An Adaptive Charging Power Supply Control Method and System

Abstract

The present application provides an adaptive charging power control method and system, which relates to the field of power control. It captures the voltage and temperature time-series signals of the battery in real time through high-precision sensors, combines the user charging behavior data, and uses a temporal convolutional network to perform multi-scale temporal modeling on the dynamic characteristics of the battery, extracting the high-frequency non-linear fluctuation characteristics of the battery state. At the same time, semantic association coding is performed on the user behavior data to establish a mapping relationship between user needs and the battery health state. Then, a cross-modal feature interaction mechanism is introduced to realize the dynamic coupling analysis of the battery state and user needs. Finally, a charging voltage regulation instruction that best matches the current battery state and meets the user scenario requirements is generated through a decoder, dynamically adjusting the charging parameters to achieve the collaborative optimization of charging efficiency, battery life, and personalized scenarios.

Description

Technical Field

[0001] This application relates to the field of power control, and more particularly, in the embodiments of this application, to an adaptive charging power control method and system. Background Art

[0002] With the wide application of lithium-ion batteries in fields such as electric vehicles, portable electronic devices, and energy storage systems, the intelligent demand for battery charging management has become increasingly prominent. The traditional constant current / constant voltage (CC / CV) charging strategy realizes charging control by presetting fixed parameters, and its core principle relies on the static model of the battery. However, the battery exhibits highly dynamic characteristics during actual use: its internal electrochemical state (such as polarization effect, capacity decay) and external working conditions (such as ambient temperature, load fluctuation) will change continuously with charge and discharge cycles. The fixed parameter strategy, due to its inability to perceive the real-time state of the battery, easily leads to overcharging, undercharging, or local overheating, exacerbating degradation phenomena such as lithium dendrite growth and SEI film thickening, and significantly shortening the battery life. In addition, users' demands for charging speed, safety, and scenario adaptability are becoming increasingly diverse (such as emergency fast charging, slow charging at night, low-temperature protection), and the rigid parameter configuration of traditional methods is difficult to meet the dynamic optimization requirements in personalized scenarios.

[0003] Therefore, an adaptive charging power control scheme is desired. Summary of the Invention

[0004] To solve the above technical problems, this application is proposed. The embodiments of this application provide an adaptive charging power control method and system, which capture the voltage and temperature time series signals of the battery in real time through high-precision sensors, combine the user charging behavior data, use a temporal convolutional network to perform multi-scale temporal modeling on the dynamic characteristics of the battery, and extract the high-frequency non-linear fluctuation characteristics of the battery state. At the same time, perform semantic association encoding on the user behavior data to establish a mapping relationship between user needs and battery health status. Then, introduce a cross-modal feature interaction mechanism to achieve dynamic coupling analysis of the battery state and user needs, and finally generate a charging voltage regulation instruction that best matches the current battery state and meets the user scenario requirements through a decoder, dynamically adjust the charging parameters, and achieve the coordinated optimization of charging efficiency, battery life, and personalized scenarios.

[0005] According to one aspect of this application, an adaptive charging power control method is provided, which includes:

[0006] Real-time monitoring of the battery state data of the power supply to be charged through a battery sensor, where the battery state data includes voltage data and temperature data;

[0007] Obtaining user charging behavior data, where the user charging behavior data includes charging time, charging frequency, charging duration, charging mode, and charging target power;

[0008] Perform time - dimension sequence fusion encoding on the battery state data to obtain a multi - dimensional fusion representation of the battery state;

[0009] Perform mapping - association encoding based on charging information on the user charging behavior data to obtain a representation of the associated encoding of the user charging behavior;

[0010] Perform cross - modal feature independent - constraint interaction analysis on the multi - dimensional fusion representation of the battery state and the representation of the associated encoding of the user charging behavior to obtain a representation of the battery state - charging behavior interaction encoding;

[0011] Based on the representation of the battery state - charging behavior interaction encoding, determine a recommended decoded value of the target charging voltage to control the charging of the power supply to be charged.

[0012] According to another aspect of the present application, an adaptive charging power supply control system is provided, which includes:

[0013] A battery state data real - time monitoring module for real - time monitoring of the battery state data of the power supply to be charged through a battery sensor, where the battery state data includes voltage data and temperature data;

[0014] A user charging behavior data acquisition module for acquiring user charging behavior data, where the user charging behavior data includes charging time, charging frequency, charging duration, charging mode, and charging target power;

[0015] A battery state sequence fusion encoding module for performing time - dimension sequence fusion encoding on the battery state data to obtain a multi - dimensional fusion representation of the battery state;

[0016] A user charging behavior associated encoding module for performing mapping - association encoding based on charging information on the user charging behavior data to obtain a representation of the associated encoding of the user charging behavior;

[0017] A battery state - charging behavior interaction encoding module for performing cross - modal feature independent - constraint interaction analysis on the multi - dimensional fusion representation of the battery state and the representation of the associated encoding of the user charging behavior to obtain a representation of the battery state - charging behavior interaction encoding;

[0018] A target charging voltage recommended decoded value determination module for determining a recommended decoded value of the target charging voltage based on the representation of the battery state - charging behavior interaction encoding to control the charging of the power supply to be charged.

[0019] Compared with the prior art, an adaptive charging power supply control method and system provided by the present application captures the voltage and temperature time-series signals of the battery in real time through high-precision sensors, combines the user charging behavior data, and uses a temporal convolutional network to perform multi-scale temporal modeling on the dynamic characteristics of the battery, extracting the high-frequency non-linear fluctuation characteristics of the battery state. At the same time, semantic association encoding is performed on the user behavior data to establish a mapping relationship between user needs and the battery health state. Then, a cross-modal feature interaction mechanism is introduced to achieve dynamic coupling analysis of the battery state and user needs. Finally, a charging voltage regulation command that best matches the current battery state and meets the user scenario requirements is generated through a decoder, dynamically adjusting the charging parameters to achieve the collaborative optimization of charging efficiency, battery life, and personalized scenarios. Description of the Drawings

[0020] The above and other objects, features, and advantages of the present application will become more apparent by describing the embodiments of the present application in more detail in conjunction with the accompanying drawings. The drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification, which are used to explain the present application together with the embodiments of the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0021] Figure 1 It is a flowchart of the adaptive charging power supply control method according to the embodiment of the present application.

[0022] Figure 2 It is a schematic diagram of the data flow of the adaptive charging power supply control method according to the embodiment of the present application.

[0023] Figure 3 It is a flowchart of performing temporal fusion encoding on the battery state data in the time dimension to obtain a multi-dimensional fusion representation of the battery state in the adaptive charging power supply control method according to the embodiment of the present application.

[0024] Figure 4 It is a flowchart of performing cross-modal feature independent constraint interaction analysis on the multi-dimensional fusion representation of the battery state and the associated encoding representation of the user charging behavior to obtain an interaction encoding representation of the battery state - charging behavior in the adaptive charging power supply control method according to the embodiment of the present application.

[0025] Figure 5 It is a system block diagram of the adaptive charging power supply control system according to the embodiment of the present application. Detailed Embodiments

[0026] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. Like reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0027] With the popularization of the application of lithium-ion batteries in fields such as electric vehicles, mobile electronic devices, and energy storage systems, the demand for intelligent battery charging management has become increasingly important. The traditional constant current / constant voltage (CC / CV) charging method relies on preset fixed parameters and is controlled based on a static battery model. However, in actual operation, the battery exhibits very dynamic characteristics: its internal electrochemical state (such as polarization phenomenon, capacity fade) and external usage conditions (such as ambient temperature, load change) will change continuously during the charge and discharge process. Since these fixed parameters cannot reflect the real-time condition of the battery, using this strategy is prone to problems such as overcharging, undercharging, or local overheating, accelerating the deterioration processes such as lithium dendrite formation and solid electrolyte interface film thickening, thus greatly shortening the battery life. At the same time, users' demands for charging speed, safety, and adaptation to specific scenarios are becoming more and more diverse (such as emergency fast charging, slow charging at night, low-temperature protection, etc.), and the fixed parameter settings of the traditional method are difficult to meet the optimization requirements in different scenarios.

