Battery life and safety collaborative prediction method and system based on improved pulse neural network algorithm

Abstract

The invention relates to a battery life and safety collaborative prediction method and system based on an improved pulse neural network algorithm, and belongs to the technical field of electric tool battery management. The method is based on a four-level SNN architecture, an input layer receives voltage, current, temperature, stress and change rate parameters of a battery, and neural pulse conversion of the parameters is realized through rate coding and time coding; the synaptic layer simulates a physical hysteresis effect between parameters by using a double-index model, and the hidden layer passes through Adaptive Iamp; the F neurons integrate historical memory and nonlinear coupling, and the output layer generates health status, residual life and risk index. The synaptic weight is optimized through a supervised STDP learning rule, the contribution degree of each parameter is quantified, a membrane potential linear accumulation mechanism of the life dimension and an abnormal pulse triggering mechanism of the safety dimension are established, and a joint early warning decision is realized. The system is deployed in a battery management system, the battery state can be monitored in real time, and accurate decision support is provided for battery full life cycle management.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of battery management of electric tools, and relates to a battery life and safety collaborative prediction method and system based on an improved pulse neural network algorithm, in particular to a multi-parameter fusion electric wrench battery life and safety prediction model based on voltage, current, temperature, stress and their change rates. BACKGROUND

[0002] With the acceleration of industrial automation and the wide application of Internet of Things technology, as a core tool in modern assembly work, the performance of the lithium battery of the electric wrench directly affects the work efficiency and safety. Precise prediction of the lithium battery life and safety of the electric wrench, realization of "knowing before being damaged and preventing from being damaged after being damaged", become the key to guarantee stable operation of equipment and safety in production. However, the traditional lithium battery state evaluation method based on a single parameter, such as relying only on voltage and current monitoring, lacks comprehensive consideration of the complex physical and chemical changes inside the lithium battery, has problems such as early warning lag, misjudgment rate and the like, and is difficult to meet the stringent requirements of industrial scenes on lithium battery reliability.

[0003] In recent years, researchers have introduced machine learning, sensor network, big data analysis and other frontier technologies into the field of lithium battery state prediction, and realized lithium battery health state evaluation through the collection of multi-dimensional data. However, existing researches mostly focus on the analysis of a single or a small number of parameters, such as predicting the life only according to the lithium battery voltage curve or the number of charge and discharge, without fully considering the coupling relationship between parameters such as temperature, internal resistance and swelling stress. For example, although the prior art with publication number CN117907872A determines the initial battery life according to the battery voltage, battery current, battery temperature and cycle number, the results obtained by the algorithm model used are not accurate, and the internal stress change of the lithium battery is not perceived, so it is difficult to issue a safety warning in time in the early stage of the internal micro-short circuit of the lithium battery. In addition, although some prediction models fuse multiple parameters, the data processing algorithm is simple, and the potential association between parameters cannot be deeply mined, so the model generalization ability is insufficient, and the prediction accuracy under complex working conditions is difficult to meet the actual application requirements SUMMARY

[0004] Therefore, the purpose of the present application is to provide a battery life and safety collaborative prediction method and system based on an improved pulse neural network algorithm, which realizes precise collaborative prediction of the battery health state, remaining life and safety risk by constructing a multi-level pulse neural network architecture and combining innovative encoding methods, neuron models and learning rules, and provides a comprehensive state evaluation scheme for the battery management system.

[0005] To achieve the above purpose, the present application provides the following technical solutions:

[0006] A battery life and safety collaborative prediction method based on an improved pulse neural network algorithm, the method comprising the following steps:

[0007] At the input layer, the basic physical parameters of the battery and their corresponding rate parameters are received, and the basic physical parameters and their rate parameters are encoded;

[0008] At the synapse layer, a double exponential synapse current model is used to simulate the physical hysteresis effect between parameters;

[0009] Through the neuron model of the hidden layer, historical memory and nonlinear coupling are integrated to quantify the dynamic response and long-term damage in the battery degradation process;

[0010] At the output layer, the state of health and remaining life of the battery are calculated, and a risk index is calculated;

[0011] A supervised STDP learning rule is used to optimize the synaptic weights to complete the training, and the parameter contribution is quantified;

[0012] Based on the life index and safety index obtained by calculation, a life and safety collaborative prediction mechanism is used for decision-making.

[0013] Further, the input layer receives eight-dimensional parameters of the battery, which are voltage, current, temperature, stress, voltage rate, current rate, temperature rate, and stress rate, wherein voltage, current, temperature, and stress are steady-state parameters, and voltage rate, current rate, temperature rate, and stress rate are transient parameters;

[0014] For steady-state parameters, speed coding is used to map steady-state parameters to pulse frequency:

[0015]

[0016] Where x i represents the steady-state physical parameters of the battery, and the value range is the actual measured value; the hyperbolic tangent function tanh maps the input value to the interval ((-1, 1)); when the parameter exceeds the safety range, tanh(x i / x max ) tends to 1; within the normal range of parameters x i <<x max , tanh(×) is approximately linear; k is a proportional coefficient, and x max is the parameter safety threshold;

[0017] For transient impact parameters, time coding is used to map transient impact parameters to pulse time to capture extreme value mutations:

[0018] t k = argmax(|x i(t)|)

[0019] where t k Indicates the time when the kth pulse is generated; x i (t) represents the measured value of parameter i at time t; argmax(|x i (t)|) represents the function |x i (t) |The time point when the maximum value is reached.

