New energy charging pile charging process early warning system based on artificial intelligence

Abstract

The invention discloses a new energy charging pile charging process early warning system based on artificial intelligence. Comprising a dynamic charging parameter real-time acquisition and multi-dimensional analysis unit, a feature deep mining and space-time correlation mapping unit, a charging state evaluation and abnormal trend pre-judgment unit, a multi-source abnormal information fusion and grading judgment unit, an intelligent early warning strategy generation and dynamic adjustment unit and an early warning information real-time push and execution feedback unit. The dynamic charging parameter real-time acquisition and multi-dimensional analysis unit acquires various types of parameters, after the parameters are processed by the feature deep mining and time-space correlation mapping unit, the charging state evaluation and abnormal trend pre-judgment unit evaluates the state and pre-judges the trend, and the multi-source abnormal information fusion and grading judgment unit performs fusion grading. The intelligent early warning strategy generation and dynamic adjustment unit generates an adjustment strategy, and the early warning information real-time push and execution feedback unit pushes information and receives feedback. The system adapts to various scenes, and the charging safety and stability are improved.

Description

Technical Field

[0001] The present invention relates to the field of new energy charging pile charging, and in particular to an artificial intelligence-based new energy charging pile charging process early warning system. Background Art

[0002] With the rapid increase in the popularity of new energy vehicles, charging piles, as key supporting facilities, are attracting increasing attention for their safety and stability during the charging process. During the charging process, voltage fluctuations, current surges, and abnormally high battery and pile temperatures frequently occur, which can easily lead to risks such as equipment failure, battery damage, and even fire. Traditional methods that rely on manual inspections or simple sensor monitoring are no longer able to cope with complex and changing charging scenarios and cannot accurately capture potential anomalies in real time. In this context, integrating artificial intelligence technology into charging pile charging process warnings, and analyzing multi-dimensional parameters through neural network models, has become an important direction for improving the efficiency and accuracy of warnings.

[0003] Existing technologies have significant deficiencies in early warning of charging pile charging processes. On the one hand, the correlation between feature extraction and status assessment is weak. Most systems only analyze a single parameter or a simple combination of parameters, failing to deeply explore the spatiotemporal correlations between multiple parameters such as voltage, current, and temperature. This leads to a one-sided assessment of the charging status and makes it difficult to identify potential complex anomalies in advance. On the other hand, the dynamic adaptability of abnormality warning strategies is insufficient. Warning thresholds and strategies are mostly fixed settings, without considering the differences between different charging pile models, battery types, and environmental conditions. When the charging scenario changes, warning lags or false alarms are prone to occur, and the warning logic cannot be adjusted in a timely manner based on real-time feedback. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides an artificial intelligence-based new energy charging pile charging process early warning system.

[0005] The technical solution adopted by the present invention is an artificial intelligence-based new energy charging pile charging process early warning system, including: a dynamic charging parameter real-time acquisition and multi-dimensional analysis unit, which is used to obtain the charging pile output voltage, charging current, battery temperature, charging time, pile body temperature and power grid fluctuation frequency; a feature deep mining and spatiotemporal correlation mapping unit, the input end of which is connected to the output end of the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit, and performs feature extraction and spatiotemporal correlation analysis on the above parameters; a charging state evaluation and abnormal trend prediction unit, the input end of which is connected to the output end of the feature deep mining and spatiotemporal correlation mapping unit, and evaluates the charging state and predicts abnormal trends based on the extracted features; A multi-source abnormal information fusion and classification judgment unit, the input end of which is connected to the output end of the charging state assessment and abnormal trend prediction unit, and fuses and classifies the abnormal information; an intelligent early warning strategy generation and dynamic adjustment unit, the input end of which is connected to the output end of the multi-source abnormal information fusion and classification judgment unit, and generates and adjusts the early warning strategy according to the classification result; an early warning information real-time push and execution feedback unit, the input end of which is connected to the output end of the intelligent early warning strategy generation and dynamic adjustment unit, and pushes early warning information and receives execution feedback, and the output end of this unit forms a closed-loop connection with the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit and the feature deep mining and spatiotemporal correlation mapping unit respectively.

[0006] Furthermore, the feature depth mining and spatiotemporal correlation mapping unit adopts an improved feature extraction model, and the model formula is: ,in, is the fusion eigenvalue at time t, 、 、 is the adaptive weight coefficient, The voltage at time t is the adaptively optimized long short-term memory neural network and current The feature extraction results of is the battery temperature at time t, Pile temperature The temperature effect function, For charging time, The grid fluctuation frequency The frequency impact function of the charging state assessment and abnormal trend prediction unit adopts a state assessment model, the formula is: ,in, is the charge state evaluation value at time t, Processing function for the improved gray wolf optimization algorithm, is the back propagation neural network function, is the weight matrix, is the bias term.

[0007] Furthermore, the feature depth mining and spatiotemporal correlation mapping unit also includes a parameter dynamic adaptation subunit, which adopts the adaptation model formula: ,in, is the network parameter adaptation value at time t, To adaptively optimize the inverse function of the long short-term memory neural network, are the partial derivatives of the fused eigenvalue with respect to voltage and current, are the changes in voltage and current at time t respectively; the charging state assessment and abnormal trend prediction unit also includes an abnormal sensitivity adjustment subunit, and the model formula is: ,in, is the abnormal sensitivity at time t, Processing function for the improved gray wolf optimization algorithm, is the charge state evaluation value at time t, is the derivative function of the back-propagation neural network, are the temperature changes of the battery and pile respectively, is the temperature influence coefficient.

[0008] Furthermore, the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit adopts a parameter analysis model, and the formula is: ,in, is the multi-dimensional analysis result at time t, is the number of parameter types, is the analytical coefficient of the i-th parameter, is the collected value of the i-th parameter at time t, is the time decay function of the i-th parameter; the feature enhancement model of the feature depth mining and spatiotemporal correlation mapping unit is: ,in, is the enhanced eigenvalue, is the enhancement coefficient, is the activation function, They are 、 The multi-dimensional analysis results of the moment, is the fusion eigenvalue at time t, The adaptively optimized long short-term memory neural network 、 Feature extraction results at that moment.

[0009] Furthermore, the abnormal trend prediction model of the charging state assessment and abnormal trend prediction unit is: ,in, for The predicted state value at time t, is the charge state evaluation value at time t, Processing function for the improved gray wolf optimization algorithm, is the back propagation neural network function, is the gradient of the state evaluation value at time t, is the prediction time interval, is the coefficient of fluctuation, is the parameter change vector at time t; the fusion model of the multi-source abnormal information fusion and classification judgment unit is: ,in, is the abnormal fusion value at time t, To predict the number of steps, is the weight of the j-th step prediction, is the temperature correction function corresponding to the j-th step prediction, is the battery temperature at time t, is the temperature of the pile at time t.

