Remote charging and discharging control and service life prediction system and method for storage battery

Abstract

The invention provides a remote charging and discharging control and life prediction system and method for a storage battery, and relates to the technical field of storage battery monitoring, the system mainly comprises a data acquisition unit, a data monitoring and testing unit, a remote control unit, an SOH prediction unit and an RUL prediction unit, the on-line monitoring module is used for carrying out on-line monitoring on the acquired operation data of the storage battery and carrying out charge-discharge control on the storage battery through a charge-discharge control circuit according to a remote control instruction so as to at least realize the capacity checking test of the storage battery; the SOH prediction unit and the RUL prediction unit carry out SOH prediction and RUL prediction based on a PINN model and a Bi-LSTM model respectively. According to the invention, remote charging and discharging control and remote capacity checking of the storage battery can be realized, monitoring and management of the storage battery can be realized, and operation safety is ensured. And the SOH prediction unit and the RUL prediction unit are used to predict the health degree and the residual life of the storage battery based on a PINN model and a Bi-LSTM model respectively.

Description

Technical Field

[0001] The present invention relates to the technical field of battery monitoring, and in particular to a battery remote charge and discharge control and life prediction system and method. Background Art

[0002] In the power industry, communication batteries and AC / DC power system operating batteries are crucial for power supply safety. They require constant health monitoring and regular capacity verification to prevent accidents caused by aging and degradation. Remote battery capacity verification and maintenance include measuring the battery's float charge voltage, total battery pack voltage, internal resistance, and capacity testing. Maintenance personnel are required to perform regular inspections and maintenance, including a checkout discharge test every two years. After several years of operation, battery packs require an annual checkout discharge test. Current remote battery capacity verification methods are often slow, incomplete, and costly. This means that abnormal cells cannot be identified promptly, and the labor involved in maintenance and management is significant. Battery capacity testing primarily relies on manual on-site testing. Workers must bring specialized testing equipment to the battery installation site, connect to the equipment, perform discharge tests, and record voltage and current data over time. The data is then used to calculate the battery capacity. This approach presents numerous challenges in data acquisition, processing, and management. During data acquisition, manual operations are susceptible to subjective factors, such as inaccurate data due to loose connections of test equipment. Data processing relies on manual calculation of complex capacity formulas, which is inefficient and prone to errors. Regarding data management, paper records are difficult to preserve, making it extremely inconvenient to query and compile historical data. Specifically, traditional manual testing has the following main drawbacks: 1. Inefficiency: Manual on-site testing / discharge testing consumes significant time and manpower, especially for widely distributed battery packs, where individual testing is time-consuming and unsuitable for large-scale testing. 2. Poor Accuracy: Manual operations introduce numerous errors, including in equipment connection and data recording, resulting in test results that do not truly reflect the actual battery capacity. 3. Lack of Real-Time Monitoring: The battery capacity status cannot be monitored in real time, making it difficult to promptly detect batteries with abnormal capacity, which can easily lead to power system failures. 4. Difficult Data Management: Paper-based data is difficult to effectively integrate, analyze, and preserve over the long term, hindering comprehensive battery lifecycle management. 5. Safety Hazards: Manual maintenance can easily lead to battery short circuits, wire slippage, sparking, dummy load heat sources, and loss of consistency.

[0003] In the prior art, a Chinese patent application with publication number "CN116125296A" provides a substation lead-acid battery monitoring device and method. The device comprises: a host computer connected to a data transmission module; a current transformer connected to a battery pack, the current transformer being used to detect the battery pack current; a data transmission module connected to the current transformer, and further connected to a parameter sensor for transmitting data detected by the current transformer to the host computer; each parameter sensor being connected to a corresponding battery pack, with multiple parameter sensors connected in series, each parameter sensor being used to detect parameters of the corresponding battery pack cells; each battery pack comprising multiple battery cells and a transfer switch, with the multiple battery packs being connected in series; and a transfer switch being used to switch the number of battery cells in the battery pack connected to the corresponding parameter sensor. This existing solution can save on the number of parameter sensors, save costs, reduce the number of wiring connections, speed up device deployment, and reduce workload, thereby resolving some of the problems associated with conventional technologies. However, this existing solution only enables status monitoring and is unable to achieve remote charge and discharge control or battery health and life prediction, thus presenting certain limitations.

[0004] Based on this, the present invention is proposed. Summary of the Invention

[0005] In order to solve at least one problem existing in the above-mentioned background technology, the purpose of the present invention is to provide a battery remote charge and discharge control and life prediction system and method, which can realize remote battery charge and discharge control, remote capacity verification, monitor and manage the battery, ensure operational safety, and predict the health and remaining life of the battery.

[0006] To achieve the above object, the present invention provides the following technical solutions: In a first aspect, the present invention provides a battery remote charge and discharge control and life prediction system, comprising: a data acquisition unit for real-time acquisition of battery operating data; a data monitoring and testing unit for online monitoring of the acquired battery operating data, and controlling the charge and discharge of the battery through a charge and discharge control circuit and according to remote control instructions, so as to at least achieve a core capacity test of the battery; a remote control unit for acquiring data uploaded by the data monitoring and testing unit and sending remote control instructions to the data monitoring and testing unit; a state of health (SOH) prediction unit for performing SOH (state of health) prediction on the acquired battery operating data based on a PINN model; and a run-of-life (RUL) prediction unit for performing RUL (remaining useful life) prediction on the acquired battery operating data based on a Bi-LSTM model.

[0007] In a second aspect, the present invention provides a preferred solution, wherein the battery operation data includes the battery's state of charge, depth of discharge, charge and discharge current, voltage, resistance, temperature, and number of cycles.

[0008] In the second aspect, the present invention provides a preferred solution, wherein the data acquisition unit includes: a plurality of single-cell battery monitoring modules, which are installed on each battery cell and monitor and collect the battery cell operation data in real time; a group-end collector, whose input end is provided with multiple acquisition channels, which are respectively connected to each single-cell battery monitoring module and are used to collect the operation data of each battery cell.

[0009] In a second aspect, the present invention provides a preferred solution, wherein the remote control unit adopts a remote monitoring platform and / or a core-capacity host; the group-end concentrator and the core-capacity host both support local visualization of the machine.

[0010] In a second aspect, the present invention provides a preferred solution, wherein the charge and discharge control circuit of the data monitoring and testing unit includes: a first load and a second load, respectively connected to a first DC bus and a second DC bus; a first rectifier and a second rectifier, used to convert AC power into DC power, and the rectified DC power is respectively connected to a first DC bus and a second DC bus; a bus tie switch, used to connect or disconnect the first DC bus and the second DC bus; a first diode, a first normally closed contactor, a second diode and a second normally closed contactor, used to connect the first battery group and the second battery group to the first DC bus and the second DC bus, respectively; a first charge and discharge switch and a second charge and discharge switch, respectively connected to the first battery group and the second battery group, for controlling the charging or discharging of the first battery group and the second battery group; a bidirectional converter, connected between the first charge and discharge switch and the second charge and discharge switch, and connected to the AC power grid, for realizing the bidirectional flow of electric energy between the first battery group or the second battery group and the AC power grid.

