Electric bicycle battery health state evaluation system based on data fusion

Abstract

The present application relates to the technical field of battery health state evaluation, in particular to an electric bicycle battery health state evaluation system based on data fusion, comprising a data acquisition module, a data alignment module, a feature extraction module and a fusion evaluation module. The data acquisition module acquires dynamic operation data and static characteristic data of the battery. The data alignment module aligns the two types of data according to timestamps to generate a synchronous space-time correlation data sequence. The feature extraction module extracts the capacity attenuation rate, internal resistance growth trend and voltage platform change characteristics from the sequence. The fusion evaluation module performs multi-source evidence fusion analysis on the above features and outputs a battery health state comprehensive score and a remaining service life prediction value. The scheme effectively improves the accuracy of state evaluation and the reliability of life prediction through accurate data fusion and multi-dimensional feature collaborative analysis.

Description

Technical Field

[0001] This invention relates to the field of battery health status assessment technology, and in particular to a data fusion-based battery health status assessment system for electric bicycles. Background Technology

[0002] Existing technologies for assessing the health status of electric bicycle batteries mostly rely on the analysis of limited static data from battery management systems or isolated dynamic data streams from onboard terminals. Conventional solutions typically process these two types of data separately or only perform coarse time-period matching, resulting in a lack of precise temporal correlation between dynamic operating parameters and the inherent static properties of the battery. This data-level fragmentation leads to temporal and state misalignment in the input information received by the model, making it difficult to accurately reflect the real-time degradation behavior of the battery under specific inherent characteristics, thus limiting the accuracy of feature extraction and the reliability of model assessment.

[0003] In terms of evaluation methods, existing technologies generally focus on single-dimensional health indicators, such as relying mainly on capacity decay curves or internal resistance change trends. This evaluation model based on a single source of evidence cannot comprehensively reflect the complex state of multiple physicochemical processes intertwined during battery aging. Because different characteristic parameters have different aging rates and sensitivities, relying on a single parameter is easily affected by measurement noise or random fluctuations, leading to significant misjudgments of health status and large deviations in remaining life prediction, resulting in insufficient robustness and comprehensiveness of the evaluation results.

[0004] A technical solution is needed to fundamentally solve the problem of accurate spatiotemporal synchronization of multi-source heterogeneous data, and to build a comprehensive assessment model that can integrate multi-dimensional degradation evidence to improve the overall accuracy of battery health status assessment and the robustness of prediction. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a data fusion-based electric bicycle battery health status assessment system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an electric bicycle battery health status assessment system based on data fusion, comprising:

[0007] The data acquisition module obtains dynamic operating data of the electric bicycle battery through the on-board terminal and static characteristic data of the battery through the battery management system.

[0008] The data alignment module creates a data fusion processing queue, aligns the dynamic operation data with the battery static characteristic data according to timestamps, and generates a synchronous spatiotemporal correlation data sequence. The synchronous spatiotemporal correlation data sequence includes the voltage value, cycle number, temperature value, current value, nominal capacity, standard internal resistance and material type corresponding to each set of timestamps.

[0009] The feature extraction module calls a preset health status assessment model and inputs the synchronous spatiotemporal correlation data sequence into the feature extraction layer of the health status assessment model. The feature extraction layer extracts a set of key parameters characterizing battery degradation features. The set of key parameters includes capacity decay rate, actual internal resistance growth trend and voltage plateau curve change characteristics.

[0010] The fusion assessment module, through the fusion assessment layer of the health status assessment model, performs multi-source evidence fusion analysis on the key parameter set to generate a comprehensive battery health status score and a predicted remaining service life, including:

[0011] An initial confidence weight is assigned to each parameter in the set of key parameters, and the initial confidence weight is adaptively adjusted based on the type identifier of the battery manufacturing material.

[0012] Fuzzy logic reasoning rules are used to process the correlation between the capacity decay rate and the voltage plateau curve change characteristics, and a fuzzy evaluation value for the degree of battery performance degradation is obtained.

[0013] Based on the particle filter algorithm, the actual internal resistance growth trend is used as the state observation value to iteratively update the parameters of the battery's equivalent circuit model and predict the internal resistance evolution path after multiple charge-discharge cycles.

[0014] The fuzzy evaluation value and the performance degradation curve obtained based on internal resistance prediction are input into a preset score mapping function to calculate the comprehensive score of the battery health status.

[0015] Based on the number of cycles at which the performance degradation curve reaches the preset failure threshold, and combined with the current number of charge-discharge cycles, the predicted value of the remaining service life is calculated.

[0016] As a further aspect of the present invention, the step of creating a data fusion processing queue, aligning the dynamic operating data and the battery static characteristic data according to timestamps, and generating a synchronized spatiotemporal correlated data sequence includes:

[0017] The dynamic operating data includes continuous real-time discharge voltage, charge-discharge cycle count, operating ambient temperature, and instantaneous output current.

[0018] Each frame of data in the dynamic operation data is assigned a high-precision timestamp generated by the vehicle system;

[0019] Extract the fixed attribute information from the static characteristic data of the battery and generate a constant timestamp sequence covering the entire evaluation period;

[0020] Establish a time synchronization window to match and associate the dynamic running data frames with the same or nearest timestamp with the battery static characteristic data attribute information;

[0021] The operating environment temperature and the instantaneous output current are subjected to moving average filtering to eliminate high-frequency noise interference;

[0022] The data, after matching, association, and filtering, are arranged in chronological order to form the synchronous spatiotemporal associated data sequence, where each record contains all data fields under the same time reference.

[0023] As a further aspect of the present invention, the step of inputting the synchronous spatiotemporal correlated data sequence into the feature extraction layer of the health status assessment model, and extracting a set of key parameters characterizing battery degradation features by the feature extraction layer, includes:

[0024] The battery static characteristic data includes the battery's nominal capacity at the factory, the battery's standard internal resistance value, and the type identification of the battery's manufacturing materials.

[0025] From the synchronous spatiotemporal correlation data sequence, voltage and current data for a complete discharge cycle are extracted, the actual discharge capacity of the complete discharge cycle is calculated, and compared with the nominal capacity of the battery corresponding to the complete discharge cycle to calculate the capacity decay rate.

[0026] During the complete discharge cycle, a specific state of charge point is selected. Based on the discharge voltage, current and temperature corresponding to the state of charge point, the DC internal resistance test method is applied to calculate the actual internal resistance of the battery at the state of charge point. This actual internal resistance is then compared with the standard internal resistance value of the battery to generate the actual internal resistance growth trend.

[0027] Plot the voltage versus discharge capacity curve during a complete discharge cycle, and identify the characteristics of the voltage plateau curve, including the decrease in plateau voltage, the shortening of the plateau curve, and the increase in the plateau slope.