[0028] In recent years, adaptive charging technology has become a research hotspot, and its core goal is to dynamically adjust charging parameters by real-time sensing the battery state to balance charging efficiency and battery health. Existing methods mainly rely on single-dimensional optimization: some studies construct state observers based on equivalent circuit models or electrochemical impedance spectroscopy (EIS), but they have strong model dependence and high computational complexity; other solutions use neural networks to predict the state of health (SOH) of the battery, but they are not deeply integrated with user behavior characteristics, resulting in insufficient flexibility of the strategy. In addition, the battery state data shows corresponding temporal variations over time, and traditional feature extraction methods are difficult to efficiently capture the high-frequency non-linear temporal fluctuation characteristics of battery states such as battery voltage / temperature.

[0029] Aiming at the problems of the limitations of static parameters and insufficient adaptability to user scenarios in traditional charging strategies, in the technical solution of this application, an adaptive charging control scheme based on collaborative parsing of multi-modal data is constructed. It captures the voltage and temperature time-series signals of the battery in real time through high-precision sensors, combines user charging behavior data (such as charging time period, mode preference, etc.), and uses a temporal convolutional network to perform multi-scale temporal modeling on the dynamic characteristics of the battery to extract the high-frequency non-linear fluctuation characteristics of the battery state. At the same time, a fully connected network is used to perform semantic association encoding on the user behavior data to establish a mapping relationship between user needs and the battery health state. On this basis, a cross-modal feature interaction mechanism is introduced to realize the dynamic coupling parsing of the battery state and user needs while preserving the semantic independence of the battery physical characteristics and user behavior. Finally, a charging voltage regulation command that best matches the current battery state and meets the user scenario requirements is generated through a decoder to dynamically adjust the charging parameters and achieve the collaborative optimization of charging efficiency, battery life, and personalized scenarios.

[0030] This application proposes an adaptive charging power supply control method. Figure 1 The flowchart of the adaptive charging power supply control method according to an embodiment of this application. Figure 2 The data flow diagram of the adaptive charging power supply control method according to an embodiment of this application. As Figure 1 and Figure 2 shown, the adaptive charging power supply control method according to an embodiment of this application includes: S110, monitoring the battery state data of the power supply to be charged in real time through a battery sensor, where the battery state data includes voltage data and temperature data; S120, obtaining user charging behavior data, where the user charging behavior data includes charging time, charging frequency, charging duration, charging mode, and charging target power; S130, performing temporal fusion encoding on the battery state data in the time dimension to obtain a multi-dimensional fusion representation of the battery state; S140, performing mapping association encoding based on charging information on the user charging behavior data to obtain a mapping association encoding representation of the user charging behavior; S150, performing cross-modal feature independent restriction interaction parsing on the multi-dimensional fusion representation of the battery state and the mapping association encoding representation of the user charging behavior to obtain an interaction encoding representation of the battery state - charging behavior; S160, based on the interaction encoding representation of the battery state - charging behavior, determining a target charging voltage recommended decoding value to control the power supply to be charged for charging.

[0031] In the above adaptive charging power supply control method, in step S110, the battery state data of the power supply to be charged is monitored in real time through a battery sensor, and the battery state data includes voltage data and temperature data. It should be understood that as lithium-ion batteries are increasingly widely used in fields such as electric vehicles, mobile electronic devices, and energy storage systems, the demand for intelligent battery charging management has become increasingly important. The traditional constant current / constant voltage (CC / CV) charging method relies on preset fixed parameters and is controlled based on a static battery model. However, in actual operation, the battery exhibits very dynamic characteristics: its internal electrochemical state (such as polarization phenomenon, capacity fade) and external usage conditions (such as ambient temperature, load changes) will change continuously during the charge and discharge process. Since these fixed parameters cannot reflect the real-time condition of the battery, using this strategy is likely to cause problems such as overcharging, undercharging, or local overheating, accelerating the deterioration processes such as lithium dendrite formation and solid electrolyte interface film thickening, thus greatly shortening the battery life. Therefore, it is necessary to monitor the battery state data of the power supply to be charged in real time through a battery sensor, especially the voltage data and temperature data. Specifically, the state of the battery, especially the voltage and temperature, is an important indicator for evaluating the battery health. The voltage directly reflects the equilibrium state of the electrochemical reaction inside the battery, while the temperature is closely related to the heat generation and heat dissipation inside the battery. By accurately measuring these two parameters, the working state of the battery can be effectively monitored, abnormal situations can be detected in time, and corresponding measures can be taken to prevent potential safety risks. For example, when it is detected that the battery voltage drops rapidly or the temperature rises sharply, this may be a sign of a short circuit or other faults inside the battery, and at this time, the charging should be stopped immediately to avoid more serious damage. To achieve real-time monitoring of the battery voltage and temperature, high-precision sensors usually need to be integrated into the battery management system. These sensors can collect the voltage signals at both ends of the battery and the temperature information on the surface or inside of the battery at a relatively high frequency, and transmit these data to the control system for processing.

[0032] In the above adaptive charging power supply control method, in step S120, user charging behavior data is obtained. The user charging behavior data includes charging time, charging frequency, charging duration, charging mode, and charging target power. It should be understood that different time periods often correspond to different electricity consumption demands and environmental conditions. For example, users may choose to charge at night when the grid load is low and the power supply is relatively stable; during the day, especially during peak hours, the grid load is large, which may affect the charging efficiency. In addition, some regions implement time-of-use electricity price policies during specific time periods to encourage users to charge during off-peak hours to save costs. By analyzing the user's charging time, their preferences can be identified, and the charging strategy can be adjusted accordingly to meet the user's needs while taking into account economy and environmental protection. Moreover, frequent charging operations will accelerate the aging process of the battery and shorten its service life. Therefore, understanding the user's charging frequency helps predict the battery's degradation rate and take corresponding maintenance measures. For example, if it is found that a certain user's charging frequency is significantly higher than the average level, the user can be advised to adopt a more reasonable charging method, such as reducing unnecessary partial charging times and avoiding keeping the battery in a high state of charge for a long time, thereby slowing down the battery aging speed. In addition, the charging duration is directly related to charging efficiency and safety. Generally speaking, although fast charging can replenish a large amount of energy for the device in a short time, this high-intensity charging process will put greater pressure on the battery, increase the risk of overheating, and even may cause safety accidents. On the contrary, slow charging, although it takes a longer time, puts less pressure on the battery and is beneficial to extending the battery life. By recording and analyzing the user's charging duration, it can help determine whether the current charging rate is appropriate and dynamically adjust the charging parameters according to the actual situation to ensure that both the user's needs are met and the safe operation of the battery is guaranteed. In addition, the choice of charging mode also has a profound impact on battery performance. Currently, the common charging modes on the market mainly include two basic forms: constant current (CC) and constant voltage (CV). In addition, there are also various variants such as fast charging and slow charging. Under different modes, the electrochemical reaction mechanisms occurring inside the battery are different, which in turn affect the battery's charging efficiency, heat generation situation, and aging rate. By collecting the user's charging mode preferences, a more suitable charging solution for the user can be better matched. For example, fast charging service can be provided when the user urgently needs to use the device, and a more gentle slow charging method can be recommended in other cases to achieve the best comprehensive benefits. Finally, the charging target power refers to the target capacity percentage to which the user hopes to charge the battery. This value is usually closely related to the user's specific application scenario. For example, long-distance travelers may hope to fully charge the battery to 100% to ensure that there is no power shortage during the journey; while daily commuters may only need to maintain at a relatively low level to meet the needs of a day. By monitoring the charging target power, the system can more accurately grasp the user's actual needs and optimize the charging process accordingly.For example, when the target battery level set by the user is relatively low, the charging rate can be appropriately reduced to reduce the stress on the battery; while in the case of a full charge required, the charging speed can be increased as much as possible within a safe range. It should be noted that all the above types of user charging behavior data have a certain degree of randomness and individual differences. This means that even the same device may exhibit completely different usage patterns in the hands of different users. Therefore, simply relying on preset fixed parameters to manage the battery is obviously insufficient. Only by fully considering and utilizing this personalized charging behavior data can we truly achieve customized services for each user, maximize the user experience, and at the same time ensure the safety and durability of the battery. Specifically, for parameters such as charging time, charging frequency, and charging duration, the charging and discharging status of the battery can be monitored in real time through built-in current sensors and voltage sensors; while for the selection of the charging mode, it often depends on the settings or operation records of the user on the device; as for the target battery level, it is usually set by the user or determined based on the system default value. All this information needs to be accurately recorded and transmitted to the corresponding processing unit for further analysis.