[0011] In a second aspect, the present invention provides a preferred solution, wherein the data monitoring and testing unit is further used to implement internal resistance testing, spectrum testing, and K1 / D1 testing of the battery.

[0012] The present invention provides a preferred solution in the second aspect, further comprising a data preprocessing unit for identifying the acquired battery operation data and eliminating missing values and abnormal data, and then normalizing the remaining data after elimination.

[0013] In the second aspect, the present invention provides a preferred solution, and the PINN model of the SOH prediction unit specifically includes: a Transformer encoder, which extracts time series features from the input battery operation data including current, voltage, resistance, temperature and number of cycles through multi-head self-attention, feedforward network, residual connection and layer normalization; a fully connected layer, which aggregates time series features and outputs the SOH prediction value; a data-driven loss calculation module, which calculates the error between the SOH prediction value and the true SOH value to obtain the data-driven loss; a physical constraint loss calculation module, which calculates the mean square error of the residual of the physical equation to obtain the physical constraint loss; a total loss function acquisition module, which sets dynamic weights according to the data-driven loss and the physical constraint loss, establishes a total loss function, and calculates the total loss; an Adam optimizer, which is used to minimize the total loss function and obtain dynamic weight parameters based on a dynamic weight adjustment strategy of uncertainty or gradient balance, and then adjusts and updates the dynamic weight parameters.

[0014] In the second aspect, the present invention provides a preferred solution, and the Bi-LSTM model of the RUL prediction unit specifically includes: an input layer, which receives a data set containing battery operation data, determines the number of neurons in the input layer according to the number of feature types in the data set, and passes the battery operation data to the hidden layer in the form of a sequence; the hidden layer includes multiple Bi-LSTM layers, which are used to capture long-term dependencies and time series information in the battery operation data; each Bi-LSTM layer consists of a forward LSTM and a reverse LSTM, and can simultaneously consider the forward and reverse information of the sequence; a Dropout layer, which is added after each Bi-LSTM layer and is used to randomly discard part of the neuron connections during the training process; a fully connected layer, which maps the features processed by the Dropout layer to the predicted target battery RUL; an output layer, which outputs the RUL predicted value of the output of the fully connected layer through an activation function; a loss calculation module, which establishes a loss function, and calculates the loss function based on the RUL predicted value and the RUL true value to obtain the loss; an Adam optimizer, which is used to minimize the loss function and update the parameters of the model.

[0015] In a second aspect, the present invention provides a method for remote charge and discharge control and life prediction of a battery, which is applied to the above-mentioned system and includes: data acquisition, real-time acquisition of battery operating data; data monitoring and testing, online monitoring of the acquired battery operating data, and charging and discharging control of the battery through a charge and discharge control circuit and according to remote control instructions, so as to at least achieve the core capacity test of the battery; remote control, acquiring data uploaded by a data monitoring and testing unit, and sending remote control instructions to the data monitoring and testing unit; SOH prediction, performing SOH (state of health) prediction on the acquired battery operating data based on a PINN model; RUL prediction, performing RUL (remaining useful life) prediction on the acquired battery operating data based on a Bi-LSTM model.

[0016] Compared with the prior art, the technical solution of the present invention has the following beneficial technical effects: 1. The present invention's remote battery charge and discharge control and life prediction system and method utilizes a data monitoring and testing unit, including a charge and discharge control circuit, and a remote control unit to enable remote battery charge and discharge control and remote capacity verification. This system also monitors and manages the battery to ensure safe operation. Furthermore, the battery's health and remaining life are predicted using a State of Health (SOH) prediction unit and a RUL prediction unit based on a PINN model and a Bi-LSTM model, respectively.

[0017] 2. The data acquisition unit of the present invention monitors voltage, temperature, internal resistance, etc. through the single-cell battery monitoring module. In combination with the group-end aggregator, it can realize the real-time collection and integration of 240 single-cell battery data, covering the full-dimensional operating parameters of the battery pack, greatly reducing the error rate, and improving the overall perception capability. The single-cell battery monitoring module, the group-end aggregator, and the core capacity host all support local machine visualization, which not only makes it convenient for operation and maintenance personnel to directly view key parameters on site, but also allows basic maintenance to be performed on site in the event of power outages or network anomalies, avoiding operation and maintenance stagnation caused by communication interruptions. At the same time, the core capacity host and the remote monitoring platform provide a unified management interface, a dual verification mechanism for local display and remote data, which reduces the risk of misjudgment and is particularly suitable for the diagnosis of occasional faults.

[0018] 3. The data monitoring and testing unit of the present invention realizes the full-process nuclear capacity test of the battery pack through the charge and discharge control circuit (including bidirectional converter) and remote command issuance, without the need for manual on-site operation, thereby improving the testing efficiency. In addition, the present invention can also realize remote internal resistance testing, spectrum testing and K1 / D1 testing.

[0019] 4. The charge-discharge control circuit of this invention utilizes a unidirectional diode and a normally closed contactor design, enabling discharge testing of the battery pack without requiring it to be offline, shortening test time and reducing the risk of power interruptions. Furthermore, a bidirectional converter enables DC / AC bidirectional conversion, adapting to different voltage levels, reducing energy loss and improving system stability. The bidirectional converter's built-in high-voltage DC circuit breaker quickly cuts off abnormal currents, reducing the risk of battery damage and high-temperature fires.

[0020] 5. The SOH prediction of the present invention is based on the PINN model, which integrates the Transformer encoder and the physical equations of battery aging. It maintains high robustness in data-scarce scenarios, significantly reducing prediction errors and improving SOH prediction accuracy. At the same time, it introduces a dynamic weight adjustment strategy to optimize model convergence speed, which can shorten model training time. The RUL prediction of the present invention uses Bi-LSTM combined with Dropout technology to capture long-term temporal dependencies, providing an early warning of the capacity decay critical point 3-6 months in advance, reducing the sudden failure rate, and enhancing the foresight of life prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0022] Figure 1 This is a module diagram of a battery remote charge and discharge control and life prediction system provided by Example 1 of the present invention; Figure 2 This is a flow chart of a method for remote battery charge and discharge control and life prediction provided by Example 1 of the present invention; Figure 3 A topological diagram of a data monitoring and testing unit of a battery remote charge and discharge control and life prediction system provided in Example 2 of the present invention; Figure 4 A circuit diagram of a bidirectional converter in a data monitoring and testing unit of a battery remote charge and discharge control and life prediction system provided in Example 2 of the present invention; Figure 5 This is a flow chart of a battery remote charge and discharge control and life prediction system provided by Example 2 of the present invention; Figure 6 This is a PINN model architecture diagram based on the SOH prediction unit of a battery remote charge and discharge control and life prediction system provided by Example 2 of the present invention; Figure 7This is a Bi-LSTM model architecture diagram based on the RUL prediction unit of a battery remote charge and discharge control and life prediction system provided in Example 2 of the present invention.

