A Collaborative Optimization Method for Individualized Battery Safety Fast Charging Based on Neural Network Models
By collecting multi-source data in real time at the charging pile and using a neural network model to predict and control the individual state of the battery, the problem of thermal runaway and accelerated aging caused by differences in individual battery aging characteristics is solved, and collaborative optimization control for safe and fast charging is achieved.
Patent Information
- Application Number
- CN202610981970.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-02
- Publication Date
- 2026-11-10
Smart Images

Figure CN123074014A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery management and intelligent charging control technology, specifically to a collaborative optimization method for individualized safe fast charging of batteries based on a neural network model. Background Technology
[0002] With the large-scale application of electric heavy-duty trucks and electric commercial vehicles in logistics, ports, mining areas, and other scenarios, the demand for rapid energy replenishment is becoming increasingly urgent. Existing fast charging technologies generally adopt a fixed charging strategy, which is executed by looking up the charging strategy table through the BMS. The charging strategy is usually obtained by the OEM or battery pack manufacturer through laboratory testing and calibration. Since laboratory testing and calibration can hardly cover the actual situation of the battery throughout its entire life cycle under different operating conditions, the full life cycle charging strategy can only be calibrated through typical operating conditions, such as 1C charging and 1C discharging cycle accelerated aging tests.
[0003] Patent CN121123458A constructs a matrix table through multiple tests and dynamically cycles through charge and discharge to achieve a fast charging strategy where the charging current I of a single lithium-ion battery cell changes in real time with SOC and temperature T. The specific method is as follows: SOC-OCV testing is performed on the lithium-ion battery to obtain a SOC-OCV-T matrix table; charging tests are conducted on the lithium-ion battery at different temperatures using three electrodes to form an I-SOC-U-Tmax matrix table at different temperatures; the I-SOC-U-Tmax matrix table is integrated to form a 0-100% SOC charging matrix table; the discharge end voltage and SOC state are found based on the SOC-OCV-T matrix table, and charging is performed in conjunction with the SOC state and the 0-100% SOC charging matrix table; the charging strategy, which uses cyclic charge and discharge, obtains the charging current I of a single lithium-ion battery cell that changes in real time with SOC and temperature T.
[0004] The lithium-ion battery charging strategy determination method described in patent CN119542586A is as follows: Based on the rated capacity of the lithium-ion battery to be charged, the target SOC range, and the target charging time, the average charging rate C' is determined, and then the charging strategy is determined. The minimum and maximum charging rates are Cmin and Cmax, respectively; Cmin < C' < Cmax; the current charging time to complete the target SOC range under the current charging strategy is determined; in the current charging strategy, n stepped charging stages are set, and the charging rate C” continuously decreases along the charging sequence; Cmin ≤ C” ≤ Cmax and at least one C” > C'; n ≥ 2; if the current charging time is less than or equal to the target charging time, the current charging strategy is the target charging strategy; if it is greater than the target charging time, correction is performed to form a new current charging strategy and the previous step is returned.
[0005] The method for determining the appropriate charging rate and segmented charging strategy for lithium-ion batteries described in patent CN119230993A is as follows: A breathing apparatus is provided: The breathing apparatus includes a battery body and a transparent tube. The transparent tube is inserted into the battery body from the outside, allowing the electrolyte level to be observed within the transparent tube. Charging test: Within a defined SOC charging range, the battery body is charged at any given rate. After charging, the electrolyte level h1 is observed immediately after charging and the electrolyte level h2 is observed after standing for 10-30 minutes. If the difference between h1 and h2 is within the range of 0-10 mm, then the rate is considered an appropriate charging rate within the defined SOC charging range; otherwise, it is considered an inappropriate charging rate.
[0006] This type of strategy does not fully consider the individualized aging characteristics of batteries due to differences in manufacturing tolerances, usage environment, and charge / discharge history, leading to the following problems: 1. Accumulated risk of thermal runaway: During high-rate fast charging, the internal resistance of the battery increases with the degree of aging, which intensifies heat generation. If the charging strategy of new batteries or general aging states is still used, it is easy to cause local overheating and excessive temperature rise rate. In particular, the risk of lithium dendrite growth increases significantly in low-temperature environments, which may lead to thermal runaway.