[0020] Furthermore, the process of simulating the physical hysteresis effect between parameters in the double exponential synaptic current model is:

[0021]

[0022] in, is a Dirac function, only at the kth pulse moment of the jth synapse The value is 1 at the moment and 0 at other moments, which is used to mark parameter mutation events; Indicates that the pulse is After the moment is triggered, the synaptic current responds over time; w ij Represents the synaptic weight, which quantifies the influence of the jth parameter on the i-th neuron; Σ j Σ k Represents the superposition of synaptic currents for multiple mutation events k of multiple parameters j.

[0023] Furthermore, the process of integrating historical memory and nonlinear coupling in the Adaptive I&F neuron model is expressed as:

[0024]

[0025] Among them, τ m Represents the membrane potential time constant, which determines the decay rate of the degradation state; V m Represents the natural attenuation term of membrane potential, simulating the slow aging of the battery when there is no abnormality; R∑ i I syn,i is the synaptic input integration term, R is the coupling resistance, and quantifies the impact of parameter anomalies on degradation. η(t) represents the random noise term, simulating the uncertainty of the battery system; a·∑ t'＜t θ(tt′) is the historical memory accumulation term, where a is the memory strength coefficient, which quantifies the long-term impact of a single anomaly. θ(tt′) represents the unit step function, which takes the value of 1 when t>t￠ and 0 otherwise, accumulating the number of historical anomalies.

[0026] Furthermore, the output layer simultaneously generates the battery life indicators health status SOH, remaining life RUL and safety index to achieve multi-dimensional status assessment; among them, the battery life includes health status SOH and remaining life RUL, and the safety index refers to the risk index RI;

[0027] For the state of health (SOH), the pulse frequency weighted average is mapped to the capacity retention rate, and the calculation formula is:

[0028]

[0029] Among them, f j It represents the pulse frequency corresponding to the j-th type of feature. The higher the frequency, the more serious the degradation. represents the degradation-related weight, which characterizes the contribution of different features to system degradation; α is the mapping coefficient, which is used to adjust the proportional relationship between the frequency-weighted average and the decline in health status;

[0030] For the remaining lifetime RUL: based on the membrane potential accumulation extrapolation, the formula is:

[0031]

[0032] in, Indicates the membrane potential threshold when the system fails; V m (t now ) represents the membrane potential at the current moment; k deg Represents the degradation rate, which is determined by the weighted contribution of each parameter;

[0033] For the risk index RI: combining abnormal pulse frequency and membrane potential mutation, the formula is:

[0034]

[0035] Among them, N anomaly is the number of abnormal pulses, N total is the total number of pulses, and the ratio of the two represents the probability of abnormality; represents the normal membrane potential threshold, It reflects the degree to which the current potential deviates from the normal; β1 and β2 are weight coefficients used to adjust the relative importance of the two types of risk factors.

[0036] Furthermore, during the supervised STDP learning rule training process, the training data is input into the improved spiking neural network algorithm to calculate the predicted output y pred ;

[0037] According to the prediction error y target -y pred Calculate the synaptic weight. The synaptic weight update formula is:

[0038] Δω ij =η·(y target -y pred )·s i (t)·h j (t)

[0039] wherein y target -y pred represents the prediction error driven weight update; s i (t) represents the impulse of the input layer parameter i; h j (t) represents the response of the hidden layer neuron j; η is the learning rate, which controls the step size of each update;

[0040] During the training process, the contribution degree C i of each parameter is calculated according to the formula

[0041]

[0042] wherein C i is the contribution degree of parameter i, which represents the relative contribution percentage of parameter i to the overall degradation; ∑ j |ω ij | represents the sum of the absolute values of the weights connected by parameter i and all hidden layer neurons; ∑ i ∑ j |ω ij | represents the sum of the absolute values of the weights connected by all parameters and all neurons;

[0043] By analyzing the contribution degree, parameters with a contribution degree that does not conform to the actual physical law or has an impact on the prediction result less than a preset degree are removed to optimize the model.

[0044] Further, the four-level pulse neural network after training is used for real-time data prediction to calculate the real-time battery life RUL and safety index RI, and the joint decision logic is used to realize the collaborative prediction of life and safety:

[0045]

[0046] In this embodiment, δ RUL represents the life safety threshold, and δ RI is the safety index threshold.

[0047] On the other hand, a battery life and safety collaborative prediction system based on the improved pulse neural network algorithm is also proposed, which comprises a data acquisition module, an encoding module, an improved pulse neural network module, a decision module and a storage module,

[0048] The data acquisition module is used to acquire the basic physical parameters and the rate parameters of the battery.

[0049] The encoding module is used for rate encoding and time encoding processing of the parameters.

[0050] The improved pulse neural network module comprises an input layer, a synapse layer, a hidden layer and an output layer, and realizes the battery life and safety collaborative prediction method based on the improved pulse neural network algorithm as described above.

[0051] The decision module generates a life warning or a safety warning according to the prediction result.

[0052] The storage module is used for storing model parameters and historical data.

[0053] In another aspect, a computer readable storage medium is also provided, and the storage medium stores a computer program, and when the computer program is executed by a processor, the battery life and safety collaborative prediction method based on the improved pulse neural network algorithm as described above is realized.

[0054] A computer program product is also provided, comprising a computer program, and when the computer program is executed by a processor, the battery life and safety collaborative prediction method based on the improved pulse neural network algorithm as described above is realized.