[0028] The calculated capacity decay rate, the actual internal resistance growth trend, and the voltage plateau curve variation characteristics are normalized and packaged, and the output is the key parameter set.

[0029] As a further aspect of the present invention, the step of performing multi-source evidence fusion analysis on the key parameter set through the fusion evaluation layer of the health status assessment model to generate a comprehensive battery health status score and a predicted value of remaining service life includes:

[0030] An initial confidence weight is assigned to each parameter in the set of key parameters, and the initial confidence weight is adaptively adjusted based on the type identifier of the battery manufacturing material.

[0031] Fuzzy logic reasoning rules are used to process the correlation between the capacity decay rate and the voltage plateau curve change characteristics, and a fuzzy evaluation value for the degree of battery performance degradation is obtained.

[0032] Based on the particle filter algorithm, the actual internal resistance growth trend is used as the state observation value to iteratively update the parameters of the battery's equivalent circuit model and predict the internal resistance evolution path after multiple charge-discharge cycles.

[0033] The fuzzy evaluation value and the performance degradation curve obtained based on internal resistance prediction are input into a preset score mapping function to calculate the comprehensive score of the battery health status.

[0034] Based on the number of cycles at which the performance degradation curve reaches the preset failure threshold, and combined with the current number of charge-discharge cycles, the predicted value of the remaining service life is calculated.

[0035] As a further aspect of the present invention, the step of employing fuzzy logic reasoning rules to process the correlation between the capacity decay rate and the voltage plateau curve variation characteristics, and obtaining a fuzzy evaluation value regarding the degree of battery performance degradation, includes:

[0036] A fuzzy linguistic variable is defined for the capacity decay rate and a fuzzy linguistic variable is defined for the voltage plateau curve variation characteristics. Each fuzzy linguistic variable contains multiple sub-states.

[0037] A fuzzy rule base is established, with each fuzzy rule taking the fuzzy sub-state of the capacity decay rate and the fuzzy sub-state of the voltage plateau curve change characteristics as premises and the fuzzy state of the degree of battery performance degradation as conclusion.

[0038] Using the triangular membership function, the calculated specific capacity decay rate and the quantized voltage plateau curve change characteristics are transformed into their membership distributions on their respective fuzzy linguistic variables.

[0039] Based on the membership distribution and the fuzzy rule base, fuzzy inference is performed to obtain the activation intensity of multiple states of battery performance degradation.

[0040] The activation intensity of all states of battery performance degradation is defuzzified to obtain a numerical output, which serves as the fuzzy evaluation value.

[0041] As a further aspect of the present invention, the step of using the particle filter algorithm to iteratively update the parameters of the battery's equivalent circuit model, taking the actual internal resistance growth trend as a state observation, and predicting the internal resistance evolution path after multiple charge-discharge cycles includes:

[0042] Initialize a set of particles, each particle representing a state vector of the equivalent circuit model parameters of the battery;

[0043] Using the historical data of the actual internal resistance growth trend as an observation sequence, for each new observation value, the likelihood probability of each particle is calculated. The likelihood probability represents the probability that the equivalent circuit model parameters of the battery represented by the particle can produce the observation value.

[0044] Based on the likelihood probability of each particle, the particle set is resampled, increasing the number of particles with high likelihood probability and decreasing the number of particles with low likelihood probability.

[0045] Using a battery aging mechanism model, the state of each resampled particle is predicted to simulate the changes in the equivalent circuit model parameters during the next charge-discharge cycle.

[0046] The steps of observation update, resampling, and state prediction are repeated. The statistical characteristics of the final retained particle set represent the optimal estimate of the parameters of the battery equivalent circuit model. Extrapolation is performed based on the optimal estimate to generate the internal resistance evolution path.

[0047] As a further aspect of the present invention, it also includes:

[0048] The model adaptive calibration module adaptively calibrates the health status assessment model based on the comprehensive battery health status score, specifically including:

[0049] After the electric bicycle completes a full charge-discharge cycle, record the measured total discharge capacity and average operating temperature of this cycle.

[0050] The actual capacity retention rate for this cycle is calculated by comparing the measured total discharge capacity with the battery's nominal capacity at the factory.

[0051] The actual capacity retention rate is compared with the predicted capacity retention rate output by the health status assessment model in the previous cycle to obtain the prediction error;

[0052] Based on the magnitude and direction of the prediction error, the parameter weights of the fusion assessment layer in the health status assessment model are adjusted in reverse, especially the confidence weights related to processing capacity decay and the parameters of the fuzzy rule base.

[0053] Using the adjusted parameter weights, the synchronous spatiotemporal correlated data sequence of the next round of input is processed to achieve online adaptive calibration of the model.

[0054] As a further aspect of the present invention, the step of comparing the measured total discharge capacity with the battery's factory nominal capacity to calculate the actual capacity retention rate for this cycle includes:

[0055] Control the electric bicycle battery to discharge at a constant power from a fully charged state to a preset termination voltage, and record the discharge duration of the entire discharge process.

[0056] The measured total discharge capacity for this cycle is calculated based on the product of the constant discharge power and the discharge duration.

[0057] Obtain the nominal factory capacity of the battery corresponding to the battery;

[0058] The actual capacity retention rate is obtained by dividing the measured total discharge capacity by the battery's nominal capacity at the factory and multiplying by a percentage base.

[0059] The calculation results, along with the average operating temperature and the number of charge-discharge cycles for this cycle, are stored as the true value data points for model calibration.

[0060] As a further aspect of the present invention, it also includes:

[0061] The maintenance suggestion module receives the comprehensive battery health status score and the predicted remaining service life value output by the health status assessment model.

[0062] Multiple health status scoring thresholds are set, and the comprehensive health status score of the battery is compared with the health status scoring thresholds to determine the health status level of the battery.

[0063] Based on the health status level of the battery, a corresponding basic maintenance suggestion template is matched from the preset maintenance strategy knowledge base;

[0064] Based on the historical statistical characteristics of the operating environment temperature and the frequency of the charge-discharge cycle count, the parameters in the basic maintenance recommendation template are fine-tuned.

[0065] The estimated battery failure date is calculated by adding the remaining lifespan prediction to the current date. This date information is then combined with the maintenance recommendations after parameter fine-tuning to generate the final personalized maintenance recommendation report.

[0066] As a further aspect of the present invention, the step of matching a corresponding basic maintenance suggestion template from a preset maintenance strategy knowledge base based on the health status level of the battery includes:

[0067] The maintenance strategy knowledge base contains multiple preset health status level ranges, and each level range is associated with one or more basic maintenance suggestion templates.

[0068] The basic maintenance suggestion template includes a text description and adjustable parameters. The text description includes suggestions for charging frequency, depth of discharge, and operating temperature range. The adjustable parameters include charging current rate and suggested idle time.