[0007] 2. Accelerated battery aging: Universal charging curves cannot adapt to the changes in electrochemical characteristics of individual batteries at different aging stages (such as capacity decay to 95%, 90%, 85%, etc.). For example, aged batteries are more sensitive to charging current in the high SOC range. Continuously using high current charging will exacerbate the thickening of the SEI film and the loss of active lithium, forming a vicious cycle of "fast charging → accelerated aging → increased internal resistance → more prone to overheating".
[0008] 3. Lack of battery identity management in battery swapping mode: In heavy-duty truck battery swapping scenarios, battery packs frequently circulate between different vehicles, and their historical usage data (such as cumulative throughput and deep cycle count) are difficult to be effectively identified and utilized by charging facilities. This results in each charging being performed according to the "most conservative" or "general" strategy, sacrificing energy replenishment efficiency or creating safety risks.
[0009] In existing technologies, some solutions attempt to adjust charging strategies based on battery health estimates from the Battery Management System (BMS). However, their adjustment logic is mostly based on simplified empirical formulas (such as linear reduction of charging rate at State of Health), lacking a precise characterization of individual battery micro-aging mechanisms (such as changes in lithium plating threshold and degradation of thermal diffusion characteristics). Other studies have proposed charging optimization based on electrochemical models, but model parameter identification is complex and computationally intensive, making real-time deployment in automotive embedded systems difficult. Furthermore, a closed-loop optimization architecture from cloud-based big data modeling to vehicle-side execution has not been established.
[0010] Therefore, there is an urgent need for a safe fast charging control method that can integrate multi-source data throughout the battery's entire life cycle, achieve individualized charging strategy adjustments, and implement the optimization results through a vehicle-charging-cloud collaborative mechanism, so as to balance charging speed, battery life, and operational safety. Summary of the Invention
[0011] To achieve the above-mentioned objectives, this invention provides the following technical solution: a battery-based individualized safe fast charging collaborative optimization method based on a neural network model, comprising the following steps: Step S1: Using the charging pile as a data interaction gateway, real-time battery operation data is collected during vehicle charging, including battery pack current, voltage, state of charge, highest single cell voltage, and temperature. Historical data is mapped in conjunction with the battery's unique identifier. At the same time, the output current at the charging pile and the request current from the battery management system (BMS) are introduced to align and fuse the timestamps of the multi-source time-series data, forming a unified time-series data tensor. Step S2: Input the time series data tensor obtained in step S1 into the cloud neural network model and output the battery state prediction vector for the next moment, including the predicted voltage, predicted temperature and predicted SOC. Step S3: Based on the battery state prediction vector, construct the battery individualized multivariate charging response function and generate a multidimensional control vector to jointly constrain and control the current and safety adjustment factor during the charging process. Step S4: Under the constraints of the multivariable charging response function, perform online collaborative optimization to solve the multidimensional control vector and generate a multidimensional charging control trajectory function; Step S5: The charging pile performs dynamic charging control based on the multi-dimensional charging control trajectory function, and corrects the control vector in real time through a closed-loop feedback mechanism to achieve coordinated optimization control for safe and fast charging of the battery.
[0012] Preferably, step S1 further includes: First, a CAN bus-based data interaction link is established between the charging pile-side embedded communication module and the vehicle battery management system (BMS). During the charging start-up and operation process, battery operating status parameters are continuously acquired at a fixed sampling period to form the original acquisition vector. In the process of forming the original acquisition vector, an active exploratory current regulation mechanism is introduced. Specifically, after the vehicle is connected to the charging pile, the cloud control platform generates an exploratory control command based on a preset strategy and sends it to the execution unit on the charging pile side. This causes the output current of the charging pile to be biased in a controlled manner relative to the current requested by the battery management system. Under different bias currents, the corresponding battery voltage, temperature and state of charge response data are collected synchronously and incorporated into the original acquisition vector for unified modeling. Subsequently, the original acquisition vectors are processed to unify the timestamps. By constructing a time alignment function, data from different sampling delay sources are dynamically corrected so that data from different sources are aligned under a unified time reference, forming a time-series data tensor. After time alignment is completed, the aligned real-time running data is bound to the battery's unique identifier, and the static embedding vector and historical charging behavior features of the corresponding battery are extracted through the cloud indexing system to characterize the battery's inherent properties and long-term aging state.