[0010] Furthermore, the strategy generation model of the intelligent early warning strategy generation and dynamic adjustment unit is: ,in, is the early warning strategy at time t, Based on the abnormal fusion value and classification thresholds The policy function, is the adjustment coefficient, To adaptively optimize the long short-term memory neural network function, for The strategy adjustment amount at the moment; the feedback processing model of the real-time push of the warning information and the execution feedback unit is: ,in, is the feedback processing result at time t, For execution results The feedback function, is the feedback coefficient, First, the product of the early warning strategy Str (t) and the feedback coefficient κ is processed by the back propagation BP algorithm, and then the result is input into the improved gray wolf optimization The algorithm performs nested operations for optimization calculations.

[0011] Furthermore, the charging state assessment and abnormal trend prediction unit includes: a state feature matching subunit, which receives feature data output from the feature deep mining and spatiotemporal correlation mapping unit, compares the feature data with the preset normal charging state feature library one by one, determines the matching degree by calculating the Euclidean distance between the feature vectors, and screens out feature combinations with a matching degree lower than a set threshold; an abnormal trend quantification subunit, which performs time series analysis on the screened low-matching feature combinations, calculates the slope and curvature of the feature as the charging time changes, and converts the abnormal trend into a quantifiable numerical indicator in combination with the rate of change of the battery temperature and the pile body temperature; an optimization evaluation subunit, which optimizes the evaluation results of the back propagation neural network using the improved gray wolf optimization algorithm, and reduces the deviation between the evaluation results and the actual charging state to the minimum range by adjusting the network weights and bias parameters; a result output subunit, which organizes the optimized charging state assessment results and abnormal trend prediction indicators according to the set data format, and transmits them to the multi-source abnormal information fusion and grading judgment unit.

[0012] Furthermore, the multi-source abnormal information fusion and grading judgment unit includes: an abnormal information receiving subunit, which receives various types of abnormal information output from the charging status assessment and abnormal trend prediction unit, including abnormal characteristics, trend indicators and evaluation scores, and verifies the integrity and format of the information to eliminate invalid information; a multi-dimensional fusion subunit, which classifies abnormal information from different sources according to the time dimension and feature dimension, and uses the weighted average method to fuse similar abnormal information at the same time point to generate a comprehensive abnormal vector; a grading threshold determination subunit, which dynamically sets different levels of abnormal thresholds according to the model of the charging pile, battery type and charging stage, and each threshold corresponds to a set warning level; a grading result output subunit, which compares the fused abnormal information with the set threshold, determines the corresponding warning level, and transmits the grading result to the intelligent warning strategy generation and dynamic adjustment unit.

[0013] Furthermore, the intelligent early warning strategy generation and dynamic adjustment unit includes: a basic strategy library calling sub-unit, which calls the corresponding initial early warning strategy from the preset basic early warning strategy library according to the early warning level output by the multi-source abnormal information fusion and hierarchical judgment unit, and the initial strategy includes the early warning method, push object and processing suggestions; a strategy dynamic adjustment sub-unit, which combines the early warning information real-time push and execution feedback results returned by the feedback unit, and uses the adaptive optimization long short-term memory neural network to adjust the initial strategy, correct the early warning frequency and push content, so that the strategy is more in line with the actual situation; a multi-scenario adaptation sub-unit, which analyzes the environmental parameters of the current charging scenario, including ambient temperature, humidity and grid stability, and performs scenario adaptation optimization on the adjusted strategy to ensure consistent early warning effects in different scenarios; a final strategy output sub-unit, which transmits the final early warning strategy after dynamic adjustment and scenario adaptation to the early warning information real-time push and execution feedback unit, and stores the strategy parameters in the historical database.

[0014] The AI-based early warning system for charging process of new energy charging piles includes: Step S1: The dynamic charging parameter real-time acquisition and multi-dimensional analysis unit continuously acquires and performs multi-dimensional analysis on the charging pile output voltage, charging current, battery temperature, charging time, pile body temperature, and grid fluctuation frequency; Step S2: transmitting the parsed parameters to the feature deep mining and spatiotemporal correlation mapping unit to perform feature deep mining and spatiotemporal correlation mapping analysis to form a feature data set; Step S3: Input the characteristic data set into the charging state assessment and abnormal trend prediction unit to perform charging state assessment and abnormal trend prediction to obtain abnormal trend information; Step S4: Send the abnormal trend information to the multi-source abnormal information fusion and classification determination unit, implement multi-source abnormal information fusion and classification determination, and generate an abnormal classification result; Step S5: Transmit the abnormality classification results to the intelligent early warning strategy generation and dynamic adjustment unit, formulate an intelligent early warning strategy and perform dynamic adjustment to produce an adaptive early warning strategy; Step S6: The adapted warning strategy is delivered to the warning information real-time push and execution feedback unit to push the warning information in real time and receive execution feedback, and the feedback information is respectively transmitted back to the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit and the feature deep mining and spatiotemporal correlation mapping unit to form a complete warning closed-loop process.

[0015] Beneficial effects: The present invention proposes an artificial intelligence-based new energy charging pile charging process warning system. The system obtains comprehensive parameters through real-time collection of dynamic charging parameters and a multi-dimensional analysis unit. The system deeply mines the spatiotemporal correlation of multiple parameters through feature deep mining and spatiotemporal correlation mapping units. The charging state assessment and abnormal trend prediction unit then accurately assesses the state and predicts the trend. The multi-source abnormal information fusion and classification judgment unit fuse the classified abnormal information. The intelligent early warning strategy generation and dynamic adjustment unit generates and adjusts the strategy. Finally, the early warning information is pushed in real time and the execution feedback unit pushes the information and forms a closed loop, significantly improving the accuracy and timeliness of the early warning. At the same time, the system uses an adaptively optimized long short-term memory neural network to conduct in-depth mining and spatiotemporal correlation analysis of multi-dimensional parameters. The improved gray wolf optimization-back propagation neural network performs state assessment and abnormal trend prediction based on this, strengthens the correlation between the two, and can identify complex abnormalities. In response to the problem of insufficient dynamic adaptability of the early warning strategy, the intelligent early warning strategy generation and dynamic adjustment unit dynamically adjusts the strategy based on the abnormal classification results and feedback information to adapt to different scenarios and avoid early warning lags or false alarms. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a diagram of the system unit composition of the present invention; Figure 2 This is a flow chart of the system operation of the present invention. DETAILED DESCRIPTION

[0017] It should be noted that, unless there is a conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The application is further described in detail below with reference to the drawings and specific embodiments.

