter models are usually designed for a specific type of battery and it is difficult to meet the requirements of multiple types of batteries simultaneously. Therefore, in the scenario of mixing multiple types of batteries, the accuracy and precision of SOC / SOH estimation will be severely affected. Therefore, a battery management system for electrical appliances integrated with a charger control is proposed. By using an adaptive multi-model fusion SOC estimation technology, which runs EKF, UKF, and particle filter in parallel and dynamically selects the optimal output, the estimation accuracy of SOC / SOH is significantly improved. This helps to more accurately understand the state of the battery and provides a reliable basis for charger collaborative control and battery equalization management. Summary of the Invention

[0004] The purpose of the present invention is to solve the deficiencies existing in the prior art, and a battery management system for electrical appliances integrated with a charger control is proposed.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions: A battery management system for electrical appliances integrated with a charger control, comprising: Master control unit: Responsible for global management, receiving parameters monitored by slave control units, estimating the State of Charge (SOC) and State of Health (SOH) of different types of battery packs, introducing an adaptive multi-model fusion SOC estimation technique, running the Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF) and particle filter in parallel, calculating confidence weights according to the matching degree between the prediction results of each model and the observed data, dynamically selecting the optimal output of different types of battery packs through confidence weights, interacting with the charger control module for control parameters and status information of the charging process, based on the inductor / capacitor topology, issuing battery balancing control commands, exchanging data with other modules through a communication network. The master control unit controls the charger control module to perform charger control. The master control unit uses Dynamic Hybrid Criticality Scheduling (DHCS) to optimize the charger control tasks, divides the tasks into "time-critical" and "computation-intensive", uses the DHCS algorithm to adjust the execution order of tasks. For "time-critical" tasks, ensure that they have the highest priority and are executed immediately. For "computation-intensive" tasks, reasonably arrange their execution time according to the current load and computing resources of the system to optimize the overall performance; Slave control unit: Responsible for monitoring key parameters such as the voltage and temperature of single cells and performing balancing control.

[0006] Charger control module: Integrates the control logic of AC / DC or DC / DC converters and is responsible for adjusting the charging voltage and current according to the real-time state of the battery; Communication network: Implements communication between modules using CAN bus or RS485; Safety protection unit: Constructs a baseline model for battery charging and discharging behaviors, calculates the Mahalanobis distance between the current operation and the baseline in real time, detects hidden faults, and uses an independent hardware circuit (such as an AFE chip) to implement overvoltage, overcurrent, and short-circuit protection.

[0007] The above technical solution further includes: Furthermore, the master control unit receives the parameters monitored by the slave control units, estimates the SOC (State of Charge) and SOH (State of Health) of the battery pack, introduces an adaptive multi-model fusion SOC estimation technique, runs the EKF, UKF and particle filter in parallel, and dynamically selects the optimal output through confidence weights, including the following steps: Data preprocessing: The slave control unit collects key parameters such as the voltage, current and temperature of the battery pack and transmits them to the master control unit through the communication network. The master control unit preprocesses the received data, and the preprocessed data is output to the EKF, UKF and particle filter; Model initialization: Initialize the three models of EKF, UKF and particle filter, set the initial state vector and covariance matrix, and set appropriate parameters for each model according to the characteristics of the battery pack (such as capacity, internal resistance, open-circuit voltage, etc.); Parallel running of filtering algorithms: EKF: Based on the linearization assumption, the SOC is iteratively calculated through prediction and update steps. In the prediction step, the battery model is used to predict the state at the next moment. The battery model describes the dynamic behavior of the battery, including equivalent circuit models, electrochemical models, and neural network-based prediction models. In the update step, the observed data is used to correct the prediction result, and the observed data represents the real-time monitored data; UKF: The probability distribution of the nonlinear function is approximated by the unscented transform (UT) to calculate the SOC, and the unscented transform (UT) is used to handle the nonlinear terms; Particle filter: Based on the Monte Carlo method, the probability distribution is represented by a set of weighted particles. Each particle represents a possible battery state, and the particle set is updated through importance sampling and resampling steps; Confidence weight calculation: According to the matching degree between the prediction results of each model and the observed data, the confidence weights are calculated. The residuals (the differences between the predicted values and the observed values) and the corresponding covariance matrix are calculated, and then the probability density function is calculated based on the Gaussian distribution. The confidence weights are used to fuse the outputs of the three models to obtain the final SOC estimation value; Online parameter identification and model update: According to the real-time data, the battery parameters (such as internal resistance, capacity attenuation coefficient, etc.) are identified online, and the model parameters are updated.