[0013] Preferably, step S2 further includes: First, the time-series data tensor generated in step S1 is input into a one-dimensional convolutional neural network structure. The convolutional kernel performs local scanning in the time dimension to extract the current mutation, voltage response and polarization feature changes of the battery in a short time scale, forming a local time-series feature representation. Subsequently, the aforementioned local features are input into the gated recurrent unit (GRU) network. Through its internal update gate and reset gate structure, the SOC evolution, thermal accumulation effect and slow voltage drift during the long-term charge and discharge process of the battery are modeled, enabling the model to have long-term dependent memory capability. Simultaneously, the static embedding vector extracted in step S1 and the historical charging behavior features are input into the multilayer perceptron coding network to form a stable battery identity representation vector. Finally, the dynamic time-series features and static health representations are concatenated and fused, and then input into the fully connected prediction layer to output a battery state prediction vector. The battery state prediction vector contains the joint prediction results of battery voltage and temperature state, which are used for subsequent safe charging strategy generation.
[0014] Preferably, step S3 further includes: Based on the battery state prediction results output by the neural network, the battery's operating capability during the current charging process is modeled individually, forming a response function to describe the change law of the battery's rechargeability under different states of charge, temperature and time conditions. Based on the response function, a multidimensional control vector is constructed for the charging process. The multidimensional control vector includes the charging current control quantity, the electrochemical inhibition adjustment factor, and the thermal safety adjustment coefficient, which are used to jointly regulate the charging process. Among them, the battery operating status is constrained and calculated through the cloud to ensure that the charging current is always within the upper limit of battery safety, and the thermal safety adjustment coefficient is dynamically adjusted according to real-time temperature changes. Finally, a multivariable safety control space is constructed based on the above multidimensional control vectors to limit the executable range of all control parameters in the subsequent optimization process, thereby ensuring that the charging process operates under electrochemical and thermal safety constraints.
[0015] Preferably, step S4 further includes: Under the constraints of multidimensional control vectors, a time-continuous optimization mechanism is established for the battery charging process. By comprehensively evaluating the combined effects of battery risk state, temperature rise change and adjustment factors, an optimization objective function is constructed to characterize the overall cost of the charging process. The objective function is used to simultaneously measure the level of safety risk, the degree of temperature change, and the degree of control deviation during the charging process. Under the premise of satisfying the upper limit constraint of battery state of charge, temperature safety constraint and control variable safety constraint, the multidimensional control vector is continuously optimized in the time dimension to obtain the optimal control evolution path in the entire charging cycle. The optimal control evolution path is used to describe the joint change trend of current control and safety adjustment factor with time during the charging process, thereby forming a cooperative control strategy for the entire charging cycle.
[0016] Preferably, step S5 further includes: During the charging execution phase, the charging pile obtains the reference control vector corresponding to the current moment from the optimal multidimensional charging control trajectory function sent from the cloud based on the time series data tensor collected in real time. The reference control vector includes the charging current control command to be executed at the moment and the corresponding electrochemical adjustment and thermal safety adjustment commands. In actual execution, the charging pile decomposes and executes the multi-dimensional control vector based on the reference control information. Among them, the charging current control quantity is used to directly control the energy input, the electrochemical inhibition adjustment factor is used to control the reaction inhibition intensity during the charging process, and the thermal safety adjustment coefficient is used to adjust the rate of temperature rise. Meanwhile, the multi-dimensional control vector being executed is corrected in real time through a closed-loop feedback mechanism, so that the actual charging process gradually approaches the optimal control path output by the cloud. When an abnormal battery operating state is detected, a safety degradation mechanism is triggered. If the electrochemical risk state exceeds the threshold, a safety degradation mechanism is triggered to reduce charging risks. The final output is the actual charging current control quantity, and the charging pile power execution module converts the control vector into a PWM signal to realize the real-time closed-loop control execution process of vehicle-pile-cloud three-terminal collaboration.