[0069] Query the maintenance strategy knowledge base to find the health status level range of the battery;

[0070] Extract all the basic maintenance suggestion templates associated with the level range;

[0071] Based on the type identifier of the battery manufacturing material, templates with matching materials are selected from all extracted basic maintenance suggestion templates as candidate basic maintenance suggestion templates.

[0072] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0073] By creating a data fusion processing queue, dynamic operational data and static characteristic data from different hardware are aligned at the millisecond level according to timestamps, generating a synchronous spatiotemporally correlated data sequence containing complete dynamic and static parameters at each moment. This technique eliminates time differences and state misalignments between data from different sources, ensuring that the input data received by subsequent models has a high degree of consistency in time and state context. The feature extraction layer works based on this fused sequence, accurately capturing the true evolution of dynamic parameters under specific static attribute backgrounds, thus building an accurate data foundation for subsequent analysis.

[0074] The feature extraction layer specifically extracts three sets of key parameters from the fused data sequence: capacity decay rate, actual internal resistance growth trend, and voltage plateau curve variation characteristics, representing different degradation mechanisms and spatiotemporal scales. The fusion evaluation layer does not simply use a weighted average; instead, it employs a multi-source evidence fusion algorithm to process these parameter sets, treating each set as an independent body of evidence for cross-validation and confidence synthesis. This approach integrates multi-dimensional information on long-term trends, medium-term performance, and short-term behavior. When generating a comprehensive health status score, it effectively offsets the interference caused by abnormal fluctuations in a single parameter. Furthermore, when predicting remaining useful life, it can extrapolate based on a more robust and consistent fused degradation trajectory, improving the reliability and prediction accuracy of the evaluation results. Attached Figure Description

[0075] Figure 1 This is a timing diagram of the data fusion-based electric bicycle battery health status assessment system described in this invention.

[0076] Figure 2 A flowchart for the integrated evaluation and analysis;

[0077] Figure 3 A graph showing the correlation between battery health score and remaining lifespan;

[0078] Figure 4 Heatmap of the adaptive calibration process for model parameters;

[0079] Figure 5 Radar charts showing battery performance indicators at different health status levels. Detailed Implementation

[0080] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0081] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0082] See Figure 1 The data acquisition module is responsible for acquiring dynamic operating data of the battery from the vehicle terminal, including parameters such as voltage, current, temperature, and cycle count that change over time. Simultaneously, it extracts static characteristic data of the battery from the battery management system, such as nominal capacity, standard internal resistance, and material type. The data alignment module creates a data fusion processing queue, aligning the dynamic operating data and static characteristic data according to timestamps to generate a synchronous spatiotemporal correlated data sequence. This sequence ensures consistency of data at each time point. The feature extraction module calls a preset health status assessment model, inputting the synchronous spatiotemporal correlated data sequence into the model's feature extraction layer. This layer extracts a set of key parameters, such as capacity decay rate, actual internal resistance growth trend, and voltage plateau curve variation characteristics. The fusion assessment module performs multi-source evidence fusion analysis on the key parameter set through the fusion assessment layer of the health status assessment model, ultimately generating a comprehensive battery health status score and a predicted remaining service life.

[0083] In one embodiment of the present invention, after the data alignment module is activated, it creates a data fusion processing queue. Dynamic running data is continuously input from the vehicle terminal in the form of a data stream. Each frame of dynamic running data contains a high-precision timestamp generated in real time by the vehicle system. The dynamic running data includes continuous real-time discharge voltage, charge-discharge cycle count, operating ambient temperature, and instantaneous output current. For example, in one operating cycle, the dynamic running data is represented by instantaneous voltage, current, temperature, and cumulative cycle count collected once per second. The high-precision timestamp is accurate to the millisecond level, ensuring the uniqueness of the time reference. Simultaneously, battery static characteristic data is read from the battery management system. The battery static characteristic data contains fixed attribute information, namely the battery's nominal capacity, standard internal resistance value, and type identifier of the battery manufacturing materials. A constant timestamp sequence covering the period from the start of the evaluation to the current moment is generated for these static attribute information. It can be understood that each static attribute information in the constant timestamp sequence is associated with the time range of the entire evaluation cycle, providing a reference for subsequent time matching. In practical implementation, to unify data from different sources, the system establishes a time synchronization window. The width of the window can be set according to the sampling frequency, for example, ±50 milliseconds. The timestamps of dynamic running data frames are compared with the constant timestamps of static attribute information within the time synchronization window. Dynamic running data frames with the same or nearest timestamps are matched and associated with the battery's static characteristic data attribute information, ensuring that at any given calculation time, the system can access the battery's inherent attributes that match the current operating state. In some embodiments, the ambient temperature and instantaneous output current in the dynamic running data undergo moving average filtering. The window length for moving average filtering is set to 10 sampling points. The average of 10 consecutive sampling values ​​is used to replace the original value of the current sampling point. This process effectively eliminates high-frequency noise introduced by sensor noise or transient interference, making the trends of ambient temperature and instantaneous output current smoother and more stable. It can be understood that the filtering process is performed after time matching and association to ensure that the aligned data sequence remains consistent in the time domain before smoothing. After completing all matching, correlation, and filtering processes, the system arranges the processed data in chronological order according to timestamps, forming a synchronous spatiotemporal correlated data sequence. Each record in the synchronous spatiotemporal correlated data sequence is a structured data object, containing all data fields under the same time base, namely voltage value, cycle count, temperature value, current value, nominal capacity, standard internal resistance, and material type. For example, a complete record can be represented as {timestamp T1, voltage U1, cycle count C1, temperature θ1, current I1, nominal capacity CN, standard internal resistance RN, material type M}.

[0084] After the feature extraction module calls the health status assessment model, the synchronous spatiotemporal correlated data sequence is input to the feature extraction layer of the health status assessment model. In specific implementation, the feature extraction operation extracts a complete discharge cycle from the synchronous spatiotemporal correlated data sequence, from a fully charged state to a preset cutoff voltage. Optionally, a complete discharge cycle can be determined by identifying continuous data segments where the voltage continuously decreases from its maximum value to the preset cutoff voltage. From the extracted complete discharge cycle data, the system calculates the actual discharge capacity of the complete discharge cycle. The calculation process is achieved by integrating the discharge current over time. The formula for calculating the actual discharge capacity is expressed as:

[0085] ;

