Rail transit high-voltage lithium battery cluster voltage equalization control method and system

Abstract

The application relates to the technical field of battery management, and discloses a high-voltage lithium battery cluster voltage equalization control method and system for rail transit. The method comprises the following steps: collecting battery voltages and tunnel environment parameters, and obtaining environment characteristic data through Kalman filtering; combining a UPS load law, and predicting a voltage deviation trend through LSTM; calculating a dynamic equalization threshold value according to the predicted value; starting a bidirectional energy transmission equalization circuit when the voltage difference value exceeds the threshold value; monitoring the state change of the battery in the equalization process, and outputting a maintenance instruction through a fault identification algorithm. The application solves the problem that the battery equalization control in the UPS system of rail transit cannot adapt to complex environments and dynamic load changes, and improves the precision and intelligent level of the battery cluster voltage equalization control.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of battery management, in particular to a high-voltage lithium battery cluster voltage equalization control method and system for rail transit. BACKGROUND

[0002] The uninterruptible power supply (UPS) of a rail transit system usually adopts lead-acid batteries or traditional lithium batteries as backup power sources to provide emergency power supply guarantee for key equipment such as signal systems, lighting systems and ventilation systems. The existing battery management technology mainly adopts a passive equalization method to consume the excess energy of high-voltage single batteries through resistance discharge, or a simple active equalization circuit to transfer energy between adjacent batteries. The equalization control strategy usually judges voltage differences and triggers equalization based on fixed thresholds.

[0003] However, the existing technology has significant deficiencies: the passive equalization method has low energy utilization rate and generates a large amount of heat, and the simple active equalization circuit has limited efficiency and slow response speed; the fixed threshold control strategy cannot adapt to the complex and variable load conditions and harsh tunnel environment of rail transit, especially under harsh conditions such as high humidity, salt spray and low temperature, the battery performance deteriorates, and the traditional equalization method is difficult to maintain the voltage consistency of the battery cluster; the existing system lacks the ability to predict load changes and cannot respond to the impact of signal system transient high-power demand on battery voltage distribution in advance. SUMMARY

[0004] The application provides a high-voltage lithium battery cluster voltage equalization control method and system for rail transit, which solves the problem that battery equalization control in the UPS system of rail transit cannot adapt to complex environments and dynamic load changes, and improves the precision and intelligent level of battery cluster voltage equalization control.

[0005] In a first aspect, the application provides a high-voltage lithium battery cluster voltage equalization control method for rail transit, which comprises:

[0006] Step S1: Collect the real-time voltage values of each single battery in the high-voltage lithium battery cluster and the temperature and humidity parameters of the rail transit tunnel environment, filter out the noise interference in the collected data through a Kalman filtering algorithm, and obtain rail transit environment characteristic data;

[0007] Step S2: Combine the rail transit environment characteristic data with the UPS load current variation law, predict the voltage deviation trend of the battery cluster under the signal system transient high-power demand through an LSTM neural network, and obtain a rail transit load adaptability voltage prediction value;

[0008] Step S3: According to the rail transit load adaptability voltage prediction value and the influence of tunnel environment temperature and humidity, the rail transit special equalization control threshold is calculated to replace the fixed threshold control strategy.

[0009] Step S4: When the single battery voltage difference exceeds the rail transit special equalization control threshold, the bidirectional energy transmission equalization circuit is started to transfer energy between high-voltage single cells and low-voltage single cells, and the voltage equalized battery cluster state is obtained.

[0010] Step S5: The voltage change trend and internal resistance growth of each single battery in the equalization process are monitored, the battery health state is judged through the fault feature recognition algorithm, and the rail transit UPS system maintenance instruction is output.

[0011] In a second aspect, the application provides a rail transit high-voltage lithium battery cluster voltage equalization control system, which comprises:

[0012] The acquisition module is used for acquiring real-time voltage values of each single battery in the high-voltage lithium battery cluster and temperature and humidity parameters of the rail transit tunnel environment, filtering out noise interference in the collected data through the Kalman filtering algorithm, and obtaining rail transit environment characteristic data.

[0013] The prediction module is used for combining the rail transit environment characteristic data with the UPS load current change rule, predicting the voltage deviation trend of the battery cluster under the instantaneous high-power demand of the signal system through the LSTM neural network, and obtaining the rail transit load adaptability voltage prediction value.

[0014] The calculation module is used for calculating the rail transit special equalization control threshold according to the rail transit load adaptability voltage prediction value and the influence of tunnel environment temperature and humidity, and replacing the fixed threshold control strategy.

[0015] The starting module is used for starting the bidirectional energy transmission equalization circuit when the single battery voltage difference exceeds the rail transit special equalization control threshold, transferring energy between high-voltage single cells and low-voltage single cells, and obtaining the voltage equalized battery cluster state.

[0016] The output module is used for monitoring the voltage change trend and internal resistance growth of each single battery in the equalization process, judging the battery health state through the fault feature recognition algorithm, and outputting the rail transit UPS system maintenance instruction.

[0017] In a third aspect, a high-voltage lithium battery cluster voltage equalization control device for rail transit is provided, comprising a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to enable the high-voltage lithium battery cluster voltage equalization control device for rail transit to perform the high-voltage lithium battery cluster voltage equalization control method for rail transit described above.

[0018] In a fourth aspect, a computer-readable storage medium is provided, the computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the high-voltage lithium battery cluster voltage equalization control method for rail transit described above.

[0019] In the technical solution provided in the present application, the Kalman filtering algorithm is used to perform adaptive filtering processing on the high-voltage lithium battery cluster voltage data and the tunnel environment parameters of rail transit, effectively eliminating the complex interference signals generated by the electromagnetic environment of rail transit, ensuring the accuracy and reliability of subsequent data processing. At the same time, the LSTM neural network prediction model combines the rail transit environment feature data and the UPS load current variation law to accurately predict the voltage deviation trend of the battery cluster under the instantaneous high-power demand of the signal system, realizing the technical leap from passive response to active prevention. The acquisition of the adaptive voltage prediction value of rail transit load enables the equalization control strategy to be forward-looking, and the dynamic calculation of the rail transit special equalization control threshold replaces the traditional fixed threshold control strategy, which is adjusted in real time according to the influence of tunnel environment temperature and humidity and load expectations, significantly improving the environmental adaptability and load adaptability of equalization control. The bidirectional energy transfer equalization circuit performs directional energy transfer between high-voltage monomers and low-voltage monomers, which greatly reduces energy loss and shortens equalization time compared with the traditional passive equalization method. The fault feature recognition algorithm realizes intelligent evaluation of the battery health state by monitoring the voltage variation trend and internal resistance growth of each monomer battery during the equalization process. The output of the rail transit UPS system maintenance instruction establishes a preventive maintenance system, effectively prolongs the service life of the battery and reduces the system failure risk.

[0020] In the specific application field of high-voltage lithium battery cluster voltage equalization control for rail transit, the adaptive characteristics of Kalman filtering algorithm are optimized for the complex electromagnetic environment of rail transit, the recursive filtering mechanism can dynamically adjust the filtering parameters to adapt to the noise characteristics under different operating conditions, the long and short term memory characteristics of LSTM neural network fully play the learning ability of periodic load mode of rail transit, by capturing the load change law of signal system, lighting system and ventilation system at different time periods, a deep correlation model between load mode and battery voltage deviation is established, the soft switching characteristics of bidirectional CLLC resonant converter show the advantages of high efficiency and low electromagnetic interference in rail transit application, the resonant frequency regulation mechanism can accurately control the energy transmission power to match the equalization demand of different voltage differences, the fault feature recognition algorithm combines support vector machine classifier and decision tree algorithm, fully utilizes the richness and regularity of rail transit operation data, through multi-dimensional feature fusion, realizes accurate identification of battery health state and intelligent decision of maintenance strategy, the algorithm optimization and model customization of the whole scheme in the special environment of rail transit significantly improve the intelligent level and reliability guarantee capability of the battery management system. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings based on these drawings without creative labor.