[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention collects battery operation data in real time from the charging pile and combines it with vehicle BMS request information and historical battery charging behavior characteristics to achieve time alignment and unified fusion input of multi-source heterogeneous data, thereby constructing a time-series state data structure with the ability to express individual differences. Compared with traditional modeling methods based only on single current or voltage data, this invention can more comprehensively characterize the dynamic changes of different batteries at different aging stages and under different operating conditions, improving the accuracy and stability of battery state characterization.
[0018] This invention outputs the upper limit of the battery's safe current, risk state parameters, and temperature rise prediction parameters through a cloud-based neural network. It further constructs a multivariate control vector to achieve the synergistic effect of current control, electrochemical suppression factors, and thermal safety regulation factors. This expands the charging process from a single current regulation to a joint control mode under multi-physics constraints. This mechanism can effectively suppress the risks of lithium plating and thermal runaway while ensuring charging efficiency, thereby improving the safety margin of the battery charging process.
[0019] This invention constructs a closed-loop control system that combines cloud-optimized charging control trajectory generation with real-time execution feedback at the charging pile. Through a real-time correction mechanism of multi-variable control vectors, the actual charging process continuously approximates the optimal control trajectory in the cloud. Compared to traditional static charging strategies or lookup table control methods, this invention possesses adaptive adjustment capabilities, dynamically correcting charging behavior based on real-time state changes, thereby achieving synergistic optimization between charging efficiency and battery life. Attached Figure Description
[0020] Figure 1 A flowchart illustrating the method steps provided in this application; Figure 2 A flowchart illustrating the process steps provided for this application. Detailed Implementation
[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0022] refer to Figures 1-2 This invention provides a battery-individualized safe fast charging collaborative optimization method based on a neural network model, comprising the following steps: Step S1: Using the charging pile as a data exchange gateway, collect battery operation data in real time during vehicle charging, including battery pack current. ,Voltage State of charge Highest single-unit voltage ,temperature Combined with the battery's unique identifier Historical data mapping is performed, and the pile end output current is introduced simultaneously. Request current from the battery management system (BMS) Timestamp alignment and fusion of multi-source time series data are performed to form a unified time series data tensor. ; Step S2: Convert the time series data tensor obtained in step S1 into a series tensor. Input a cloud-based neural network model and output a battery state prediction vector for the next time step. This includes predicting voltage, predicting temperature, and predicting SOC; Step S3: Based on the battery state prediction vector, construct the individualized multivariate charging response function for the battery. And generate multidimensional control vectors. The charging process involves joint constraint control of the current and safety adjustment factor. Step S4: Under the constraints of the multivariable charging response function, adjust the multidimensional control vector. Online collaborative optimization is performed to generate a multi-dimensional charging control trajectory function. ; Step S5: The charging pile is based on the multi-dimensional charging control trajectory function. Dynamic charging control is implemented, and the control vector is corrected in real time through a closed-loop feedback mechanism to achieve coordinated optimization control for safe and fast charging of batteries.
[0023] In step S1, firstly, a CAN bus-based data interaction link is established between the charging pile-side embedded communication module and the vehicle's battery management system (BMS). During charging startup and operation, battery operating status parameters are continuously acquired at a fixed sampling period to form an original acquisition vector. : ; in, Indicates battery pack current, Indicates voltage, Indicates the state of charge, Indicates the highest single-unit voltage, Indicates temperature. Indicates the output current at the pile end. This indicates a current request from the vehicle's battery management system (BMS). Subsequently, the original acquisition vector Timestamp unification is performed by constructing a time alignment function to dynamically correct data from different sampling delay sources, ensuring that data from different sources are aligned under a unified time reference. ; in, Indicates the time decay coefficient. This indicates the actual time point at which the data was collected. The time reference point is generated by the charging pile's main control clock to ensure data consistency across devices. After completing time alignment, the aligned real-time operating data will be compared with the battery's unique identifier. Binding is performed, and the static embedding vector of the corresponding battery is extracted through the cloud indexing system. and historical charging behavior characteristics It is used to characterize the inherent properties and long-term aging state of batteries; Finally, real-time data, static features, and historical features are uniformly spliced and fused to form a time-series data tensor. : .