[0023] Figure numerals: data acquisition unit 100, data monitoring and testing unit 200, first-stage load 211, second-stage load 212, first rectifier 221, second rectifier 222, bus tie switch 230, first diode 241, second diode 242, first normally closed contactor 251, second normally closed contactor 252, first charge and discharge switch 261, second charge and discharge switch 262, bidirectional converter 270, active inverter discharge module 271, DC power supply 272, remote control unit 300, SOH prediction unit 400, RUL prediction unit 500, first battery pack 61, second battery pack 62, AC power grid 7. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Example 1

[0025] Please refer to Figure 1 This embodiment provides a battery remote charge and discharge control and life prediction system, which mainly includes the following parts: a data acquisition unit 100, which is used to collect battery operating data in real time; a data monitoring and testing unit 200, which is used to monitor the acquired battery operating data online, and control the battery charge and discharge through a charge and discharge control circuit and according to remote control instructions to at least achieve the core capacity test of the battery; a remote control unit 300, which is used to acquire data uploaded by the data monitoring and testing unit 200 and send remote control instructions to the data monitoring and testing unit 200; a state of health (SOH) prediction unit 400, which is used to perform state of health (SOH) prediction on the acquired battery operating data based on a PINN model; and a load-limiting (RUL) prediction unit 500, which is used to perform load-limiting (RUL) prediction on the acquired battery operating data based on a Bi-LSTM model.

[0026] The present invention's remote battery charge and discharge control and life prediction system and method utilizes a data monitoring and testing unit 200, including a charge and discharge control circuit, and a remote control unit 300 to enable remote battery charge and discharge control and remote capacity verification. This system also monitors and manages the battery to ensure safe operation. Furthermore, the battery's health and remaining life are predicted using the PINN model and the Bi-LSTM model, respectively, through the SOH prediction unit 400 and RUL prediction unit 500.

[0027] Please refer to Figure 2 Accordingly, this embodiment provides a method for remote battery charge and discharge control and life prediction, which is applied to the above system and is mainly implemented through the following steps: S100. Data acquisition, real-time collection of battery operation data; S200 data monitoring and testing, to obtain the battery operating data online monitoring, through a charge and discharge control circuit and according to remote control instructions to control the battery charge and discharge, in order to achieve at least the battery nuclear capacity test; S300 remote control, obtain data monitoring and testing unit 200 uploaded data, and sends remote control instructions to the data monitoring and testing unit 200; S400.SOH prediction: performs SOH prediction based on the acquired battery operation data based on the PINN model; S500.RUL prediction: performs RUL prediction based on the acquired battery operation data based on the Bi-LSTM model.

[0028] This embodiment of the remote battery charge and discharge control and life prediction system and method utilizes a data monitoring and testing unit 200 and a remote control unit 300, including a charge and discharge control circuit, to implement remote battery charge and discharge control and remote capacity verification. This system also monitors and manages the battery to ensure safe operation. Furthermore, the state of health (SOH) prediction unit 400 and the residual life (RUL) prediction unit 500 predict the battery's health and remaining life based on the PINN model and the Bi-LSTM model, respectively. Example 2

[0029] Based on Example 1, a more preferred remote battery charge and discharge control and life prediction system is provided. Specifically, for the data acquisition unit 100, this embodiment adopts a single-cell monitoring module + group-end aggregator architecture. As an example, two battery packs are provided in this embodiment, and each single cell in each battery pack is equipped with a single-cell monitoring module for real-time monitoring and acquisition of battery cell operating data, including battery pack charge state, battery pack discharge depth, battery pack charge and discharge current, voltage (DC bus voltage, battery pack voltage, single cell voltage), resistance (cell internal resistance), temperature (cell negative electrode temperature), and cycle count. The group-end aggregator has multiple acquisition channels at its input, connected to each single-cell monitoring module, for aggregating individual battery cell operating data.

[0030] In a preferred embodiment, the data monitoring and testing unit 200 is mainly implemented by a charge and discharge control circuit, please refer to Figure 3 and Figure 4 The specific components of the charge and discharge control circuit are as follows: a first-stage load 211 and a second-stage load 212 are connected to a first-stage DC bus and a second-stage DC bus, respectively; a first rectifier 221 and a second rectifier 222 are used to convert AC power into DC power, and the rectified DC power is connected to a first-stage DC bus and a second-stage DC bus, respectively; a bus tie switch 230 is used to connect or disconnect the first-stage DC bus and the second-stage DC bus; a first diode 241, a first normally closed contactor 251, a second diode 242, and a second normally closed contactor are used to connect the first battery pack 61 and the second battery pack 62 to the first-stage DC bus and the second-stage DC bus, respectively; the first diode 241 and the first normally closed contactor 251 form K01, and the second diode 242 and the second normally closed contactor form K02. The first charge / discharge switch 261 (K1) and the second charge / discharge switch 262 (K2) are connected to the first battery pack 61 and the second battery pack 62, respectively, to control the charging or discharging of the first battery pack 61 and the second battery pack 62. A bidirectional converter 270 is connected between the first charge / discharge switch 261 and the second charge / discharge switch 262 and to the AC grid 7, enabling bidirectional flow of electrical energy between the first battery pack 61 or the second battery pack 62 and the AC grid 7. In this embodiment, the bidirectional converter 270 includes an active inverter discharge module 271 and a DC power supply 272. It utilizes the principle of controlling battery line switching and unidirectional diode conduction to achieve seamless switching and ensure the batteries are always online.

[0031] In summary, the charge and discharge control circuit of this embodiment uses a diode circuit and a normally closed contactor to implement real-time online discharge testing of the battery. The diode's unidirectional conduction principle enables seamless switching, ensuring the battery's real-time online operation and providing protection against overvoltage, overtemperature, and short circuits. A built-in bidirectional converter 270 controls the battery's charging and discharging processes, enabling bidirectional flow of electrical energy. This converter can convert stored DC power to a voltage level suitable for use or transmission, helping to improve the overall efficiency and stability of the energy storage system. The bidirectional converter 270 is equipped with a high-voltage DC circuit breaker. In the event of DC input or output anomalies, it can quickly disconnect the battery pack and busbar interconnection unit, automatically closing or disconnecting the first and second busbar connections based on the on-site battery test status.