[0055] The beneficial effects of the present application are as follows:

[0056] (1) The present application deeply fuses the physical mechanism of the battery by the double exponential synapse model and the Adaptive I&F neuron model, embeds the parameter hysteresis effect, historical damage accumulation and other characteristics into the network architecture, so that the model can accurately capture the physical influence of temperature sudden change, overcurrent impact and other transient processes on the battery; at the same time, a multi-dimensional collaborative prediction system of life and safety is constructed, and life indexes such as SOH and RUL and RI risk index are output synchronously, the joint evaluation of long-term degradation trend and transient failure is realized through a three-level warning mechanism, and the misjudgment limitation of a single index is avoided.

[0057] (2) The present application quantifies the influence weight of each physical parameter on the battery state through a parameter contribution quantification algorithm, which can directly locate the main cause of failure such as high temperature and overcharge, and provide data support for battery design optimization and operation; based on the dynamic weight updating mechanism of the STDP learning rule, the prediction model can continuously optimize in different use scenarios, maintain a prediction error within 5% without manual retraining, and significantly improve the real-time performance and robustness in engineering applications.

[0058] Other advantages, objects, and features of the present application will be apparent to those skilled in the art from the following specification, and will be learned from the study of the following text, or will be taught from the practice of the present application. The objects and other advantages of the present application can be achieved and obtained by the following specification. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to make the purposes, technical solutions and advantages of the present application clearer, the preferred detailed description of the present application will be given below with reference to the drawings, in which:

[0060] Figure 1 The whole flowchart of the battery life and safety collaborative prediction method based on the improved pulse neural network algorithm under the embodiment of the present application is shown in the figure.

[0061] Figure 2 The processing flowchart of the double exponential synaptic current model under the embodiment of the present application is shown in the figure.

[0062] Figure 3 The supervised STDP learning rule training process of the improved pulse neural network under the embodiment of the present application is shown in the figure.

[0063] Figure 4 The whole architecture of the battery life and safety collaborative prediction system based on the improved pulse neural network algorithm under another embodiment of the present application is shown in the figure.

[0064] Figure 5 The specific deployment mode of the battery life and safety collaborative prediction system based on the improved pulse neural network algorithm under another embodiment of the present application is shown in the figure.

[0065] Figure 6 The battery SOH prediction comparison under another embodiment of the present application is shown in the figure.

[0066] Figure 7 The battery RUL prediction comparison under another embodiment of the present application is shown in the figure.

[0067] Figure 8 The battery RI prediction comparison under another embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0068] The embodiments of the present application are described below through specific and concrete examples, and those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in the specification. The present application can also be implemented or applied through other different specific embodiments, and each detail in the specification can be modified or changed based on different views and applications without departing from the spirit of the present application. It should be noted that the figures provided in the following embodiments only illustrate the basic concept of the present application in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0069] In the drawings, only for example, the representation is a schematic diagram, not a physical diagram, and cannot be understood as a limitation of the present application; in order to better illustrate the embodiments of the present application, some components of the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0070] The same or similar reference numerals in the drawings of the embodiments of the present application correspond to the same or similar components; in the description of the present application, it should be understood that if the terms "upper", "lower", "left", "right", "front", "back" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, only for the convenience of describing the present application and simplifying the description, and not to indicate or imply that the device or element referred to must have a particular orientation, structure and operation, therefore the positional relationship described in the drawings is only for example, and cannot be understood as a limitation of the present application, for those skilled in the art, the specific meaning of the above terms can be understood according to the specific situation.

[0071] Please refer to Figures 1-8 , a battery life and safety collaborative prediction method and system based on an improved pulse neural network algorithm.

[0072] Embodiment 1

[0073] This embodiment first provides a detailed process of a battery life and safety collaborative prediction method based on an improved pulse neural network algorithm, as shown in Figure 1 , which performs the following steps based on the constructed four-level pulse neural network architecture:

[0074] S1, receiving the basic physical parameters of the battery and the corresponding change rate parameters in the input layer, and encoding the basic physical parameters and the change rate parameters;

[0075] S2, using a double exponential synaptic current model to simulate the physical hysteresis effect between parameters in the synaptic layer;

[0076] S3, integrating historical memory and nonlinear coupling through the neuron model of the hidden layer, quantifying the dynamic response and long-term damage in the battery degradation process;

[0077] S4, calculating the health state and remaining life of the battery in the output layer, and calculating the risk index;

[0078] S5, using supervised STDP learning rules to optimize synaptic weights to complete training, and quantifying the contribution of parameters;

[0079] S6, based on the life index and safety index obtained by calculation, making decisions according to the collaborative prediction mechanism of life and safety.

[0080] In step S1 of the embodiment, the input layer receives eight-dimensional parameters of the battery, which are basic physical parameters of the battery, such as voltage (V), current (I), temperature (T), stress (σ), and their rate parameters, specifically, the rate parameters include voltage rate, current rate, temperature rate, and stress rate parameters. Then, for the input eight-dimensional parameters, the steady-state parameters are encoded by using rate coding, and the transient impact parameters are encoded by using time coding. The rate coding hyperbolic tangent function tanh maps the parameters to the interval (-1, 1), suppresses the unbounded increase of pulse frequency when the parameters exceed the safety threshold, and ensures the effective encoding of small amplitude changes. The time coding marks the extreme value mutation event by the pulse time, and accurately captures the instantaneous influence of abnormal transient on the battery. In the above eight-dimensional parameters, the steady-state parameters include: voltage, current, temperature, stress; the transient impact parameters include: voltage rate, current rate, temperature rate, and stress rate.