[0024] In step S2, firstly, the time series data tensor generated in step S1 is... The input is fed into a one-dimensional convolutional neural network structure, which performs local scanning in the time dimension through convolutional kernels to extract the battery's current mutations, voltage responses, and polarization feature changes within a short time scale, forming a local temporal feature representation. : ; Convolution operations are used to capture transient physical changes during the charging process. This represents the convolution weight matrix in a one-dimensional convolutional neural network. Indicates the convolutional layer bias term; Subsequently, the aforementioned local features are input into a gated recurrent unit (GRU) network. Through its internal update and reset gate structures, the system can model the state of charge (SOC) evolution, thermal accumulation effect, and slow voltage drift during long-term charge and discharge of the battery, enabling the model to have long-term dependent memory capabilities. The state update process is as follows: ; ; ; ; in, This represents the update gate vector, used to control the proportion of historical information retained. This represents the reset gate vector, used to control the degree of involvement of historical states. Indicates the candidate hidden state. This represents the weight matrix input to each gate structure. Indicates hidden state Weight matrices and hidden states of each structure It is used to characterize the overall dynamic health status of the battery at the current charging moment; At the same time, the static embedding vector extracted in step S1 Characteristics of historical charging behavior Inputting the data into a multilayer perceptron coding network forms a stable battery identity representation vector. : ; Used to characterize the inherent differences and degradation levels between different individual batteries; Finally, the dynamic temporal features and static health representations are concatenated and fused, and then input into a fully connected prediction layer to output a battery state prediction vector. : ; in, Indicates the output layer bias term. Also includes battery voltage State of charge (SOC) and temperature state The joint prediction results are used to generate subsequent safe charging strategies.
[0025] In step S3, based on the battery state prediction vector output by the neural network, an individualized response model is performed on the current operating state of the battery in the cloud to construct an individualized charging response function for the battery. The charging response function is used to continuously characterize the charging capability of the battery under different states of charge, temperature and time evolution conditions, thereby forming a dynamic charging boundary mapping relationship of the battery. In the charging response function Based on this, a multi-dimensional control vector is further constructed. It is used to achieve multi-dimensional coordinated control during the charging process, and multi-dimensional control vector. The formula is as follows: ; in: This indicates the charging current control quantity, used to represent the actual energy input intensity; This represents the electrochemical inhibition regulation factor, used to dynamically adjust the degree of inhibition of lithium plating risk inside the battery; This represents the thermal safety adjustment coefficient, which is used to adjust the battery's thermal diffusion response and temperature rise control intensity. In the specific implementation process, the cloud-based current upper limit parameter is based on the neural network output. Boundary constraints are applied to the current control quantity, and coupled control is achieved through the following relationship: ; At the same time, based on the real-time temperature status By mapping the predicted temperature rise parameters, the thermal safety adjustment coefficient is obtained: ; in: Indicates the temperature response sensitivity coefficient; Indicates the thermal control reference threshold; Furthermore, a multivariate safety control space is constructed based on the above-mentioned control variable constraints. This multivariate safety control space is used to define the physical boundaries and safety constraints of all feasible control paths in the subsequent optimization process. .
[0026] In step S4, in the multidimensional control vector Under the constraints, a time-continuous optimization model is established for the battery charging process to achieve a synergistic balance between charging safety and charging efficiency. First, a system-level comprehensive optimization objective function is constructed: ; in: This represents the comprehensive cost function of fast charging throughout the battery's entire lifecycle, used to measure the safety and loss level during the charging process; Represents the weights, a set of coefficients used to adjust the weights of different physical mechanisms; This represents the electrochemical risk state output by the neural network. ; This indicates the real-time temperature rise of the battery. During the optimization process, multi-level constraints are introduced to limit the control variables, including: (1) Control variable constraints, ; (2) SOC constraint, ; (3) Temperature safety constraints, ; Based on satisfying the above constraints, the multidimensional control vector By performing continuous optimization in the time domain, the optimal charging control trajectory function is obtained: ; The multidimensional charging control trajectory function This represents the joint evolution path of the current control quantity, electrochemical regulation factor, and thermal regulation factor throughout the entire charging cycle. Its essence is a multi-variable collaborative dynamic control strategy, rather than a single current optimization result, thereby achieving system-level collaborative optimization of the charging process.