[0086] in: This represents the actual discharge capacity for a complete discharge cycle. This represents the discharge current value at the k-th sampling point within a complete discharge cycle. Indicates the sampling time interval. This represents the total number of sampling points within a complete discharge cycle. After calculating the actual discharge capacity, it is compared with the battery's nominal capacity corresponding to the complete discharge cycle. The capacity decay rate is calculated through difference or ratio calculation. The capacity decay rate can be expressed as the ratio of (battery nominal capacity - actual discharge capacity) / battery nominal capacity to the number of completed cycles, used to quantify the rate of capacity loss per unit cycle. During a complete discharge cycle, the system selects one or more specific state of charge (SOC) points, such as the moment when the SOC is 50%. Based on the discharge voltage, discharge current, and battery temperature corresponding to the SOC point, the actual internal resistance of the battery at the SOC point is calculated using the DC internal resistance test method. The DC internal resistance test method solves for the internal resistance by measuring the ratio of the change in voltage to the change in current over a short time interval. The calculated actual internal resistance is compared with the battery's standard internal resistance value. By comparing the difference or its trend, the actual internal resistance growth trend is generated. The actual internal resistance growth trend can be expressed as the difference between the current actual internal resistance and the battery's standard internal resistance, or as the rate of change of the actual internal resistance with the number of cycles. The system plots the voltage-to-capacity change curve throughout a complete discharge cycle, identifying voltage plateau curve characteristics that characterize the battery's electrochemical plateau properties from the voltage-capacity curve. These characteristics include the decrease in plateau voltage (the difference between the current average voltage and the initial plateau voltage), the shortening of the plateau curve (the reduction in the projected length of the plateau on the capacity axis), and the increase in plateau slope (the change in the inclination of the voltage plateau segment). In practice, the calculated capacity decay rate, actual internal resistance growth trend, and identified voltage plateau curve characteristics (including the decrease in plateau voltage, the shortening of the plateau curve, and the increase in plateau slope) are normalized. Normalization maps parameters of different dimensions and magnitudes to the interval [0,1]. After normalization, these parameters are encapsulated into a structured dataset; this structured dataset, containing various battery degradation characteristics, is the final output set of key parameters.

[0087] See Figure 2In one embodiment of the present invention, the fusion evaluation layer of the health status assessment model receives a set of key parameters output from the feature extraction layer. The key parameter set includes the capacity decay rate, the actual internal resistance growth trend, and the voltage plateau curve variation characteristics. In specific implementation, the fusion evaluation layer assigns an initial confidence weight to each parameter in the key parameter set. The allocation of the initial confidence weight is adaptively adjusted based on the type identifier of the battery manufacturing material. For example, when the battery manufacturing material type identifier is "lithium-ion battery ternary material," the system assigns a higher initial confidence weight to the capacity decay rate and a medium initial confidence weight to the actual internal resistance growth trend; when the battery manufacturing material type identifier is "lithium-ion battery lithium iron phosphate material," the system assigns a higher initial confidence weight to the actual internal resistance growth trend and a medium initial confidence weight to the voltage plateau curve variation characteristics. It can be understood that the initial confidence weight reflects the difference in the reliability of different degradation characteristics in characterizing the health status of batteries made of different materials. In some embodiments, fuzzy logic inference rules are used to handle the correlation between the capacity decay rate and the voltage plateau curve variation characteristics. The fuzzy logic inference system predefines fuzzy sets of input variables and fuzzy sets of output variables, and establishes a fuzzy rule base containing a series of "if-then" rules. For example, a fuzzy rule could be "If the capacity decay rate is 'high' and the voltage plateau curve variation characteristics are 'significant,' then the battery performance degradation degree is 'severe.'" The system inputs the normalized specific value of the capacity decay rate and the quantified specific value of the voltage plateau curve variation characteristics into the fuzzy logic inference system. Through the processes of fuzzification, rule evaluation, and defuzzification, a deterministic numerical output regarding the degree of battery performance degradation is obtained. This numerical output is the fuzzy evaluation value. The fuzzy evaluation value integrates information from both capacity decay and voltage plateau distortion. In a specific implementation, the actual internal resistance growth trend is handled based on the particle filter algorithm. The particle filter algorithm uses the battery's equivalent circuit model parameters as the system's state vector and the observed actual internal resistance growth trend as the state observation value. The system initializes a set of particles, each carrying a set of random values ​​for the equivalent circuit model parameters, representing a possible state of the parameters. The particle filter algorithm estimates the state by iteratively performing prediction and update steps. In the update step, the algorithm compares historical observations of the actual internal resistance growth trend with the predicted observations for each particle, calculating the likelihood probability of each particle. The likelihood probability represents the probability that the parameters of the equivalent circuit model represented by the particle can produce the actually observed internal resistance growth trend. In the resampling step, the particle set is resampled based on the particle likelihood probabilities, increasing the number of particles with high likelihood probabilities and decreasing the number of particles with low likelihood probabilities.In the prediction step, a battery aging mechanism model, which describes how the battery's internal parameters evolve with charge-discharge cycles, is used to predict the state vector of each resampled particle one step ahead, simulating the changes in the equivalent circuit model parameters during the next charge-discharge cycle. This iterative process of observation update, resampling, and state prediction is repeated until the statistical characteristics of the final particle set, such as the mean of the particle states, represent the optimal estimate of the battery's equivalent circuit model parameters. Based on this optimal estimate, extrapolation through the battery aging mechanism model generates the internal resistance evolution path after multiple future charge-discharge cycles. This path depicts the possible trend of internal resistance change with increasing cycle count.

[0088] The fusion evaluation layer inputs the fuzzy evaluation value derived from fuzzy logic reasoning and the performance degradation curve derived from the internal resistance evolution path into a preset score mapping function. The score mapping function is a mathematical function that maps multidimensional inputs to a single health score. One specific form of the score mapping function is a weighted linear combination, and its formula is as follows:

[0089] ;

[0090] in: This represents the final calculated overall score of battery health. This represents the fuzzy evaluation value output by the fuzzy logic reasoning system. This represents the internal resistance value predicted based on the internal resistance evolution path at a specific number of future cycles (e.g., predicting the end of life). It is a predictor of internal resistance value A function mapped to a performance degradation score. and These are the fusion weighting coefficients. and The value of is related to the previously assigned initial confidence weights and can be adjusted based on feedback from the model adaptive calibration module. The overall battery health score is a value between 0 and 100, with higher values ​​indicating better battery health. Simultaneously with calculating the overall battery health score, the system estimates the remaining lifespan based on the number of cycles at which the performance degradation curve derived from the internal resistance evolution path reaches a preset failure threshold. The preset failure threshold is a critical value for parameters such as internal resistance or capacity; when the performance degradation curve (e.g., the internal resistance growth curve) reaches this critical value, the battery lifespan is considered to have ended. The performance degradation curve depicts the change of parameters with the number of cycles. Assuming the performance degradation curve indicates that the internal resistance will reach the failure threshold on the Nth cycle, and the battery has currently completed n charge-discharge cycles, then the predicted remaining lifespan is _____. The remaining useful life prediction can be given in the form of the number of charge-discharge cycles, or it can be converted into the remaining usage time in days or months based on the user's average usage frequency.