[0081] More specifically, in the embodiment, in the rate coding process, the steady-state parameters are mapped to pulse frequency, which is suitable for slowly changing features:

[0082]

[0083] where x i represents the steady-state physical parameters of the battery, and the value range is the actual measured value; the hyperbolic tangent function tanh maps the input value to the interval ((-1, 1)). When the parameter exceeds the safety range (for example, x i >x max ), tanh(x i / x max ) tends to 1, avoiding unbounded increase of pulse frequency; within the normal range of the parameter x i <<x max , tanh(x) is approximately linear, ensuring that small amplitude changes can be effectively encoded; k is the proportional coefficient, and x max is the parameter safety threshold (for example, V max = 4.5V).

[0084] In the time coding process, the transient impact parameters are mapped to pulse time, and the extreme value mutation is captured:

[0085] t k = argmax(|x i (t)|)

[0086] where t k represents the time (unit: second) when the kth pulse is generated; x i (t) represents the measured value of the parameter i at time t; argmax(|x i (t)|) represents the function |xi (t)| time point when the maximum is reached.

[0087] In step S2 of the present embodiment, a double exponential synapse current model is adopted, parameter mutation events (such as current overshoot) are marked by Dirac functions, physical hysteresis effects between parameters are simulated, and the influence of transient impact on the battery is captured. As shown in Figure 2 , in the double exponential synapse current model, the processing process includes:

[0088] (1) Input parameter pulse sequence:

[0089] The system receives the pulse sequence data input externally, which contains the core parameters such as the time stamp of the pulse and the frequency, as the original input for subsequent calculation.

[0090] (2) Synaptic weight matrix loading:

[0091] The synaptic weight matrix is read from the storage unit (or pre-trained model), which describes the strength (weight value) of the connection between neurons, and is a key parameter for simulating synaptic transmission characteristics.

[0092] (3) Pulse time point recognition:

[0093] The input pulse sequence is parsed to extract the occurrence time (time stamp) of each pulse, and the positioning of the pulse on the time axis is determined, providing a time reference for subsequent modeling of the pulse signal.

[0094] (4) Dirac delta function pulse generation:

[0095] Based on the recognized pulse time point, the Dirac delta function is used to simulate the “instantaneous” characteristics of the pulse to generate an idealized pulse signal.

[0096] (5) Rising edge exponential calculation & decay edge exponential calculation:

[0097] Rising edge exponential calculation: Calculate the time constant τ rise of the rising phase of the pulse, which describes the rate of pulse amplitude rising over time.

[0098] Decay edge exponential calculation: Calculate the time constant τ decay of the decay phase of the pulse, which describes the rate of pulse amplitude decay over time.

[0099] (6) Rising edge function and decay edge function

[0100] Rising edge function: According to τ rise , an exponential rising model is constructed to simulate the amplitude variation law of the rising phase of the pulse.

[0101] Attenuation edge function: According to τ decay An exponential decay model is constructed to simulate the amplitude variation law during the pulse decay stage.

[0102] (7) Double exponential difference operation:

[0103] The rise-edge function and the decay-edge function are differentially calculated to simulate the "rise-decay" dynamic characteristics of the synaptic current changing with time.

[0104] (8) Weighted summation of synaptic weights:

[0105] The double exponential difference result of step (7) is multiplied element by element with the synaptic weight matrix of step (2), and then the results of all synaptic connections are summed to reflect the difference in contribution of different synaptic connections to the current.

[0106] (9) Synaptic current accumulation calculation:

[0107] The weighted summation results in step (8) are accumulated in the time dimension to calculate the total synaptic current received by a single neuron (integrating the contribution of all input pulses).

[0108] (10) Output to hidden layer neurons:

[0109] The total synaptic current calculated in step 9 is used as the input signal and transmitted to the hidden layer neurons of the neural network to drive the calculation of the next layer.

[0110] Based on the above process, the process of simulating the physical hysteresis effect between parameters of the double exponential synaptic current model is expressed as:

[0111]

[0112] in, is a Dirac function, only at the kth pulse moment of the jth synapse The value is 1 at the moment and 0 at other moments, which is used to mark parameter mutation events (such as current overshoot moments); Indicates that the pulse is After the moment is triggered, the synaptic current responds over time; w ij Represents the synaptic weight, which quantifies the influence of the jth parameter on the i-th neuron; Σ j Σ k Represents the superposition of synaptic currents for multiple mutation events k of multiple parameters j.

[0113] In step S3 of the embodiment, the Adaptive I&F neuron model is introduced to integrate historical memory and nonlinear coupling, and to quantify the dynamic response and long-term damage in the battery degradation process through the membrane potential time constant, synaptic input integration term, random noise term and historical memory accumulation term. The process of integrating historical memory and nonlinear coupling in the Adaptive I&F neuron model is represented as:

[0114]

[0115] wherein τ m represents the membrane potential time constant, determining the decay rate of the degradation state; V m represents the membrane potential natural decay term, simulating the slow aging of the battery in the absence of abnormalities (such as self-discharge during normal storage); R∑ i I syn,i is the synaptic input integration term, R is the coupling resistance, quantifying the influence strength of parameter abnormalities on degradation. η(t) represents the random noise term, simulating the uncertainty of the battery system, which is usually set as Gaussian white noise: η(t) ~ N(0, σ 2 ); a·∑ t'＜t θ(t-t′) is the historical memory accumulation term, wherein a is the memory strength coefficient, quantifying the long-term influence of a single abnormality, and θ(t-t′) represents the unit step function, which takes the value of 1 when t>t￠, and 0 otherwise, accumulating the number of historical abnormalities.