[0008] Furthermore, based on the inductor / capacitor topology, the main control unit issues battery equalization control commands, including the following steps: Monitoring the battery state: The main control unit monitors the key parameters such as the voltage and temperature of each single battery in the battery pack in real time through the slave control unit. The key parameters are used to evaluate the imbalance degree of the battery pack and serve as the basis for equalization control; Determining the equalization demand: According to the monitored key parameters, the main control unit calculates the single battery that needs to be equalized and its target voltage or SOC value. The equalization target is usually set as the average value of the voltages or SOCs of all single batteries in the battery pack or a preset threshold; Selecting the equalization strategy: According to the equalization demand and the inductor / capacitor topology, the main control unit selects the equalization strategy. The inductor-based equalization strategy involves the on / off control of switching devices to achieve energy transfer; while the capacitor-based equalization strategy uses the charge and discharge characteristics of the capacitor to balance the voltage; Calculating the control parameters: According to the selected equalization strategy and the battery state, the main control unit calculates the parameters required for equalization control, such as the on / off time of the switching device, the charge and discharge time of the capacitor, etc.; Issuing an equalization control command: The master control unit sends an equalization control command to the slave control unit through the communication network, including the selection of the equalization strategy and the values of the control parameters. The slave control unit controls the corresponding switching devices or capacitors to perform equalization operations according to the received command; Monitoring the equalization process: During the equalization process, the master control unit continues to monitor the battery status and adjusts the equalization strategy and control parameters according to the actual situation; Ending the equalization operation: When the degree of imbalance of the battery pack is reduced below the preset threshold, the master control unit ends the equalization operation. At this time, it returns to the normal monitoring state and waits for the next equalization requirement to appear.

[0009] Furthermore, the master control unit optimizes the charger control task by using dynamic hybrid criticality scheduling (DHCS), including the following steps; Step 1: Initialize the system, set the task priorities and scheduling policies; Step 2: Monitor the battery status in real time, including parameters such as voltage, current, and temperature; Step 3: Calculate the current charging strategy according to the battery status, including the charging voltage and current. The formula is expressed as , where is the charging voltage, SOC is the remaining charge, T is the temperature, and aging_model is the battery aging model; Step 4: Divide the charging strategy tasks into time-critical and compute-intensive tasks; Time-critical tasks: Immediately perform battery protection actions; Compute-intensive tasks: Detect hidden faults and optimize the charging strategy; Step 5: Use the DHCS algorithm to dynamically adjust the execution order of tasks and the allocated computing resources; Step 6: Execute the tasks and monitor the system status in real time to ensure the safety and efficiency of the charging process.

[0010] Furthermore, the specific steps for the master control unit to formulate an adaptive charging strategy based on the adaptive control algorithm of the battery aging model: Establishing the battery aging model: Collect the charging data of the battery at different aging stages (such as new battery, mild aging, severe aging), including charging voltage, charging current, charging time, etc., and establish a battery aging model that can reflect the relationship between the battery aging degree and the charging parameters; Real-time aging degree evaluation: During the charging process, collect the battery parameters in real time, input the collected battery parameters into the battery aging model, and calculate the current aging degree of the battery; Adaptive charging strategy formulation: Dynamically adjust the charging strategy according to the aging degree of the battery. For example, for a battery with a higher aging degree, the charging current and voltage can be reduced to reduce the internal thermal stress and mechanical stress of the battery, thereby extending the battery life. Set different charging stages, including constant current charging stage, constant voltage charging stage, and floating charging stage, and adjust the charging parameters of each stage according to the battery aging degree; Charging process control: Convert the adaptive charging strategy into specific charging control instructions and send them to the charger control module. The charger control module adjusts the charging voltage and current according to the instructions to achieve the control of the charging process.