[0091] In one embodiment of the present invention, when using fuzzy logic reasoning rules to handle the correlation between capacity decay rate and voltage plateau curve variation characteristics, it is necessary to define fuzzy linguistic variables and establish a fuzzy rule base. In specific implementations, fuzzy linguistic variables are defined for both capacity decay rate and voltage plateau curve variation characteristics. The fuzzy linguistic variable for capacity decay rate can be named "capacity decay rate," and the fuzzy linguistic variable for voltage plateau curve variation characteristics can be named "plateau distortion degree." Each fuzzy linguistic variable contains multiple sub-states, which are linguistic values ​​describing the degree of the variable. For example, the sub-states of the fuzzy linguistic variable for capacity decay rate can be set to "very low," "low," "medium," "high," and "very high," and the sub-states of the fuzzy linguistic variable for plateau distortion degree can be set to "small," "slight," "moderate," "significant," and "severe." A fuzzy rule base is established, containing multiple fuzzy rules. Each fuzzy rule uses the fuzzy sub-states of capacity decay rate and voltage plateau curve variation characteristics as preconditions and the fuzzy state of battery performance degradation degree as the conclusion. The fuzzy state of battery performance degradation is also a fuzzy linguistic variable, and its sub-states can be set as "excellent," "good," "average," "poor," and "failed." A specific fuzzy rule is expressed as "if the capacity decay rate is 'high' and the plateau distortion degree is 'medium,' then the battery performance degradation degree is 'poor.'" The fuzzy rule base typically contains rules covering all possible input combinations; for example, with 5 capacity decay rate sub-states and 5 plateau distortion degree sub-states, the fuzzy rule base can contain up to 25 rules. In some embodiments, a triangular membership function is used to transform the calculated specific capacity decay rate value and the quantized voltage plateau curve change characteristic value into their membership degree distributions on each sub-state of their respective fuzzy linguistic variables. The triangular membership function is determined by three parameters, corresponding to the abscissa values ​​of the left endpoint, vertex, and right endpoint of the triangle's base, respectively. For a specific capacity decay rate value, the triangular membership function corresponding to all sub-states of the fuzzy linguistic variable of the input capacity decay rate is calculated to obtain the membership degree of this value to the sub-states of "very low", "low", "medium", "high", and "very high". The membership degree is a value between 0 and 1, indicating the degree to which the value belongs to a certain fuzzy sub-state. The quantized voltage plateau curve change characteristic value is also calculated in the same way to obtain its membership degree distribution to sub-states such as "tiny", "slight", "moderate", "significant", and "severe". Based on the calculated membership degree distribution and the pre-established fuzzy rule base, fuzzy inference operations are performed.The fuzzy inference operation employs a maximum-minimum synthesis method. For each rule in the fuzzy rule base, the activation degree of its precondition "capacity decay rate is A and plateau distortion degree is B" is obtained by taking the smaller value between the membership degree of the capacity decay rate value to sub-state A and the membership degree of the voltage plateau curve change characteristic value to sub-state B. This activation degree is then used to truncate or scale the fuzzy set of battery performance degradation degree in the rule's conclusion. The activation intensity of all sub-states of battery performance degradation degree is defuzzified. The defuzzification process merges multiple activated and scaled fuzzy conclusion sets into a total fuzzy set. Then, by calculating the centroid or area center of this total fuzzy set, a specific and deterministic numerical output is obtained. This numerical output is the fuzzy evaluation value. It can be understood that the fuzzy evaluation value is a fusion index that integrates information on capacity decay rate and voltage plateau curve change characteristics.

[0092] In practical implementation, when predicting the internal resistance evolution path based on the particle filter algorithm, a set of particles is first initialized. The number of particles can be set according to the computational accuracy and resources, for example, 1000 particles. Each particle represents a possible state vector of the battery's equivalent circuit model parameters. The state vector can contain parameters such as ohmic internal resistance, polarization internal resistance, and polarization capacitance. The state vector of each particle is randomly initialized within a preset parameter value range, and each particle is assigned the same initial weight. In practical implementation, historical data of the actual internal resistance growth trend is used as an observation sequence, which is a series of internal resistance observations recorded over time. For each new internal resistance observation, the likelihood probability of each particle is calculated. The likelihood probability represents the probability of observing the current internal resistance observation value under the current equivalent circuit model parameter state represented by the particle. The calculation of the likelihood probability is usually based on an observation noise model. Assuming that the observation noise follows a Gaussian distribution with a mean of zero, the smaller the difference between the predicted internal resistance value obtained by the particle state through the battery aging mechanism model and the actual observed value, the higher the likelihood probability of the particle. The formula for calculating the likelihood probability of each particle is expressed as:

[0093] ;

[0094] in: Let represent the likelihood probability (i.e., the non-normalized weight) of the i-th particle at time t. This represents the internal resistance value actually observed at time t. This represents the predicted internal resistance value calculated from the state vector of the i-th particle using the observation model. This represents the standard deviation of the observation noise. The particle ensemble is resampled based on the likelihood probability of each particle. The resampling process involves drawing new particles with replacement from the current ensemble using the normalized likelihood probabilities of the particles as the probability distribution. Particles with high likelihood probabilities have a greater chance of being drawn multiple times, while particles with low likelihood probabilities may not be drawn. After resampling, the total number of particles remains unchanged, but the number of high-likelihood particles increases, the number of low-likelihood particles decreases, and the particle weights are reset to a uniform distribution. A battery aging mechanism model is used to predict the state of each resampled particle. This model describes how the parameters of the equivalent circuit model evolve with charge-discharge cycles. The state prediction simulates the change in the parameter state represented by each particle in the next charge-discharge cycle. For example, a simplified battery aging mechanism model can be expressed as internal resistance increasing linearly with the number of cycles. By repeating the steps of observation update, resampling, and state prediction, with each new internal resistance observation input, the particle ensemble gradually converges to the parameter state that best explains the actual observation sequence. Ultimately, the statistical characteristics of the particle set retained after multiple iterations, such as the mean of all particle state vectors, represent the optimal estimate of the battery equivalent circuit model parameters at the current moment. Based on this optimal estimate, extrapolation through the battery aging mechanism model can generate the internal resistance evolution path after multiple charge-discharge cycles. The internal resistance evolution path is a prediction of the future trend of internal resistance value changes.

[0095] See Figure 3 This is a correlation analysis chart between battery health score and remaining lifespan. It visually demonstrates the synchronous decay trend of the overall health score and remaining lifespan with the number of charge-discharge cycles, verifying the strong correlation between the two. The health score linearly decreases from nearly 100 points to below 10 points, while the remaining lifespan decreases from 800 cycles to 0, clearly reflecting the entire life cycle of the battery from "new" to "failure." Through the linkage between the health score and remaining lifespan, the system can generate personalized maintenance recommendations. For example, after the score drops to 60 points, it is recommended to reduce the charging frequency and avoid deep discharge to delay performance degradation. The smoothness and consistency of the curves demonstrate the effectiveness of fuzzy logic reasoning and particle filtering algorithms in the fusion evaluation. The rate of decline in the health score and the prediction accuracy of the remaining lifespan can be used to reverse-calibrate the health status assessment model, improving the long-term reliability of the system.