[0022] Figure 1 An embodiment of the high-voltage lithium battery cluster voltage equalization control method for rail transit in the present application is shown in the figure.

[0023] Figure 2 An embodiment of the high-voltage lithium battery cluster voltage equalization control system for rail transit in the present application is shown in the figure.

[0024] Figure 3 The structure schematic diagram of the high-voltage lithium battery cluster voltage equalization control device for rail transit in the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0025] The embodiment of the present application provides a rail transit high-voltage lithium battery cluster voltage equalization control method and system. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0026] For ease of understanding, the specific process of the embodiment of the present application is described below. Please refer to Figure 1 One embodiment of the rail transit high-voltage lithium battery cluster voltage equalization control method in the present application includes:

[0027] Step S1: Collecting real-time voltage values of each single battery in the high-voltage lithium battery cluster and temperature and humidity parameters of the rail transit tunnel environment, filtering out noise interference in the collected data through Kalman filtering algorithm to obtain rail transit environment characteristic data;

[0028] Step S2: Combining the rail transit environment characteristic data with the UPS load current variation law, predicting the voltage deviation trend of the battery cluster under the instantaneous high-power demand of the signal system through the LSTM neural network to obtain the rail transit load adaptability voltage prediction value;

[0029] Step S3: According to the rail transit load adaptability voltage prediction value and the influence of the tunnel environment temperature and humidity, calculating the rail transit special equalization control threshold to replace the fixed threshold control strategy;

[0030] Step S4: When the voltage difference of the single battery exceeds the rail transit special equalization control threshold, starting the bidirectional energy transmission equalization circuit to transfer energy between the high-voltage single battery and the low-voltage single battery to obtain the voltage equalized battery cluster state;

[0031] Step S5: Monitoring the voltage variation trend and internal resistance growth of each single battery in the equalization process, judging the battery health state through the fault feature recognition algorithm, and outputting the rail transit UPS system maintenance instruction.

[0032] It can be understood that the execution subject of the present application can be a rail transit high-voltage lithium battery cluster voltage equalization control system, and can also be a terminal or a server, which is not limited here. The embodiment of the present application takes the server as the execution subject for example.

[0033] Specifically, the distributed voltage sensor array monitors the voltage of each monomer battery in the battery cluster through a high-precision sampling circuit. The voltage sensor uses a 12-bit ADC converter, and the sampling frequency is set to 100 Hz. At the same time, the tunnel environment data is collected through a PT1000 temperature sensor and a SHT30 humidity sensor. The raw data matrix contains three types of parameters: battery voltage, environmental temperature, and environmental humidity. The parameters are aligned according to the time stamp to form a multi-dimensional data sequence. Kalman filtering algorithm plays a key role in this link. By establishing a state prediction model and an observation update model, the raw data collected is recursively filtered. The Kalman filter first predicts the state at the current time based on the state estimate at the previous time, and then corrects the prediction result using the observation value at the current time, effectively filtering out the high-frequency noise generated by the rail transit electromagnetic environment and the sensor drift.

[0034] An LSTM neural network prediction model is established, which is specifically designed for the load characteristics of rail transit UPS systems. The UPS load current time series database stores the current variation of the signal system, lighting system, and ventilation system at different operating periods. These historical data are associated and matched with the filtered environmental feature data according to the time dimension to form a training sample set. The LSTM neural network can capture long-term dependencies in time series through its unique gating mechanism. The input layer receives a multi-dimensional feature vector containing voltage, environmental, and load information. The bidirectional LSTM hidden layer contains 128 neurons, which extract time series features through forward and backward propagation mechanisms. During network training, the training samples are windowed according to 20 time steps, each window containing continuous voltage and load variation information. After 1000 iterations of training, the model can predict the voltage deviation trend of each monomer battery in the future time period based on the current environmental feature data and operating schedule information.

[0035] A dynamic threshold calculation is implemented, which combines the predicted voltage deviation data with environmental factors. First, the future voltage deviation amplitude data of each monomer battery is extracted from the rail transit load adaptive voltage prediction value, and the overall voltage dispersion index of the battery cluster is calculated by standard deviation. The calculation of the environmental correction factor involves the comprehensive consideration of the temperature correction coefficient and the humidity correction coefficient. The temperature correction coefficient is calculated based on the deviation of the tunnel temperature from the standard temperature, and the humidity correction coefficient reflects the influence of humidity on the internal resistance and electrochemical reaction rate of the battery. The dynamic weight distribution algorithm adjusts the contribution proportion of each influencing factor according to the severity of the current environmental conditions. In high-humidity and high-temperature environments, the weight of the environmental correction factor is increased, and in the case of large load fluctuations, the weight of the voltage dispersion index is increased. The final rail transit dedicated equalization control threshold is obtained through weighted operation, which can adapt to the environmental changes and load demands of different operating periods.

[0036] The bidirectional energy transfer equalization circuit adopts a CLLC resonant converter topology, which has the advantages of high efficiency and low electromagnetic interference in rail transit applications. When the voltage deviation of a certain single battery is monitored to exceed the dynamic threshold, the control algorithm first identifies the high-voltage single battery and the low-voltage single battery that need to transfer energy, and then determines the optimal energy transfer path through a pairing algorithm. The switching control logic of the CLLC resonant converter adjusts the resonant frequency and power transfer size according to the voltage difference, and the parameter design of the resonant inductance and resonant capacitance ensures the maintenance of soft switching characteristics in a wide load range. During the energy transfer process, the controller monitors the transmission current and the voltage change rate of each single battery in real time, and dynamically adjusts the switching frequency and duty cycle parameters through closed-loop feedback control. When the voltage difference of all single batteries converges to within the threshold range, the equalization circuit stops working.

[0037] A machine learning-based fault prediction system is constructed, which can simultaneously evaluate the battery health state during the equalization process. The extraction of voltage trend characteristic parameters is achieved by linear regression analysis of time series voltage data, calculating the voltage change slope and fluctuation variance of each single battery. The slope reflects the battery capacity attenuation trend, and the variance reflects the battery consistency change. The internal resistance measurement is achieved by integrating an alternating current impedance test function in the equalization circuit, injecting a small amplitude sinusoidal current signal into the battery, measuring the corresponding voltage response, and calculating the battery internal resistance value through impedance calculation. The support vector machine classification algorithm takes the voltage trend parameters and internal resistance growth rate as feature vectors, and classifies the health state of each single battery through the trained classification model, outputting three levels of normal, warning, and fault. The decision tree algorithm generates specific maintenance strategy schemes according to the fault risk level and the requirements of rail transit operation safety, including battery replacement timing, maintenance cycle adjustment, monitoring frequency setting, etc.

[0038] In a specific embodiment, step S1 comprises:

[0039] The instantaneous voltage data of each single battery in the high-voltage lithium battery cluster is collected by a distributed voltage sensor array, and a battery cluster voltage sampling sequence is obtained;

[0040] The real-time values of tunnel temperature and humidity are collected by tunnel temperature and humidity sensors, and a tunnel environment parameter sequence is obtained;

[0041] The battery cluster voltage sampling sequence and the tunnel environment parameter sequence are aligned and merged according to the timestamp to form an original data matrix containing three-dimensional information of voltage-temperature-humidity;

[0042] The Kalman filtering algorithm is used to filter the jump data and drift noise in the original data matrix, so as to remove the interference signals generated by the electromagnetic environment of the rail transit and obtain the rail transit environment feature data.