[0011] Furthermore, the main control unit uses LSTM to establish a battery aging model, including the following steps: Data preparation: Collect the charge and discharge data of the battery, clean the collected data to remove noise and outliers, then normalize the data, screen out the features related to battery aging to construct a data set, and divide the data set into a training set, a validation set, and a test set with a ratio of 70%, 15%, 15%; Model construction: Use LSTM to construct a model: Forget gate: ; Among them, is the output of the forget gate (a vector between 0 and 1, 1 means completely retain, 0 means completely forget), is the weight matrix of the forget gate, is the bias term, is the sigmoid function, is the previous hidden state, indicating that and are concatenated into a vector; Input gate: The input gate is responsible for updating the cell state. The input gate consists of two parts: one is the sigmoid layer, which determines which information will be updated; the other is the tanh layer, which creates a new candidate value vector. The new candidate value vector is added to the cell state. Finally, the two pieces of information are multiplied to update the cell state; Sigmoid layer: ; Tanh layer: ; Cell state update: ; Among them, is the output of the input gate sigmoid layer (controlling the proportion of new information written), is the candidate cell state (new information generated by the tanh layer), , is the input gate weight matrix, , is the input gate bias term, is the hyperbolic tangent function, is the cell state at the current time step (long-term memory carrier), is the cell state at the previous time step, is the Hadamard product (element-wise multiplication); Output gate: The output gate determines which part of the information based on the cell state is used for output. According to the current input , the state of the hidden layer at the previous time step and the latest cell state , through the combined action of the sigmoid function and the tanh function, determines the output at the current time step ; Sigmoid layer: ; Output hidden state: ; Among them, is the output of the output gate sigmoid layer, is the hidden state at the current time step (short-term memory output), is the output gate weight matrix, is the output gate bias term; Training process: Forward propagation: For the input at each time step, calculate the hidden state and output according to the RNN; Calculate the loss: Use the loss function to measure the difference between the model prediction and the actual label; Backward propagation: Calculate the gradient of the loss with respect to the model parameters through the Backpropagation Through Time algorithm; Parameter update: Use Adam to update the model parameters according to the gradient; Testing and inference: Evaluate the model performance on the test set. During inference, input the battery parameters collected in real time into the trained model. The model outputs the probability distribution of each category, and take the category with the highest probability as the classification result.

[0012] Furthermore, the specific steps for the safety protection unit to construct a baseline model for battery charging and discharging behavior and calculate the Mahalanobis distance between the current operation and the baseline in real time to detect hidden faults are as follows: Data collection: Collect key parameter data such as voltage, current, and temperature of the battery during normal charging and discharging. The data should cover various working conditions, such as different charging rates, discharge depths, and environmental temperatures; Baseline model construction: pre-process the collected data, remove noise and outliers, and build a baseline model of battery charging and discharging behavior. The baseline model reflects the normal behavior characteristics of the battery under different working conditions; Real-time calculation and monitoring: During the operation of the system, the key parameter data of the battery is collected in real time, and the Mahalanobis distance between the current operation and the baseline model is calculated to measure the difference between the current operation and the baseline model. The calculation formula of the Mahalanobis distance is: , where x is the feature vector of the current operation, is the mean eigenvector of the baseline model, is the transpose of the feature deviation vector, is the inverse of the covariance matrix; Hidden fault detection: Set a threshold. When the Mahalanobis distance exceeds the threshold, the battery is considered to behave abnormally and there is a hidden fault.

[0013] Furthermore, the safety protection unit uses the DBSCAN algorithm to construct a battery charging and discharging behavior baseline model. Given a battery data set D={x1, x2, ..., xn}, where each data point xi represents a d-dimensional vector, for data point xi, the data point xi The neighborhood contains the distance to xi that is less than or equal to All data points of ,in, Indicates xi Neighborhood, core object: If the data point xi If the neighborhood contains at least MinPts data points (including xi itself), then xi is a core object and can be directly density-reachable: for data points xi and xj, if xj is in xi In the neighborhood, and xi is a core object, then xj is said to be directly density-reachable to xi, density-reachable: For data points xi and xj, if there exists a data point sequence p1, p2, ..., pl, where p1=xi, pl=xj, and pi+1 is directly density-reachable to pi, then xj is said to be density-reachable to xi, density-connected: For data points xi and xj, if there exists a data point xk such that both xi and xj are density-reachable to xk, then xi and xj are said to be density-connected; Initialization: mark all data points as unclassified; For each data point xi, determine whether it is a core object. If so, build a new cluster and group xi and its All data points in the neighborhood are added to the cluster; For each data point of the core object, recursively add the density-reachable data points to the same cluster; Continue to perform the above operations on the unclassified data points until all data points are classified; Repeat the above process until all core objects are processed.

[0014] The present invention has the following beneficial effects: 1. In the present invention, an adaptive multi-model fusion SOC estimation technology is adopted. By running EKF, UKF, and particle filtering in parallel and dynamically selecting the optimal output, the estimation accuracy of SOC / SOH is significantly improved. This helps to more accurately understand the state of the battery and provides a reliable basis for the coordinated control of the charger and the battery equalization management.