[0116] In step S4 of the embodiment, the output layer synchronously generates the battery life indicators (state of health SOH, remaining useful life RUL) and safety indicators (risk index RI), realizing multi-dimensional state evaluation.

[0117] For the state of health SOH: the capacity retention rate is mapped through pulse frequency weighted average, and the calculation formula is:

[0118]

[0119] wherein f j represents the pulse frequency corresponding to the jth feature, and the higher the frequency, the more likely it represents more severe degradation; represents the degradation-related weight, representing the contribution of different features to system degradation (such as different weights of parameters such as temperature and current); α is the mapping coefficient, used to adjust the proportional relationship between the frequency weighted average and the state of health decline.

[0120] For the remaining useful life RUL: based on the membrane potential accumulation extrapolation, the formula is:

[0121]

[0122] wherein, represents the membrane potential threshold at system failure; Vm (t now ) represents the current membrane potential; k deg represents the degradation rate, which is determined by the contribution of each parameter.

[0123] For the risk index RI: combining the abnormal pulse frequency and the membrane potential mutation, the formula is:

[0124]

[0125] Where N anomaly is the number of abnormal pulses, N total is the total number of pulses, and the ratio of the two represents the probability of abnormality; represents the normal membrane potential threshold, reflects the degree of deviation of the current potential from the normal, and the denominator is normalized to the proportion within the failure range; β1, β2 are weight coefficients, used to adjust the relative importance of the two types of risk factors.

[0126] In step S5 of the embodiment, the supervised STDP learning rule integrates error correction and time dependence, as shown in Figure 3 Prepare the data first, then input the training data into the improved pulse neural network algorithm during the training process, and calculate the predicted output y pred . According to the prediction error y target -y pred , the synaptic weight is calculated, and the synaptic weight update formula is:

[0127] Δω ij = η·(y target -y pred )·s i (t)·h j (t)

[0128] Where y target -y pred represents the prediction error driving weight update; s i (t) represents the pulse of input layer parameter i; h j (t) represents the response of hidden layer neuron j; η is the learning rate, which controls the step size of each update.

[0129] During the training process, the contribution C i of each parameter is calculated according to the formula:

[0130]

[0131] Where C i is the contribution of parameter i, which represents the relative contribution percentage of parameter i to the overall degradation; ∑ j |ω ij| represents the sum of the absolute values ​​of the weights connecting parameter i (such as temperature T) and all hidden layer neurons; ∑ i ∑ j |ω ij | represents the sum of the absolute values ​​of the weights connecting all parameters to all neurons;

[0132] Analyze the contribution of parameters. If it is found that the contribution of certain parameters is inconsistent with the actual physical laws or has little impact on the prediction results, the model can be optimized.

[0133] In step S6 of this embodiment, the trained four-layer spiking neural network is used for real-time data prediction to calculate the real-time battery life RUL and safety index RI. A joint decision logic is used to achieve a coordinated prediction of life and safety:

[0134]

[0135] In this embodiment, δ RUL represents the life safety threshold, δ RI is the safety index threshold; in this embodiment, the life safety threshold δ RUL It can be taken as 80%, and the safety index threshold δ RI You can take 70%.

[0136] Example 2

[0137] This embodiment establishes a system for executing the method in embodiment 1, namely a battery life and safety collaborative prediction system based on an improved pulse neural network algorithm, such as Figure 4 As shown, it includes a data acquisition module, an encoding module, an improved pulse neural network module, a decision module and a storage module.

[0138] Data acquisition module, used to obtain basic physical parameters and change rate parameters of the battery;

[0139] An encoding module, configured to perform rate encoding and time encoding processing on the parameters;

[0140] An improved spiking neural network module, comprising an input layer, a synaptic layer, a hidden layer, and an output layer, implements the aforementioned battery life and safety collaborative prediction method based on the improved spiking neural network algorithm;

[0141] Decision module, which generates life warning or safety warning based on the prediction results;

[0142] Storage module, used to store model parameters and historical data.

[0143] The system is deployed in the battery management system (BMS) to achieve real-time monitoring and prediction of battery status.

[0144] In this embodiment, from hardware building, data processing, model training and optimization, prediction execution and system deployment, etc. to realize the accurate prediction and effective management of the battery state.

[0145] As shown in Figure 5 , first, hardware building is to be carried out, and the data acquisition module is responsible for obtaining the basic physical parameters and the rate parameters of the battery. High-precision sensors are used to realize parameter acquisition: voltage parameters are collected by voltage Hall sensors, with a measurement range of 0-1000V and an accuracy of ±0.1%; current parameters are collected by current Hall sensors, with a range of ±25A and an accuracy of ±0.5%; temperature parameters are collected by multiple NTC thermistors, which are arranged at key positions such as the battery cell and the tab to obtain temperature information at different positions, with an accuracy of ±0.3℃; stress parameters are collected by micro pressure sensors, which are pasted on the battery shell and can sense the thermal stress changes of the battery caused by the charging and discharging process, with an accuracy that meets the engineering requirements. At the same time, in order to calculate the rate of change of parameters, a high-speed data acquisition card is used, which has a sampling frequency of up to 50kHz and can quickly collect data and calculate the voltage rate of change, current rate of change, temperature rate of change and stress rate of change through the built-in algorithm. The acquisition card is connected to the host control device through the interface to ensure the stability and high speed of data transmission. The encoding module and the improved pulse neural network algorithm module hardware are realized based on field programmable gate array (FPGA) and digital signal processor (DSP). The FPGA integrated with ARM processor and programmable logic resources can realize high-speed parallel computing. The hardware logic circuit is built on the FPGA to complete the rate encoding and time encoding processing, and the collected analog signals are converted into pulse signals.