[0043] Specifically, the distributed voltage sensor array acquisition process adopts a modular design, each voltage sensor module integrates a high-precision ADC converter and a signal conditioning circuit, and is directly connected to the positive and negative terminals of the single battery. The sensor array is distributed according to the physical layout of the battery cluster, each sensor module has an independent address code, and communicates with the data acquisition controller through the CAN bus. During voltage sampling, the ADC converter converts the analog voltage signal into a digital signal, with a sampling accuracy of millivolts and a sampling frequency of 100 times per second, ensuring that the instantaneous changes of the battery voltage can be captured. The data acquisition controller polls all single batteries according to the preset sampling time sequence, encapsulates the collected voltage values and corresponding timestamp information into data packets, and forms a battery cluster voltage sampling sequence. The sequence contains the voltage change information of each single battery at consecutive time points, and the data format is a three-tuple structure of timestamp-battery number-voltage value.

[0044] The temperature sensor and the humidity sensor in the rail transit tunnel adopt industrial-grade environmental monitoring equipment. The temperature sensor is based on the platinum resistance temperature measurement principle, with a measurement accuracy of 0.1 degrees Celsius, and the humidity sensor uses a capacitive humidity sensor, with a measurement accuracy of 1%RH. The arrangement position of the sensor considers the air flow characteristics and temperature and humidity distribution law in the tunnel, and sets the main monitoring points near the battery cluster and the auxiliary monitoring points at the entrance and exit of the tunnel. The temperature sensor reflects the environmental temperature through resistance change, and the control circuit converts the resistance change into a standard voltage signal, and obtains the digitized temperature value through ADC conversion. The humidity sensor reflects the environmental humidity through capacitance change, and the signal processing circuit converts the capacitance change into a frequency signal, and then obtains the humidity value through a frequency counter. The environmental parameter acquisition controller synchronously acquires temperature and humidity data, records data at the same time interval as voltage sampling, and forms a tunnel environmental parameter sequence with a data format of timestamp-temperature value-humidity value.

[0045] The timestamp alignment and merging process is a key step in data fusion, which needs to handle the time synchronization problem between different sensors. The data processing unit receives data streams from the voltage sensor array and environmental sensors, first sorts all data according to timestamps, and then associates voltage data and environmental data within the same time window through a time window matching algorithm. The width of the time window is set to half of the sampling period, ensuring the time consistency of the data. For data with incomplete timestamp matching, a linear interpolation method is used for time alignment processing, and the interpolation calculation is based on the data change rate of adjacent time points. The structure of the merged raw data matrix is a two-dimensional array of N rows and M columns, where N represents the length of the time series, and M represents the data dimension, including parameters such as single battery voltage, environmental temperature, and environmental humidity. Each row in the matrix represents the complete state information at a certain time, and each column represents the change trajectory of a certain parameter over time.

[0046] The adaptive filtering process of the Kalman filter algorithm is specially optimized for the complex interference characteristics of the rail transit electromagnetic environment. The algorithm establishes a state space model, taking battery voltage and environmental parameters as state variables and sensor measurements as observation variables. The state transition model considers the natural change law of battery voltage and the slow change characteristics of environmental parameters, and determines the parameters of the state transition matrix through historical data statistical analysis. The observation model describes the relationship between sensor measurements and true states, considering the influence of sensor noise and electromagnetic interference. The prediction step of the filter calculates the state prediction value at the current time according to the state estimation at the previous time and the state transition model, and updates the state covariance matrix. The update step uses the observation value at the current time to correct the prediction result, and adjusts the weight distribution of the prediction value and the observation value through the Kalman gain matrix. The adaptive adjustment mechanism of the noise covariance matrix dynamically adjusts the filter parameters according to the statistical characteristics of the observation residual, increases the process noise covariance when a larger observation residual is detected, and reduces the observation noise covariance when the observation residual is smaller.

[0047] The identification and processing of jump data uses an abnormal detection method based on statistical analysis, which establishes a normal range by calculating the mean and standard deviation of the data sequence, and marks data points that exceed three times the standard deviation of the normal range as jump data. The processing of drift noise combines trend analysis and high-pass filtering, first extracts the long-term trend component of the data, and then filters out the low-frequency drift component through a high-pass filter. The interference signals generated by the rail transit electromagnetic environment mainly exhibit high-frequency pulses and periodic interference, and the frequency characteristics of the interference signals are identified through frequency domain analysis, and corresponding notch filters are designed for interference suppression. The rail transit environmental feature data after filtering processing retains the original time series structure, but the noise component is effectively suppressed, and the signal-to-noise ratio of the data is significantly improved.

[0048] In a specific embodiment, step S2 comprises:

[0049] extracting the load current change data of the signal system, lighting system and ventilation system from the historical operation records of the rail transit UPS system, and constructing a UPS load current time series database;

[0050] associating and matching the rail transit environment feature data with the UPS load current time series database according to the time dimension to form a training sample set containing a three-layer coupling relationship of voltage-environment-load;

[0051] segmenting the training sample set according to a time window and inputting the segmented training sample set into an LSTM neural network for deep learning training, and establishing a prediction model of the rail transit load mode and the battery voltage deviation correlation;

[0052] Based on the current rail transit operation timetable and the instantaneous power demand of the signal system, the voltage deviation change trend of each single battery cluster in the future time period is calculated through the prediction model to obtain the rail transit load adaptive voltage prediction value.

[0053] Specifically, the process of extracting load current data from the historical operation records of the rail transit UPS system needs to access the historical operation logs stored in the database through the data interface of the UPS monitoring system. The construction process of the UPS load current time series database first classifies the load data according to the subsystem type. The signal system load current includes the current consumption records of the train automatic control system, the communication system and the monitoring system. The lighting system load current includes the current change data of the tunnel lighting, the emergency lighting and the identification lighting. The ventilation system load current includes the current fluctuation information of the tunnel ventilator, the exhaust fan and the air conditioning equipment. The data extraction algorithm filters out the current data in the specified time period from the UPS operation logs through the SQL query statement, and establishes the index structure according to the time sequence and the load type. The time sequence characteristics of the load current are reflected in the current change law in different operation periods. The signal system load current is high during the morning peak period, the ventilation system load current dominates during the night maintenance period, and the overall load current is relatively low during holidays. The abnormal current records caused by equipment failure or measurement error are identified and removed through the abnormal value detection algorithm in the data preprocessing link, and then the missing time point data is filled through the data interpolation method, finally forming a complete and continuous UPS load current time series database.

[0054] The association matching of rail transit environmental feature data and UPS load current time series database is processed based on a time dimension alignment algorithm, which considers the time synchronization deviation and sampling frequency difference between different data sources. The association matching process first re-samples the environmental feature data and load current data according to a unified time reference, and the sampling interval is set to the minute level to balance data accuracy and computational efficiency. The time window matching strategy combines and associates the battery voltage data, environmental temperature and humidity data, and each subsystem load current data in the same time period through sliding window technology. The establishment of the three-layer coupling relationship reflects the real-time state of the battery cluster in the voltage layer, the influence of external conditions on the battery performance in the environment layer, and the power demand of the UPS system on the battery in the load layer. The data fusion algorithm extracts the correlation features between the layers of data, calculates the correlation coefficient between the battery voltage and the load current, analyzes the influence of environmental temperature and humidity on the battery internal resistance, and identifies the disturbance law of the load change on the battery voltage distribution. The data structure of the training sample set adopts a multi-dimensional tensor form, each sample contains a multi-dimensional feature vector in a specific time window, and the dimensions of the feature vector include the product of the number of batteries and the number of sampling points, the product of the number of environmental parameters and the number of sampling points, and the product of the number of load types and the number of sampling points.