[0015] 2. In the present invention, by introducing RTOS task scheduling optimization and dynamic hybrid criticality scheduling (DHCS), the system can manage tasks more efficiently, ensure that critical tasks are processed in a timely manner, and at the same time optimize the execution efficiency of computationally intensive tasks. This significantly improves the intelligence and overall operating efficiency of the system. Brief Description of the Drawings

[0016] Figure 1 It is a system block diagram of a battery management system for electrical appliances integrating charger control proposed by the present invention. Detailed Embodiments

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] Please refer to Figure 1 As shown, the present invention is a battery management system for electrical appliances integrating charger control, including: Master control unit: Responsible for global management, receiving parameters monitored by slave control units, estimating the State of Charge (SOC) and State of Health (SOH) of different types of battery packs, introducing an adaptive multi-model fusion SOC estimation technique, running the Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF) and particle filter in parallel, calculating confidence weights based on the matching degree between the prediction results of each model and the observed data, dynamically selecting the optimal output of different types of battery packs through the confidence weights, interacting with the charger control module for control parameters and status information of the charging process, based on the inductance / capacitance topology, issuing battery equalization control commands, exchanging data with other modules through the communication network. The master control unit controls the charger control module to perform charger control. The master control unit uses Dynamic Hybrid Criticality Scheduling (DHCS) to optimize the charger control tasks, divides the tasks into "time-critical" and "computation-intensive", uses the DHCS algorithm to adjust the execution order of the tasks. For "time-critical" tasks, ensure that they have the highest priority and are executed immediately. For "computation-intensive" tasks, reasonably arrange their execution time according to the current load and computing resources of the system to optimize the overall performance; Slave control unit: Responsible for monitoring key parameters such as the voltage and temperature of single cells and performing equalization control.

[0019] Charger control module: Integrates the control logic of AC / DC or DC / DC converters and is responsible for adjusting the charging voltage and current according to the real-time state of the battery; Communication network: Implements communication between modules using CAN bus or RS485; Safety protection unit: Builds a baseline model of battery charge and discharge behavior, calculates the Mahalanobis distance between the current operation and the baseline in real time, detects hidden faults, and uses an independent hardware circuit (such as an AFE chip) to achieve overvoltage, overcurrent and short-circuit protection.

[0020] In one embodiment, the master control unit receives the parameters monitored by the slave control unit, estimates the State of Charge (SOC) and State of Health (SOH) of the battery pack, introduces an adaptive multi-model fusion SOC estimation technique, runs the EKF, UKF and particle filter in parallel, and dynamically selects the optimal output through the confidence weight, including the following steps: Data preprocessing: The slave control unit collects key parameters such as the voltage, current and temperature of the battery pack and transmits them to the master control unit through the communication network. The master control unit preprocesses the received data, and the preprocessed data is output to the EKF, UKF and particle filter; Model initialization: Initialize the three models of EKF, UKF and particle filter, set the initial state vector and covariance matrix, and set appropriate parameters for each model according to the characteristics of the battery pack (such as capacity, internal resistance, open circuit voltage, etc.); Run the filtering algorithms in parallel: EKF: Based on the linearization assumption, the SOC is iteratively calculated through prediction and update steps. In the prediction step, a battery model is used to predict the state at the next moment. The battery model describes the dynamic behavior of the battery, including equivalent circuit models, electrochemical models, and prediction models based on neural networks. In the update step, the prediction result is corrected using the observed data, which represents the data monitored in real time; UKF: Calculate the SOC by approximating the probability distribution of the nonlinear function through unscented transformation (UT), and use unscented transformation (UT) to handle the nonlinear terms; Particle filter: Based on the Monte Carlo method, represent the probability distribution through a set of weighted particles. Each particle represents a possible battery state, and the particle set is updated through importance sampling and resampling steps; Confidence weight calculation: Calculate the confidence weights according to the matching degree between the prediction results of each model and the observed data. Calculate the residuals (the difference between the predicted value and the observed value) and the corresponding covariance matrix, and then calculate the probability density function based on the Gaussian distribution. The confidence weights are used to fuse the outputs of the three models to obtain the final SOC estimation value; Online parameter identification and model update: According to the real-time data, online identify the battery parameters (such as internal resistance, capacity attenuation coefficient, etc.) and update the model parameters.