[0096] In one embodiment of the present invention, the model adaptive calibration module adaptively calibrates the health status assessment model based on the comprehensive battery health status score. The model adaptive calibration module is triggered after each complete full-charge-discharge cycle of the electric bicycle. In specific implementations, the model adaptive calibration module records the measured total discharge capacity and average operating temperature of this complete full-charge-discharge cycle. Obtaining the measured total discharge capacity requires controlling the electric bicycle battery to discharge at a constant power from a fully charged state to a preset termination voltage, and recording the discharge duration of the entire discharge process. It can be understood that the preset termination voltage is the discharge cutoff voltage specified by the battery manufacturer to protect the battery from over-discharge. In specific implementations, the measured total discharge capacity of this cycle is calculated by multiplying the set constant discharge power by the recorded discharge duration. For example, if the constant discharge power is 350 watts and the discharge duration is 3600 seconds, then the measured total discharge capacity is calculated as 350 watts * 3600 seconds = 1,260,000 joules, which is then converted to ampere-hours based on the voltage. The battery's nominal factory capacity is obtained from the battery management system. Divide the calculated total discharge capacity by the battery's nominal factory capacity, and multiply by a percentage base of 100 to obtain the actual capacity retention rate for this cycle. The actual capacity retention rate can be calculated using the following formula:

[0097] ;

[0098] in: This represents the calculated actual capacity retention rate. This represents the measured total discharge capacity for this cycle. This represents the battery's nominal factory capacity. This calculated result, along with the average operating temperature and number of charge-discharge cycles for this cycle, is stored as the true value data point for model calibration. In practice, the model adaptive calibration module compares the calculated actual capacity retention rate with the predicted capacity retention rate output by the health status assessment model in the previous cycle to obtain the prediction error. The prediction error is the difference between the actual capacity retention rate and the predicted capacity retention rate. For example, if the actual capacity retention rate for this cycle is 92.5%, while the health status assessment model predicted a capacity retention rate of 94.0% at the end of the previous cycle, then the prediction error is -1.5%. See Table 1.

[0099] Table 1: Model Calibration Data Recording Table

[0100] Cycle number (times) Measured total discharge capacity (Ah) Battery factory nominal capacity (Ah) Actual capacity retention rate (%) Model predicted capacity retention rate (%) Prediction error (%) Average working temperature (°C) 150 12.30 13.20 93.18 94.00 -0.82 28.5 151 12.25 13.20 92.80 93.50 -0.70 29.1

[0101] Based on the magnitude and direction of the prediction error, the model adaptive calibration module adjusts the parameter weights of the fusion assessment layer in the health status assessment model in reverse. The magnitude of the prediction error indicates the degree of bias in the model's prediction, while the direction of the prediction error indicates whether the model's prediction is too high or too low. In specific implementations, the adjustment targets are particularly the confidence weights related to capacity decay and the parameters of the fuzzy rule base. For example, in the fusion assessment layer, the initial confidence weight for handling the capacity decay rate may be adjusted. If the prediction error remains negative (i.e., the model's predicted value is consistently higher than the actual value), the confidence weight of the capacity decay rate in the fusion assessment may be reduced, or the output of rules related to "high capacity decay rate" in the fuzzy rule base may be adjusted. It is understood that the adjustment can be achieved through gradient descent, heuristic rules, or lookup tables. In some embodiments, the adjustment of the fuzzy rule base parameters can be reflected in adjusting the membership function parameters of the conclusion fuzzy set. After completing the parameter weight adjustment, the adjusted parameter weights are used to process the next round of synchronous spatiotemporal correlated data sequences input from the data alignment module, realizing online adaptive calibration of the health status assessment model. Online adaptive calibration enables the model to dynamically adjust its evaluation logic based on the actual degradation performance of the battery, reducing evaluation bias caused by individual battery differences and changes in usage conditions. The model's adaptive calibration module operates periodically, providing a calibration opportunity with each complete charge-discharge cycle, allowing the health status assessment model to continuously track changes in battery characteristics.

[0102] See Figure 4 This is a heatmap of the adaptive calibration process for model parameters. It visually illustrates the changes in the adjustment values ​​of four key model parameters at different calibration cycles. The color gradient from blue to red clearly reflects the direction and magnitude of correction for each parameter at different calibration stages. As the number of calibration cycles increases, the adjustment magnitude of each parameter generally shows a convergence trend, especially at 300 cycles, where the adjustment values ​​of all parameters approach 0, proving the effectiveness of the adaptive calibration mechanism. This convergence indicates that after multiple rounds of calibration, the model parameters have reached their optimal state and can stably and accurately predict the battery health status. By analyzing the timing and magnitude of parameter adjustments, calibration strategies can be optimized in a targeted manner. For example, for parameter 4, which shows a large negative adjustment in the early stages, the initial parameter settings can be optimized to reduce the pressure on subsequent calibrations. At the same time, computational resources can be rationally allocated based on parameter sensitivity, prioritizing the calibration of parameters that have the greatest impact on model performance, thereby improving calibration efficiency.

[0103] In one embodiment of the present invention, the maintenance suggestion module receives a comprehensive battery health status score and a predicted remaining service life value output by a health status assessment model. The maintenance suggestion module internally sets multiple health status score thresholds, which divide continuous scores into discrete health status levels. For example, health status score thresholds can be set to 90, 80, 60, and 40, corresponding to health status level ranges of: Excellent (score >= 90), Good (80 <= score < 90), Average (60 <= score < 80), Poor (40 <= score < 60), and Failed (score < 40). In specific implementations, the comprehensive battery health status score is compared with the preset health status score thresholds to determine the battery's current health status level. If the comprehensive battery health status score is 85 points, falling between 80 and 90, the battery's current health status level is determined to be "Good." Based on the battery's health status level, the maintenance suggestion module matches the corresponding basic maintenance suggestion template from a preset maintenance strategy knowledge base. The maintenance strategy knowledge base is a database of rules and templates stored in the system. It contains multiple preset health status level ranges, each associated with one or more basic maintenance suggestion templates. These templates include text descriptions and adjustable parameters. The text descriptions include recommendations for charging frequency, depth of discharge, and operating temperature range. Adjustable parameters include charging current rate and recommended idle time. For example, for a health status level of "Good," the associated basic maintenance suggestion template might include text descriptions such as "It is recommended to charge when the battery level is 20%-30% to avoid complete depletion" and "Avoid prolonged use or charging in ambient temperatures above 45°C or below 0°C." Adjustable parameters might be set to a charging current rate not exceeding 0.5C and a recommendation to maintain a battery state of charge of approximately 50% when the vehicle is idle for more than 7 days.