[0055] The sliding window technique is used in the process of dividing the training sample set according to the time window, and the time window length is set to 20 time steps, corresponding to 20 minutes of historical data. The window sliding step is set to 1 time step, ensuring that the training samples have sufficient overlap. The deep learning training process of the LSTM neural network includes data preprocessing, network construction, parameter initialization, forward propagation, backward propagation, weight update, etc. In the data preprocessing stage, the normalization algorithm is used to map input data of different dimensions to the same numerical range, eliminating the influence of data scale difference on network training. In the network construction process, a bidirectional LSTM architecture is designed, the forward LSTM layer processes the forward information of the time series, and the backward LSTM layer processes the reverse information of the time series. The hidden state vectors of the two directions are concatenated and input into the fully connected layer for feature fusion. The gating mechanism of the LSTM unit includes three control units: forget gate, input gate, and output gate. The forget gate determines which historical information to discard, the input gate determines which new information to store, and the output gate determines which information to output to the next moment. In the training process, the network parameters are optimized by the stochastic gradient descent algorithm, the learning rate is dynamically adjusted using the exponential decay strategy, and the batch size is balanced according to the memory capacity and training efficiency.

[0056] The prediction model associated with the deviation of the rail transit load mode and the battery voltage is established by a supervised learning method. The input variables of the model include time series of historical voltage data, environmental data, and load data, and the output variable is the voltage deviation prediction value of each single battery in the future time period. The model training uses time series cross-validation method to divide the historical data into training set and validation set in time sequence to avoid data leakage problem. The loss function uses mean square error form to measure the difference between the predicted voltage deviation and the actual voltage deviation. The model convergence is judged based on the change trend of the validation loss function. When the validation loss of continuous multiple training rounds does not decrease, the training is stopped. The overfitting prevention strategy is realized by dropout technology and early stopping mechanism. Dropout randomly turns off part of the neurons in the training process to reduce the complexity of the model, and early stopping mechanism terminates the training in advance when the validation loss starts to rise.

[0057] The prediction process based on the current rail transit operation timetable and the prediction of the instantaneous power demand of the signal system is combined with real-time operation scheduling information for dynamic prediction. The operation timetable data includes train departure interval, station stop time, peak period distribution and other information, which directly affects the power demand change mode of the signal system. The instantaneous power demand of the signal system is obtained in real time by power monitoring equipment, including power consumption data of train positioning system, communication system and control system. The prediction model receives the current operation state information as the initial input, calculates the load power demand at each future time point by recursive prediction method, and then predicts the voltage variation trend of each single battery according to the correlation between power demand and battery voltage deviation. The rail transit load adaptive voltage prediction value reflects the expected variation amplitude of the battery voltage under different load conditions. The battery discharge current is large in high load period, resulting in increased voltage drop amplitude, and the battery is in floating state in low load period, and the voltage remains relatively stable.

[0058] In a specific embodiment, the execution step of dividing the training sample set according to the time window and inputting the LSTM neural network for deep learning training can specifically include the following steps:

[0059] The time series division processing is performed on the training sample set, and the sample data is cut into a window unit of 20 time steps to form an input sequence matrix of the LSTM network;

[0060] The voltage data, environmental data and load data in the input sequence matrix are normalized respectively to eliminate the influence of numerical differences between different dimension parameters, and a standardized training data set is obtained;

[0061] A neural network architecture including an input layer, a bidirectional LSTM hidden layer and a fully connected output layer is constructed, the number of hidden layer neurons is set to 128, the activation function is tanh function, and the loss function is mean square error function;

[0062] The standardized training data set is divided into training set and validation set in the ratio of 8 to 2, the network weight parameter is iteratively updated by back propagation algorithm, and the training is stopped when the validation set loss function converges and the training round reaches 1000 rounds;

[0063] The trained LSTM network is tested for rail transit special load mode, the environmental characteristic data under the typical signal system starting and braking working condition are input, and the voltage deviation prediction accuracy is verified;

[0064] The verified network model is packaged as a prediction model for the correlation between rail transit load mode and battery voltage deviation, which is used for real-time prediction of battery cluster voltage trend.

[0065] Specifically, the time series division processing of the training sample set adopts sliding window cutting technology, and the continuous time series data is divided according to the fixed length window. The setting of 20 time step window unit is based on the time characteristics of rail transit operation, each time step corresponds to 1 minute sampling interval, and 20 minute window length covers the complete operation cycle of train from entering station to leaving station. The window cutting process is realized by array index operation, starting from the starting position of the original time series, each time a continuous data segment with length of 20 is extracted as a sample, then the window is slid forward by one time step, the next sample is extracted, and the whole time series is traversed. The structure of LSTM network input sequence matrix is three-dimensional tensor, the first dimension represents the number of samples, the second dimension represents the time step, and the third dimension represents the number of features. The number of features includes the voltage values of each single battery in the battery cluster, the temperature and humidity parameters of the tunnel environment, and the load current values of each subsystem of the UPS system, forming a multi-dimensional feature vector. The arrangement of matrix elements follows the time sequence, each row represents the complete state information at a certain time, and each column represents the change trajectory of a certain feature in the time sequence.

[0066] The normalization of input sequence matrix is to standardize the numerical range of different dimension parameters. The numerical range of voltage data is usually between 3 volts and 4 volts, the numerical range of ambient temperature data is between minus 40 degrees Celsius and plus 60 degrees Celsius, and the numerical range of load current data is between 0 amperes and 50 amperes. The dimensions and numerical scales of these parameters differ significantly. The normalization algorithm uses the min-max normalization method to calculate the minimum and maximum values of each feature dimension, and then maps the original data to the interval of 0 to 1. The normalization transformation formula is: normalized value equals original value minus the difference between the minimum value divided by the difference between the maximum value and the minimum value, where the original value represents the data point to be processed, the minimum value and the maximum value represent the extreme values of the feature dimension. The standardized training data set retains the original time series structure, but eliminates the dimensional influence between different features, so that the neural network can equally learn the change rule of each feature. The storage mechanism of the normalization parameter ensures that the same standardization processing can be performed on new input data during the model prediction stage, and the prediction results are denormalized to the original dimension space.

[0067] The construction process of the neural network architecture includes network level design and parameter configuration. The input layer receives the standardized multi-dimensional time series data, the input dimension is equal to the number of features, and the input sequence length is equal to the time window length. The bidirectional LSTM hidden layer includes two sub-layers, the forward LSTM and the backward LSTM. The forward LSTM processes the input sequence in time order, and the backward LSTM processes the input sequence in reverse time order. The hidden state vectors of the two directions are spliced at each time step. The internal structure of the LSTM unit includes two state vectors, cell state and hidden state. The cell state is responsible for the storage and transmission of long-term information, and the hidden state is responsible for the output and transmission of short-term information. The gating mechanism controls the information flow through the combination of sigmoid activation function and tanh activation function. The output of the sigmoid function ranges from 0 to 1, which is used to control the passing ratio of information, and the output of the tanh function ranges from negative 1 to positive 1, which is used for bidirectional adjustment of information. The number of hidden layer neurons is set to 128, which determines the representation ability and computational complexity of the network. Too few will result in insufficient representation ability, and too many will increase the risk of overfitting. The fully connected output layer maps the hidden state of the bidirectional LSTM to the prediction target space, and the output dimension is equal to the number of batteries to be predicted. Each output node corresponds to a single battery voltage deviation prediction value.