[0021] In one embodiment, the main control unit issues a battery balancing control command based on the inductor / capacitor topology, including the following steps: Monitor the battery state: The main control unit monitors the key parameters such as the voltage and temperature of each single battery in the battery pack in real time through the slave control unit. The key parameters are used to evaluate the imbalance degree of the battery pack and serve as the basis for balancing control; Determine the balancing requirement: According to the monitored key parameters, the main control unit calculates the single battery that needs to be balanced and its target voltage or SOC value. The balancing target is usually set as the average value of the voltages or SOCs of all single batteries in the battery pack or a preset threshold; Select the balancing strategy: According to the balancing requirement and the inductor / capacitor topology, the main control unit selects the balancing strategy. The inductor-based balancing strategy involves the on / off control of switching devices to achieve energy transfer; while the capacitor-based balancing strategy uses the charge and discharge characteristics of the capacitor to balance the voltage; Calculate the control parameters: According to the selected balancing strategy and the battery state, the main control unit calculates the parameters required for balancing control, such as the on / off time of the switching device, the charge and discharge time of the capacitor, etc.; Issue the balancing control command: The main control unit sends the balancing control command to the slave control unit through the communication network, including the selection of the balancing strategy and the values of the control parameters. The slave control unit controls the corresponding switching devices or capacitors to perform the balancing operation according to the received command; Monitoring the equalization process: During the equalization process, the master control unit continues to monitor the battery status and adjusts the equalization strategy and control parameters according to the actual situation; Ending the equalization operation: When the degree of imbalance of the battery pack drops below the preset threshold, the master control unit ends the equalization operation. At this time, it returns to the normal monitoring state and waits for the next equalization requirement to occur.

[0022] In one embodiment, the master control unit optimizes the charger control task using dynamic hybrid criticality scheduling (DHCS), including the following steps; Step 1: Initialize the system, set task priorities and scheduling strategies.

[0023] Step 2: Monitor the battery status in real time, including parameters such as voltage, current, and temperature.

[0024] Step 3: Calculate the current charging strategy based on the battery status, including charging voltage and current, expressed by the formula , where is the charging voltage, SOC is the remaining battery capacity, T is the temperature, and aging_model is the battery aging model; Step 4: Divide the charging strategy tasks into time-critical and compute-intensive tasks; Time-critical tasks: Immediately perform battery protection actions; Compute-intensive tasks: Detect latent faults and optimize the charging strategy; Step 5: Use the DHCS algorithm to dynamically adjust the execution order of tasks and the allocated computing resources; Step 6: Execute the tasks and monitor the system status in real time to ensure the safety and efficiency of the charging process.

[0025] In one embodiment, the specific steps for the master control unit to formulate an adaptive charging strategy based on the adaptive control algorithm of the battery aging model: Establishing the battery aging model: Collect charging data of the battery at different aging stages (such as new battery, slightly aged, severely aged), including charging voltage, charging current, charging time, etc., and establish a battery aging model that can reflect the relationship between the degree of battery aging and charging parameters; Real-time aging degree evaluation: During the charging process, collect battery parameters in real time, input the collected battery parameters into the battery aging model, and calculate the current aging degree of the battery; Adaptive charging strategy formulation: Dynamically adjust the charging strategy according to the aging degree of the battery. For example, for a battery with a higher aging degree, the charging current and voltage can be reduced to reduce the internal thermal stress and mechanical stress of the battery, thereby extending the battery life. Set different charging stages, including constant current charging stage, constant voltage charging stage, and floating charging stage, and adjust the charging parameters of each stage according to the battery aging degree; Charging process control: Convert the adaptive charging strategy into specific charging control instructions and send them to the charger control module. The charger control module adjusts the charging voltage and current according to the instructions to achieve control of the charging process.

[0026] In one embodiment, the main control unit uses LSTM to establish a battery aging model, including the following steps: Data preparation: Collect the charge and discharge data of the battery, clean the collected data to remove noise and outliers, then normalize the data, screen out the features related to battery aging to construct a data set, and divide the data set into a training set, a validation set, and a test set, with a ratio of 70%, 15%, 15%; Model construction: Use LSTM to construct a model: Forget gate: ; Among them, is the output of the forget gate (a vector between 0 and 1, 1 means completely retained, 0 means completely forgotten), is the weight matrix of the forget gate, is the bias term, is the sigmoid function, is the previous hidden state, indicating that and are concatenated into a vector; Input gate: The input gate is responsible for updating the cell state. The input gate consists of two parts: one is the sigmoid layer, which determines which information will be updated; the other is the tanh layer, which creates a new candidate value vector. The new candidate value vector is added to the cell state. Finally, the two pieces of information are multiplied to update the cell state; Sigmoid layer: ; Tanh layer: ; Cell state update: ; Among them, is the output of the input gate sigmoid layer (controlling the proportion of new information written), is the candidate cell state (new information generated by the tanh layer), , is the input gate weight matrix, , is the input gate bias term, is the hyperbolic tangent function, is the cell state at the current time step (long-term memory carrier), is the cell state at the previous time step, is the Hadamard product (element-wise multiplication); Output gate: The output gate determines which part of the information based on the cell state is used for output, according to the current input , the state of the hidden layer at the previous time step and the latest cell state , through the combined action of the sigmoid function and the tanh function, determines the output at the current time step ; Sigmoid layer: ; Output hidden state: ; Among them, is the output of the output gate sigmoid layer, is the hidden state at the current time step (short-term memory output), is the output gate weight matrix, is the output gate bias term; Training process: Forward propagation: For the input at each time step, calculate the hidden state and output according to the RNN; Calculate the loss: Use the loss function to measure the difference between the model prediction and the actual label; Backward propagation: Calculate the gradient of the loss with respect to the model parameters through the Backpropagation Through Time algorithm; Parameter update: Use Adam to update the model parameters according to the gradient; Testing and inference: Evaluate the model performance on the test set. During inference, input the battery parameters collected in real time into the trained model. The model outputs the probability distribution of each class, and take the class with the highest probability as the classification result.