[0033] Figure 3 A flowchart for performing time - dimension sequence fusion encoding on the battery state data to obtain a multi - dimensional fusion representation of the battery state in the adaptive charging power control method according to an embodiment of the present application. As Figure 3 shown, in the embodiment of the present application, step S130 of performing time - dimension sequence fusion encoding on the battery state data to obtain a multi - dimensional fusion representation of the battery state includes: S131, after integrating the battery state data into a voltage data time - sequence vector and a temperature data time - sequence vector according to the time dimension, respectively passing the voltage data time - sequence vector and the temperature data time - sequence vector through a time - sequence encoder based on a time - convolutional network to obtain a voltage data time - sequence encoding vector and a temperature data time - sequence encoding vector; S132, performing feature - level fusion on the voltage data time - sequence encoding vector and the temperature data time - sequence encoding vector to obtain a multi - dimensional fusion representation vector of the battery state as the multi - dimensional fusion representation of the battery state.

[0034] Specifically, in step S131, after integrating the battery state data into a voltage data time series vector and a temperature data time series vector according to the time dimension, the voltage data time series vector and the temperature data time series vector are respectively passed through a time series encoder based on a temporal convolutional network to obtain a voltage data time series encoded vector and a temperature data time series encoded vector. It should be understood that considering that traditional charging strategies are difficult to capture the dynamic time series characteristics of battery voltage / temperature during charging and discharging due to relying on static parameter models, and degradation phenomena such as internal polarization effects and capacity attenuation in the battery often reflect in the voltage / temperature time series signals through high-frequency nonlinear fluctuations. Existing time series modeling and feature extraction methods are limited in their ability to extract features from long sequence data due to the gradient attenuation problem, and it is difficult to effectively analyze the multi-scale correlation features in the battery transient response. Therefore, in the technical solution of this application, after integrating the battery state data into a voltage data time series vector and a temperature data time series vector according to the time dimension, the voltage data time series vector and the temperature data time series vector are respectively passed through a time series encoder based on a temporal convolutional network to obtain a voltage data time series encoded vector and a temperature data time series encoded vector. It is worth mentioning that through the causal dilated convolution structure of the temporal convolutional network, local features at different time scales are extracted layer by layer, and its parallel computing advantage is used to overcome the time series modeling bottleneck of traditional recurrent neural networks. This step aims to establish a high-precision representation of the dynamic evolution of the battery state, and capture short-term mutations (such as voltage jumps at the moment of charging) and long-term trends (such as temperature rise drift caused by cyclic aging) in the voltage / temperature sequence through the dilated convolution kernel of the temporal convolutional network, so as to reveal the implicit relationship between the real-time health state of the battery and the external operating conditions. In this way, it helps to improve the expression ability of battery state features, enabling the subsequent fusion and decision-making module to dynamically optimize the charging parameters based on finer-grained time series information, avoiding control biases caused by overfitting static models, and providing a physically interpretable feature basis for cross-modal interaction analysis, ultimately achieving the synchronous adaptation of the charging strategy to the battery degradation dynamics and user scenario requirements.

[0035] Specifically, in step S132, the voltage data time-series encoded vector and the temperature data time-series encoded vector are subjected to feature-level fusion to obtain a multi-dimensional fusion representation vector of the battery state as the multi-dimensional fusion representation of the battery state. It should be understood that considering that voltage changes directly reflect the dynamic balance of the electrochemical reactions inside the battery (such as the lithium-ion insertion / extraction rate), while temperature fluctuations are closely related to ohmic heat, polarization heat, and environmental heat exchange, the two together constitute a multi-dimensional observation window for the battery health state. However, traditional methods often split the potential correlations between features (such as the voltage sudden drop caused by increased polarization at high temperatures) when independently processing the time-series correlations of voltage and temperature data, making it difficult to construct a global state representation. Therefore, in the technical solution of this application, the voltage data time-series encoded vector and the temperature data time-series encoded vector are subjected to feature-level fusion to obtain a multi-dimensional fusion representation vector of the battery state. By performing feature-level fusion to jointly model the time-series encoded vectors of voltage and temperature in the latent space, it aims to break through the limitations of single-signal analysis and utilize the complementarity of voltage-temperature to reveal the coupling relationship between the battery degradation mechanism (such as the temperature rise lag effect accompanied by the thickening of the SEI film) and external operating conditions (such as the co-temperature and pressure oscillation caused by rapid charging). This step non-linearly correlates the transient response features in the voltage sequence (such as the voltage inflection point at the end of charging) with the thermal inertia features in the temperature sequence (such as the temperature accumulation caused by heat dissipation delay) through tensor splicing or attention weighting mechanism, thereby constructing a fusion representation covering multiple dimensions of electro-thermal-aging. This design effectively enhances the model's ability to identify complex hidden fault modes (such as abnormal voltage caused by local overheating), enabling subsequent cross-modal interaction analysis to generate charging strategies based on more complete battery state information, avoiding control risks caused by misjudgment of single signals (such as ignoring potential thermal runaway hazards only relying on voltage), and precisely balancing the charging rate and temperature rise constraint through electro-thermal coupling analysis, ultimately realizing the upgrade of battery health management from "single-dimensional threshold control" to "multi-physical field collaborative optimization".

[0036] In an embodiment of the present application, step S140, which performs mapping and association encoding based on charging information on the user charging behavior data to obtain a user charging behavior association encoding representation, includes: passing the user charging behavior data through a charging information mapping and association encoder based on a fully connected layer to obtain a user charging behavior association encoding vector as the user charging behavior association encoding representation. It should be understood that since user charging behavior data (such as charging period preference, target battery level setting) contains personalized requirements and scenario constraint information, traditional charging strategies are difficult to convert discrete charging habits (such as frequent fast charging, maintaining low battery level at night) into quantifiable control parameters due to the lack of semantic parsing of user behavior. Although existing solutions attempt to introduce user preferences, they often directly use raw data splicing or simple rule mapping, resulting in the implicit association between user intentions and battery states not being deeply mined (for example, the 80% target battery level set by the user may correspond to different health protection priorities, depending on their charging frequency and usage scenario). Therefore, in the technical solution of the present application, the user charging behavior data is passed through a charging information mapping and association encoder based on a fully connected layer to obtain a user charging behavior association encoding vector. By constructing a charging information mapping and association encoder through a fully connected layer, multi-dimensional and heterogeneous user behavior data (including temporal attributes such as charging time, statistical features such as charging frequency) is mapped to a unified semantic space, aiming to deconstruct the potential decision-making logic behind user behavior (such as the "emergency fast charging" mode implies a strong demand for charging speed, while the "low temperature protection" mode gives priority to temperature rise constraints). This step captures the complex interactions between user behavior features through non-linear transformation (such as high-frequency short-time charging is often negatively correlated with high target battery level demand), generating a low-dimensional encoding vector with semantic coherence, enabling user requirements to be embedded in the subsequent cross-modal interaction process in an interpretable form. In practical applications, this design breaks through the rigid mapping limitations of traditional rule engines, enabling the charging strategy to not only recognize explicit instructions (such as charging mode selection), but also infer implicit requirements (such as predicting the user's future usage scenario based on the historical charging period distribution), ultimately achieving the dynamic alignment of the battery health management strategy and user behavior habits - for example, when it is recognized that the user has long adopted the "slow charging at night" mode, the charging curve is automatically optimized to utilize the valley electricity period to extend the trickle charging stage, thereby maximizing the delay of battery capacity decay while meeting the user's scenario preferences.