[0068] The division of the training set and the validation set adopts a time series cross-validation strategy to avoid data leakage problems while maintaining the continuity of the time series. The 8-to-2 ratio means that the first 80% of the time series data is used for model training, and the last 20% of the time series data is used for model validation. The iterative update process of the back propagation algorithm optimizes the network weight parameters through gradient descent. First, forward propagation is performed to calculate the predicted output and loss function value, then backward propagation is performed to calculate the gradient value of each layer weight, and finally the weight parameters are updated according to the learning rate. The mean square error loss function calculates the average of the square difference between the predicted value and the true value, and this function gives higher penalty weight to samples with larger prediction error. The training round number reaches 1000 rounds based on experience value and convergence analysis, and too few training rounds will cause model underfitting, and too many training rounds will cause model overfitting. The judgment standard for the convergence of the validation set loss function is that the change amplitude of the validation loss of consecutive multiple training rounds is less than the pre-set threshold value, and the validation loss no longer shows a downward trend.

[0069] The rail transit special load mode test verifies the prediction performance of the model by constructing typical working condition scenarios. The signal system startup working condition simulates the power mutation of the signal device when the train enters the station, corresponding to the pulse increase of the UPS load current and the instantaneous increase of the battery discharge current. The braking working condition simulates the influence of regenerative braking energy feedback on the power system when the train brakes, corresponding to the fluctuation of the load power and the switching of the battery charge and discharge state. The test data set contains historical operation data under these typical working conditions, and the prediction accuracy is evaluated by comparing the model prediction output with the actual measured value. The quantitative indicators of prediction accuracy include mean absolute error, root mean square error, correlation coefficient, etc. These indicators reflect the prediction accuracy and stability of the model from different angles.

[0070] The encapsulation process of the network model packages the trained LSTM network parameters and normalization parameters into an independent prediction module, which receives real-time rail transit operation data as input and outputs the voltage deviation prediction value of each single battery in the future time period. The model deployment adopts a modular design, supporting online updating and offline maintenance. When sufficient new operation data is accumulated, the model parameters are updated through incremental learning method to maintain the synchronization of the prediction model with the actual operation working condition.

[0071] In a specific embodiment, step S3 comprises:

[0072] The future voltage deviation amplitude data of each single battery is extracted from the rail transit load adaptability voltage prediction value, and the voltage dispersion index of the battery cluster as a whole is calculated;

[0073] According to the tunnel temperature and humidity parameters in the rail transit environment feature data, the temperature correction coefficient and the humidity correction coefficient are calculated respectively, and the two correction coefficients are multiplied to obtain the environmental comprehensive correction factor;

[0074] The preset rail transit basic equalization threshold is weighted with the environmental comprehensive correction factor and the voltage dispersion index, and the contribution proportions of various influencing factors are adjusted through a dynamic weight distribution algorithm.

[0075] Based on the weighted operation result and the safety margin requirement of the rail transit UPS system, a rail transit dedicated equalization control threshold is generated for the current environmental conditions and load expectations, covering the threshold variation range in different operation periods.

[0076] Specifically, the process of extracting future voltage deviation amplitude data from the rail transit load adaptive voltage prediction value involves secondary processing of the prediction result through statistical analysis. The voltage deviation amplitude data reflects the deviation of each single battery from the average voltage of the battery cluster. The extraction process first calculates the mean of the voltage of each single battery in the prediction time window, and then calculates the absolute value of the difference between each single battery voltage and the mean, obtaining the voltage deviation amplitude sequence of each single battery. The battery cluster overall voltage dispersion index is obtained through the variance calculation method, which quantifies the unevenness of the voltage distribution within the battery cluster. The calculation formula is the sum of the squares of all single battery voltage deviation amplitudes divided by the number of batteries, and then the square root is taken to obtain the dispersion index in the form of standard deviation. The larger the dispersion index value, the more obvious the voltage difference within the battery cluster, and the more stringent the equalization control strategy. The smaller the value, the more uniform the voltage distribution of the battery cluster, and the more relaxed the equalization control requirement. The calculation process of the index takes into account the influence of the number of batteries, and through normalization processing ensures the comparability of the dispersion index between battery clusters of different sizes.

[0077] The correction coefficient calculation of tunnel temperature and humidity parameters is based on the quantitative analysis of the influence of rail transit environment on battery performance. The calculation of temperature correction coefficient considers the influence law of temperature change on battery internal resistance and electrochemical reaction rate. The increase of battery internal resistance in low temperature environment leads to the increase of voltage drop, while the increase of battery activity in high temperature environment leads to the increase of safety risk. The temperature correction coefficient calculation uses a piecewise linear function. When the tunnel temperature is lower than the reference temperature, the correction coefficient is greater than 1, increasing the equalization threshold. When the tunnel temperature is higher than the reference temperature, the correction coefficient is less than 1, reducing the equalization threshold. The calculation of humidity correction coefficient considers the influence of humidity change on battery sealing performance and electrical insulation performance. High humidity environment increases the risk of battery leakage and corrosion, and requires more conservative equalization control strategy. The humidity correction coefficient calculation is based on an exponential function model. As the humidity increases, the correction coefficient gradually increases, reflecting the trend that the equalization threshold needs to be increased accordingly. The environmental comprehensive correction factor is obtained by multiplying the two correction coefficients. This factor comprehensively reflects the joint influence of temperature and humidity environment on the equalization control strategy, and the multiplication operation reflects the coupling effect of the two environmental factors.

[0078] The preset rail transit basic equalization threshold is taken as a control reference under standard environmental conditions, and the threshold is determined based on battery manufacturer specifications and rail transit safety standards. The weighting operation process linearly combines the basic equalization threshold, the environmental comprehensive correction factor, and the voltage dispersion index according to predetermined weights, and the weight distribution reflects the contribution degree of each factor to the final threshold. The dynamic weight distribution algorithm dynamically adjusts the weight coefficients of each factor according to the current operating conditions, increases the weight of the environmental correction factor when the environment is severe, increases the weight of the voltage dispersion index when the load fluctuates sharply, and gives priority to the basic equalization threshold when the operation is normal. The weight distribution algorithm adopts a fuzzy logic control method to establish a fuzzy mapping relationship between the severity of the environment and the degree of load fluctuation and the weight coefficients, and realizes the smooth adjustment of the weights through a fuzzy inference mechanism. The mathematical expression of the weighted operation is that the final threshold is equal to the basic threshold multiplied by weight 1, plus the environmental correction factor multiplied by weight 2, plus the dispersion index multiplied by weight 3, and the sum of the three weight coefficients is equal to 1 to ensure the rationality of the threshold.

[0079] The safety margin requirement of the rail transit UPS system is reflected in the safety boundary constraint of the equalization control threshold, which ensures that the battery cluster can still maintain safe operation under the most severe operating conditions. The safety margin calculation considers the superimposed effects of various adverse factors such as battery aging, environmental deterioration, and load mutation, and determines the safety coefficient through reliability analysis methods. The threshold generation process applies a safety coefficient correction based on the weighted operation result, automatically limits the calculated threshold to the safety range when it exceeds the upper safety limit, and automatically raises it to the minimum safety level when it is lower than the lower safety limit. The generation result of the rail transit special equalization control threshold is customized for the current environmental conditions and load expectations, unlike the single value of the traditional fixed threshold control strategy. This threshold has time-varying and adaptive characteristics.

[0080] The coverage of the threshold variation range in different operating periods is achieved through period division and threshold mapping. The operating periods are divided into typical periods such as morning peak, evening peak, flat peak, and night maintenance according to passenger flow rules and load characteristics. Each period corresponds to specific environmental conditions and load expectations, and the threshold variation range is determined by statistical analysis of historical data to determine the threshold distribution interval for each period. The threshold mapping algorithm establishes a correspondence between the period identifier and the threshold range, and the real-time operation scheduling automatically selects the corresponding threshold range according to the current period, ensuring the matching of the equalization control strategy with the actual operation.

[0081] In a specific embodiment, step S4 comprises:

[0082] The real-time voltages of each single battery in the high-voltage lithium battery cluster are calculated by difference processing, and each single voltage is subtracted from the average voltage of the battery cluster to obtain the voltage deviation value of each single battery.