[0032] This embodiment uses the unidirectional conduction characteristics of the diode in combination with a normally closed contactor to complete the discharge test without disconnecting the battery from the load, avoiding the risk of power supply interruption caused by offline operation in traditional nuclear capacity testing. The bidirectional converter 270 of this embodiment adopts a V2G bidirectional converter 270, which supports dynamic adaptation of the input / output voltage level, reduces energy loss and improves system stability. This shortens the online test time, and the voltage fluctuation range of the power system is controlled within ±1% during the test. The high-voltage DC circuit breaker of the bidirectional converter 270 is a fast tripping mechanism that can cut off abnormal current within 10ms, reducing battery pack damage and the risk of high-temperature fire.

[0033] Next, this embodiment will combine Figure 3 The specific functions, implementation processes and circuit diagram changes of various tests under different tests are explained: 1. Core capacity test: By selecting the battery pack for core capacity test, the core capacity test is performed according to the set control parameters (such as discharge current, discharge capacity, discharge time, single cell voltage lower limit, charging current, etc.). First, discharge is performed. When the discharge time or discharge capacity is reached, it is switched to pre-charge state. After charging is completed, it is switched to float charge state, and the core capacity test ends.

[0034] 2. Internal resistance test. Select the battery pack for internal resistance test. After the test is completed, you can view the internal resistance test details of each single cell in this test on the display screen of the core capacity host or the platform. When doing the core capacity test, the discharge stage: K01 / K02 status is open and closed, the charge and discharge switch K1 / K2 is closed, and the bidirectional converter 270 discharge module discharges the battery pack. The charging stage: K01 / K02 status is open and closed, the charge and discharge switch K1 / K2 is closed, and the bidirectional converter 270 charging module charges the battery pack. The floating charge stage: K01 / K02 status is closed, the charge and discharge switch K1 / K2 is open and closed, and the rectifier charges the battery. In the process of doing the internal resistance test, Figure 3 The circuit diagram in remains unchanged.

[0035] 3. Spectrum test. Select the battery pack for spectrum test. After the test is completed, you can view the electrochemical impedance spectrum of each single cell tested on the display screen of the core host or the platform. During the spectrum test, Figure 3 The circuit diagram in remains unchanged.

[0036] 4. K1 / D1 Test. Check whether the actual operating status of the normally closed switch is consistent with the displayed result. Check the bus system voltage to ensure it does not fluctuate and that the bus system equipment can operate normally. During the K1 / D1 test, the circuit diagram K01 / K02 status is open or closed, and the charge and discharge switches K1 / K2 are open or closed.

[0037] The bidirectional converter 270 automatically controls the charging, discharging, and core capacity testing of the battery pack according to specific instructions issued by the remote control unit 300. In a preferred embodiment, the data monitoring and testing unit 200 also performs internal resistance testing, pattern testing, and K1 / D1 testing on the battery pack according to specific instructions issued by the remote control unit 300. In a preferred embodiment, the bidirectional converter 270 can automatically adjust test priorities based on historical alarm information. For example, if a cell frequently experiences abnormal internal resistance, the pattern test or K1 / D1 test will be triggered first. This dynamic testing strategy improves core capacity testing efficiency and avoids the additional loss of battery life caused by redundant testing. In a preferred embodiment, the system also includes a data correlation analysis unit that performs multi-dimensional cross-analysis on acquired historical data, such as comparing temperature-internal resistance curves to identify degradation trends of different battery batches. This provides a statistical basis for optimizing operation and maintenance strategies. Historical data backtracking can identify batch-specific quality issues, such as abnormal collective temperature rise in a manufacturer's batteries, promoting closed-loop quality management in the supply chain.

[0038] In a preferred embodiment, the remote control unit 300 utilizes a remote monitoring platform and a core capacity host. The single-cell battery monitoring module, the group-end concentrator, and the core capacity host all support local machine visualization. This not only allows maintenance personnel to directly view key parameters such as single-cell voltage and temperature deviation on-site, but also allows for basic maintenance on-site during power outages or network anomalies, avoiding operational downtime caused by communication interruptions. The core capacity host also provides a unified management interface via a 7-inch touchscreen and remote monitoring platform. This dual verification mechanism of local display and remote data reduces the risk of misjudgment, making it particularly suitable for diagnosing occasional faults.

[0039] The core capacity host can be considered a central monitoring system, featuring a built-in centralized monitoring unit and display and interactive modules. For example, it utilizes a 7-inch color touchscreen to receive and display aggregated data such as individual battery voltage, internal resistance, temperature, and charge / discharge current. It also displays the status of each unit in the system, supports on-screen configuration of monitoring parameters, alarm thresholds, and historical alarm information. This data is then uploaded to the remote monitoring platform / server program, including the SOH prediction unit 400 and RUL prediction unit 500, using the IEC 61850 MMS protocol for data access, device control, and data transmission. The core capacity host supports exporting historical battery data and issuing real-time alarms, ensuring safe and efficient battery operation. Using the IEC 61850 MMS protocol, it directly connects to the standardized power system communication framework and supports customized interfaces for adapting to non-standard systems. This provides excellent communication protocol compatibility, eliminating the need to modify existing power monitoring platforms during system deployment, reducing integration development costs. This also reduces the risk of hardware damage during actual operation due to operational errors, such as reverse connection. The present invention forms three layers of protection (single-cell level, group-end level, and system level), covering the entire link of electrical faults, from the reverse connection protection of the single-cell battery monitoring module to the automatic segmentation of the busbar connection unit. In addition, for the interface design of the 7-inch touch screen, in a preferred embodiment, industry-standard icons and hierarchical menus can be used. For example, the "alarm threshold setting" is hidden in the advanced menu by default to avoid parameter tampering caused by accidental touch. Through the interface design of the interactive module and the display module, the status visualization is enhanced. The following scheme can be adopted: the charge and discharge current is displayed in the form of a dynamic curve superimposed on a bar chart to intuitively reflect the consistency of the battery pack. For example, a sudden drop in the current of a single cell indicates poor contact. The cell with excessive internal resistance can also be quickly located by color marking, reducing the time for abnormal location.