[0018] like Figure 1 As shown in the figure, the artificial intelligence-based new energy charging pile charging process early warning system includes: Dynamic charging parameter real-time acquisition and multi-dimensional analysis unit, used to obtain charging pile output voltage, charging current, battery temperature, charging time, pile body temperature and grid fluctuation frequency; Specifically, the real-time collection and multi-dimensional analysis unit for dynamic charging parameters serves as the data input core of the entire early warning system. Its role is to provide comprehensive and accurate raw data support for subsequent feature mining, status assessment, and other steps. The technical parameters covered by this unit include charging pile output voltage (typically ranging from 200V to 1000V), charging current (generally 10A to 200A), battery temperature (normally between 25°C and 45°C), charging time (ranging from a few minutes to several hours), pile body temperature (typically no more than 60°C), and grid fluctuation frequency (approximately 50Hz, with an allowable fluctuation range of ±0.5Hz). These parameters directly reflect the power transmission status, equipment operating conditions, and real-time battery response during the charging process. Their accuracy and timeliness are crucial to the performance of the entire early warning system. The collection and analysis of these multi-dimensional parameters provides the system with a rich information foundation, enabling more accurate detection of potential anomaly risks.

[0019] The specific implementation process of this unit is as follows: First, various sensors deployed inside the charging pile and on the connecting lines collect electrical parameters such as output voltage and charging current in real time at sampling intervals of 10ms-50ms. Simultaneously, temperature sensors installed on the battery pack surface and the charging pile housing collect battery and pile body temperatures at sampling intervals of 1s-5s. Grid fluctuation frequency is collected by a frequency monitoring module connected to the grid at sampling intervals of 0.1s. After filtering, the collected raw data is transmitted to the analysis module. According to preset rules, the analysis module converts voltage and current data into derived parameters such as power and power, temperature data into characteristic quantities such as temperature change rate, and grid fluctuation frequency into indicators such as fluctuation amplitude. After analysis, all parameters are integrated according to timestamps to form a structured dataset. This dataset is sent to the feature deep mining and spatiotemporal correlation mapping unit via the data transmission interface and backed up to a local storage device for 30 days to meet the needs of data traceability and system optimization.

[0020] A feature deep mining and spatiotemporal correlation mapping unit, the input of which is connected to the output of the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit, and performs feature extraction and spatiotemporal correlation analysis on the above parameters; Specifically, the feature deep mining and spatiotemporal correlation mapping unit is a key link in the system's data processing and feature extraction. Its core function is to deeply mine the multi-dimensional parameters transmitted by the real-time acquisition of dynamic charging parameters and the multi-dimensional analysis unit, extract information that can reflect the essential characteristics of the charging process, and establish spatiotemporal correlations between these features. The adaptively optimized long short-term memory neural network adopted by this unit can effectively solve the gradient vanishing or gradient explosion problems that occur in traditional neural networks when processing time series data by optimizing the gating mechanism and memory units within the network. It is especially suitable for parameter analysis with time dependence in the charging process. Through feature mining and spatiotemporal correlation mapping of parameters such as voltage, current, and temperature, the original, scattered parameter data can be converted into feature vectors with clear physical meaning and strong correlation, providing high-quality input for subsequent state assessment and abnormality prediction, and significantly improving the system's ability to recognize complex abnormal patterns.

[0021] The specific implementation process of this unit is as follows: First, it receives the structured data set sent by the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit, and performs time alignment processing on the data to ensure that all parameters are consistent in the time dimension. Then, the processed data set is divided into multiple data windows according to the time series, and the length of each window is set to 10s-60s according to the stability of the charging process. Next, each data window is input into the adaptive optimized long short-term memory neural network. Through the coordinated action of the input gate, forget gate, and output gate, the network extracts features such as the voltage change trend, current fluctuation pattern, and temperature rise rate within the window. At the same time, it learns the correlation between different parameters, such as the phase relationship between voltage and current, and the dynamic response relationship between battery temperature and charging current. During the network training process, the network parameters are continuously adjusted through the backpropagation algorithm, so that the network can adapt to the parameter characteristics of different charging pile models, different battery types, and different environmental conditions. The extracted feature vectors and spatiotemporal correlation information are standardized to form a feature matrix of fixed dimension, which is transmitted to the charging status assessment and abnormal trend prediction unit. At the same time, key parameters in the feature extraction process, such as the network's activation function output value and memory unit status, are recorded for subsequent system debugging and performance optimization.

[0022] A charging state assessment and abnormal trend prediction unit, the input of which is connected to the output of the feature deep mining and spatiotemporal correlation mapping unit, and assesses the charging state and predicts abnormal trends based on the extracted features; Specifically, the charging status assessment and abnormal trend prediction unit in the system bears the important responsibility of accurately assessing the charging status and predicting abnormal trends in advance. This unit uses the improved gray wolf optimization algorithm to optimize the initial weights and biases of the back-propagation neural network, effectively solving the problems of slow convergence and easy falling into local optimality of traditional back-propagation neural networks, and improving the learning efficiency and generalization ability of the network. By taking the feature vectors extracted by the feature deep mining and spatiotemporal correlation mapping unit as input, the unit can quantitatively evaluate the normal state, slight abnormal state and serious abnormal state during the charging process, and predict possible abnormal situations 5 minutes to 30 minutes in advance based on historical data and current trends. This assessment and prediction capability enables the system to issue an early warning before the abnormality occurs, buying valuable time for taking preventive measures, thereby reducing the probability of equipment damage and safety accidents.

[0023] The specific implementation process of this unit is as follows: First, it receives the feature matrix transmitted by the Feature Deep Mining and Spatiotemporal Correlation Mapping unit and divides it into a training set and a test set. The training set accounts for 70%-80% and is used for network training, while the test set is used to verify network performance. Next, the backpropagation neural network structure is initialized, with the number of input layer nodes set to the dimension of the feature matrix, the number of hidden layer nodes set to 20-50 based on feature complexity, and the number of output layer nodes set to 3, corresponding to the evaluation values ​​of normal state, slightly abnormal state, and severely abnormal state, respectively. Next, the network's initial weights and biases are optimized using a modified gray wolf optimization algorithm. During the optimization process, the network's prediction error is used as the fitness function. The optimal parameter combination is found by simulating the wolf pack's search, encirclement, and attack behaviors. The number of iterations is set to 50-100. After optimization, the network is trained using the training set, adjusting the weights and biases through the backpropagation algorithm until the network's prediction error is less than a preset threshold (typically 0.01-0.05). The trained network processes the test set and outputs the charging status evaluation value at each time point and the abnormal trend prediction results for a period of time in the future. These results are sent to the multi-source abnormal information fusion and classification judgment unit through the data interface. At the same time, the network's evaluation and prediction records, including evaluation values, prediction times, and error values, are stored for subsequent analysis and optimization.