[0027] In one embodiment, the safety protection unit constructs a battery charge and discharge behavior baseline model, and calculates the Mahalanobis distance between the current operation and the baseline in real time. The specific steps for detecting latent faults are as follows: Data collection: Collect key parameter data such as voltage, current, and temperature of the battery during normal charge and discharge processes. The data should cover various working conditions, such as different charging rates, discharge depths, and ambient temperatures; Baseline model construction: pre-process the collected data, remove noise and outliers, and build a baseline model of battery charging and discharging behavior. The baseline model reflects the normal behavior characteristics of the battery under different working conditions; Real-time calculation and monitoring: During the operation of the system, the key parameter data of the battery is collected in real time, and the Mahalanobis distance between the current operation and the baseline model is calculated to measure the difference between the current operation and the baseline model. The calculation formula of the Mahalanobis distance is: , where x is the feature vector of the current operation, is the mean eigenvector of the baseline model, is the transpose of the feature deviation vector, is the inverse of the covariance matrix; Hidden fault detection: Set a threshold. When the Mahalanobis distance exceeds the threshold, the battery is considered to behave abnormally and there is a hidden fault.

[0028] In one embodiment, the safety protection unit uses the DBSCAN algorithm to construct a battery charging and discharging behavior baseline model. Given a battery data set D={x1, x2, ..., xn}, where each data point xi represents a d-dimensional vector, for data point xi, the data point xi The neighborhood contains the distance to xi that is less than or equal to All data points of ,in, Indicates xi Neighborhood, core object: If the data point xi If the neighborhood contains at least MinPts data points (including xi itself), then xi is a core object and can be directly density-reachable: for data points xi and xj, if xj is in xi In the neighborhood, and xi is a core object, then xj is said to be directly density-reachable to xi, density-reachable: For data points xi and xj, if there exists a data point sequence p1, p2, ..., pl, where p1=xi, pl=xj, and pi+1 is directly density-reachable to pi, then xj is said to be density-reachable to xi, density-connected: For data points xi and xj, if there exists a data point xk such that both xi and xj are density-reachable to xk, then xi and xj are said to be density-connected; Initialization: mark all data points as unclassified; For each data point xi, determine whether it is a core object. If so, build a new cluster and group xi and its All data points in the neighborhood are added to the cluster; For each data point of the core object, recursively add the data points that are density-reachable to the same cluster; Continue to perform the above operations on the unclassified data points until all data points are classified; Repeat the above process until all core objects are processed.

[0029] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An electrical appliance battery management system integrated with a charger control, characterized in that, Including: Main control unit: Responsible for global management, receiving parameters monitored by slave control units, estimating the SOC and SOH of different types of battery packs, running EKF, UKF, and particle filtering in parallel, calculating confidence weights according to the matching degree between the prediction results of each model and the observed data, dynamically selecting the optimal output of different types of battery packs through confidence weights, interacting with the charger control module for control parameters and status information of the charging process, based on the inductor / capacitor topology, issuing battery balancing control commands, exchanging data with other modules through a communication network. The main control unit controls the charger control module to perform charger control. The main control unit optimizes the charger control task using dynamic hybrid criticality scheduling, divides the tasks into "time-critical" and "computation-intensive", and adjusts the execution order of tasks using the DHCS algorithm; Slave control unit: Responsible for monitoring key parameters and performing balancing control; Charger control module: Responsible for adjusting the charging voltage and current according to the real-time state of the battery; Communication network: Implements communication between modules using CAN bus or RS485; Safety protection unit: Constructs a baseline model of battery charging and discharging behavior, calculates the Mahalanobis distance between the current operation and the baseline in real time, detects hidden faults, and uses an independent hardware circuit to implement overvoltage, overcurrent, and short-circuit protection.