[0040] More specifically, the group-end aggregator used in this embodiment can support data aggregation of up to 240 single cells, provide an RJ45 interface upward, use the RS485 communication bus to communicate with the intelligent core capacity host, provide 4 RJ11 interfaces downward, support series use, support the automatic coding function of the single cell monitoring module, automatically number the battery, support current acquisition and one-way group-end voltage acquisition, support reverse connection protection and overvoltage protection, and the above data is uploaded to the centralized monitoring unit of the core capacity host via wired mode and serial port protocol. This embodiment uses a multi-channel acquisition design, and the group-end aggregator supports series use through 4 RJ11 interfaces, and the automatic coding function can achieve one-stop management of 240 single cells, greatly reducing the wiring complexity and hardware cost in multi-battery group scenarios. RS485 bus communication has long-distance transmission capability and anti-electromagnetic interference characteristics, combined with reverse connection protection, overvoltage protection ( Figure 4The bidirectional converter 270 in the system has overvoltage, overtemperature, short circuit and other protections), ensuring the anti-interference, stability and real-time performance of data acquisition. Therefore, the system of this embodiment can support the rapid deployment of large-scale battery groups, such as substations and data centers, reducing on-site construction time. It can be flexibly expanded to multiple battery groups in parallel or series, and is compatible with the needs of energy storage systems of different scales. Please refer to Figure 5 In a preferred embodiment, the specific data transmission, control and processing flow of the battery remote charge and discharge control and life prediction system of this embodiment is as follows: S1. (Single battery monitoring module) monitors and collects battery cell operating data in real time; S2. (Group-end concentrator) collects the operating data of each battery cell; S3. (Core capacity host) receives, stores and displays the collected operating data of each single battery; S4. (Remote monitoring platform / server program) receives and stores the operating data of each battery cell; S5. (Remote monitoring platform / server program) determines whether to generate a remote test operation instruction; S6. (Remote monitoring platform / server program) performs Ukey / authorization verification when generating remote test operation instructions.

[0041] S7. (Remote monitoring platform / server program) preprocesses the acquired operating data and performs feature extraction and model training when no remote test operation instructions are received; S8. (SOH prediction unit 400 and RUL prediction unit 500 of the remote monitoring platform / server program) performs SOH prediction and RUL prediction based on the preprocessed operation data; S9. (Remote monitoring platform / server program) determines whether the verification is successful; S10. (Remote monitoring platform / server program) sends remote control instructions after verification is successful; if verification fails, the process ends. S11. (Core capacity host) sends test instructions; S12 (data monitoring and testing unit 200) performs specific tests according to instructions, and uploads the test process and results; S13. (Core capacity host) Receive and store the test process and results.

[0042] In a preferred embodiment, the system is also provided with a data preprocessing unit for identifying the acquired battery operation data and eliminating missing values and abnormal data, and then normalizing the remaining data after elimination. Specifically: For the historical operation data of the battery collected by the data acquisition module, including parameters such as the battery's charge state, discharge depth, charge and discharge current, voltage, resistance, temperature, and number of cycles, etc., preprocessing is required, including: checking the integrity of the data, processing missing values, and identifying and eliminating data points that obviously deviate from the normal range through statistical analysis or setting a reasonable threshold range to avoid adverse effects on model training. Each feature data is normalized so that the data falls within the interval [0, 1]. The normalization method adopted in this embodiment is minimum-maximum normalization, and the formula is: ,in, is the original data, and are the minimum and maximum values of the feature, respectively. The characteristics here refer to each specific characteristic parameter in the battery's historical operating data, such as the battery's state of charge, depth of discharge, charge and discharge current, voltage, resistance, temperature, and number of cycles. For each such characteristic parameter, its minimum value is used. and maximum value The advantages of using this minimum-maximum normalization are: unifying the data range, reflecting the relative relationship of the data, and enhancing the generalization ability of the model.

[0043] In a preferred embodiment, the SOH prediction unit 400 is set in the remote monitoring platform / server program and is used to perform SOH prediction on the acquired battery operating data based on the PINN model. The battery life prediction module uses the PINN algorithm that embeds physical laws into neural networks. Compared with traditional neural networks, it may perform better under physical constraints, especially when data is insufficient. Please refer to Figure 6 The PINN model of the SOH prediction unit 400 specifically consists of a Transformer encoder, a fully connected layer, a data-driven loss calculation module, a physical constraint loss calculation module, a total loss function acquisition module, and an Adam optimizer. First, physical modeling introduces the battery aging equation: , where K is the attenuation coefficient and N is the number of cycles, is the initial value of the battery health status. Secondly, the specific implementation principles and processes of each part of the PINN model are explained as follows: The Transformer encoder performs Min-Max normalization on voltage, current, and temperature. It extracts time series features from input battery operating data, including current, voltage, resistance, temperature, and cycle count, using multi-head self-attention, a feedforward network, residual connections, and layer normalization. The Transformer uses several key formulas when processing time series data, primarily including the attention mechanism, feedforward network, residual connections, and layer normalization.

[0044] The calculation formula of multi-head self-attention is:

[0045] represents the output of the first attention head, represents the output of the second attention head, h represents the total number of attention heads, which is a hyperparameter and is usually set to values such as 4, 8, or 16.

[0046] Represents the output projection matrix (Output projection matrix), the dimension is , : The dimension of each attention head, : The hidden layer dimension of the model.

[0047] Specific calculation (combined with battery voltage, current, and temperature time series data): Assume that the input is a single batch of battery charge and discharge cycle data, the sequence length is 5 time steps (corresponding to 5 cycles), and each time step contains 4 features: Current It (A), voltage Vt (V), temperature Tt (°C), number of cycles Ct.

[0048] Input data X (normalized):

[0049] Here, the shape: [5,4] indicates that the data is a two-dimensional array with 5 rows and 4 columns. The 5 rows correspond to the 5 time steps mentioned above (that is, 5 cycles), and the 4 columns correspond to the 4 features contained in each time step (current, voltage, temperature, and number of cycles).

[0050] Example of actual input data obtained:

[0051] Generation of queries (Q), keys (K), and values (V); Feature Dimension , the number of multiple heads h=2, key (K), value (V) dimensions of each head .

[0052] Calculation process:

[0053] Among them, W Q is the Query matrix, W K is the Key matrix, W V is the Value matrix.

[0054] Weight matrix:

[0055] Calculation results:

[0056] Multi-head segmentation and attention calculation Split into two heads

[0057] are the first 2 columns (dimension = 2) The last two columns (dimension = 2) Formula for calculating attention score

[0058] =2 (dimensions of each head) Score matrix calculation:

[0059]

[0060] Attention output:

[0061] Final attention output:

[0062] The calculation formula of the feedforward network is: ; where x is the input vector, is the weight matrix of the first layer linear transformation, is the bias vector of the first layer linear transformation, is the bias vector of the second layer linear transformation, The weight matrix of the second layer linear transformation, max(0,.): activation function, sets negative values to zero and retains positive values.

[0063] Specific calculation: Assume there is an input vector x∈R4 and the hidden layer dimension of the feedforward network is 8, then:

[0064] After calculation through the feedforward network, the final output is a vector of dimension 4, which is used for subsequent residual connection and layer normalization operations. Residual connection and layer normalization are used to assist in training. The formula is: ; where x is the input, Is the output of the sublayer (feedforward neural network layer), first add the input and sublayer output, and then perform layer normalization. The calculation formula for layer normalization is: ; where μ and are the mean and variance of the input x in each dimension, and are learnable parameters, is a small constant (usually set to or ), used to prevent the denominator from being zero.