[0037] Figure 4 It is a flowchart for performing cross-modal feature independent restriction interaction parsing of battery state - charging behavior on the multi-dimensional fusion representation of the battery state and the user charging behavior association encoding representation to obtain a battery state - charging behavior interaction encoding representation in the adaptive charging power control method according to an embodiment of the present application. As Figure 4As shown, in an embodiment of the present application, the step S150 performs battery state-charging behavior cross-modal feature independence restriction interactive analysis on the multidimensional fusion representation of the battery state and the user charging behavior associated coding representation to obtain a battery state-charging behavior interactive coding representation, including: S151, transforming the multidimensional fusion representation vector of the battery state to obtain a set of battery state principal component linear transformation feature coding vectors; S152, performing inter-modal independence constraint analysis on the set of battery state principal component linear transformation feature coding vectors and the user charging behavior associated coding vector to determine a set of battery state-charging behavior principal component inter-modal independence soft constraint factors; S153, based on the set of battery state-charging behavior principal component inter-modal independence soft constraint factors, dynamically and adaptively aggregate the set of battery state principal component linear transformation feature coding vectors and the user charging behavior associated coding vector to obtain a battery state-charging behavior interactive coding vector as the battery state-charging behavior interactive coding representation. It should be understandable that the multi-dimensional fusion representation vector of the battery state and the user charging behavior associated coding vector respectively carry two types of heterogeneous modal information: physical characteristics and semantic needs: the former characterizes the dynamic health status of the battery through electro-thermal coupling analysis, and the latter reveals the user's scenario-based charging intention through behavior mapping. However, traditional cross-modal interaction methods often simply splice or weightedly fuse, ignoring the essential differences between the two types of data in information density (high-frequency sampling of battery time series data and sparse event records of user behavior) and semantic levels (electrochemical degradation mechanisms and user subjective preferences), resulting in redundant features interfering with effective information transmission during the interaction process (such as the charging mode set by the user may mask the actual temperature rise pressure that the battery is currently under). Therefore, in the technical solution of the present application, the multi-dimensional fusion representation vector of the battery state and the user charging behavior associated coding vector are further subjected to feature interaction analysis of independent constraint representation between modalities to obtain the battery state-charging behavior interaction coding vector. Specifically, the process of feature interaction analysis of independent constraint representation between modalities is to decouple the multi-dimensional fusion representation vector of battery state through principal component analysis (PCA), remove minor variation components (such as random noise) and extract core electro-thermal coupling principal components (such as the correlation pattern between voltage inflection point and temperature rise rate), generate low-dimensional orthogonal feature subspace, and eliminate internal redundancy; then align the numerical scale of battery state principal components and user behavior encoding through linear transformation (for example, normalize the voltage fluctuation amplitude to a range close to the user's charging frequency) to avoid imbalance of modal contribution due to magnitude difference. In other words, by constructing an interactive feature space of battery state and user behavior, the independence relationship between modalities is explicitly modeled at the same time. The complementarity and independence boundaries of the two types of modalities in the long-range correlation global structure are captured through the intrinsic normative potential representation of the topological order matrix (for example, the dynamic balance relationship between the user's fast charging demand and the physical constraint of the maximum allowable charging current of the battery).The dynamic calculation mechanism of the soft constraint factor further realizes the adaptive adjustment of the fusion weight: when the battery health state conflicts highly with the user's needs (such as the user initiating fast charging in a high-temperature environment), the independent coding matrix suppresses high-risk interaction paths through stability optimization (reducing the contribution weight of the fast charging mode to the voltage boost instruction), while strengthening the effective feature correlation within the safety boundary (increasing the charging efficiency within the allowable temperature rise range). In terms of actual effects, this design breaks through the rigid limitations of traditional hard constraint fusion, enabling the charging strategy to have the ability of multi-objective dynamic trade-off. For example, when the user selects the "low-temperature protection" mode, the temperature rise rate feature in the main components of the battery state enhances its influence on the decision-making through independent soft constraints, dynamically suppressing the charging current fluctuations that may cause lithium plating in a low-temperature environment; while in the "emergency fast charging" scenario, the semantic weight of the user behavior coding is increased, driving the decoder to maximize the charging power within the battery safety threshold. The finally generated battery state-charging behavior interaction coding vector mines the potential optimization space through fine-grained response interaction (such as identifying the conditions for extending the trickle charging stage during night charging to utilize the natural heat dissipation of the temperature), ensuring both battery health and accurately adapting to the user scenario requirements, realizing the paradigm upgrade from "single-modal dominance" to "multi-modal game optimization".

[0038] In an embodiment of the present application, the step S151 of performing transformation processing on the battery state multi-dimensional fusion characterization vector to obtain a set of battery state principal component linear transformation feature coding vectors includes: S1511, performing principal component analysis on the battery state multi-dimensional fusion characterization vector to obtain a set of battery state principal component feature coding vectors; S1512, performing linear transformation on each battery state principal component feature coding vector in the set of battery state principal component feature coding vectors to obtain a set of battery state principal component linear transformation feature coding vectors, wherein each battery state principal component linear transformation feature coding vector in the set of battery state principal component linear transformation feature coding vectors has the same feature scale as the user charging behavior association coding vector.

[0039] Specifically, the step S1511 of performing principal component analysis on the battery state multi-dimensional fusion characterization vector to obtain a set of battery state principal component feature coding vectors is represented by the battery state principal component analysis formula as follows: Wherein, is the battery state multi-dimensional fusion characterization vector, is the principal component analysis, is the covariance matrix of the battery state multi-dimensional fusion characterization vector, is the battery state multi-dimensional feature principal component orthogonal matrix, , and are respectively the 1st, 2nd, and A multi-dimensional feature principal component coding vector of the battery state, is a diagonal matrix of the multi-dimensional features of the battery state, , and are respectively , and corresponding eigenvalues, is 's transpose matrix, The number of eigenvalues in the multi-dimensional fusion characterization vector of the battery state, Indicates extracting diagonal elements. It should be understood that the multi-dimensional fusion characterization vector of the battery state is a high-dimensional feature set of voltage and temperature data generated through time-series fusion coding, and there are naturally problems of information redundancy and multicollinearity. Specifically, physical parameters such as voltage and temperature will form complex non-linear coupling relationships during the dynamic charge and discharge process, resulting in a large number of inefficient or repetitive associated dimensions in the original feature space. Such high-dimensional redundant features not only increase the computational complexity but may also obscure the key variation features of the true degradation mode of the battery, thereby affecting the accuracy of subsequent cross-modal interaction analysis. Principal component analysis can, through orthogonal transformation means, strip off secondary noise interference from a mathematical level, project the multi-dimensional fusion characterization vector of the battery state into the principal component space composed of orthogonal basis vectors, thereby extracting a low-dimensional feature set that characterizes the core dynamic characteristics of the battery state, providing a redundancy-removed and high-information-density input basis for subsequent cross-modal interaction. In this way, a compact characterization form of the battery state features is constructed, and through principal component analysis, the dimensionality reduction reconstruction and information purification of the feature space are realized to eliminate the linear correlation between multi-dimensional time-series features such as voltage and temperature, and map the original high-dimensional fusion vector to a group of orthogonal principal component axes, so that a few leading principal components can centrally reflect the main variation modes in the dynamic characteristics of the battery (such as polarization effect, temperature rise rate change, etc.). This process not only optimizes the dimensionality structure of the feature space, but more importantly, selects the key principal components sensitive to the battery health state through the principle of maximizing information entropy, enabling the subsequent cross-modal interaction mechanism to focus on the core variation direction of the battery physical characteristics and avoiding the interference of invalid features on the user behavior semantic analysis. After implementing principal component analysis, the multi-dimensional fusion characterization vector of the battery state is transformed into a set of low-dimensional orthogonal principal component feature coding vectors. This operation significantly reduces the feature dimension, alleviates the "curse of dimensionality" problem, and at the same time improves the feature independence by eliminating multicollinearity. The set of battery state principal component feature coding vectors can more efficiently characterize the core modes of dynamic behaviors such as battery voltage fluctuations and temperature drifts by retaining the components with the largest variance contribution in the original data.