[0024] Multi-source abnormal information fusion and classification judgment unit, the input end of which is connected to the output end of the charging state assessment and abnormal trend prediction unit to fuse and classify the abnormal information; Specifically, the multi-source abnormal information fusion and grading judgment unit is an important component of the system's abnormal information integration and risk level classification. Its main function is to perform multi-dimensional and multi-level fusion of the abnormal trend prediction results output by the charging status assessment and abnormal trend prediction unit, and to grade the abnormal risk based on the fused information. Since abnormal information during the charging process may come from multiple parameters and multiple evaluation results, a single information source may have errors or one-sidedness. Through multi-source information fusion, various factors can be comprehensively considered to improve the accuracy and reliability of abnormal identification. Grading judgment can take corresponding early warning measures according to the severity of the abnormality, avoid unnecessary panic and waste of resources, and ensure that serious abnormalities can be handled in a timely manner. By effectively fusing and grading abnormal information, this unit provides a clear basis for the generation of intelligent early warning strategies, making early warnings more accurate and efficient.

[0025] The specific implementation process of this unit is as follows: First, it receives abnormal trend prediction results from the Charging State Assessment and Abnormal Trend Prediction Unit, including information such as abnormal probability at different time points, abnormal parameter combinations, and abnormal development rate. Simultaneously, it obtains the latest raw parameter data from the Dynamic Charging Parameter Real-time Collection and Multi-dimensional Analysis Unit as auxiliary verification information. Then, a weighted fusion algorithm is used to fuse the multi-source abnormal information, with the abnormal trend prediction results weighted at 0.7-0.8 and the raw parameter data weighted at 0.2-0.3. The weights are dynamically adjusted based on the importance and reliability of the parameters. During the fusion process, consistency checks are performed on the information from different sources, obvious contradictions are eliminated, and any discrepancies are corrected. After the fusion is complete, the abnormality risk is classified into four levels according to a pre-set grading standard: Level 1 (minor abnormality, no immediate action required, only enhanced monitoring required); Level 2 (general abnormality, requiring attention from operations and maintenance personnel); Level 3 (more serious abnormality, requiring prompt on-site inspection by operations and maintenance personnel); and Level 4 (serious abnormality, requiring immediate charging suspension and emergency measures). The thresholds for the grading standards are set based on the charging pile model, battery tolerance, and safety regulations. For example, when the battery temperature exceeds 45°C and the rate of increase is greater than 2°C / min, it is considered a Level 3 abnormality; when the pile temperature exceeds 60°C, it is considered a Level 4 abnormality. After the grading results are generated, they are transmitted to the intelligent early warning strategy generation and dynamic adjustment unit. Detailed records of the fusion process and grading results, including fusion weights, test results, and grading basis, are stored for subsequent query and analysis.

[0026] An intelligent early warning strategy generation and dynamic adjustment unit, the input of which is connected to the output of the multi-source abnormal information fusion and classification judgment unit, and generates and adjusts the early warning strategy based on the classification results; Specifically, the intelligent early warning strategy generation and dynamic adjustment unit is the core module for the system to formulate and optimize early warning plans. Its main function is to generate corresponding early warning strategies based on the abnormal risk level output by the multi-source abnormal information fusion and hierarchical judgment unit, and dynamically adjust the strategy based on real-time feedback information. The rationality and adaptability of the early warning strategy directly affect the effectiveness of the early warning. Different levels of abnormalities require matching different early warning methods, push objects and processing suggestions to ensure that the early warning information can be correctly received and effectively processed. At the same time, since the charging scenario and equipment status are constantly changing, fixed early warning strategies may not be able to adapt to actual needs. Through dynamic adjustment, the strategy can always maintain the best early warning effect. By generating targeted early warning strategies and performing real-time optimization, this unit ensures that the early warning system can flexibly respond to various complex situations and improves the practicality and effectiveness of the early warning.

[0027] The specific implementation process of this unit is as follows: First, it receives the anomaly risk level transmitted by the Multi-Source Anomaly Information Fusion and Grading Determination Unit, calls upon a pre-set policy template library, and generates an initial warning strategy based on the different levels. For example, the warning strategy for a Level 1 anomaly is to display a notification on the charging station's local display screen, updated every 5 minutes; the strategy for a Level 2 anomaly is to send an alert message to the operator's mobile app with detailed anomaly parameters; the strategy for a Level 3 anomaly is to send an emergency notification to the operator and dispatch the nearest personnel to the site, while also notifying the vehicle owner via text message; and the strategy for a Level 4 anomaly is to immediately trigger the charging station's emergency stop device, disconnecting the charging circuit, sending an alarm signal to the operation and maintenance center, and simultaneously calling the vehicle owner to notify them. Next, it receives feedback from the real-time push and execution feedback unit, including the status of the warning message, the operator's response time, and the results of the processing. Based on this feedback, the warning strategy is evaluated. If the warning message is not received or processed in a timely manner, the strategy is adjusted, such as changing the push channel, increasing the push frequency, or adjusting the processing suggestions. During the adjustment process, the optimal strategy for similar scenarios in historical data is referenced and adaptive algorithms are used to optimize strategy parameters. The adjustment cycle is 10-30 minutes. The adjusted early warning strategy is transmitted to the real-time early warning information push and execution feedback unit. The strategy template library is updated simultaneously, and the effective adjustment plan is saved as a new template for subsequent use in similar situations. The entire process of strategy generation and adjustment, including the initial strategy content, feedback information, and adjustment details, is recorded for continuous system improvement.

[0028] The warning information real-time push and execution feedback unit has an input end connected to the output end of the intelligent warning strategy generation and dynamic adjustment unit, pushes warning information and receives execution feedback, and the output end of the unit forms a closed-loop connection with the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit and the feature deep mining and spatiotemporal correlation mapping unit respectively.

[0029] Specifically, the real-time push of warning information and the execution feedback unit are key links in the system's implementation of warning information transmission and closed-loop control. Its main function is to convert the warning strategy formulated by the intelligent warning strategy generation and dynamic adjustment unit into specific warning information, and push it to relevant personnel and equipment in real time through multiple channels. At the same time, it collects the execution status and processing results of the warning information to form feedback data. This unit is a bridge for the system to interact with the outside world. The timeliness and accuracy of its information push directly affect the efficiency of exception handling, and feedback data provides an important basis for the system's self-optimization. By pushing warning information in real time and collecting execution feedback, it can ensure that abnormal situations receive timely attention and processing, while enabling the system to continuously adjust and improve according to actual conditions, forming a complete closed-loop control system.