[0083] The voltage deviation value is compared with the rail transit dedicated equalization control threshold to identify high-voltage single cells and low-voltage single cells whose voltage deviation exceeds the threshold, and form an equalization object pairing list.

[0084] According to the voltage difference degree in the equalization object pairing list, the switching control logic of the bidirectional CLLC resonant converter adjusts the energy transmission power to establish a directional energy transmission channel from the high-voltage single cell to the low-voltage single cell.

[0085] Based on the current monitoring data and voltage change rate during energy transmission, the working frequency and duty cycle parameters of the equalization circuit are dynamically adjusted, and when the voltage difference of each single cell converges to the threshold range, the energy transfer is ended, and the battery cluster state after voltage equalization is obtained.

[0086] Specifically, the difference calculation process of the real-time voltage of each single cell in the high-voltage lithium battery cluster adopts a method combining real-time data acquisition and statistical analysis. The voltage sensor array continuously monitors the terminal voltage of each single cell, and the data acquisition system records the voltage value at a fixed sampling interval and stores it in the cache array. The calculation process of the average voltage of the battery cluster adds the real-time voltages of all single cells by summation operation, and then divides the total number of batteries to obtain the average value, which represents the overall voltage level of the battery cluster. The calculation of the voltage deviation value of each single cell is realized by subtraction operation, which subtracts the real-time voltage of each single cell from the just calculated average voltage of the battery cluster. A positive value indicates that the voltage of the single cell is higher than the average level, and a negative value indicates that the voltage of the single cell is lower than the average level. The absolute value of the deviation value reflects the deviation degree of the single cell from the overall state of the battery cluster. The larger the absolute value, the more serious the deviation, and the equalization process needs to be prioritized. The data processing process adopts a loop traversal algorithm, which calculates the voltage deviation of each single cell in sequence according to the battery number, and stores the calculation results in the deviation array. The array index is consistent with the battery number, which is convenient for subsequent lookup and matching.

[0087] The comparison of the voltage deviation value and the rail transit dedicated equalization control threshold is classified and processed by conditional branching logic. The comparison algorithm first reads the voltage deviation value of each single battery from the deviation array, and then compares the value with the dynamically calculated equalization control threshold. The identification condition of high-voltage single battery is that the voltage deviation value is greater than the positive threshold, and the identification condition of low-voltage single battery is that the voltage deviation value is less than the negative threshold. The single battery within the threshold range is marked as normal state and does not need to be equalized. The formation process of the equalization object pairing list is realized by the pairing algorithm. The algorithm sorts the identified high-voltage single battery and low-voltage single battery according to the voltage difference degree, preferentially pairs the single battery with the largest voltage difference, and ensures the most significant equalization effect. The pairing list uses a two-dimensional array structure to store each row containing a pair of single batteries that need to be equalized, as well as their voltage difference value and estimated energy transfer demand. The pairing algorithm considers the topology of the battery cluster and the connection mode of the equalization circuit, and preferentially selects single batteries with short physical distance and convenient electrical connection for pairing, reducing the loss of energy transfer path.

[0088] The switch control logic of the bidirectional CLLC resonant converter adjusts the energy transfer power according to the voltage difference degree in the equalization object pairing list. The CLLC resonant converter is a bidirectional DC-DC converter with soft switching characteristics and high efficiency advantages, especially suitable for battery equalization applications. The switch control logic controls the switching frequency and duty cycle of the converter through pulse width modulation technology. The adjustment of the switching frequency changes the working state of the resonant circuit, and the adjustment of the duty cycle controls the direction and size of energy transfer. The calculation of energy transfer power is based on the voltage difference degree and the target equalization time. The greater the voltage difference, the more energy needs to be transferred, and the shorter the target equalization time, the greater the transfer power required. The establishment of the directional energy transfer channel is realized by the selective conduction of the switch matrix. When energy needs to be transferred from a high-voltage single battery to a low-voltage single battery, the controller closes the corresponding switch device to form a complete current loop. The path selection of the transfer channel considers the constraint conditions of the circuit topology, avoids mutual interference between multiple transfer channels, and ensures the stability and controllability of energy transfer.

[0089] The acquisition of current monitoring data in the energy transmission process is realized by a Hall current sensor, which is installed at a key node of each transmission channel to measure the current size and direction through the channel in real time. The calculation of the voltage change rate is realized by solving the time derivative of the voltage of the single battery, and the change rate is obtained by dividing the difference between the voltage values at adjacent sampling times by the sampling interval using the numerical differentiation method. The dynamic adjustment algorithm corrects the control parameters of the equalization circuit according to the feedback information of the current monitoring data and the voltage change rate. When the monitored transmission current is too large, the switching frequency is reduced to reduce the transmission power. When the monitored voltage change rate is too fast, the duty cycle is adjusted to control the equalization speed. The adjustment of the working frequency and the duty cycle parameters adopts a PID control algorithm, which realizes accurate adjustment of the parameters through the combination of the proportional, integral and differential control links. The proportional link adjusts the control amount according to the current deviation, the integral link adjusts the control amount according to the historical deviation, and the differential link adjusts the control amount according to the deviation trend. The convergence judgment condition is that the voltage difference of each single battery is less than the equalization control threshold. When the convergence condition is met, the controller stops the energy transmission process and disconnects the switching device to end the equalization operation.

[0090] In a specific embodiment, step S5 comprises:

[0091] The voltage time series data of each single battery in the state of the voltage equalized battery cluster are processed for trend analysis, the voltage change slope and fluctuation variance of each single battery are calculated, and the battery voltage trend characteristic parameters are obtained;

[0092] The change of the internal resistance of each single battery before and after the equalization process is measured and calculated by AC impedance test, the measured internal resistance value is compared with the reference internal resistance, and the internal resistance growth rate data matrix is obtained;

[0093] The voltage trend characteristic parameters and the internal resistance growth rate data matrix are combined to form a fault feature vector, the health state of each single battery is classified and identified by a support vector machine classification algorithm, and a battery fault risk level is output;

[0094] According to the battery fault risk level and the operation safety requirement of the rail transit UPS system, a maintenance strategy scheme including replacement time, maintenance cycle and monitoring frequency is generated by a decision tree algorithm, and a rail transit UPS system maintenance instruction is obtained.

[0095] Specifically, the trend analysis of the voltage time series data of each single battery in the balanced battery cluster state is a mathematical modeling of the voltage change trajectory after balancing by time series analysis method. The voltage time series data is collected during the continuous monitoring period after the balancing process is completed, the data collection interval is set to once every minute, and the continuous collection time is one complete operation cycle to obtain sufficient statistical samples. The calculation of the voltage change slope uses the least squares linear regression analysis, taking time as the independent variable and voltage as the dependent variable. The slope coefficient of the linear equation is obtained by regression fitting, which reflects the trend of battery voltage change with time. A positive slope indicates an upward trend in voltage, a negative slope indicates a downward trend in voltage, and the absolute value of the slope reflects the speed of change. The calculation of the fluctuation variance is realized by the statistical variance formula. First, calculate the mean of the voltage time series data, then calculate the square of the difference between each sampling point voltage value and the mean, sum all the square differences, and divide by the sample size minus one to get the sample variance. The variance value reflects the dispersion of the voltage data around the mean. The larger the variance, the more volatile the voltage, and the smaller the variance, the more stable the voltage change. The battery voltage trend characteristic parameters consist of slope and variance, forming a two-dimensional feature vector to describe the voltage behavior characteristics of each single battery.