[0040] Specifically, in step S1512, linear transformation is performed on each battery state principal component feature coding vector in the set of battery state principal component feature coding vectors to obtain a set of battery state principal component linear transformation feature coding vectors, where each battery state principal component linear transformation feature coding vector in the set of battery state principal component linear transformation feature coding vectors has the same feature scale as the user charging behavior associated coding vector, and is represented by the battery state linear transformation formula: Wherein, denotes performing linear transformation on each battery state principal component feature coding vector in , and are respectively the 1st, 2nd, th, and th battery state principal component linear transformation feature coding vectors in the set of battery state principal component linear transformation feature coding vectors, It is a set of linear transformation feature encoding vectors of the main components of the battery state. It should be understood that the main component feature encoding vectors of the battery state and the associated encoding vectors of the user's charging behavior are respectively derived from two types of heterogeneous modal data, namely physical signals and user behaviors, and there are significant differences in the original feature distribution scales. The main component features of the battery (such as the main components of voltage fluctuations and the main components of temperature rise trends) are usually generated by dimensionality reduction of high-sampling-rate time-series data, and the numerical range is affected by the physical dimensions of electrochemical parameters; while the user behavior encoding vectors (such as charging frequency and target battery level) are generated based on the statistics of discrete behavior events, and the numerical distribution is constrained by semantic mapping rules. If cross-modal interaction is directly carried out, the scale difference will cause the model to be overly sensitive to high-magnitude modalities (such as the main components of voltage amplitude), while the contribution of low-magnitude modalities (such as charging mode encoding) is suppressed, leading to information bias and decision-making bias in the feature interaction process. Through scale normalization and space mapping, the dimensionality barrier between the main component features of the battery and the user behavior encoding can be eliminated, and a fairness foundation for cross-modal interaction can be constructed. Specifically, by applying a linear transformation matrix to the main component feature encoding vectors of the battery, they are projected onto a feature scale space that matches the user behavior encoding vectors, so that the numerical ranges and distribution variances of the two modalities tend to be consistent. This operation not only achieves alignment at the numerical level, but also preliminarily establishes a correlation mapping channel between the physical characteristics of the battery and the semantics of user behavior through the adjustment of the geometric structure of the feature space, providing a compatibility guarantee for the collaborative fusion of multi-dimensional features in subsequent cross-modal interaction analysis, and avoiding the problem of unbalanced modal contributions caused by scale differences. After linear transformation, the set of main component feature encoding vectors of the battery and the associated encoding vectors of the user behavior are strictly aligned in terms of feature scale. The numerical ranges of the main component features of the battery (such as the low-order main components reflecting polarization effects and the high-order main components representing temperature rise inertia) are compressed or expanded to the same magnitude interval as the user behavior encoding (such as the preference encoding of charging time periods and the quantified values of target battery levels), ensuring that the weight distribution of the two types of features during cross-modal interaction is not interfered by the original dimensions. The feature space with unified scale enables the interaction model to fairly evaluate the game relationship between the battery health constraints and the user scenario requirements. For example, when the fast charging mode is activated, the main component feature of battery temperature rise and the user's charging duration encoding can dynamically balance the charging rate and the risk of thermal runaway at the same scale, thereby improving the robustness and interpretability of the charging strategy in multi-objective optimization scenarios.

[0041] In an embodiment of the present application, step S152, performing an inter-modal independence constraint analysis on the set of battery state principal component linear transformation feature coding vectors and the user charging behavior associated coding vectors to determine a set of battery state-charging behavior principal component inter-modal independence soft constraint factors, includes: S1521, inputting the user charging behavior associated coding vector and each battery state principal component linear transformation feature coding vector in the set of battery state principal component linear transformation feature coding vectors into an inter-modal independence modeling unit to obtain a set of battery state-charging behavior principal component inter-modal independence coding matrices; S1522, performing a canonical stability optimization on each battery state-charging behavior principal component inter-modal independence coding matrix in the set of battery state-charging behavior principal component inter-modal independence coding matrices to obtain a set of battery state-charging behavior principal component inter-modal independence canonical optimization coding matrices, where the process of the canonical stability optimization is determined by a battery state-charging behavior eigenvector composed of each eigenvalue of the battery state-charging behavior principal component inter-modal independence coding matrix; S1523, calculating a set of battery state-charging behavior principal component inter-modal independence soft constraint factors based on the set of battery state-charging behavior principal component inter-modal independence canonical optimization coding matrices.

[0042] Specifically, step S1521, inputting the user charging behavior associated coding vector and each battery state principal component linear transformation feature coding vector in the set of battery state principal component linear transformation feature coding vectors into an inter-modal independence modeling unit to obtain a set of battery state-charging behavior principal component inter-modal independence coding matrices, which is represented by an inter-modal independence modeling formula as: Where and are feature mapping functions, such as linear mapping or non-linear kernel functions, is the user charging behavior associated coding vector, is the th length of the battery state principal component linear transformation feature coding vector, is the Battery state - charging behavior principal component modal independence encoding matrix. It should be understood that considering the linear transformation feature encoding vectors of the battery state principal components (representing electrochemical dynamic characteristics) and the associated encoding vectors of the user's charging behavior (reflecting the user's scenario requirements), although they have been processed by scale alignment, the two essentially belong to the physical modality and the behavior modality, and their deep - level association is highly non - linear and asymmetric. Traditional cross - modal fusion methods rely on implicitly learning the relationship between modalities and are vulnerable to interference from redundant features (such as the unnecessary association between the battery temperature rise trend and the user's charging time preference), resulting in the masking of key complementary information by noise during the interaction process. By explicitly modeling modal independence, the inherent attributes and interaction dependencies of the two modalities can be separated, avoiding decision - making biases caused by information redundancy during the feature coupling process (for example, misjudging the user's fast - charging demand as the battery's high - rate tolerance). In this way, a measurable modal independence constraint space is constructed, and the complementary boundary between the battery's physical state and the user's behavior requirements is dynamically analyzed through a learnable encoding matrix. Specifically, the modal independence modeling unit does not simply calculate statistical independence indicators, but maps the interaction relationship between the battery principal component features (such as the principal component of voltage fluctuation, the principal component of temperature rise rate) and the user behavior encoding (such as charging mode preference, target charge setting) to a low - dimensional orthogonal subspace by parameterizing the generation of the battery state - charging behavior principal component modal independence encoding matrix. This explicit modeling mechanism can quantify the dynamic balance between modal independence and correlation (for example, identifying the positive synergistic effect between the user's fast - charging demand and the battery's polarization voltage recovery ability), providing a directional guidance for subsequent cross - modal aggregation and ensuring that the interaction process focuses on feature combinations with physical meaning and scenario adaptability. After modal independence modeling, a set of battery state - charging behavior principal component modal independence encoding matrices is output, which is essentially a dynamic weight map describing the interaction intensity of the two types of modal features. Each battery state - charging behavior principal component modal independence encoding matrix corresponds to the independence quantification result of a specific principal component and the user behavior encoding (such as the negative correlation intensity between a certain temperature rise principal component and the user's charging frequency), and the interaction stability is controlled by constraining the matrix spectral radius or eigenvalue distribution.

[0043] Specifically, in step S1522, each battery state - charging behavior principal component modal independence encoding matrix in the set of battery state - charging behavior principal component modal independence encoding matrices is subjected to canonical stability optimization to obtain a set of battery state - charging behavior principal component modal independence canonical optimization encoding matrices, where the process of the canonical stability optimization is determined by the battery state - charging behavior eigenvector composed of the respective eigenvalues of the battery state - charging behavior principal component modal independence encoding matrix, and is expressed by the canonical stability optimization formula as: where is the Each eigenvalue of the inter-modal independence encoding matrix of the battery state-charging behavior principal component, is the battery state-charging behavior eigenvector composed of the above eigenvalues, is matrix multiplication, is the battery state-charging behavior canonical association encoding vector, The first set of encoding matrices in the set of battery state-charging behavior principal component modal independence norm optimization The battery state-charging behavior main component modal independence norm optimization coding matrix. It should be understood that the battery state-charging behavior main component modal independence coding matrix, as a topological order matrix, has a potential risk of instability in the modal interaction structure due to its asymmetry and long-range correlation characteristics. The eigenvalue distribution of the original battery state-charging behavior main component modal independence coding matrix may cause the eigenvector phase accumulation offset due to local disturbances (such as instantaneous fluctuations in battery voltage or sudden changes in user charging mode), destroying the global topological consistency of modal interaction. For example, the strong asymmetric coupling between fast charging demand and temperature rise principal component may cause the battery state-charging behavior main component modal independence coding matrix to have eigenvalue divergence in dynamic scenarios, resulting in violent oscillation of cross-modal feature aggregation weights, affecting the stability and continuity of the charging strategy. This step reconstructs the geometric structure of the battery state-charging behavior main component modal independence coding matrix to enhance its anti-interference ability by introducing topological invariant constraints based on the eigennormal potential. Specifically, the battery state-charging behavior eigenvector composed of the eigenvalues ​​is used as the mathematical basis of the canonical potential to map the global topological association of the asymmetric matrix to the canonical association coding vector space. The phase accumulation is calibrated by the first-order canonical potential term to eliminate the interference of the high-order perturbation term on the long-range correlation structure, and the structural stability optimization of the battery state-charging behavior main component modal independence coding matrix under asymmetric conditions is achieved. In this way, it is ensured that local characteristic disturbances (such as temporary adjustment of the charging target by the user) during the modal interaction process will not destroy the global balance relationship between the battery health constraint and the user's needs, and maintain the smooth transition of the cross-modal fusion weights. After the canonical stability optimization, the set of the battery state-charging behavior main component modal independence canonical optimization coding matrices obtained constrains the spectral distribution of its autocorrelation matrix through the eigennormal potential, so that the eigenvalue modulus and phase angle of the matrix meet the stability boundary conditions (such as limiting the real part of the eigenvalue to be negative to ensure the convergence of the dynamic system). This structural stability ensures the robustness of the cross-modal interaction mechanism, enabling the charging strategy to maintain a dynamic balance between battery health and user needs under complex working conditions, and avoiding control command oscillations or strategy failures caused by local data fluctuations.