[0030] The specific implementation process of this unit is as follows: First, it receives the warning strategy transmitted by the intelligent warning strategy generation and dynamic adjustment unit and analyzes the policy parameters, including push targets, content, method, and frequency. Based on the analysis results, warning information is pushed through various channels: For local charging station notifications, the control display screen displays abnormal information and handling suggestions, adjusts brightness to maximum, and displays the frequency according to the policy settings. For mobile app push notifications, the warning information is packaged into a standard format message and sent via the mobile network to the designated operation and maintenance personnel and the vehicle owner's app, ensuring delivery within 10 seconds. For emergency notifications, a text message is sent through the SMS gateway and a voice call is made to confirm receipt. Equipment control commands, such as emergency stop signals, are sent to the charging station control module via the internal communication bus, ensuring execution within 1 second. Then, the push status of the warning information is monitored in real time, including whether it is successfully sent and whether it has been read. Through interaction with the operation and maintenance personnel's app and the charging station control module, execution feedback information is collected, such as the operator's departure time, arrival time, handling measures and results, and the charging station execution status. The feedback information is sorted in chronological order to form a feedback report, which is sent to the intelligent early warning strategy generation and dynamic adjustment unit and the dynamic charging parameter real-time collection and multi-dimensional analysis unit through the data interface. At the same time, the push records and feedback reports are stored in the database for a period of 1 year for data statistics and system evaluation.

[0031] Preferably, the feature depth mining and spatiotemporal correlation mapping unit adopts an improved feature extraction model, and the model formula is: ,in, is the fusion eigenvalue at time t, 、 、 is the adaptive weight coefficient, The voltage at time t is the adaptively optimized long short-term memory neural network and current The feature extraction results of is the battery temperature at time t, Pile temperature The temperature effect function, For charging time, The grid fluctuation frequency The frequency impact function of the charging state assessment and abnormal trend prediction unit adopts a state assessment model, the formula is: ,in, is the charge state evaluation value at time t, Processing function for the improved gray wolf optimization algorithm, is the back propagation neural network function, is the weight matrix, is the bias term.

[0032] Specifically, the improved feature extraction model employed by the Deep Feature Mining and Spatiotemporal Correlation Mapping unit uses adaptive weighting coefficients to comprehensively consider the neural network-processed voltage and current features, as well as the influence of battery temperature, charging pile temperature, charging duration, and grid fluctuation frequency. This model more comprehensively extracts fused feature values ​​reflecting the charging state, enhancing the relevance of these features to the essence of the charging process. In this model, the temperature and frequency influence functions are set based on the actual variations of charging pile temperature and grid fluctuation frequency, respectively, ensuring that these parameters receive appropriate weight in feature extraction. The state assessment model in the Charging State Assessment and Abnormal Trend Prediction unit utilizes an improved Grey Wolf Optimization algorithm to optimize the results of backpropagation neural network processing. By properly setting the weight matrix and bias term, the mapping relationship between features and charging state assessment values ​​is strengthened, improving assessment accuracy. During implementation, the value range and function form of each parameter are first determined. The model is then trained using extensive historical data to determine the optimal values ​​for the adaptive weighting coefficients, weight matrix, and bias term. This ensures that the model consistently outputs reliable fused feature values ​​and state assessment values, providing a solid foundation for subsequent anomaly detection.

[0033] Preferably, the feature depth mining and spatiotemporal correlation mapping unit further includes a parameter dynamic adaptation subunit, and the adaptation model formula adopted by the unit is: ,in, is the network parameter adaptation value at time t, To adaptively optimize the inverse function of the long short-term memory neural network, are the partial derivatives of the fused eigenvalue with respect to voltage and current, are the changes in voltage and current at time t respectively; the charging state assessment and abnormal trend prediction unit also includes an abnormal sensitivity adjustment subunit, and the model formula is: ,in, is the abnormal sensitivity at time t, Processing function for the improved gray wolf optimization algorithm, is the charge state evaluation value at time t, is the derivative function of the back-propagation neural network, are the temperature changes of the battery and pile respectively, is the temperature influence coefficient.

[0034] Specifically, the dynamic parameter adaptation subunit of the Feature Deep Mining and Spatiotemporal Correlation Mapping unit calculates network parameter adaptation values ​​to achieve real-time adjustment of neural network parameters as voltage and current change, ensuring that the network maintains good feature extraction performance despite parameter fluctuations. This adaptation model quantifies the impact of parameter changes on feature extraction by fusing the partial derivatives of the feature values ​​with respect to voltage and current. It then adjusts network parameters based on the parameter changes, enhancing the network's adaptability to dynamic charging processes. The abnormality sensitivity adjustment subunit of the Charging State Assessment and Abnormal Trend Prediction unit calculates abnormality sensitivity, enabling the state assessment to dynamically adjust its sensitivity to abnormalities based on battery and charging pile temperature changes. This sensitivity is increased during rapid temperature fluctuations to detect abnormalities earlier, while it is reduced during stable temperature to reduce false positives. During implementation, the partial derivative calculation method and the appropriate value of the temperature impact coefficient must be determined experimentally. The model is then trained using actual charging data to ensure that the network parameter adaptation values ​​and abnormality sensitivity accurately reflect parameter changes and temperature conditions, enhancing the system's flexibility in detecting abnormalities.

[0035] Preferably, the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit adopts a parameter analysis model, and the formula is: ,in, is the multi-dimensional analysis result at time t, is the number of parameter types, is the analytical coefficient of the i-th parameter, is the collected value of the i-th parameter at time t, is the time decay function of the i-th parameter; the feature enhancement model of the feature depth mining and spatiotemporal correlation mapping unit is: ,in, is the enhanced eigenvalue, is the enhancement coefficient, is the activation function, They are 、 The multi-dimensional analysis results of the moment, is the fusion eigenvalue at time t, The adaptively optimized long short-term memory neural network 、 Feature extraction results at that moment.

[0036] Specifically, the parameter analysis model of the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit weights and time-decays the collected values ​​of different parameters by setting analysis coefficients and time decay functions, highlighting the impact of important parameters at the current moment while weakening the interference of historical data, making the analysis results more consistent with the real-time charging status. The analysis coefficient is determined based on the parameter's impact on charging safety, and the time decay function is set according to the parameter's timeliness, ensuring that recent data has a higher weight in the analysis. The feature enhancement model of the feature deep mining and spatiotemporal correlation mapping unit introduces an enhancement coefficient and activation function, using the analysis results of the previous two moments to enhance the current fused feature value, strengthen the historical trend information contained in the feature, and make the feature more representative. During implementation, the specific form of the analysis coefficient, time decay function, enhancement coefficient, and activation function must be determined through data analysis. After multiple iterative optimizations, the analysis results can accurately reflect the real-time status of the parameters, and the enhanced features can more effectively support subsequent status assessments.