[0027] In step S5, step S5 further includes: During the charging execution phase, the charging pile uses the currently collected real-time time-series data tensor. The optimal multidimensional charging control trajectory function is sent from the cloud. Obtain the reference control vector corresponding to the current time step: ; in: ; In actual execution, the charging pile control unit decomposes and executes the multivariable control vector. Among them, the current control part is used to directly adjust the power output, the electrochemical suppression adjustment factor is used to adjust the current change slope, and the thermal safety adjustment coefficient is used to adjust the temperature rise response rate. Then, a closed-loop feedback update mechanism is constructed: ; in: This represents the current actual control vector being executed; This represents the convergence coefficient of the control system, used to adjust the control response speed; Represents the target control vector; During operation, when the system detects an abnormal operating state, the electrochemical risk state... Exceeding the threshold This will trigger the security downgrade mechanism: ; The final output is the actual charging current control quantity. The power execution module of the charging pile converts the control vector into a PWM signal, realizing a real-time closed-loop control execution process that coordinates the vehicle, the charging pile, and the cloud.
[0028] It should be noted that, unless otherwise specified, the embodiments and features and technical solutions in the present invention can be combined with each other.
[0029] Obviously, the embodiments described above are merely some embodiments of the present invention, not all embodiments. The accompanying drawings show preferred embodiments of the present invention, but do not limit the patent scope of the present invention. The present invention can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the patent protection scope of this invention.
Claims
1. A battery-individualized safe fast charging collaborative optimization method based on a neural network model, characterized in that, Includes the following steps: Step S1: Using the charging pile as a data interaction gateway, real-time battery operation data is collected during vehicle charging, including battery pack current, voltage, state of charge, highest single cell voltage, and temperature. Historical data is mapped in conjunction with the battery's unique identifier. At the same time, the output current at the charging pile and the request current from the battery management system (BMS) are introduced to align and fuse the timestamps of the multi-source time-series data, forming a unified time-series data tensor. Step S2: Input the time series data tensor obtained in step S1 into the cloud neural network model and output the battery state prediction vector for the next moment, including the predicted voltage, predicted temperature and predicted SOC. Step S3: Based on the battery state prediction vector, construct the battery individualized multivariate charging response function and generate a multidimensional control vector to jointly constrain and control the current and safety adjustment factor during the charging process. Step S4: Under the constraints of the multivariable charging response function, perform online collaborative optimization to solve the multidimensional control vector and generate a multidimensional charging control trajectory function; Step S5: The charging pile performs dynamic charging control based on the multi-dimensional charging control trajectory function, and corrects the control vector in real time through a closed-loop feedback mechanism to achieve coordinated optimization control for safe and fast charging of the battery.
2. The battery individualized safe fast charging collaborative optimization method based on a neural network model according to claim 1, characterized in that, Step S1 further includes: First, a CAN bus-based data interaction link is established between the charging pile-side embedded communication module and the vehicle battery management system (BMS). During the charging start-up and operation process, battery operating status parameters are continuously acquired at a fixed sampling period to form the original acquisition vector. In the process of forming the original acquisition vector, an active exploratory current regulation mechanism is introduced. Specifically, after the vehicle is connected to the charging pile, the cloud control platform generates an exploration control command based on a preset strategy and sends it to the execution unit on the charging pile side. This causes the output current of the charging pile to be biased relative to the current requested by the battery management system. Under different bias currents, the corresponding battery voltage, temperature and state of charge response data are collected synchronously and incorporated into the original acquisition vector for unified modeling. Subsequently, the original acquisition vectors are processed to unify the timestamps. By constructing a time alignment function, data from different sampling delay sources are dynamically corrected so that data from different sources are aligned under a unified time reference, forming a time series data tensor. After time alignment is completed, the aligned real-time running data is bound to the battery's unique identifier, and the static embedding vector and historical charging behavior features of the corresponding battery are extracted through the cloud indexing system to characterize the battery's inherent properties and long-term aging state.