[0044] Specifically, in step S1523, based on the independence criterion between the battery state-charging behavior principal component modes, optimize the set of encoding matrices, and calculate the set of soft constraint factors for the independence between the battery state-charging behavior principal component modes, which is expressed by the calculation formula of the soft constraint factor for the independence between the battery state-charging behavior principal component modes as follows: Wherein, represents the square of the matrix F norm, Denote the set of soft constraint factors for the independence between the battery state - charging behavior principal component modes. It should be understood that although the optimized battery state - charging behavior principal component mode - independence optimized coding matrix ensures the structural stability of modal interaction through topological constraints, its essence is still a static mathematical representation and is difficult to directly adapt to the real - time changes in battery dynamic characteristics and user behavior. For example, when there is a sudden change in the battery health state (such as a sharp increase in the temperature rise rate) or the user temporarily switches the charging mode (such as from regular charging to fast charging), the modal independence constraint with fixed weights may lead to decision - making lag or rigidity and cannot accurately capture the impact of scenario dynamics on the interaction intensity. Therefore, it is necessary to convert the modal independence information contained in the optimized battery state - charging behavior principal component mode - independence optimized coding matrix into dynamically adjustable constraint parameters to cope with the non - linear evolution of the multi - modal game relationship under complex working conditions. Specifically, through the quantitative extraction of the soft constraint factors for the independence between the battery state - charging behavior principal component modes, an adaptive mapping mechanism between the modal independence constraint and the dynamic feature interaction can be established. Among them, based on the eigen - canonical potential distribution of the optimized battery state - charging behavior principal component mode - independence optimized coding matrix (such as the stability margin reflected by the real part of the eigenvalue and the phase accumulation rate corresponding to the imaginary part), a differentiable calculation function for the soft constraint factors for the independence between the battery state - charging behavior principal component modes is designed to convert the topological invariant features of the optimized battery state - charging behavior principal component mode - independence optimized coding matrix (such as the long - range correlation strength and the local perturbation suppression ability) into dynamic weight coefficients that can act on the feature aggregation process. Its core goal is to translate the static topological structure of the optimized battery state - charging behavior principal component mode - independence optimized coding matrix into flexible constraint rules, so that cross - modal interaction can dynamically balance the game relationship between independence maintenance and correlation utilization according to real - time scenario requirements (such as battery polarization degree and user charging urgency), and avoid the sub - optimal decision - making problem caused by hard threshold segmentation. After the soft constraint factors for the independence between the battery state - charging behavior principal component modes, each optimized battery state - charging behavior principal component mode - independence optimized coding matrix is mapped to the dynamic constraint strength parameters between the corresponding principal component and the user behavior mode. For example, when the independence between a certain temperature - rise principal component of the battery and the user's fast - charging demand coding is relatively high (manifested as a negative shift in the real part of the coding matrix eigenvalue), the soft constraint factors for the independence between the battery state - charging behavior principal component modes automatically enhance the independence weight of this principal component and suppress the excessive demand for the charging current in the fast - charging mode; conversely, when the independence between the two decreases (the eigenvalue approaches neutrality), the soft constraint factors for the independence between the battery state - charging behavior principal component modes moderately relax the constraint and allow the user's demand to dominate the charging strategy within the safety threshold.This dynamic adjustment mechanism enables the charging control system to flexibly adjust the decision boundary of multimodal fusion according to the coupling strength between the real-time degradation state of the battery and the user's scenario intention, preventing the control risks caused by overfitting a single modality (such as overcharging caused by blindly following the user's demand), making full use of the modality complementarity to improve the charging efficiency and scenario adaptability, and finally achieving the adaptive convergence of the global optimization goal.

[0045] In an embodiment of the present application, in step S153, based on the set of independence soft constraint factors between the battery state-charging behavior principal component modalities, dynamically and adaptively aggregating the set of battery state principal component linear transformation feature encoding vectors and the user charging behavior association encoding vector to obtain a battery state-charging behavior interaction encoding vector as the battery state-charging behavior interaction encoding representation, includes: S1531, inputting the user charging behavior association encoding vector and each battery state principal component linear transformation feature encoding vector in the set of battery state principal component linear transformation feature encoding vectors into a feature interaction response unit to obtain a set of battery state-charging behavior principal component modality fine-grained response interaction encoding vectors; S1532, based on the set of independence soft constraint factors between the battery state-charging behavior principal component modalities, dynamically and adaptively aggregating the set of battery state-charging behavior principal component modality fine-grained response interaction encoding vectors to obtain the battery state-charging behavior interaction encoding vector.

[0046] Specifically, in step S1531, inputting the user charging behavior association encoding vector and each battery state principal component linear transformation feature encoding vector in the set of battery state principal component linear transformation feature encoding vectors into a feature interaction response unit to obtain a set of battery state-charging behavior principal component modality fine-grained response interaction encoding vectors, which is represented by the feature interaction response formula as: Where, represents concatenation processing, 、 and represent element-wise multiplication, element-wise addition, and element-wise division by position, and represent trainable weight matrices and trainable bias vectors, represents the The fine-grained response interaction coding vector between the battery state-charging behavior principal component modes. It should be understood that traditional coarse-grained fusion methods (such as weighted summation or simple concatenation) are difficult to capture the implicit coupling mechanism between the battery principal components (such as the polarization voltage recovery rate, temperature rise trend) and user behaviors (such as charging period preference, target battery charge setting). For example, the dynamic game relationship between the transient response of specific battery principal components when the fast charging mode is activated and the user's charging duration requirement. Directly performing shallow interaction easily leads to the loss of key complementary information and cannot accurately quantify the differential influence weight of user scenario requirements on the charging strategy. Through the feature interaction response unit, the fine-grained semantic association parsing of cross-modal features is realized, and a deep interaction network between the battery physical characteristics and user behavior intentions is constructed from the microscopic dimension. Specifically, the feature interaction response unit uses non-linear mapping functions (such as cross-attention mechanism, tensor product operation) to gradually explore the dependence and complementarity of the battery principal component features (such as the time series pattern of the voltage fluctuation principal component) and user behavior encoding (such as the charging frequency statistical feature) in the subspace, and generates the fine-grained response interaction coding vector between the battery state-charging behavior principal component modes. In this way, translating the user scenario requirement (such as emergency fast charging) into the dynamic constraint of the battery principal component (such as the maximum allowable charging current slope) can establish an interpretable feature-level interaction rule, providing a decision basis with both physical rationality and scenario adaptability for the adaptive charging strategy. After being processed by the feature interaction response unit, the set of fine-grained response interaction coding vectors between the battery state-charging behavior principal component modes can accurately depict the dynamic association strength between user behaviors and battery principal components. For example, when the user selects the "low temperature protection" mode, the fine-grained response interaction coding vector between the battery state-charging behavior principal component modes can strengthen the suppression effect of the temperature rise principal component on the charging power and weaken the weight of the voltage principal component related to the charging speed; while in the "efficient energy replenishment" scenario, the synergistic effect of the high-rate tolerance principal component of the battery and the user's target battery charge is preferentially activated. This fine-grained interaction mechanism enables the charging strategy to dynamically allocate the game priority between the battery health constraint and user requirements in the feature dimension according to the real-time working conditions, avoiding the strategy rigidity caused by global hard rules and exploring potential optimization space through local feature interaction (such as moderately increasing the trickle charging stage voltage by using the battery polarization relaxation characteristic), and finally achieving the adaptive balance of charging efficiency, battery life and scenario requirements.