[0104] In some embodiments, the maintenance strategy knowledge base is queried to find the health status level range of the battery, and all basic maintenance suggestion templates associated with the level range are extracted. Then, based on the battery manufacturing material type identifier, templates matching the material are selected from all extracted basic maintenance suggestion templates as candidate basic maintenance suggestion templates. It is understood that the battery manufacturing material type identifier is, for example, "lithium-ion battery ternary material NCM" or "lithium-ion battery lithium iron phosphate material LFP," and different types of battery materials have different sensitivities to maintenance. For example, a template for "lithium-ion battery ternary material NCM" may emphasize avoiding high-rate charging, while a template for "lithium-ion battery lithium iron phosphate material LFP" may focus more on the impact of full-charge storage. The maintenance suggestion module fine-tunes the parameters in the candidate basic maintenance suggestion templates by combining the historical statistical characteristics of the operating environment temperature and the frequency of charge-discharge cycles. The historical statistical characteristics of the operating environment temperature may include the average operating temperature, the highest operating temperature, and the temperature fluctuation range over a past period (e.g., the past 30 days). The frequency of charge-discharge cycles may refer to the average number of equivalent complete cycles completed per day. For example, if historical statistics show a high average operating temperature, the suggested charging current rate parameter in the template can be further lowered, or the suggested maximum operating temperature threshold can be lowered. If the charge / discharge cycle frequency is high, more frequent checks of battery connection status can be suggested. After fine-tuning the parameters, the maintenance recommendation module adds the remaining lifespan prediction to the current date to calculate the estimated battery failure date. The remaining lifespan prediction may be a value in "days". The formula for calculating the estimated battery failure date is expressed as:

[0105] ;

[0106] in: This indicates the estimated battery failure date obtained from the calculation. Indicates the current date. This represents the predicted remaining battery life in days. For example, if the current date is May 20, 2024, and the predicted remaining battery life is 180 days, then the estimated battery failure date is November 16, 2024. The maintenance recommendation module combines this date information with the adjusted maintenance recommendation text and parameters to generate a final personalized maintenance recommendation report. The final personalized maintenance recommendation report includes the battery's current health status level, specific maintenance operation suggestions, adjusted parameter values, and the estimated battery failure date. Optionally, the report can be displayed in text form on the user's mobile application or sent to the service station in structured data form. The generation logic of the maintenance recommendation module allows maintenance recommendations to combine the battery's real-time health score, usage history (temperature, cycle time), and inherent properties (materials) to provide targeted, personalized guidance, rather than generic, static recommendations.

[0107] See Figure 5 This is a radar chart of battery performance indicators at different health status levels. It visually presents the battery's overall performance across five health levels: "Excellent," "Good," "Average," "Poor," and "Failure," based on five core dimensions: capacity retention, internal resistance growth, voltage stability, cycle life, and temperature adaptability. The score (0-100) for each dimension clearly quantifies battery performance, forming a complete performance radar chart and avoiding the limitations of single-indicator evaluation. In the evolution from "Excellent" to "Failure," the decline in capacity retention and cycle life is the most significant, representing the core driving factors for battery health deterioration. Simultaneously, the increase in internal resistance and the deterioration in voltage stability are direct causes of capacity and lifespan degradation, providing clear direction for fault diagnosis. By comparing the current battery's radar chart with the baseline charts for each health level, the battery's current health stage can be quickly identified.

[0108] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A data fusion-based electric bicycle battery health status assessment system, characterized in that, The system includes: The data acquisition module obtains dynamic operating data of the electric bicycle battery through the on-board terminal and static characteristic data of the battery through the battery management system. The data alignment module creates a data fusion processing queue, aligns the dynamic operation data with the battery static characteristic data according to timestamps, and generates a synchronous spatiotemporal correlation data sequence. The synchronous spatiotemporal correlation data sequence includes the voltage value, cycle number, temperature value, current value, nominal capacity, standard internal resistance and material type corresponding to each set of timestamps. The feature extraction module calls a preset health status assessment model and inputs the synchronous spatiotemporal correlation data sequence into the feature extraction layer of the health status assessment model. The feature extraction layer extracts a set of key parameters characterizing battery degradation features. The set of key parameters includes capacity decay rate, actual internal resistance growth trend and voltage plateau curve change characteristics. The fusion assessment module, through the fusion assessment layer of the health status assessment model, performs multi-source evidence fusion analysis on the key parameter set to generate a comprehensive battery health status score and a predicted remaining service life, including: An initial confidence weight is assigned to each parameter in the set of key parameters, and the initial confidence weight is adaptively adjusted based on the type identifier of the battery manufacturing material. Fuzzy logic reasoning rules are used to process the correlation between the capacity decay rate and the voltage plateau curve change characteristics, and a fuzzy evaluation value for the degree of battery performance degradation is obtained. Based on the particle filter algorithm, the actual internal resistance growth trend is used as the state observation value to iteratively update the parameters of the battery's equivalent circuit model and predict the internal resistance evolution path after multiple charge-discharge cycles. The fuzzy evaluation value and the performance degradation curve obtained based on internal resistance prediction are input into a preset score mapping function to calculate the comprehensive score of the battery health status. Based on the number of cycles at which the performance degradation curve reaches the preset failure threshold, and combined with the current number of charge-discharge cycles, the predicted value of the remaining service life is calculated.

2. The electric bicycle battery health status assessment system based on data fusion according to claim 1, characterized in that, The creation of the data fusion processing queue aligns the dynamic operating data and the battery static characteristic data according to timestamps, generating a synchronized spatiotemporal correlated data sequence, including: The dynamic operating data includes continuous real-time discharge voltage, charge-discharge cycle count, operating ambient temperature, and instantaneous output current. Each frame of data in the dynamic operation data is assigned a high-precision timestamp generated by the vehicle system; Extract the fixed attribute information from the static characteristic data of the battery and generate a constant timestamp sequence covering the entire evaluation period; Establish a time synchronization window to match and associate each frame of data in the dynamic running data with the same or nearest timestamp with the attribute information of the battery static characteristic data; The operating environment temperature and the instantaneous output current are subjected to moving average filtering to eliminate high-frequency noise interference; The data, after matching, association, and filtering, are arranged in chronological order to form the synchronous spatiotemporal associated data sequence, where each record contains all data fields under the same time reference.