2. The battery management system for an electrical appliance integrated with a charger control according to claim 1, wherein The main control unit receives the parameters monitored by the slave control unit, estimates the SOC and SOH of the battery pack, introduces an adaptive multi-model fusion SOC estimation technology, runs EKF, UKF, and particle filtering in parallel, and dynamically selects the optimal output through confidence weights, including the following steps: Data preprocessing: The slave control unit collects the key parameters of the battery pack and transmits them to the main control unit through the communication network. The main control unit preprocesses the received data, and the preprocessed data is output to EKF, UKF, and particle filtering; Model initialization: Initialize the three models of EKF, UKF, and particle filtering, and set the initial state vector and covariance matrix; Run the filtering algorithm in parallel: EKF: Based on the linearization assumption, iteratively calculate the SOC through prediction and update steps. The prediction step uses the battery model to predict the state at the next moment. The battery model describes the dynamic behavior of the battery, including the equivalent circuit model, electrochemical model, and prediction model based on neural network. The update step uses the observed data to correct the prediction result. The observed data represents the real-time monitored data; UKF: Approximate the probability distribution of the nonlinear function through unscented transformation, and use unscented transformation to process the nonlinear terms; Particle filtering: Based on the Monte Carlo method, represent the probability distribution through a set of weighted particles. Each particle represents a battery state, and the particle set is updated through importance sampling and resampling steps; Confidence weight calculation: Calculate the confidence weights according to the matching degree between the prediction results of each model and the observed data, calculate the residual and the corresponding covariance matrix, and then calculate the probability density function based on the Gaussian distribution. The confidence weights are used to fuse the outputs of the three models to obtain the final SOC estimation value; Online parameter identification and model update: According to real-time data, online identify battery parameters and update model parameters.

3. An electrical appliance battery management system integrated with a charger control, characterized in that, The master control unit issues a battery balancing control command based on the inductor / capacitor topology, including the following steps: Monitor battery status: The master control unit monitors the key parameters of each single battery in the battery pack in real time through the slave control unit. The key parameters are used to evaluate the imbalance degree of the battery pack; Determine the balancing requirement: According to the monitored key parameters, the master control unit calculates the battery cells that need to be balanced and their target voltage or SOC value; Select a balancing strategy: According to the balancing requirement and the inductor / capacitor topology, the master control unit selects a balancing strategy; Calculate control parameters: According to the selected balancing strategy and battery status, the master control unit calculates the parameters required for balancing control; Issue a balancing control command: The master control unit sends a balancing control command to the slave control unit through the communication network, including the selection of the balancing strategy and the value of the control parameters. The slave control unit controls the corresponding switching device or capacitor to perform the balancing operation according to the received command; Monitor the balancing process: During the balancing process, the master control unit continues to monitor the battery status and adjusts the balancing strategy and control parameters according to the actual situation; End the balancing operation: When the imbalance degree of the battery pack drops below the preset threshold, the master control unit ends the balancing operation. At this time, it returns to the normal monitoring state and waits for the next balancing requirement to appear.

4. An electrical appliance battery management system integrated with a charger control according to claim 1, characterized in that, The master control unit adopts a dynamic hybrid criticality scheduling to optimize the charger control task, including the following steps; Step 1: Initialize the system, set task priorities and scheduling strategies; Step 2: Monitor the battery status in real time; Step 3: Calculate the current charging strategy according to the battery state, including the charging voltage and current, which is expressed by the formula as , where is the charging voltage, SOC is the remaining battery capacity, T is the temperature, and aging_model is the battery aging model; Step 4: Divide the charging strategy tasks into time-critical and computationally intensive tasks; Time-critical task: Immediately perform battery protection actions; Computationally intensive task: Detect latent faults and optimize the charging strategy; Step 5: Adopt the DHCS algorithm to dynamically adjust the execution order of tasks and the allocated computing resources; Step 6: Execute tasks and monitor the system status in real time.

5. An electrical appliance battery management system integrated with a charger control, characterized in that, The specific steps for the master control unit to formulate an adaptive charging strategy based on the adaptive control algorithm of the battery aging model: Establish a battery aging model: Collect charging data of the battery at different aging stages and establish a battery aging model; Evaluate the real-time aging degree: During the charging process, collect battery parameters in real time, input the collected battery parameters into the battery aging model, and calculate the current aging degree of the battery; Formulate an adaptive charging strategy: According to the aging degree of the battery, dynamically adjust the charging strategy, set different charging stages, including constant current charging stage, constant voltage charging stage and floating charge charging stage, and adjust the charging parameters of each stage according to the battery aging degree; Control the charging process: Convert the adaptive charging strategy into specific charging control instructions and send them to the charger control module. The charger control module adjusts the charging voltage and current according to the instructions to achieve the control of the charging process.