[0065] The fully connected layer aggregates time series features and outputs a predicted SOH value. Specifically, the fully connected layer receives the layer-normalized output from the Transformer encoder. This output provides a standardized data foundation for time series feature aggregation by eliminating feature dimensionality differences and stabilizing distribution. At this stage, the fully connected layer uses multiple layers of nonlinear transformations to gradually compress and map the high-dimensional time series features extracted by the encoder into the battery state of health (SOH) prediction space, ultimately outputting a continuous SOH prediction value. This process not only preserves the Transformer's ability to model long-range dependencies but also further enhances the model's capture of local time series patterns through the fully connected layer's feature reorganization capabilities.

[0066] The data-driven loss calculation module calculates the error between the SOH prediction value and the actual SOH value to obtain the data-driven loss. Data loss (MSE) : The error between the predicted SOH and the true SOH. Specifically, the data-driven loss calculation module uses the prediction value of the fully connected layer as input and calculates the mean square error (MSE) with the true SOH value, which directly reflects the numerical deviation of the model prediction. The physical constraint loss module calculates the residual of the battery electrochemical equation, constraining the predicted value to conform to physical laws (such as the correlation between the capacity decay rate and temperature and voltage).

[0067] The physical constraint loss calculation module calculates the mean square error of the physical equation residual to obtain the physical constraint loss. The formula for the mean square error of the physical equation residual is: ; Where M is the total number of samples, is the model prediction value, is the theoretical value of the physical equation.

[0068] The total loss function acquisition module sets dynamic weights based on data-driven loss and physical constraint loss, establishes a total loss function, and calculates the total loss. The total loss calculation formula is: ;in, For data loss, For physical losses, , is a hyperparameter. In this formula, automatic differentiation is also used to calculate the derivative terms in the physical equations (such as the partial derivative of voltage with respect to SOC) to ensure that the model conforms to physical laws.

[0069] Adam optimizer is used to minimize the total loss function and obtain dynamic weight parameters based on the dynamic weight adjustment strategy of uncertainty or gradient balance, and then adjust and update the dynamic weight parameters. Using Adam optimizer, dynamic adjustment and The total loss function combines the two according to the dynamic weights and Weighted combination, the Adam optimizer is used to update the network parameters and weight coefficients synchronously. When the battery operating conditions are complex, the dynamic weight strategy will increase the weight of the physical constraint loss. , suppressing the prediction bias caused by data noise; when there is sufficient data and low noise, it focuses on data-driven loss To improve prediction sensitivity. The Adam optimizer adjusts the learning rate adaptively and balances the parameter update step size to ensure that the model achieves the optimal balance between convergence speed and stability, ultimately achieving dual fitting of physical laws and data characteristics.

[0070] In summary, the data acquisition module in the system is used to collect data on the battery charge and discharge cycles, including voltage, current, temperature, and the actual capacity measurement value after each cycle as labels, which are normalized and processed after missing values, and then the training set and test set are split. The battery charge and discharge data are collected in real time, input into the trained PINN model, and the SOH estimation value is output. In a preferred embodiment, an alarm mechanism is set up, and when the SOH is lower than the set threshold, a battery replacement or maintenance alarm is triggered. For labels, the actual capacity measurement value (for example, the battery capacity is 290Ah after a certain cycle) is the label, which indicates the actual performance of the battery under specific working conditions. The input data (voltage, current, temperature, etc.) are the features of the model (Features) used to predict labels.

[0071] In a preferred embodiment, the RUL prediction unit 500 is set in the remote monitoring platform / server program and is used to perform RUL prediction on the acquired battery operation data based on the Bi-LSTM model. Figure 7The Bi-LSTM model of the RUL prediction unit 500 specifically consists of an input layer, a hidden layer, a Dropout layer, a fully connected layer, an output layer, a loss calculation module, and an Adam optimizer. The specific implementation principles and processes of each part of the Bi-LSTM model are described below: (1) Data preparation The preprocessed dataset is divided into a training set, a validation set, and a test set according to a certain ratio. The training set is used to train the model, the validation set is used to adjust the model's hyperparameters and monitor the model's training process, and the test set is used to evaluate the model's final performance.

[0072] (2) Model construction The input layer receives a dataset containing battery operating data. The number of neurons in the input layer is determined based on the number of feature types in the dataset. The input layer then passes the battery operating data as a sequence to the hidden layer. Specifically, the number of neurons in the input layer is determined by the number of features in the dataset. For example, if the dataset contains four features: state of charge, depth of discharge, voltage, and temperature, the number of neurons in the input layer is four. The input layer receives the normalized battery operating data sequence and passes it to the hidden layer.

[0073] The hidden layer, consisting of multiple Bi-LSTM layers, is used to capture long-term dependencies and time series information in battery operating data. Each Bi-LSTM layer consists of a forward LSTM and a backward LSTM, capable of simultaneously considering both forward and backward information in the sequence. The number of neurons in each Bi-LSTM layer is determined appropriately, as increasing the number of neurons in the hidden layer improves the model's expressiveness but also increases computational complexity and the risk of overfitting. Specifically, a dynamic adjustment strategy is used to appropriately determine the number of neurons in each Bi-LSTM layer, using the following two methods: (a) Incremental Expansion and Early Stopping Initial setup: Start with a smaller size (e.g. 32 / 64 neurons / layer) and gradually increase.

[0074] Early stopping monitoring: Track the validation set loss during training and stop if there is no improvement in consecutive NN rounds to avoid overfitting.

[0075] Python code example: for units in [32, 64, 128]: model = build_model(units) history = model.fit(X_train, Y_train, validation_split=0.2,callbacks=[EarlyStopping(patience=5)]) if validation_accuracy_plateau: break (b) Inflection point analysis Plot the number of neurons vs. validation accuracy and select the critical point where the accuracy improvement slows down.

[0076] A Dropout layer is added after each Bi-LSTM layer to randomly drop some neuron connections during training. Dropout is a regularization technique that reduces the risk of overfitting and improves the model's generalization ability by randomly dropping a certain percentage of neuron connections during training. In this example, the Dropout rate is set to 0.3.

[0077] The fully connected layer maps the features processed by the Dropout layer to the predicted target battery RUL. After processing through multiple Bi-LSTM layers and Dropout layers, the output is passed to the fully connected layer. The fully connected layer maps the output features of the Bi-LSTM layer to the predicted target (battery RUL).

[0078] The output layer passes the output of the fully connected layer through an activation function to output the RUL prediction value. The activation function of the output layer usually chooses a linear function to directly output the predicted battery RUL value.

[0079] (3) Model training 1. Loss function selection: The loss calculation module establishes a loss function and calculates the loss function based on the RUL predicted value and the RUL true value to obtain the loss. For regression problems, the loss function used in this embodiment is the mean square error (MSE), and its calculation formula is: Where n is the number of samples, is the true value, and is the predicted value. MSE measures the mean squared error between the predicted value and the true value, which can reflect the prediction accuracy of the model.