[0096] The AC impedance test obtains the battery internal resistance information by injecting a small amplitude sinusoidal AC current signal into the single battery and measuring the corresponding voltage response. The impedance test frequency is selected at 1 kHz, and the impedance value measured at this frequency mainly reflects the ohmic resistance of the battery. The injected current amplitude is controlled at 0.1C of the battery rated capacity to avoid damage to the battery. The measurement and calculation process analyzes the amplitude and phase of the voltage response signal by Fourier transform. The real part of the impedance corresponds to the series resistance of the battery, and the imaginary part of the impedance corresponds to the capacitance characteristics of the battery. The internal resistance measurement before and after the balancing process is carried out before the balancing starts and after the balancing ends, respectively. The time interval between the two measurements ensures that the battery reaches a steady state to avoid measurement errors. The reference internal resistance value is derived from the standard test data of the battery at the factory or the baseline measurement data at the initial stage of the battery's use, which serves as a reference standard for evaluating internal resistance changes. The internal resistance growth rate is calculated in percentage form, with the formula being the difference between the current internal resistance and the reference internal resistance divided by the reference internal resistance multiplied by 100%. A positive growth rate indicates an increase in internal resistance, a negative growth rate indicates a decrease in internal resistance, and the absolute value of the growth rate reflects the degree of internal resistance change. The internal resistance growth rate data matrix is arranged in a two-dimensional array structure according to the battery number and measurement time, facilitating subsequent data processing and analysis.

[0097] The composition of the fault feature vector is obtained by fusing the voltage trend feature parameters and the internal resistance growth rate data through a feature engineering method. The dimension of the feature vector includes three main components, i.e., the voltage change slope, the voltage fluctuation variance, and the internal resistance growth rate, to form a three-dimensional feature space for describing the health state of the single battery. The support vector machine classification algorithm adopts a radial basis function as a kernel function, and the original feature space is mapped to a high-dimensional space through the kernel function for classification decision. The training sample set includes historical battery data with known health state labels, and the health state labels are divided into three levels, i.e., normal, early warning, and fault. The training process of the classification algorithm finds the optimal classification hyperplane by maximizing the classification interval, and the position and direction of the hyperplane are determined by the support vectors, which are the training sample points closest to the classification boundary. In the classification and identification process, the fault feature vector to be classified is input into the trained support vector machine model, and the model outputs the classification result and the confidence score, which reflects the reliability of the classification result. The battery fault risk level is determined comprehensively according to the classification result and the confidence score, i.e., a high-risk level corresponds to a high-confidence fault classification result, a medium-risk level corresponds to a low-confidence early warning classification result, and a low-risk level corresponds to a normal classification result.

[0098] The decision tree algorithm generates differentiated maintenance strategy schemes according to the battery fault risk level and the operation safety requirement of the rail transit UPS system. The construction of the decision tree is based on historical maintenance data and operation experience, the root node of the tree is the fault risk level judgment, the internal node is the maintenance strategy selection condition, and the leaf node is the specific maintenance action scheme. The decision logic of the replacement timing considers multiple factors such as the battery fault risk level, the remaining service life, and the operation safety requirement. The high-risk level battery is arranged to be replaced at the next planned shutdown, the medium-risk level battery is arranged to be replaced at the quarterly maintenance window, and the low-risk level battery is executed according to the normal replacement period. The adjustment strategy of the maintenance period dynamically adjusts the inspection frequency according to the fault risk level. The maintenance period of the high-risk battery is shortened to half of the normal period, the maintenance period of the medium-risk battery is shortened to two-thirds of the normal period, and the maintenance period of the low-risk battery remains unchanged. The setting strategy of the monitoring frequency associates the data acquisition frequency with the fault risk level. The monitoring frequency of the high-risk battery is increased to once an hour, the monitoring frequency of the medium-risk battery is set to once every four hours, and the monitoring frequency of the low-risk battery remains the standard frequency of once a day. The generation process of the rail transit UPS system maintenance instruction converts the maintenance strategy scheme output by the decision tree into a specific work task sheet, which includes detailed information such as the battery number, the maintenance type, the execution time, the required resources, and the safety precautions.

[0099] The above describes the voltage equalization control method for high-voltage lithium battery clusters for rail transit in the embodiments of the present application, and the voltage equalization control system for high-voltage lithium battery clusters for rail transit in the embodiments of the present application is described below. Please refer to Figure 2The high-voltage lithium battery cluster voltage equalization control system for rail transit in the embodiment of the application comprises the following modules:

[0100] The acquisition module is configured to acquire real-time voltage values of each single battery in the high-voltage lithium battery cluster and temperature and humidity parameters of the tunnel environment of rail transit, filter noise interference in the acquired data by using a Kalman filtering algorithm, and obtain environmental characteristic data of rail transit.

[0101] The prediction module is configured to combine the environmental characteristic data of rail transit with a change rule of the UPS load current, predict a voltage deviation trend of the battery cluster under instantaneous high-power demand of the signal system by using an LSTM neural network, and obtain a rail transit load adaptability voltage prediction value.

[0102] The calculation module is configured to calculate a rail transit special equalization control threshold according to the rail transit load adaptability voltage prediction value and the influence of the tunnel environment temperature and humidity, and replace a fixed threshold control strategy.

[0103] The starting module is configured to start a bidirectional energy transmission equalization circuit when the voltage difference of the single battery exceeds the rail transit special equalization control threshold, perform energy transfer between high-voltage single batteries and low-voltage single batteries, and obtain a state of the battery cluster after voltage equalization.

[0104] The output module is configured to monitor voltage change trends and internal resistance growth of each single battery in the equalization process, judge a battery health state by using a fault feature recognition algorithm, and output a rail transit UPS system maintenance instruction.

[0105] The above Figure 2 The high-voltage lithium battery cluster voltage equalization control system for rail transit in the embodiment of the application is described in detail from the perspective of a modular functional entity, and the high-voltage lithium battery cluster voltage equalization control device for rail transit in the embodiment of the application is described in detail from the perspective of hardware processing.

[0106] Reference Figure 3 The embodiment of the application further provides a high-voltage lithium battery cluster voltage equalization control device for rail transit, which can be a server, and the internal structure thereof can be as shown in Figure 3The rail transit high-voltage lithium battery cluster voltage equalization control device shown in the figure includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide calculation and control ability. The memory of the rail transit high-voltage lithium battery cluster voltage equalization control device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the rail transit high-voltage lithium battery cluster voltage equalization control device is used to store the corresponding data in this embodiment. The network interface of the rail transit high-voltage lithium battery cluster voltage equalization control device is used to communicate with the external terminal through the network connection. The computer program is executed by the processor to realize the above method.

[0107] Those skilled in the art can understand that, Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the application, and does not constitute a limitation on the rail transit high-voltage lithium battery cluster voltage equalization control device to which the scheme of the application is applied.

[0108] The application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium or a volatile computer readable storage medium. The computer readable storage medium stores instructions, and when the instructions run on the computer, the computer executes the steps of the rail transit high-voltage lithium battery cluster voltage equalization control method.