3. The battery individualized safe fast charging collaborative optimization method based on a neural network model according to claim 1, characterized in that, Step S2 further includes: First, the time-series data tensor generated in step S1 is input into a one-dimensional convolutional neural network structure. The convolutional kernel performs local scanning in the time dimension to extract the current mutation, voltage response and polarization feature changes of the battery in a short time scale, forming a local time-series feature representation. Subsequently, the aforementioned local features are input into the gated recurrent unit (GRU) network. Through its internal update and reset gate structures, the SOC evolution, thermal accumulation effect, and slow voltage drift during the long-term charge and discharge process of the battery are modeled, enabling the model to have long-term dependent memory capabilities. Simultaneously, the static embedding vector extracted in step S1 and the historical charging behavior features are input into the multilayer perceptron coding network to form a stable battery identity representation vector. Finally, the dynamic time-series features and static health representations are concatenated and fused, and then input into the fully connected prediction layer to output a battery state prediction vector. The battery state prediction vector contains the joint prediction results of battery voltage and temperature state, which are used for subsequent safe charging strategy generation.
4. The battery individualized safe fast charging collaborative optimization method based on a neural network model according to claim 1, characterized in that, Step S3 further includes: Based on the battery state prediction results output by the neural network, the battery's operating capability during the current charging process is modeled individually, forming a response function to describe the change law of the battery's rechargeability under different states of charge, temperature and time conditions. Based on the response function, a multidimensional control vector is constructed for the charging process. The multidimensional control vector includes the charging current control quantity, the electrochemical inhibition adjustment factor, and the thermal safety adjustment coefficient, which are used to jointly regulate the charging process. Among them, the battery operating status is constrained and calculated through the cloud to ensure that the charging current is always within the upper limit of battery safety, and the thermal safety adjustment coefficient is dynamically adjusted according to real-time temperature changes. Finally, a multivariable safety control space is constructed based on the above multidimensional control vectors to limit the executable range of all control parameters in the subsequent optimization process, thereby ensuring that the charging process operates under electrochemical and thermal safety constraints.
5. The battery individualized safe fast charging collaborative optimization method based on a neural network model according to claim 1, characterized in that, Step S4 further includes: Under the constraints of multidimensional control vectors, a time-continuous optimization mechanism is established for the battery charging process. By comprehensively evaluating the combined effects of battery risk state, temperature rise change and adjustment factors, an optimization objective function is constructed to characterize the overall cost of the charging process. The objective function is used to simultaneously measure the level of safety risk, the degree of temperature change, and the degree of control deviation during the charging process. Under the premise of satisfying the upper limit constraint of battery state of charge, temperature safety constraint and control variable safety constraint, the multidimensional control vector is continuously optimized in the time dimension to obtain the optimal control evolution path in the entire charging cycle. The optimal control evolution path is used to describe the joint change trend of current control and safety adjustment factor with time during the charging process, thereby forming a cooperative control strategy for the entire charging cycle.
6. The battery individualized safe fast charging collaborative optimization method based on a neural network model according to claim 1, characterized in that, Step S5 further includes: During the charging execution phase, the charging pile obtains the reference control vector corresponding to the current moment from the optimal multidimensional charging control trajectory function sent from the cloud based on the time series data tensor collected in real time. The reference control vector includes the charging current control command to be executed at the moment and the corresponding electrochemical adjustment and thermal safety adjustment commands. In actual execution, the charging pile decomposes and executes the multi-dimensional control vector based on the reference control information. Among them, the charging current control quantity is used to directly control the energy input, the electrochemical inhibition adjustment factor is used to control the reaction inhibition intensity during the charging process, and the thermal safety adjustment coefficient is used to adjust the rate of temperature rise. Meanwhile, the multi-dimensional control vector being executed is corrected in real time through a closed-loop feedback mechanism, so that the actual charging process gradually approaches the optimal control path output by the cloud. When an abnormal battery operating state is detected, a safety degradation mechanism is triggered. If the electrochemical risk state exceeds the threshold, a safety degradation mechanism is triggered to reduce charging risks. The final output is the actual charging current control quantity, and the charging pile power execution module converts the control vector into a PWM signal to realize the real-time closed-loop control execution process of vehicle-pile-cloud three-terminal collaboration.
Citation Information
Patent Citations
Method for determining proper charging rate and segmented charging strategy of lithium ion battery
CN119230993A
Lithium ion battery charging strategy determination method and battery charging method
CN119542586A
Lithium ion battery charging strategy optimization method
CN121123458A