[0146] For the calculation of the pulse neural network, the parallel processing capability of the FPGA can accelerate the operations such as synaptic current calculation and neuron state update. At the same time, with the chip, the powerful digital signal processing capability is used to perform complex mathematical operations, such as solving differential equations in the Adaptive I&F neuron model and weight update calculation in the supervised STDP learning rule. The FPGA and the DSP interact with each other through a high-speed serial interface to realize efficient collaborative work. The decision module uses an ARM series processor to generate life warning or safety warning according to the prediction results. The processor has a rich set of peripheral interfaces and can communicate with the battery management system (BMS) and other devices to timely transmit warning information. The storage module uses a large-capacity non-volatile storage device to store model parameters, historical data and other information, ensuring that data is not lost after power failure, and supporting fast data read and write operations to meet the data access needs of model training and real-time prediction.

[0147] Next, the collected basic physical parameters and rate parameters are encoded. For steady-state parameters such as the average value of voltage and current, a rate coding method is used. Taking voltage as an example, the voltage value V is mapped to the pulse frequency f V , and the mapping formula is f V =k V *V, where k V is a proportional coefficient adjusted according to the voltage measurement range and the pulse frequency range, ensuring that the pulse frequency is within a reasonable range when the voltage is within the normal working range, facilitating subsequent neural network processing. For transient impact parameters such as parameter changes during overcurrent and overvoltage, a time coding method is used. When the current exceeds the set overcurrent threshold I th , the time t over of overcurrent occurrence is taken as time coding information, and the neuron generates a pulse output at that time, thereby accurately capturing transient events.

[0148] Before encoding, the collected data needs to be preprocessed. First, filter processing is performed to remove noise and improve data accuracy using the Kalman filter algorithm. Taking current data as an example, the Kalman filter performs optimal estimation on the true value of the current by establishing state equations and observation equations, effectively suppressing the interference of measurement noise. Then, normalization processing is performed to map the values of each parameter to the [0, 1] interval, eliminating the influence of different parameter dimensions and making the data comparable. For voltage parameter V, the normalization formula is V norm =(V-V min ) / (V max -V min ), where V min and V max are the minimum and maximum values of the voltage, respectively.

[0149] Next, a four-layer improved pulse neural network architecture is constructed, including an input layer, a synaptic layer, a hidden layer, and an output layer. The number of input layer neurons is determined according to the number of collected parameters, with 8 neurons corresponding to voltage (V), current (I), temperature (T), stress (σ), and their rate of change (dV / dt, dI / dt, dT / dt, dσ / dt). The synaptic layer uses a double exponential synaptic current model, and the initial rise time constant τ rise and decay time constant τ decay , and the initial value of synaptic weight w ij are set according to experience or preliminary experiments, which can be set to small random numbers such as within the interval [-0.1, 0.1]. The hidden layer uses the Adaptive I&F neuron model, and the membrane time constant τ m, film resistance R, memory strength factor a, and other parameters. The number of output layer neurons is determined according to the generated indicators, such as 2 neurons corresponding to the remaining useful life (RUL) and risk index (RI) outputs.

[0150] Then the processed data is labeled, and the target output y is determined according to the actual life of the battery, the health state evaluation result, etc. target , such as the true value of the health state (SOH), the actual remaining time of the remaining useful life, etc. The data is divided into training set, validation set and test set, usually in the ratio of 7:1:2, to ensure effective training and evaluation of the model on different data sets.

[0151] Next, the supervised STDP learning rule is used to optimize the synaptic weight. In the training process, the training data is input into the improved pulse neural network algorithm to calculate the predicted output y pred . According to the prediction error y target -y pred , the synaptic weight w ij is updated according to the synaptic weight update formula Δω target = η·(y pred -y i )·s j (t)·h ij (t), where the learning rate η is initially set to 0.01 and dynamically adjusted according to the training situation. After each training batch is completed, the loss function value on the validation set, such as mean square error (MSE), is calculated to evaluate the prediction error of SOH and RUL, and cross-entropy loss is used to evaluate the prediction error of RI. When the loss function value on the validation set no longer decreases or the decrease is very small, the model is considered to have reached convergence and training is stopped. During training, the contribution C i of each parameter is calculated according to the formula . Analyze the parameter contribution, if some parameters do not match the actual physical law or have little effect on the prediction result, the model can be optimized. For example, adjust the model architecture to reduce dependence on parameters with low contribution, or reset related parameters, such as adjusting the time constant of the double exponential synaptic current model to better simulate the physical lag effect between parameters.

[0152] Finally, based on the life indicators and safety indicators obtained from the calculation output, the decision is made according to the collaborative prediction mechanism of life and safety. In the life dimension, the aging degree of the battery is evaluated through the linear accumulation model of membrane potential . In the safety dimension, when an abnormal parameter is detected, such as voltage exceeding the safety threshold, the corresponding synaptic weight is suddenly increased to make the membrane potential jump nonlinearly, quickly reflecting potential risks.