[0037] Preferably, the abnormal trend prediction model of the charging state assessment and abnormal trend prediction unit is: ,in, for The predicted state value at time t, is the charge state evaluation value at time t, Processing function for the improved gray wolf optimization algorithm, is the back propagation neural network function, is the gradient of the state evaluation value at time t, is the prediction time interval, is the coefficient of fluctuation, is the parameter change vector at time t; the fusion model of the multi-source abnormal information fusion and classification judgment unit is: ,in, is the abnormal fusion value at time t, To predict the number of steps, is the weight of the j-th step prediction, is the temperature correction function corresponding to the j-th step prediction, is the battery temperature at time t, is the temperature of the pile at time t.

[0038] Specifically, the abnormal trend prediction model of the charging state assessment and abnormal trend prediction unit predicts the charging state at future moments by incorporating the prediction interval, the gradient of the state assessment value, the fluctuation coefficient, and the parameter change vector. The gradient reflects the changing trend of the current state, the parameter change vector reflects the impact of parameter fluctuations on the future state, and the fluctuation coefficient adjusts the degree of this impact, ensuring that the prediction results can proactively reflect potential abnormalities. The fusion model of the multi-source abnormal information fusion and classification determination unit performs a weighted fusion of multi-step predicted state values ​​and incorporates a temperature correction function to comprehensively consider the reliability of different prediction steps and the impact of temperature on anomalies. This ensures that the fused anomaly value comprehensively reflects abnormal conditions across multiple time periods. In implementation, the prediction interval, fluctuation coefficient, number of prediction steps, and weights for each step are determined based on the dynamic characteristics of the charging process. The temperature correction function is then trained on a large amount of sample data to determine its form. This ensures that the predicted state values ​​and fused anomaly values ​​accurately reflect future abnormal trends, providing a reliable basis for classification determination.

[0039] Preferably, the strategy generation model of the intelligent early warning strategy generation and dynamic adjustment unit is: ,in, is the early warning strategy at time t, Based on the abnormal fusion value and classification thresholds The policy function, is the adjustment coefficient, To adaptively optimize the long short-term memory neural network function, for The strategy adjustment amount at the moment; the feedback processing model of the real-time push of the warning information and the execution feedback unit is: ,in, is the feedback processing result at time t, For execution results The feedback function, is the feedback coefficient, First, the product of the early warning strategy Str (t) and the feedback coefficient κ is processed by the back propagation BP algorithm, and then the result is input into the improved gray wolf optimization The algorithm performs nested operations for optimization calculations.

[0040] Specifically, the strategy generation model of the intelligent early warning strategy generation and dynamic adjustment unit combines anomaly fusion values ​​and classification thresholds to determine the basic early warning strategy. It also incorporates neural network processing of adjustment coefficients and previous strategy adjustments, enabling the strategy to be optimized based on historical adjustment experience, enhancing its relevance to varying anomaly levels. This model uses a function to transform the anomaly fusion values ​​and classification thresholds into an initial strategy. The neural network then learns from historical adjustment values ​​to ensure the strategy remains applicable despite dynamic changes. The feedback processing model of the real-time early warning information push and execution feedback unit links the early warning strategy with execution results through feedback function processing and neural network optimization, ensuring that feedback results accurately reflect the strategy's effectiveness. During implementation, appropriate ranges for the classification thresholds and adjustment coefficients must be set. The form of the strategy function and feedback function must be determined through real-world case studies. The neural network is then trained using historical strategy adjustment data and execution feedback data. This allows the generated early warning strategy to be dynamically optimized based on anomaly conditions and feedback information, ensuring continuous improvement in early warning effectiveness.

[0041] Preferably, the charging state assessment and abnormal trend prediction unit includes: a state feature matching subunit, which receives feature data output from the feature depth mining and spatiotemporal correlation mapping unit, compares the feature data with the preset normal charging state feature library one by one, determines the matching degree by calculating the Euclidean distance between the feature vectors, and screens out feature combinations with a matching degree lower than a set threshold; an abnormal trend quantification subunit, which performs time series analysis on the screened low-matching feature combinations, calculates the slope and curvature of the feature as the charging time changes, and converts the abnormal trend into a quantifiable numerical indicator in combination with the rate of change of the battery temperature and the pile body temperature; an optimization assessment subunit, which optimizes the assessment results of the back propagation neural network using the improved gray wolf optimization algorithm, and reduces the deviation between the assessment results and the actual charging state to the minimum range by adjusting the weights and bias parameters of the network; a result output subunit, which organizes the optimized charging state assessment results and abnormal trend prediction indicators according to the set data format, and transmits them to the multi-source abnormal information fusion and grading judgment unit.

[0042] Preferably, the multi-source abnormal information fusion and grading judgment unit includes: an abnormal information receiving subunit, which receives various types of abnormal information output from the charging state evaluation and abnormal trend prediction unit, including abnormal characteristics, trend indicators and evaluation scores, and verifies the integrity and format of the information to eliminate invalid information; a multi-dimensional fusion subunit, which classifies abnormal information from different sources according to the time dimension and feature dimension, and uses the weighted average method to fuse similar abnormal information at the same time point to generate a comprehensive abnormal vector; a grading threshold determination subunit, which dynamically sets different levels of abnormal thresholds according to the model of the charging pile, the battery type and the charging stage, and each threshold corresponds to a set warning level; a grading result output subunit, which compares the fused abnormal information with the set threshold, determines the corresponding warning level, and transmits the grading result to the intelligent warning strategy generation and dynamic adjustment unit.

[0043] Preferably, the intelligent early warning strategy generation and dynamic adjustment unit includes: a basic strategy library calling subunit, which calls the corresponding initial early warning strategy from the preset basic early warning strategy library according to the early warning level output by the multi-source abnormal information fusion and hierarchical judgment unit, and the initial strategy includes the early warning method, push object and processing suggestion; a strategy dynamic adjustment subunit, which combines the early warning information real-time push and the feedback result returned by the execution feedback unit, and uses the adaptive optimization long short-term memory neural network to adjust the initial strategy, correct the early warning frequency and push content, so that the strategy is more in line with the actual situation; a multi-scenario adaptation subunit, which analyzes the environmental parameters of the current charging scene, including ambient temperature, humidity and grid stability, and performs scene adaptation optimization on the adjusted strategy to ensure consistent early warning effects in different scenarios; a final strategy output subunit, which transmits the final early warning strategy after dynamic adjustment and scene adaptation to the early warning information real-time push and execution feedback unit, and stores the strategy parameters in the historical database.