[0109] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-mentioned system, system and unit can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0110] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or the entire or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing an orbital traffic high-voltage lithium battery cluster voltage equalization control device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0111] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A high-voltage lithium battery cluster voltage equalization control method for rail transit, characterized in that, The method comprises: Step S1: Collecting real-time voltage values of each single battery in the high-voltage lithium battery cluster and temperature and humidity parameters of the tunnel environment of rail transit, filtering out noise interference in the collected data by a Kalman filtering algorithm, and obtaining rail transit environment characteristic data; Step S2: Combining the rail transit environment characteristic data with the UPS load current variation law, predicting the voltage deviation trend of the battery cluster under the instantaneous high-power demand of the signal system by an LSTM neural network, and obtaining a rail transit load adaptability voltage prediction value; Step S3: According to the rail transit load adaptability voltage prediction value and the influence of the tunnel environment temperature and humidity, calculating a rail transit special equalization control threshold to replace the fixed threshold control strategy; Step S4: When the voltage difference of the single battery exceeds the rail transit special equalization control threshold, starting a bidirectional energy transmission equalization circuit to transfer energy between high-voltage single batteries and low-voltage single batteries, and obtaining a voltage-equalized battery cluster state; Step S5: Monitoring the voltage variation trend and internal resistance growth of each single battery in the equalization process, judging the battery health state by a fault feature recognition algorithm, and outputting rail transit UPS system maintenance instructions. 2.The rail transit high-voltage lithium battery cluster voltage equalization control method according to claim 1, characterized in that, The step S1 comprises: Collecting instantaneous voltage data of each single battery in the high-voltage lithium battery cluster by a distributed voltage sensor array, and obtaining a battery cluster voltage sampling sequence; Collecting real-time values of tunnel temperature and humidity by a temperature sensor and a humidity sensor in the tunnel, and obtaining a tunnel environment parameter sequence; Aligning and merging the battery cluster voltage sampling sequence and the tunnel environment parameter sequence according to time stamps to form an original data matrix containing three-dimensional information of voltage-temperature-humidity; Based on the Kalman filtering algorithm, performing adaptive filtering processing on the jump data and drift noise in the original data matrix, eliminating interference signals generated by the electromagnetic environment of rail transit, and obtaining the rail transit environment characteristic data. 3.The rail transit high-voltage lithium battery cluster voltage equalization control method according to claim 1, characterized in that, The step S2 comprises: Extracting load current variation data of the signal system, lighting system and ventilation system from the historical operation records of the rail transit UPS system, and constructing a UPS load current time series database; Associating and matching the rail transit environment characteristic data and the UPS load current time series database according to the time dimension to form a training sample set containing a three-layer coupling relationship of voltage-environment-load; Dividing the training sample set according to a time window and inputting it into an LSTM neural network for deep learning training to establish a prediction model of the association between rail transit load patterns and battery voltage deviation; Based on the current rail transit operation timetable and the instantaneous power demand of the signal system, calculating the voltage deviation variation trend of each single battery in the battery cluster in the future time period by the prediction model, and obtaining the rail transit load adaptability voltage prediction value. 4.The rail transit high-voltage lithium battery cluster voltage equalization control method according to claim 3, characterized in that, The training sample set is divided according to a time window and input into an LSTM neural network for deep learning training to establish a prediction model of the association between rail transit load patterns and battery voltage deviation, comprising: Time series division processing is performed on the training sample set, and the sample data is divided into a window unit of 20 time steps to form an LSTM network input sequence matrix; The voltage data, environmental data, and load data in the input sequence matrix are normalized to eliminate the numerical difference between different dimension parameters and obtain a standardized training data set; A neural network architecture including an input layer, a bidirectional LSTM hidden layer, and a fully connected output layer is constructed, the number of hidden layer neurons is set to 128, the activation function is a tanh function, and the loss function is a mean square error function; The standardized training data set is divided into a training set and a validation set in a ratio of 8:2, network weight parameter iterative updating is performed through a back propagation algorithm, and training is stopped when the validation set loss function converges and the training round number reaches 1000 rounds; The trained LSTM network is tested for rail transit special load mode, environmental characteristic data in typical signal system startup and braking working conditions are input, and voltage deviation prediction accuracy is verified; The verified network model is packaged as the prediction model of the rail transit load mode and the battery voltage deviation, which is used for real-time prediction of the battery cluster voltage trend. 5.The rail transit high-voltage lithium battery cluster voltage equalization control method according to claim 1, characterized in that, The step S3 comprises: The future voltage deviation amplitude data of each single battery is extracted from the rail transit load adaptive voltage prediction value, and the battery cluster overall voltage dispersion index is calculated; According to the tunnel temperature and humidity parameters in the rail transit environmental characteristic data, a temperature correction coefficient and a humidity correction coefficient are respectively calculated, and the two correction coefficients are multiplied to obtain an environmental comprehensive correction factor; The preset rail transit basic equalization threshold value is weighted with the environmental comprehensive correction factor and the voltage dispersion index, and the contribution proportion of each influencing factor is adjusted through a dynamic weight distribution algorithm; Based on the weighted operation result and the safety margin requirement of the rail transit UPS system, the rail transit special equalization control threshold value for the current environmental working condition and the expected load is generated, covering the threshold value change range in different operation periods. 6.The rail transit high-voltage lithium battery cluster voltage equalization control method according to claim 1, characterized in that, The step S4 comprises: The real-time voltage of each single battery in the high-voltage lithium battery cluster is subjected to difference calculation processing, each single battery voltage is subtracted from the battery cluster average voltage to obtain the voltage deviation value of each single battery; The voltage deviation value is compared with the rail transit special equalization control threshold value to identify high-voltage single batteries and low-voltage single batteries whose voltage deviation exceeds the threshold value, and form an equalization object pairing list; According to the voltage difference degree in the equalization object pairing list, the switching control logic of the bidirectional CLLC resonant converter is adjusted to adjust the energy transmission power, and a directional energy transmission channel from the high-voltage single battery to the low-voltage single battery is established; Based on the current monitoring data and voltage change rate in the energy transmission process, the working frequency and duty cycle parameters of the equalization circuit are dynamically adjusted, and the energy transfer is ended when the voltage difference of each single battery converges to the threshold value range, and the voltage equalized battery cluster state is obtained. 7.The rail transit high-voltage lithium battery cluster voltage equalization control method according to claim 1, characterized in that, The step S5 comprises: The voltage trend characteristic parameters and the internal resistance growth rate data matrix are combined to form a fault feature vector, and a support vector machine classification algorithm is used to classify and identify the health state of each single battery, and a battery fault risk level is output. According to the battery fault risk level and the operation safety requirement of the rail transit UPS system, a decision tree algorithm is used to generate a maintenance strategy scheme including replacement time, maintenance cycle and monitoring frequency, and a rail transit UPS system maintenance instruction is obtained. The rail transit high-voltage lithium battery cluster voltage equalization control system comprises: A collection module is configured to collect real-time voltage values of each single battery in a high-voltage lithium battery cluster and temperature and humidity parameters of a rail transit tunnel environment, filter out noise interference in the collected data by using a Kalman filtering algorithm, and obtain rail transit environment characteristic data.

8. A high-voltage lithium battery cluster voltage equalization control system for rail transit, characterized in that, A prediction module is configured to combine the rail transit environment characteristic data with a UPS load current variation law, predict a voltage deviation trend of the battery cluster under a signal system instantaneous high-power demand by using an LSTM neural network, and obtain a rail transit load adaptability voltage prediction value. A calculation module is configured to calculate a rail transit special equalization control threshold according to the rail transit load adaptability voltage prediction value and the influence of the tunnel environment temperature and humidity, and replace a fixed threshold control strategy. A starting module is configured to start a bidirectional energy transfer equalization circuit when a single battery voltage difference value exceeds the rail transit special equalization control threshold, transfer energy between high-voltage single batteries and low-voltage single batteries, and obtain a voltage equalized battery cluster state. An output module is configured to monitor voltage variation trends and internal resistance growth of each single battery during the equalization process, determine a battery health state by using a fault feature recognition algorithm, and output a rail transit UPS system maintenance instruction. The computer program causes the processor to execute the rail transit high-voltage lithium battery cluster voltage equalization control method according to any one of claims 1 to 7 when the computer program is run on the processor. The computer program causes the processor to execute the rail transit high-voltage lithium battery cluster voltage equalization control method according to any one of claims 1 to 7 when the computer program is run on the processor.

9. A high-voltage lithium battery cluster voltage equalization control device for rail transit, characterized in that, ​ 10. A computer-readable storage medium having stored thereon a computer program, characterized in that, ​

Images