[0080] 2. Adam Optimizer Selection: The Adam optimizer is used to minimize the loss function and update the model parameters. Select an appropriate optimization algorithm to update the model parameters to minimize the loss function. This example uses the Adam optimization algorithm, which combines the advantages of Adagrad and Adadelta, with an adaptive learning rate and momentum term, and generally enables faster convergence during training.

[0081] 3. Training Process: The training set data is fed into the constructed Bi-LSTM model. Forward propagation calculations are performed batch by batch according to the set batch size to obtain prediction results. Loss is calculated based on the prediction results and the true labels. Backpropagation is then used to calculate the gradient, and the optimizer is used to update the model parameters. These steps are repeated until the model's performance on the validation set stops improving or the preset number of training rounds is reached.

[0082] (IV) Model evaluation and optimization 1. Root Mean Square Error (RMSE): For regression problems, RMSE is an important metric for evaluating model prediction accuracy. It is the square root of the mean square error (MSE). The smaller the RMSE, the closer the model's predicted value is to the true value, and the higher the prediction accuracy.

[0083] 2. Mean Absolute Error (MAE): MAE is also a commonly used evaluation indicator in regression problems. It calculates the average of the absolute values of the differences between the predicted values and the true values. MAE is insensitive to outliers and can more intuitively reflect the average prediction error of the model.

[0084] 3. Hyperparameter Tuning: Optimize the model's hyperparameters using methods such as grid search, random search, or Bayesian optimization. In this example, hyperparameters include the number of neurons in the Bi-LSTM layer, dropout rate, learning rate, and batch size. The optimal hyperparameter configuration is selected by evaluating the model's performance under different hyperparameter combinations on a validation set.

[0085] 4. Model Structure Adjustment: Based on the model's performance on the validation set and analysis of the data characteristics, consider adjusting the model structure. For example, increasing or decreasing the number of Bi-LSTM layers, changing the number of neurons in the hidden layer, and trying different activation functions can improve the model's performance and generalization capabilities.

[0086] 5. Ensemble learning: Ensemble learning methods are used to combine multiple different Bi-LSTM models or other machine learning models (such as support vector machines and decision trees) to obtain the final prediction results through voting and averaging. Ensemble learning can improve model stability and prediction accuracy.

[0087] (V) Result calculation and evaluation 1. RUL prediction process 1) Input sequence: voltage, current, temperature, SOC of the past 50 cycles.

[0088] 2) Model output: predicted value of remaining life of the current cycle.

[0089] 3) Error calculation: Compare with the actual RUL (calculated using the capacity decay formula).

[0090] 2. Evaluation Metrics MAE (mean absolute error): reflects the absolute size of the prediction deviation.

[0091] RMSE (Root Mean Square Error): Penalizes large errors and focuses more on accuracy.

[0092] R² Score: Measures the explanatory power of the model (the closer to 1, the better).

[0093] Example results (see Table 1) Table 1: Comparison of prediction results evaluation indicators of the traditional LSTM model and the Bi-LSTM model of the present invention index Traditional LSTM Bi-LSTM of the present invention Improvement MAE 4.2 3.1 ↓26.2% RMSE 5.8 4.5 ↓22.4% R² 0.87 0.93 ↑6.9% Table 1 is analyzed as follows: MAE (Mean Absolute Error): Measures the average deviation between the predicted value and the true value. The smaller the value, the higher the accuracy.

[0094] Bi-LSTM reduces MAE from 4.2 to 3.1, indicating that the predicted value is closer to the true remaining life (RUL).

[0095] RMSE (Root Mean Square Error): Comprehensively reflects the size of the prediction error and is sensitive to outliers.

[0096] The RMSE of Bi-LSTM decreased by 22.4%, indicating that the model has a stronger ability to fit fluctuating data.

[0097] R² (coefficient of determination): represents the proportion of the variance of the target variable explained by the model. The closer to 1, the better.

[0098] The R² improves to 0.93, proving that it can better capture the battery degradation trend.

[0099] Improved prediction accuracy: The Bi-LSTM's MAE is reduced by 26.2%, reducing the risk of over-replacement or premature retirement during battery maintenance and saving approximately 15% to 20% in operation and maintenance costs (based on industry average data).

[0100] Enhanced reliability: R² exceeds 0.93, indicating that the model is more robust to complex operating conditions (such as temperature fluctuations and changes in charge and discharge rates).

[0101] Conclusion: The Bi-LSTM model proposed in this paper achieves a leap in accuracy and stability in the task of battery remaining life prediction through a bidirectional information flow design. Compared with traditional LSTM, its core advantages are reflected in: It can provide more accurate trend prediction: effectively capture the asymmetric characteristics of battery degradation, and has stronger anti-interference capabilities: reduce the impact of noise and abnormal data on the results, and provide more reliable decision support for the system.

[0102] RUL prediction part code: import torchimport torch.nn as nn class BatteryRULModel(nn.Module): def __init__(self, input_dim=5, hidden_dim1=128, hidden_dim2=64): super().__init__() self.bi_lstm1 = nn.LSTM(input_dim, hidden_dim1, bidirectional=True, batch_first=True) self.dropout1 = nn.Dropout(0.3) self.bi_lstm2 = nn.LSTM(hidden_dim1*2, hidden_dim2,bidirectional=True, batch_first=True) self.dropout2 = nn.Dropout(0.4) self.fc = nn.Sequential( nn.Linear(hidden_dim2*2, 32), nn.ReLU(), nn.Linear(32, 1) ) def forward(self, x): # x shape: (batch, time_steps, features) h1, _ = self.bi_lstm1(x) h1 = self.dropout1(h1[:, -1, :]) # Take the last time step output h2, _ = self.bi_lstm2(h1.unsqueeze(1)) h2 = self.dropout2(h2[:, 0, :]) return self.fc(h2) # Initialize the model model = BatteryRULModel(input_dim=5) criterion = nn.MSELoss() optimizer = torch.optim.Adam(model.parameters(), lr=0.001, weight_decay=1e-5) (6) Model application 1. Real-time prediction: In actual operating environments, real-time battery operating data is preprocessed and fed into a loaded Bi-LSTM model for prediction, yielding the battery's RUL prediction. Based on these predictions, timely maintenance measures or replacement plans can be implemented to avoid equipment downtime or other losses caused by battery failure.

[0103] 2. Model monitoring and updating: Establish a model monitoring mechanism to regularly evaluate the performance of the model in real-world applications. As time passes and new data accumulates, if the model performance degrades or deviates, retrain and update the model using the latest data to ensure its accuracy and effectiveness.