[0044] like Figure 2 As shown, the new energy charging pile charging process early warning system based on artificial intelligence includes: Step S1: The dynamic charging parameter real-time acquisition and multi-dimensional analysis unit continuously acquires and performs multi-dimensional analysis on the charging pile output voltage, charging current, battery temperature, charging time, pile body temperature, and grid fluctuation frequency; Step S2: transmitting the parsed parameters to the feature deep mining and spatiotemporal correlation mapping unit to perform feature deep mining and spatiotemporal correlation mapping analysis to form a feature data set; Step S3: Input the characteristic data set into the charging state assessment and abnormal trend prediction unit to perform charging state assessment and abnormal trend prediction to obtain abnormal trend information; Step S4: Send the abnormal trend information to the multi-source abnormal information fusion and classification determination unit, implement multi-source abnormal information fusion and classification determination, and generate an abnormal classification result; Step S5: Transmit the abnormality classification results to the intelligent early warning strategy generation and dynamic adjustment unit, formulate an intelligent early warning strategy and perform dynamic adjustment to produce an adaptive early warning strategy; Step S6: The adapted warning strategy is delivered to the warning information real-time push and execution feedback unit to push the warning information in real time and receive execution feedback, and the feedback information is respectively transmitted back to the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit and the feature deep mining and spatiotemporal correlation mapping unit to form a complete warning closed-loop process.

[0045] The AI-based early warning system for charging piles in new energy vehicles features powerful multi-parameter collaborative analysis capabilities. The real-time acquisition and multi-dimensional analysis unit for dynamic charging parameters comprehensively captures information such as the charging pile's output voltage, current, battery and pile body temperature, charging duration, and grid fluctuation frequency. Through deep feature mining and spatiotemporal correlation mapping, it can deeply explore the spatiotemporal correlations between various parameters, providing precise feature support for subsequent evaluations. Based on these features, the charging status assessment and abnormal trend prediction unit enables detailed assessment of the charging status and accurate prediction of abnormal trends, significantly improving the ability to identify anomalies in complex charging scenarios.

[0046] The system effectively overcomes the weak correlation between feature extraction and condition assessment in existing technologies. Traditional techniques primarily analyze single parameters or simple combinations of parameters, ignoring inter-parameter correlations and leading to one-sided assessments. However, this system uses adaptively optimized long-short-term memory neural networks to deeply mine multiple parameters and analyze spatiotemporal correlations, transforming dispersed parameters into highly correlated features. An improved Gray Wolf Optimization-Backpropagation neural network then uses these correlated features to perform condition assessment and predict abnormal trends. This seamlessly links feature extraction and condition assessment, enabling accurate identification of complex issues caused by multi-parameter anomalies.

[0047] At the same time, the system addresses the lack of dynamic adaptability of existing warning strategies. Traditional warning strategies are often fixed, making them difficult to adapt to different charging station models, battery types, and environmental conditions. In this system, the intelligent warning strategy generation and dynamic adjustment unit continuously adjusts the warning logic based on the results of the multi-source abnormal information fusion and classification judgment unit, combined with feedback from the real-time push of warning information and the execution feedback unit. Through this dynamic adjustment mechanism, the strategy can flexibly adapt to various scenarios, avoiding warning lags or false alarms caused by scenario changes, and ensuring that effective warning effects are maintained under different conditions.

[0048] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to mechanical connections or electrical connections; they may refer to direct connections or indirect connections through an intermediate medium; and they may refer to internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0049] While embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The artificial intelligence-based new energy charging pile charging process early warning system is characterized by: include: The dynamic charging parameter real-time acquisition and multi-dimensional analysis unit is used to obtain the charging pile output voltage, charging current, battery temperature, charging time, pile body temperature and grid fluctuation frequency; the feature deep mining and spatiotemporal correlation mapping unit, the input end of which is connected to the output end of the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit, and performs feature extraction and spatiotemporal correlation analysis on the parameters; the charging state evaluation and abnormal trend prediction unit, the input end of which is connected to the output end of the feature deep mining and spatiotemporal correlation mapping unit, and evaluates the charging state and predicts abnormal trends based on the extracted features; the multi-source abnormal information fusion and hierarchical judgment unit, the output end of which is connected to the output end of the feature deep mining and spatiotemporal correlation mapping unit, and evaluates the charging state and predicts abnormal trends based on the extracted features; The input end is connected to the output end of the charging status assessment and abnormal trend prediction unit to fuse and classify the abnormal information; the intelligent early warning strategy generation and dynamic adjustment unit, the input end of which is connected to the output end of the multi-source abnormal information fusion and classification judgment unit, and generates and adjusts the early warning strategy according to the classification result; the early warning information real-time push and execution feedback unit, the input end of which is connected to the output end of the intelligent early warning strategy generation and dynamic adjustment unit, pushes the early warning information and receives the execution feedback, and the output end of this unit forms a closed-loop connection with the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit and the feature deep mining and spatiotemporal correlation mapping unit respectively.

2. The artificial intelligence-based new energy charging pile charging process early warning system according to claim 1 is characterized in that: The feature depth mining and spatiotemporal correlation mapping unit adopts an improved feature extraction model, and the model formula is: ,in, is the fusion eigenvalue at time t, 、 、 is the adaptive weight coefficient, The voltage at time t is the adaptively optimized long short-term memory neural network and current The feature extraction results of is the battery temperature at time t, Pile temperature The temperature effect function, For charging time, The grid fluctuation frequency The charging state assessment and abnormal trend prediction unit adopts a state assessment model, the formula is: ,in, is the charge state evaluation value at time t, Processing function for the improved gray wolf optimization algorithm, is the back propagation neural network function, is the weight matrix, is the bias term.

3. The artificial intelligence-based new energy charging pile charging process early warning system according to claim 1 is characterized in that: The feature depth mining and spatiotemporal correlation mapping unit also includes a parameter dynamic adaptation subunit, which adopts the adaptation model formula: ,in, is the network parameter adaptation value at time t, To adaptively optimize the inverse function of the long short-term memory neural network, are the partial derivatives of the fused eigenvalue with respect to voltage and current, are the changes in voltage and current at time t respectively; the charging state assessment and abnormal trend prediction unit also includes an abnormal sensitivity adjustment subunit, and the model formula is: ,in, is the abnormal sensitivity at time t, Processing function for the improved gray wolf optimization algorithm, is the charge state evaluation value at time t, is the derivative function of the back-propagation neural network, are the temperature changes of the battery and pile respectively, is the temperature influence coefficient.