3. The electric bicycle battery health status assessment system based on data fusion according to claim 2, characterized in that, The synchronous spatiotemporal correlated data sequence is input into the feature extraction layer of the health status assessment model, and the feature extraction layer extracts a set of key parameters characterizing battery degradation features, including: The battery static characteristic data includes the battery's nominal capacity at the factory, the battery's standard internal resistance value, and the type identification of the battery's manufacturing materials. From the synchronous spatiotemporal correlation data sequence, voltage and current data for a complete discharge cycle are extracted, the actual discharge capacity of the complete discharge cycle is calculated, and compared with the nominal capacity of the battery corresponding to the complete discharge cycle to calculate the capacity decay rate. During the complete discharge cycle, a specific state of charge point is selected. Based on the discharge voltage, current and temperature corresponding to the state of charge point, the DC internal resistance test method is applied to calculate the actual internal resistance of the battery at the state of charge point. This actual internal resistance is then compared with the standard internal resistance value of the battery to generate the actual internal resistance growth trend. Plot the voltage versus discharge capacity curve during a complete discharge cycle, and identify the characteristics of the voltage plateau curve, including the decrease in plateau voltage, the shortening of the plateau curve, and the increase in the plateau slope. The calculated capacity decay rate, the actual internal resistance growth trend, and the voltage plateau curve variation characteristics are normalized and packaged, and the output is the key parameter set.

4. The electric bicycle battery health status assessment system based on data fusion according to claim 1, characterized in that, The process employs fuzzy logic reasoning rules to process the correlation between the capacity decay rate and the voltage plateau curve variation characteristics, deriving a fuzzy evaluation value regarding the degree of battery performance degradation, including: A fuzzy linguistic variable is defined for the capacity decay rate and a fuzzy linguistic variable is defined for the voltage plateau curve variation characteristics. Each fuzzy linguistic variable contains multiple sub-states. A fuzzy rule base is established, with each fuzzy rule taking the fuzzy sub-state of the capacity decay rate and the fuzzy sub-state of the voltage plateau curve change characteristics as premises and the fuzzy state of the degree of battery performance degradation as conclusion. Using the triangular membership function, the calculated specific capacity decay rate and the quantized voltage plateau curve change characteristics are transformed into their membership distributions on their respective fuzzy linguistic variables. Based on the membership distribution and the fuzzy rule base, fuzzy inference is performed to obtain the activation intensity of multiple states of battery performance degradation. The activation intensity of all states of battery performance degradation is defuzzified to obtain a numerical output, which serves as the fuzzy evaluation value.

5. The electric bicycle battery health status assessment system based on data fusion according to claim 4, characterized in that, The particle filter algorithm uses the actual internal resistance growth trend as a state observation to iteratively update the parameters of the battery's equivalent circuit model, predicting the internal resistance evolution path after multiple charge-discharge cycles, including: Initialize a set of particles, each particle representing a state vector of the equivalent circuit model parameters of the battery; Using the historical data of the actual internal resistance growth trend as an observation sequence, for each new observation value, the likelihood probability of each particle is calculated. The likelihood probability represents the probability that the equivalent circuit model parameters of the battery represented by the particle can produce the observation value. Based on the likelihood probability of each particle, the particle set is resampled, increasing the number of particles with high likelihood probability and decreasing the number of particles with low likelihood probability. Using a battery aging mechanism model, the state of each resampled particle is predicted to simulate the changes in the equivalent circuit model parameters during the next charge-discharge cycle. The steps of observation update, resampling, and state prediction are repeated. The statistical characteristics of the final retained particle set represent the optimal estimate of the parameters of the battery equivalent circuit model. Extrapolation is performed based on the optimal estimate to generate the internal resistance evolution path.

6. The electric bicycle battery health status assessment system based on data fusion according to claim 3, characterized in that, Also includes: The adaptive calibration module performs adaptive calibration on the health status assessment model based on the comprehensive battery health status score, specifically including: After the electric bicycle completes a full charge-discharge cycle, record the measured total discharge capacity and average operating temperature of this cycle. The actual capacity retention rate for this cycle is calculated by comparing the measured total discharge capacity with the battery's nominal capacity at the factory. The actual capacity retention rate is compared with the predicted capacity retention rate output by the health status assessment model in the previous cycle to obtain the prediction error; Based on the magnitude and direction of the prediction error, the parameter weights of the fusion assessment layer in the health status assessment model are adjusted in reverse, especially the confidence weights related to processing capacity decay and the parameters of the fuzzy rule base. Using the adjusted parameter weights, the synchronous spatiotemporal correlated data sequence of the next round of input is processed to achieve online adaptive calibration of the model.

7. The electric bicycle battery health status assessment system based on data fusion according to claim 6, characterized in that, The step of comparing the measured total discharge capacity with the battery's factory nominal capacity to calculate the actual capacity retention rate for this cycle includes: Control the electric bicycle battery to discharge at a constant power from a fully charged state to a preset termination voltage, and record the discharge duration of the entire discharge process. The measured total discharge capacity for this cycle is calculated based on the product of the constant discharge power and the discharge duration. Obtain the nominal factory capacity of the battery corresponding to the battery; The actual capacity retention rate is obtained by dividing the measured total discharge capacity by the battery's nominal capacity at the factory and multiplying by a percentage base. The calculation results, along with the average operating temperature and the number of charge-discharge cycles for this cycle, are stored as the true value data points for model calibration.

8. The electric bicycle battery health status assessment system based on data fusion according to claim 7, characterized in that, Also includes: The maintenance suggestion module receives the comprehensive battery health status score and the predicted remaining service life value output by the health status assessment model. Multiple health status scoring thresholds are set, and the comprehensive health status score of the battery is compared with the health status scoring thresholds to determine the health status level of the battery. Based on the health status level of the battery, a corresponding basic maintenance suggestion template is matched from the preset maintenance strategy knowledge base; Based on the historical statistical characteristics of the operating environment temperature and the frequency of the charge-discharge cycle count, the parameters in the basic maintenance recommendation template are fine-tuned. The estimated battery failure date is calculated by adding the remaining lifespan prediction to the current date. This date information is then combined with the maintenance recommendations after parameter fine-tuning to generate the final personalized maintenance recommendation report.

9. The electric bicycle battery health status assessment system based on data fusion according to claim 8, characterized in that, The step involves matching a corresponding basic maintenance suggestion template from a preset maintenance strategy knowledge base based on the battery's health status level, including: The maintenance strategy knowledge base contains multiple preset health status level ranges, and each level range is associated with one or more basic maintenance suggestion templates. The basic maintenance suggestion template includes a text description and adjustable parameters. The text description includes suggestions for charging frequency, depth of discharge, and operating temperature range. The adjustable parameters include charging current rate and suggested idle time. Query the maintenance strategy knowledge base to find the health status level range of the battery; Extract all the basic maintenance suggestion templates associated with the level range; Based on the type identifier of the battery manufacturing material, templates with matching materials are selected from all extracted basic maintenance suggestion templates as candidate basic maintenance suggestion templates.

Images