6. An electrical appliance battery management system integrated with a charger control, characterized in that, The master control unit uses LSTM to establish a battery aging model, including the following steps: Data preparation: Collect the charge and discharge data of the battery, clean the collected data to remove noise and outliers, then normalize the data, screen out the features related to battery aging to construct a data set, and divide the data set into a training set, a validation set, and a test set with a ratio of 70%, 15%, and 15%; Model construction: Use LSTM to construct a model: Forgotten Gate: ; Among them, is the output of the forget gate, is the weight matrix of the forget gate, is the bias term, is the sigmoid function, is the hidden state at the previous moment, indicating that and are concatenated into a vector; Input gate: The input gate is responsible for updating the cell state. The input gate consists of two parts: one is the sigmoid layer, which determines which information will be updated; the other is the tanh layer, which creates a new candidate value vector. The new candidate value vector is added to the cell state. Finally, the two pieces of information are multiplied to update the cell state; Sigmoid layer: ; tanh layer: ; Cell status update: ; Among them, is the output of the input gate sigmoid layer, is the candidate cell state, , is the input gate weight matrix, , is the input gate bias term, is the hyperbolic tangent function, is the current cell state, is the previous cell state, is the Hadamard product; Output Gate: The output gate determines which part of the information based on the cell state is used for output, according to the current input , the state of the hidden layer at the previous moment and the latest cell state , through the combined action of the sigmoid function and the tanh function, determines the output at the current moment ; Sigmoid layer: ; Output hidden state: ; Among them, is the output of the output gate sigmoid layer, is the hidden state at the current moment, is the weight matrix of the output gate, is the bias term of the output gate; Training process: Forward propagation: For the input at each time step, calculate the hidden state and output according to the RNN; Calculate the loss: Use the loss function to measure the difference between the model prediction and the actual label; Backward propagation: Calculate the gradient of the loss with respect to the model parameters through the Backpropagation Through Time algorithm; Parameter update: Use Adam to update the model parameters according to the gradient; Testing and inference: Evaluate the model performance on the test set. During inference, input the battery parameters collected in real time into the trained model. The model outputs the probability distribution of each category, and select the category with the highest probability as the classification result.

7. An electrical appliance battery management system integrated with a charger control, characterized in that, The safety protection unit constructs a baseline model for battery charge and discharge behavior, and calculates the Mahalanobis distance between the current operation and the baseline in real time. The specific steps for detecting latent faults are as follows: Data collection: Collect the key parameter data of the battery during normal charge and discharge; Baseline model construction: Preprocess the collected data to remove noise and outliers, and construct a baseline model for battery charge and discharge behavior. The baseline model reflects the normal behavior characteristics of the battery under different working conditions; Real-time calculation and monitoring: During the operation of the system, key parameter data of the battery are collected in real time, and the Mahalanobis distance between the current operation and the baseline model is calculated to measure the degree of difference between the current operation and the baseline model. The calculation formula of the Mahalanobis distance is , where x is the feature vector of the current operation, is the mean of the feature vectors of the baseline model, is the transpose of the feature deviation vector, is the inverse of the covariance matrix; Latent fault detection: Set a threshold. When the Mahalanobis distance exceeds this threshold, it is considered that the battery behavior is abnormal and there is a latent fault.

8. An electrical appliance battery management system integrated with a charger control, characterized in that, The safety protection unit uses the DBSCAN algorithm to build a battery charging and discharging behavior baseline model. Given a battery data set D={x1, x2, ..., xn}, where each data point xi represents a d-dimensional vector, for data point xi, the data point xi The neighborhood contains the distance to xi that is less than or equal to All data points of ,in, Indicates xi Neighborhood, core object: If the data point xi If the neighborhood contains at least MinPts data points (including xi itself), then xi is a core object and can be directly density-reachable: for data points xi and xj, if xj is in xi In the neighborhood, and xi is a core object, then xj is said to be directly density-reachable to xi, density-reachable: For data points xi and xj, if there exists a data point sequence p1, p2, ..., pl, where p1=xi, pl=xj, and pi+1 is directly density-reachable to pi, then xj is said to be density-reachable to xi, density-connected: For data points xi and xj, if there exists a data point xk such that both xi and xj are density-reachable to xk, then xi and xj are said to be density-connected; Initialization: Mark all data points as unclassified; For each data point xi, determine whether it is a core object. If so, construct a new cluster and add xi and all data points within its neighborhood to this cluster; For each data point of the core object, recursively add the data points that are density-reachable to the same cluster; Continue to perform the above operations on the unclassified data points until all data points are classified; Repeat the above process until all core objects are processed.

Images