[0047] Specifically, in step S1532, based on the set of independence soft constraint factors between the battery state-charging behavior principal component modes, dynamic adaptive aggregation is performed on the set of fine-grained response interaction coding vectors between the battery state-charging behavior principal component modes to obtain the battery state-charging behavior interaction coding vector, which is represented by the dynamic adaptive aggregation formula: Where, is function, is the number of vectors in the set of fine-grained response interaction coding vectors between the battery state-charging behavior principal component modes, is the battery state-charging behavior interaction coding vector. It should be understood that through the independence soft constraint factor-driven dynamic aggregation mechanism between the battery state-charging behavior principal component modes, selective fusion of fine-grained interaction features and global information optimization integration can be achieved. Specifically, based on the set of independence soft constraint factors between the battery state-charging behavior principal component modes (reflecting the independence strength between each principal component and the user behavior coding), the fine-grained response interaction coding vectors between the battery state-charging behavior principal component modes are weighted and aggregated, dynamically adjusting the contribution weights of different interaction paths, so as to retain cross-modal complementary information while eliminating redundant noise through soft constraints, generating a battery state-charging behavior interaction coding vector with both high information density and low redundancy, providing an accurate basis for optimizing the charging strategy for the decoder.

[0048] In the embodiment of the present application, the step S160, based on the battery state-charging behavior interaction coding representation, determining the target charging voltage recommended decoding value to control the charged power supply to charge, includes: S161, passing the battery state-charging behavior interaction coding vector through a charging voltage controller based on a decoder to obtain the target charging voltage recommended decoding value; S162, based on the target charging voltage recommended decoding value, controlling the charged power supply to charge.

[0049] Specifically, in the step S161, passing the battery state-charging behavior interaction coding vector through a charging voltage controller based on a decoder to obtain the target charging voltage recommended decoding value. It should be understood that although the battery state-charging behavior interaction coding vector, as an abstract representation of multi-modal deep fusion, contains the dynamic game relationship between battery health constraints and user scenario requirements, its high-dimensional non-linear characteristics cannot be directly mapped into executable charging voltage control instructions. Traditional control strategies rely on threshold triggering or linear regression and are difficult to analyze the multi-objective optimization logic (such as the trade-off relationship between fast charging demand and temperature rise constraint) hidden in such complex interaction features. Therefore, in the technical solution of the present application, the battery state-charging behavior interaction coding vector is further passed through a charging voltage controller based on a decoder to obtain the target charging voltage recommended decoding value. Specifically, the charging voltage controller based on a decoder decodes the compressed complementary information (such as the association mode between the user's fast charging intention and the battery polarization voltage recovery rate) in the battery state-charging behavior interaction coding vector into a continuously adjustable charging voltage trajectory through a multi-layer transposed convolution or fully connected network, thereby obtaining the target charging voltage recommended decoding value.

[0050] Specifically, in step S162, based on the target charging voltage recommended decoding value, the power supply to be charged is controlled to charge. It should be understood that after determining the target charging voltage recommended decoding value, precise control of the power supply to be charged can be implemented accordingly. This means that during the entire charging cycle, the charging parameters will be dynamically adjusted according to the target charging voltage recommended decoding value to adapt to the changes in the battery state. For example, if it is detected that the battery temperature rises too fast, the system may temporarily reduce the charging current and gradually resume it after the temperature stabilizes; if it is found that the battery voltage is approaching the full charge threshold, it will automatically switch to the trickle charging mode to prevent damage caused by overcharging. This real-time response mechanism ensures that the battery can obtain the most suitable charging conditions regardless of its state. Compared with the traditional fixed-parameter charging method, the advantage of this method lies in its flexibility and self-adaptability. It can respond in real time to the small changes in the battery state, timely adjust the charging strategy, and avoid problems caused by improper parameter settings. At the same time, since the charging voltage is dynamically generated according to the actual needs of the battery, unnecessary energy loss can be minimized to improve the overall charging efficiency. Moreover, the charging control method based on the target charging voltage recommended decoding value helps to extend the service life of the battery. A reasonable charging strategy can not only avoid damage to the battery caused by overcharging or over-discharging, but also effectively alleviate various adverse phenomena during the battery aging process, such as lithium dendrite growth and solid electrolyte interface film thickening. By precisely controlling the charging voltage, the electrochemical balance inside the battery can be maintained to the greatest extent, and the decline rate of the battery performance can be delayed. This is particularly important for application scenarios that require long-term stable operation, such as electric vehicles and energy storage power stations.

[0051] In summary, the adaptive charging power supply control method based on the embodiments of the present application is clarified. It captures the voltage and temperature time series signals of the battery in real time through high-precision sensors, combines user charging behavior data, and uses a temporal convolutional network to perform multi-scale temporal modeling on the dynamic characteristics of the battery to extract the high-frequency non-linear fluctuation characteristics of the battery state. At the same time, semantic association coding is performed on the user behavior data to establish a mapping relationship between user needs and the battery health state. Then, a cross-modal feature interaction mechanism is introduced to realize the dynamic coupling analysis of the battery state and user needs. Finally, a charging voltage regulation instruction that best matches the current battery state and meets the user scenario needs is generated through a decoder, and the charging parameters are dynamically adjusted to achieve the coordinated optimization of charging efficiency, battery life, and personalized scenarios.

[0052] Figure 5 It is a system block diagram of the adaptive charging power supply control system according to the embodiments of the present application. As Figure 5As shown, the adaptive charging power supply control system 100 according to an embodiment of the present application includes: a battery state data real-time monitoring module 110 for real-time monitoring of the battery state data of the power supply to be charged through a battery sensor, where the battery state data includes voltage data and temperature data; a user charging behavior data acquisition module 120 for acquiring user charging behavior data, where the user charging behavior data includes charging time, charging frequency, charging duration, charging mode, and charging target power; a battery state time-series fusion encoding module 130 for performing time-dimension time-series fusion encoding on the battery state data to obtain a multi-dimensional fusion representation of the battery state; a user charging behavior association encoding module 140 for performing mapping association encoding based on charging information on the user charging behavior data to obtain a user charging behavior association encoding representation; a battery state-charging behavior interaction encoding module 150 for performing cross-modal feature independent restriction interaction analysis on the multi-dimensional fusion representation of the battery state and the user charging behavior association encoding representation to obtain a battery state-charging behavior interaction encoding representation; and a target charging voltage recommended decoding value determination module 160 for determining a target charging voltage recommended decoding value based on the battery state-charging behavior interaction encoding representation to control the charging of the power supply to be charged.

[0053] Here, those skilled in the art can understand that the specific operations of each step in the above adaptive charging power supply control system have been introduced in detail in the description of the Figures 1 to 4 adaptive charging power supply control method above, and thus, the repeated description thereof will be omitted.

[0054] In summary, the adaptive charging power supply control system based on the embodiment of the present application is clarified. It captures the voltage and temperature time-series signals of the battery in real time through high-precision sensors, combines user charging behavior data, and uses a temporal convolutional network to perform multi-scale temporal modeling on the dynamic characteristics of the battery to extract the high-frequency non-linear fluctuation characteristics of the battery state. At the same time, semantic association encoding is performed on user behavior data to establish a mapping relationship between user needs and the battery health state. Then, a cross-modal feature interaction mechanism is introduced to achieve dynamic coupling analysis of the battery state and user needs. Finally, a charging voltage regulation instruction that best matches the current battery state and meets the user scenario requirements is generated through a decoder to dynamically adjust the charging parameters and achieve the coordinated optimization of charging efficiency, battery life, and personalized scenarios.