[0104] In summary, this embodiment uses a Bi-LSTM (bidirectional long short-term memory) network combined with Dropout technology to build a prediction model for battery remaining useful life (RUL) prediction, aiming to improve the accuracy and reliability of the prediction. Traditional methods based on physical models or empirical rules have limited prediction accuracy in complex environments. Compared with traditional models, the present invention uses a Bi-LSTM model for life prediction, which reduces the error rate and can warn of the critical point of battery capacity degradation 3-6 months in advance, leaving a buffer period for replacement decisions and reducing the spare parts scheduling costs caused by sudden failures.

[0105] In this embodiment, the SOH prediction results and RUL prediction results obtained by the above-mentioned SOH prediction unit 400 and RUL prediction unit 500 are respectively displayed visually in the remote monitoring platform / server program.

[0106] In summary, the above embodiments of the present invention can achieve the following technical effects: 1. Realize intelligent control and local visualization of battery pack charging and discharging (remote core capacity).

[0107] 2. Estimate the battery's percentage of health (SOH) and remaining life (RUL).

[0108] 3. Monitor key parameters such as DC bus voltage, battery pack voltage, battery pack charge and discharge current, battery pack charge and discharge status, cell voltage, cell negative pole temperature, cell internal resistance, and provide alarms.

[0109] 4. Build an efficient data management system to facilitate the storage, query, statistics and analysis of battery historical data.

[0110] 5. Provide scientific basis for battery maintenance and replacement.

[0111] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0112] The technical features of the above-described embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above-described embodiments only express several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. For those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A battery remote charge and discharge control and life prediction system, characterized in that: include: Data acquisition unit, used to collect battery operation data in real time; A data monitoring and testing unit, configured to monitor the acquired battery operating data online, and control the charge and discharge of the battery through a charge and discharge control circuit and according to remote control instructions, so as to at least perform a core capacity test of the battery; A remote control unit is used to obtain data uploaded by the data monitoring and testing unit and send remote control instructions to the data monitoring and testing unit; An SOH prediction unit, used to perform SOH prediction on the acquired battery operation data based on a PINN model; The RUL prediction unit is used to perform RUL prediction on the acquired battery operation data based on the Bi-LSTM model.

2. The battery remote charge and discharge control and life prediction system according to claim 1, characterized in that: The battery operation data includes the battery's state of charge, depth of discharge, charge and discharge current, voltage, resistance, temperature, and cycle number.

3. The battery remote charge and discharge control and life prediction system according to claim 1, characterized in that: The data acquisition unit includes: a plurality of single cell monitoring modules, which are installed on each battery cell and monitor and collect the battery cell operation data in real time; a group end collector, whose input end is provided with multiple acquisition channels, which are respectively connected to each single cell monitoring module and used to collect the operation data of each battery cell.

4. The battery remote charge and discharge control and life prediction system according to claim 3, characterized in that: The remote control unit adopts a remote monitoring platform and / or a core-capacity host; both the group-end concentrator and the core-capacity host support local visualization of the machine.

5. The battery remote charge and discharge control and life prediction system according to claim 1, characterized in that: The charge and discharge control circuit of the data monitoring and testing unit includes: The first-stage load and the second-stage load are connected to the first-stage DC bus and the second-stage DC bus respectively; The first rectifier and the second rectifier are used to convert AC power into DC power, and the rectified DC power is connected to the first DC bus and the second DC bus respectively; The bus tie switch is used to connect or disconnect one section of the DC bus with the other section of the DC bus; The first diode, the first normally closed contactor, the second diode and the second normally closed contactor are used to connect the first battery pack and the second battery pack to the first DC bus segment and the second DC bus segment respectively; A first charge and discharge switch and a second charge and discharge switch are connected to the first battery pack and the second battery pack, respectively, for controlling the charging or discharging of the first battery pack and the second battery pack; The bidirectional converter is connected between the first charge and discharge switch and the second charge and discharge switch and is connected to the AC power grid, and is used to realize the bidirectional flow of electric energy between the first battery pack or the second battery pack and the AC power grid.

6. The battery remote charge and discharge control and life prediction system according to claim 1, characterized in that: The data monitoring and testing unit is also used to implement internal resistance testing, spectrum testing and K1 / D1 testing of the battery.

7. The battery remote charge and discharge control and life prediction system according to claim 1, characterized in that: It also includes a data preprocessing unit for identifying the acquired battery operation data and eliminating missing values and abnormal data, and then normalizing the remaining data after elimination.

8. The battery remote charge and discharge control and life prediction system according to claim 1, characterized in that: The PINN model of the SOH prediction unit specifically includes: The Transformer encoder extracts time series features from the input battery operation data including current, voltage, resistance, temperature, and cycle number through multi-head self-attention, feedforward network, residual connection, and layer normalization; Fully connected layer, aggregates time series features and outputs SOH prediction value; The data-driven loss calculation module calculates the error between the predicted SOH value and the actual SOH value to obtain the data-driven loss; The physical constraint loss calculation module calculates the mean square error of the physical equation residual to obtain the physical constraint loss; The total loss function acquisition module sets dynamic weights based on data-driven loss and physical constraint loss, establishes a total loss function, and calculates the total loss; Adam optimizer is used to minimize the total loss function and obtain dynamic weight parameters based on a dynamic weight adjustment strategy of uncertainty or gradient balance, and then adjust and update the dynamic weight parameters.

9. The battery remote charge and discharge control and life prediction system according to claim 1, characterized in that: The Bi-LSTM model of the RUL prediction unit specifically includes: The input layer receives a data set containing battery operating data, determines the number of neurons in the input layer based on the number of feature types in the data set, and passes the battery operating data to the hidden layer in the form of a sequence; The hidden layer includes multiple Bi-LSTM layers, which are used to capture long-term dependencies and time series information in battery operation data. Each Bi-LSTM layer consists of a forward LSTM and a backward LSTM, which can simultaneously consider the forward and backward information of the sequence. Dropout layer, which is added after each Bi-LSTM layer to randomly discard some neuron connections during training; The fully connected layer maps the features processed by the Dropout layer to the predicted target battery RUL; The output layer outputs the RUL prediction value through the activation function of the output of the fully connected layer; The loss calculation module establishes a loss function and calculates the loss function based on the RUL predicted value and the RUL true value to obtain the loss; Adam optimizer is used to minimize the loss function and update the parameters of the model.

10. A method for remote battery charge and discharge control and life prediction, applied to the system according to any one of claims 1 to 9, characterized in that: include: Data acquisition: real-time collection of battery operation data; Data monitoring and testing: online monitoring of acquired battery operating data, and control of battery charge and discharge through a charge and discharge control circuit and according to remote control instructions, so as to at least achieve battery core capacity testing; Remote control, obtaining data uploaded by the data monitoring and testing unit, and sending remote control instructions to the data monitoring and testing unit; SOH prediction: SOH prediction is performed on the acquired battery operation data based on the PINN model; RUL prediction: Based on the Bi-LSTM model, RUL prediction is performed on the acquired battery operation data.

Images