[0153] The decision module makes judgments according to the joint decision logic:

[0154]

[0155] The system is equipped with an efficient storage module designed specifically for storing model parameters and historical data, ensuring data integrity and traceability. Model parameters record learning outcomes and optimization status during system operation, while historical data covers key indicators during battery use, providing a rich information foundation for subsequent analysis and prediction.

[0156] Deployed within a battery management system (BMS), the system enables real-time monitoring and accurate prediction of battery status. Through continuous data collection and analysis, the system can promptly detect subtle changes in battery performance and provide early warning of potential failures, effectively extending battery life and improving the stability and safety of overall system operations.

[0157] In this embodiment, the prediction method of the present invention is compared with the traditional prediction method, and the following results are obtained: Figure 6-8 Schematic diagram of the comparison of prediction results, where Figure 6 This is a schematic diagram of battery SOH prediction comparison; Figure 7 This is a comparison diagram of battery RUL prediction; Figure 8 Schematic diagram for battery RI prediction comparison.

[0158] exist Figure 6 In the figure, the blue solid line represents the real health state of the battery, which shows a gradually decreasing trend over time (hours), reflecting the gradual decline in battery performance during actual use. The red dotted line represents the battery health state predicted by the present invention. It can be seen that the prediction curve of the present invention is very close to the true value curve, and the predicted values ​​at each time point can well follow the changes in the true value, and the prediction accuracy is high. The green dotted line represents the battery health state predicted by the traditional method. There is a relatively obvious deviation between the prediction curve of the traditional method and the true value curve, especially in the later period, the predicted value decreases relatively slowly, and the gap with the true value gradually increases.

[0159] exist Figure 7 In the figure, the blue solid line represents the actual remaining service life of the battery. As time goes by, the remaining service life gradually decreases, showing a linear downward trend. The red dotted line is the remaining service life of the battery predicted by the present invention. The prediction curve of the present invention almost coincides with the true value curve, and the predicted values ​​at each time point are very close to the true values. The green dotted line represents the remaining service life of the battery predicted by the traditional method. There is a certain deviation between the prediction curve of the traditional method and the true value curve, and the predicted value decreases relatively slowly, resulting in the predicted remaining service life being larger than the true value.

[0160] exist Figure 8In the figure, the black solid line represents the real risk index of the battery, which is at a relatively low and stable level as a whole, but some fluctuations occur in the later period. The red dashed line represents the predicted battery risk index of the present application. The prediction curve of the present application can better follow the changes of the real value, especially in the fluctuation part, can timely respond to the changes of the real value, the predicted value is close to the real value, which shows that the present application can accurately predict the changes of the battery risk index. The blue dashed line represents the predicted battery risk index of the traditional method. The prediction curve of the traditional method deviates greatly from the real value curve, the predicted value is significantly higher than the real value, and it cannot accurately capture the fluctuations of the real value, which shows that the traditional method is not good at predicting the battery risk index.

[0161] From Figure 6 , Figure 7 and Figure 8 , it can be seen that the present application has significant advantages in predicting the battery health state, remaining useful life and risk index. Compared with the traditional method, the present application can more accurately follow the changes of the real value, has higher prediction accuracy, can provide more reliable basis for the management and maintenance of the battery, helps to find potential problems of the battery in advance, takes timely measures, and improves the use efficiency and safety of the battery.

[0162] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the purpose and scope of the technical solutions, which should be covered in the scope of the claims of the present application.

Claims

1. A battery life and safety collaborative prediction method based on an improved spiking neural network algorithm, characterized by: The method comprises the following steps: The input layer receives the basic physical parameters of the battery and their corresponding rate of change parameters, and encodes the basic physical parameters and their rate of change parameters; A double exponential synaptic current model is used at the synaptic layer to simulate the physical hysteresis effect between parameters; By integrating historical memory and nonlinear coupling through a hidden layer neuron model, the dynamic response and long-term damage of the battery degradation process are quantified; The output layer calculates the battery's health status and remaining life, and calculates the risk index; The supervised STDP learning rule is used to optimize the salient weights to complete the training and quantify the parameter contribution; Based on the life and safety indicators obtained from the calculation output, decisions are made according to the collaborative prediction mechanism of life and safety.

2. The method for collaboratively predicting battery life and safety based on an improved pulse neural network algorithm according to claim 1, characterized in that: The input layer receives eight parameters of the battery, namely the battery voltage, current, temperature, stress, voltage change rate, current change rate, temperature change rate, and stress change rate. Among them, voltage, current, temperature, and stress are steady-state parameters, while voltage change rate, current change rate, temperature change rate, and stress change rate are transient parameters. For steady-state parameters, velocity encoding is used to map the steady-state parameters to pulse frequency: Among them, x i Represents the steady-state physical parameters of the battery, and the value range is the actual measured value; the hyperbolic tangent function tanh maps the input value to the ((-1,1)) interval; when the parameter exceeds the safe range, tanh(x i / x max ) approaches 1; within the normal range of parameters x i < <x max , tanh(×) is approximately linear; k is the proportional coefficient, x max is the parameter safety threshold; For transient impact parameters, time coding is used to map the transient impact parameters to pulse moments to capture extreme value mutations: t k =argmax(|x i (t)|) where t k Indicates the time when the kth pulse is generated; x i (t) represents the measured value of parameter i at time t; argmax(|x i (t)|) represents the function |x i (t) |The time point when the maximum value is reached.