Claims

1. An adaptive charging power supply control method, characterized in that Including: Real-time monitoring of the battery state data of the power supply to be charged through a battery sensor, where the battery state data includes voltage data and temperature data; Obtaining user charging behavior data, where the user charging behavior data includes charging time, charging frequency, charging duration, charging mode, and charging target power; Performing time-series fusion encoding on the battery state data in the time dimension to obtain a multi-dimensional fusion representation of the battery state; Performing mapping association encoding on the user charging behavior data based on charging information to obtain a user charging behavior association encoding representation; Performing cross-modal feature independent constraint interaction analysis on the multi-dimensional fusion representation of the battery state and the user charging behavior association encoding representation to obtain a battery state-charging behavior interaction encoding representation; Based on the battery state-charging behavior interaction encoding representation, determining a target charging voltage recommended decoding value to control the charging of the power supply to be charged; Wherein, based on the battery state-charging behavior interaction encoding representation, determining a target charging voltage recommended decoding value to control the charging of the power supply to be charged includes: Passing the battery state-charging behavior interaction encoding vector through a charging voltage controller based on a decoder to obtain a target charging voltage recommended decoding value; Based on the target charging voltage recommended decoding value, controlling the charging of the power supply to be charged; Wherein, the charging voltage controller based on the decoder decodes the complementary information compressed in the battery state-charging behavior interaction encoding vector into a continuously adjustable charging voltage trajectory through a multi-layer transposed convolution or fully connected network, thereby obtaining a target charging voltage recommended decoding value.

2. The adaptive charging power supply control method according to claim 1, wherein Performing time-series fusion encoding on the battery state data in the time dimension to obtain a multi-dimensional fusion representation of the battery state, including: After integrating the battery state data into a voltage data time-series vector and a temperature data time-series vector in the time dimension, passing the voltage data time-series vector and the temperature data time-series vector through a time-series encoder based on a time convolutional network to obtain a voltage data time-series encoding vector and a temperature data time-series encoding vector; Performing feature-level fusion on the voltage data time-series encoding vector and the temperature data time-series encoding vector to obtain a multi-dimensional fusion representation vector of the battery state as the multi-dimensional fusion representation of the battery state.

3. The adaptive charging power supply control method according to claim 2, characterized in that Performing mapping association encoding on the user charging behavior data based on charging information to obtain a user charging behavior association encoding representation, including: Passing the user charging behavior data through a charging information mapping association encoder based on a fully connected layer to obtain a user charging behavior association encoding vector as the user charging behavior association encoding representation.

4. The adaptive charging power supply control method according to claim 3, wherein Performing cross-modal feature independent constraint interaction analysis on the multi-dimensional fusion representation of the battery state and the user charging behavior association encoding representation to obtain a battery state-charging behavior interaction encoding representation, including: Performing transformation processing on the multi-dimensional fusion representation vector of the battery state to obtain a set of battery state principal component linear transformation feature encoding vectors; Perform an inter-modal independence constraint analysis on the set of battery state principal component linear transformation feature encoding vectors and the user charging behavior associated encoding vectors to determine a set of battery state-charging behavior principal component inter-modal independence soft constraint factors; Based on the set of battery state-charging behavior principal component inter-modal independence soft constraint factors, perform dynamic adaptive aggregation on the set of battery state principal component linear transformation feature encoding vectors and the user charging behavior associated encoding vectors to obtain a battery state-charging behavior interaction encoding vector as the battery state-charging behavior interaction encoding representation.

5. The adaptive charging power supply control method according to claim 4, wherein, Perform a transformation process on the battery state multi-dimensional fusion representation vector to obtain a set of battery state principal component linear transformation feature encoding vectors, including: Perform principal component analysis on the battery state multi-dimensional fusion representation vector to obtain a set of battery state principal component feature encoding vectors; Perform a linear transformation on each battery state principal component feature encoding vector in the set of battery state principal component feature encoding vectors to obtain a set of battery state principal component linear transformation feature encoding vectors, where each battery state principal component linear transformation feature encoding vector in the set of battery state principal component linear transformation feature encoding vectors has the same feature scale as the user charging behavior associated encoding vector.

6. The adaptive charging power supply control method according to claim 5, wherein Perform an inter-modal independence constraint analysis on the set of battery state principal component linear transformation feature encoding vectors and the user charging behavior associated encoding vectors to determine a set of battery state-charging behavior principal component inter-modal independence soft constraint factors, including: Input the user charging behavior associated encoding vector and each battery state principal component linear transformation feature encoding vector in the set of battery state principal component linear transformation feature encoding vectors into an inter-modal independence modeling unit to obtain a set of battery state-charging behavior principal component inter-modal independence encoding matrices; Perform canonical stability optimization on each battery state-charging behavior principal component inter-modal independence encoding matrix in the set of battery state-charging behavior principal component inter-modal independence encoding matrices to obtain a set of battery state-charging behavior principal component inter-modal independence canonical optimization encoding matrices, where the process of the canonical stability optimization is determined by the battery state-charging behavior eigenvectors composed of the respective eigenvalues of the battery state-charging behavior principal component inter-modal independence encoding matrix; Based on the set of battery state-charging behavior principal component inter-modal independence canonical optimization encoding matrices, calculate the set of battery state-charging behavior principal component inter-modal independence soft constraint factors.

7. The adaptive charging power supply control method according to claim 6, characterized in that, Based on the set of battery state-charging behavior principal component inter-modal independence soft constraint factors, perform dynamic adaptive aggregation on the set of battery state principal component linear transformation feature encoding vectors and the user charging behavior associated encoding vectors to obtain a battery state-charging behavior interaction encoding vector as the battery state-charging behavior interaction encoding representation, including: Input each battery state principal component linear transformation feature encoding vector in the set of the user charging behavior associated encoding vector and the battery state principal component linear transformation feature encoding vectors into the feature interaction response unit to obtain a set of battery state-charging behavior principal component mode fine-grained response interaction encoding vectors; Based on the set of battery state-charging behavior principal component mode independence soft constraint factors, perform dynamic adaptive aggregation on the set of battery state-charging behavior principal component mode fine-grained response interaction encoding vectors to obtain the battery state-charging behavior interaction encoding vector.

8. An adaptive charging power supply control system, characterized in that, Including: A battery state data real-time monitoring module, configured to real-time monitor the battery state data of the power supply to be charged through a battery sensor, where the battery state data includes voltage data and temperature data; A user charging behavior data acquisition module, configured to acquire user charging behavior data, where the user charging behavior data includes charging time, charging frequency, charging duration, charging mode, and charging target power; A battery state time-series fusion encoding module, configured to perform time-series fusion encoding on the battery state data in the time dimension to obtain a battery state multi-dimensional fusion representation; A user charging behavior associated encoding module, configured to perform mapping association encoding on the user charging behavior data based on charging information to obtain a user charging behavior associated encoding representation; A battery state-charging behavior interaction encoding module, configured to perform battery state-charging behavior cross-modal feature independent restriction interaction parsing on the battery state multi-dimensional fusion representation and the user charging behavior associated encoding representation to obtain a battery state-charging behavior interaction encoding representation; A target charging voltage recommended decoding value determination module, configured to determine a target charging voltage recommended decoding value based on the battery state-charging behavior interaction encoding representation to control the power supply to be charged to charge; Wherein, determining a target charging voltage recommended decoding value based on the battery state-charging behavior interaction encoding representation to control the power supply to be charged to charge includes: Pass the battery state-charging behavior interaction encoding vector through a charging voltage controller based on a decoder to obtain a target charging voltage recommended decoding value; Based on the target charging voltage recommended decoding value, control the power supply to be charged to charge; Wherein, the charging voltage controller based on a decoder decodes the compressed complementary information in the battery state-charging behavior interaction encoding vector into a continuously adjustable charging voltage trajectory through a multi-layer transposed convolution or fully connected network, so as to obtain a target charging voltage recommended decoding value.

9. The adaptive charging power supply control system according to claim 8, wherein The battery state time-series fusion encoding module is configured to: After integrating the battery state data into a voltage data time-series vector and a temperature data time-series vector in the time dimension, respectively pass the voltage data time-series vector and the temperature data time-series vector through a time-series encoder based on a time convolutional network to obtain a voltage data time-series encoding vector and a temperature data time-series encoding vector; Perform feature-level fusion on the voltage data time-series encoding vector and the temperature data time-series encoding vector to obtain a battery state multi-dimensional fusion representation vector as the battery state multi-dimensional fusion representation.

Images