4. The artificial intelligence-based new energy charging pile charging process early warning system according to claim 1 is characterized in that: The dynamic charging parameter real-time acquisition and multi-dimensional analysis unit adopts a parameter analysis model, and the formula is: ,in, is the multi-dimensional analysis result at time t, is the number of parameter types, is the analytical coefficient of the i-th parameter, is the collected value of the i-th parameter at time t, is the time decay function of the i-th parameter; the feature enhancement model of the feature depth mining and spatiotemporal correlation mapping unit is: ,in, is the enhanced eigenvalue, is the enhancement coefficient, is the activation function, They are 、 The multi-dimensional analysis results of the moment, is the fusion eigenvalue at time t, For the adaptively optimized long short-term memory neural network 、 Feature extraction results at that moment.

5. The artificial intelligence-based new energy charging pile charging process early warning system according to claim 1 is characterized in that: The abnormal trend prediction model of the charging state evaluation and abnormal trend prediction unit is: ,in, for The predicted state value at time t, is the charge state evaluation value at time t, Processing function for the improved gray wolf optimization algorithm, is the back propagation neural network function, is the gradient of the state evaluation value at time t, is the prediction time interval, is the coefficient of fluctuation, is the parameter change vector at time t; the fusion model of the multi-source abnormal information fusion and classification judgment unit is: ,in, is the abnormal fusion value at time t, To predict the number of steps, is the weight of the prediction at step j, is the temperature correction function corresponding to the j-th step prediction, is the battery temperature at time t, is the temperature of the pile at time t.

6. The artificial intelligence-based new energy charging pile charging process early warning system according to claim 1 is characterized in that: The strategy generation model of the intelligent early warning strategy generation and dynamic adjustment unit is: ,in, is the early warning strategy at time t, Based on the abnormal fusion value and classification thresholds The policy function, is the adjustment coefficient, To adaptively optimize the long short-term memory neural network function, for The strategy adjustment amount at the moment; the feedback processing model of the real-time push of the warning information and the execution feedback unit is: ,in, is the feedback processing result at time t, For execution results The feedback function, is the feedback coefficient, First, the product of the early warning strategy Str (t) and the feedback coefficient κ is processed by the back propagation BP algorithm, and then the result is input into the improved gray wolf optimization The algorithm performs nested operations for optimization calculations.

7. The artificial intelligence-based new energy charging pile charging process warning system according to claim 1 is characterized in that: The charging state assessment and abnormal trend prediction unit includes: a state feature matching subunit, which receives feature data output from the feature deep mining and spatiotemporal correlation mapping unit, compares the feature data with a preset normal charging state feature library one by one, determines the matching degree by calculating the Euclidean distance between feature vectors, and screens out feature combinations with a matching degree lower than a set threshold; an abnormal trend quantification subunit, which performs time series analysis on the screened low-matching feature combinations, calculates the slope and curvature of the feature as the charging time changes, and converts the abnormal trend into a quantifiable numerical indicator in combination with the rate of change of battery temperature and pile body temperature; an optimization assessment subunit, which optimizes the assessment results of the back propagation neural network using an improved gray wolf optimization algorithm, and reduces the deviation between the assessment results and the actual charging state to a minimum range by adjusting the network weights and bias parameters; a result output subunit, which organizes the optimized charging state assessment results and abnormal trend prediction indicators according to the set data format, and transmits them to the multi-source abnormal information fusion and classification judgment unit.

8. The artificial intelligence-based new energy charging pile charging process early warning system according to claim 1 is characterized in that: The multi-source abnormal information fusion and grading judgment unit includes: an abnormal information receiving subunit, which receives various types of abnormal information output by the charging state assessment and abnormal trend prediction unit, including abnormal characteristics, trend indicators and evaluation scores, and verifies the integrity and format of the information and eliminates invalid information; a multi-dimensional fusion subunit, which classifies abnormal information from different sources according to the time dimension and feature dimension, and uses the weighted average method to fuse similar abnormal information at the same time point to generate a comprehensive abnormal vector; a grading threshold determination subunit, which dynamically sets different levels of abnormal thresholds based on the model of the charging pile, the battery type and the charging stage, and each threshold corresponds to a set warning level; a grading result output subunit, which compares the fused abnormal information with the set threshold, determines the corresponding warning level, and transmits the grading result to the intelligent warning strategy generation and dynamic adjustment unit.

9. The artificial intelligence-based new energy charging pile charging process early warning system according to claim 1 is characterized in that: The intelligent early warning strategy generation and dynamic adjustment unit includes: a basic strategy library calling subunit, which calls the corresponding initial early warning strategy from the preset basic early warning strategy library according to the early warning level output by the multi-source abnormal information fusion and hierarchical judgment unit. The initial early warning strategy includes the early warning method, push object and processing suggestions; a strategy dynamic adjustment subunit, which combines the feedback results returned by the early warning information real-time push and execution feedback unit, and uses the adaptive optimization long short-term memory neural network to adjust the initial early warning strategy, correct the early warning frequency and push content, and make the strategy more in line with the actual situation; a multi-scenario adaptation subunit, which analyzes the environmental parameters of the current charging scenario, including ambient temperature, humidity and power grid stability, and performs scenario adaptation optimization on the adjusted strategy to ensure consistent early warning effects in different scenarios; a final strategy output subunit, which transmits the final early warning strategy after dynamic adjustment and scenario adaptation to the early warning information real-time push and execution feedback unit, and stores the strategy parameters in the historical database.

10. The artificial intelligence-based new energy charging pile charging process warning system according to any one of claims 1 to 9, characterized in that: The system operation includes: Step S1: The dynamic charging parameter real-time acquisition and multi-dimensional analysis unit continuously acquires and performs multi-dimensional analysis on the charging pile output voltage, charging current, battery temperature, charging time, pile body temperature, and grid fluctuation frequency; Step S2: transmitting the parsed parameters to the feature deep mining and spatiotemporal correlation mapping unit to perform feature deep mining and spatiotemporal correlation mapping analysis to form a feature data set; Step S3: Input the characteristic data set into the charging state assessment and abnormal trend prediction unit to perform charging state assessment and abnormal trend prediction to obtain abnormal trend information; Step S4: Send the abnormal trend information to the multi-source abnormal information fusion and classification determination unit, implement multi-source abnormal information fusion and classification determination, and generate an abnormal classification result; Step S5: Transmit the abnormality classification results to the intelligent early warning strategy generation and dynamic adjustment unit, formulate an intelligent early warning strategy and perform dynamic adjustment to produce an adaptive early warning strategy; Step S6: The adapted warning strategy is delivered to the warning information real-time push and execution feedback unit to push the warning information in real time and receive execution feedback, and the feedback information is respectively transmitted back to the dynamic charging parameter real-time acquisition and multi-dimensional analysis unit and the feature deep mining and spatiotemporal correlation mapping unit to form a complete warning closed-loop process.

Images