3. The method for collaboratively predicting battery life and safety based on an improved pulse neural network algorithm according to claim 2, characterized in that: The process of simulating the physical hysteresis effect between parameters in the double exponential synaptic current model is: in, is a Dirac function, only at the kth pulse moment of the jth synapse The value is 1 at the moment and 0 at other moments, which is used to mark parameter mutation events; Indicates that the pulse is After the moment is triggered, the synaptic current responds over time; w ij Represents the synaptic weight, which quantifies the influence of the jth parameter on the i-th neuron; Σ j Σ k Represents the superposition of synaptic currents for multiple mutation events k of multiple parameters j.

4. The method for collaboratively predicting battery life and safety based on an improved pulse neural network algorithm according to claim 3, characterized in that: The process of integrating historical memory and nonlinear coupling in the Adaptive I&F neuron model is expressed as: Among them, τ m Represents the membrane potential time constant, which determines the decay rate of the degradation state; V m Represents the natural attenuation term of membrane potential, simulating the slow aging of the battery when there is no abnormality; R∑ i I syn,i is the synaptic input integration term, R is the coupling resistance, and quantifies the impact of parameter anomalies on degradation. η(t) represents the random noise term, simulating the uncertainty of the battery system; a·∑ t'＜t θ(tt′) is the historical memory accumulation term, where a is the memory strength coefficient, which quantifies the long-term impact of a single anomaly. θ(tt′) represents the unit step function, which takes the value of 1 when t>t￠ and 0 otherwise, accumulating the number of historical anomalies.

5. The method for collaboratively predicting battery life and safety based on an improved pulse neural network algorithm according to claim 4, characterized in that: The output layer simultaneously generates battery life indicators such as the state of health (SOH), remaining life (RUL), and safety indicators to achieve multi-dimensional status assessment. Battery life includes the state of health (SOH) and remaining life (RUL), while the safety indicator refers to the risk index (RI). For the state of health (SOH), the pulse frequency weighted average is mapped to the capacity retention rate, and the calculation formula is: Among them, f j It represents the pulse frequency corresponding to the j-th type of feature. The higher the frequency, the more serious the degradation. represents the degradation-related weight, which characterizes the contribution of different features to system degradation; α is the mapping coefficient, which is used to adjust the proportional relationship between the frequency-weighted average and the decline in health status; For the remaining lifetime RUL: based on the membrane potential accumulation extrapolation, the formula is: in, Indicates the membrane potential threshold when the system fails; V m (t now ) represents the membrane potential at the current moment; k deg Represents the degradation rate, which is determined by the weighted contribution of each parameter; For the risk index RI: combining abnormal pulse frequency and membrane potential mutation, the formula is: Among them, N anomaly is the number of abnormal pulses, N total is the total number of pulses, and the ratio of the two represents the probability of abnormality; represents the normal membrane potential threshold, It reflects the degree to which the current potential deviates from the normal; β1 and β2 are weight coefficients used to adjust the relative importance of the two types of risk factors.

6. The method for collaboratively predicting battery life and safety based on an improved spiking neural network algorithm according to claim 5, characterized in that: During the supervised STDP learning rule training process, the training data is input into the improved spiking neural network algorithm to calculate the predicted output y pred ; According to the prediction error y target -y pred Calculate the synaptic weight. The synaptic weight update formula is: Give ij =η·(y target -y pred )·s i (t)·h j (t) Among them, y target -y pred Indicates that the prediction error drives the weight update; s i (t) represents the pulse of input layer parameter i; h j (t) represents the response of hidden layer neuron j; η is the learning rate, which controls the step size of each update; During the training process, the contribution C of each parameter is calculated according to the formula i : Among them, C i is the contribution of parameter i, which represents the relative contribution percentage of parameter i to the overall degradation; ∑ j |ω ij | represents the sum of the absolute values ​​of the weights connecting parameter i and all hidden layer neurons; ∑ i ∑ j |ω ij | represents the sum of the absolute values ​​of the weights connecting all parameters to all neurons; Analyze the contribution of parameters and eliminate parameters whose contribution does not conform to the actual physical laws or whose impact on the prediction results is less than the preset level to optimize the model.

7. The method for collaboratively predicting battery life and safety based on an improved pulse neural network algorithm according to claim 6, characterized in that: The trained four-layer spiking neural network is used for real-time data prediction to calculate the real-time battery life RUL and safety index RI. Joint decision-making logic is used to achieve coordinated prediction of life and safety: In this embodiment, δ RUL represents the life safety threshold, δ RI is the safety indicator threshold.

8. A battery life and safety collaborative prediction system based on an improved pulse neural network algorithm, characterized by: The system includes: a data acquisition module, an encoding module, an improved pulse neural network module, a decision module and a storage module. Data acquisition module, used to obtain basic physical parameters and change rate parameters of the battery; An encoding module, configured to perform rate encoding and time encoding processing on the parameters; An improved pulse neural network module, comprising an input layer, a synaptic layer, a hidden layer, and an output layer, for implementing the battery life and safety collaborative prediction method based on the improved pulse neural network algorithm as described in any one of claims 1 to 7; Decision module, which generates life warning or safety warning based on the prediction results; Storage module, used to store model parameters and historical data.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by the processor, the battery life and safety collaborative prediction method based on the improved pulse neural network algorithm as described in any one of claims 1 to 7 is implemented.

10. A computer program product, characterized in that: It includes a computer program, which, when executed by a processor, implements the battery life and safety collaborative prediction method based on the improved pulse neural network algorithm as described in any one of claims 1 to 7.

Images