Distributed monitoring and warning system for the electromagnetic environment of vehicles

Abstract

This invention relates to the field of vehicle electromagnetic environment monitoring, specifically a distributed monitoring and alarm system for vehicle electromagnetic environments. The system includes a host computer, an in-vehicle router, and multiple electromagnetic acquisition nodes. These nodes collect electromagnetic environment data and send it to the host computer. The host computer processes the received electromagnetic environment data and uses the processed data to train an electromagnetic environment alarm model. During normal vehicle operation, electromagnetic environment data is collected by each acquisition node and sent to the host computer via the in-vehicle router. The host computer analyzes the electromagnetic data collected by each node in real time using the trained electromagnetic environment alarm model. If the current test data is similar to electromagnetic data collected under a past fault environment, or if the current test data deviates significantly from the dataset included in the model, the electromagnetic environment alarm model uses resources from the host computer to issue an alarm to the vehicle operator. This invention is applicable to distributed monitoring and alarming of vehicle electromagnetic environments.

Description

Technical Field

[0001] This invention relates to the field of vehicle electromagnetic environment monitoring, specifically to a distributed monitoring and alarm system for vehicle electromagnetic environment. Background Technology

[0002] In the process of electrification and intelligentization, the variety of electronic components and electrical devices used in automobiles is rapidly increasing, and the internal circuitry of vehicles is becoming increasingly integrated. Besides the surge in electronic components and electrical devices, the number of cables in vehicles is also increasing. These highly integrated electronic components and electrical devices include wiring harnesses carrying various signals, including high-voltage, low-voltage, and transient signals. Furthermore, various signals from the surrounding environment during vehicle operation create an extremely complex electromagnetic environment for intelligent electric vehicles. Various electronic components and electrical devices on the vehicle, especially the core system modules that maintain normal vehicle operation, including the control system, power system, sensors, and communication system, may be subject to various forms of electromagnetic interference from both inside and outside the vehicle. This interference may be picked up by the electronic components, electrical devices, and cables of the intelligent electric vehicle, thereby affecting the signal logic of the intelligent driving system. This can lead to false triggering and misoperation of the intelligent electric vehicle, affecting passenger comfort to the point of causing serious accidents with fatalities.

[0003] Therefore, it is necessary to monitor the electromagnetic environment in which the vehicle is located over a long period of time and issue timely warnings to prevent the vehicle from accidentally causing a major accident due to electromagnetic interference. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a distributed monitoring and alarm system for vehicle electromagnetic environment, which greatly improves the accuracy of alarms for abnormal vehicle electromagnetic environment.

[0005] This invention achieves the above objectives by adopting the following technical solution: This invention provides a distributed monitoring and alarm system for vehicle electromagnetic environment, comprising: a host computer, an in-vehicle router, and multiple electromagnetic acquisition nodes distributed throughout the vehicle. Each electromagnetic acquisition node includes a communication module and an MCU (Microcontroller Unit). The vehicle consists of a microcontroller unit (MCU), a data acquisition module, and acquisition probes. A communication module connects to the host computer via an onboard router, receiving instructions from the host computer and sending them to the MCU. The MCU controls the data acquisition module to collect electromagnetic environment data via the acquisition probes. After formatting the collected electromagnetic environment data, the MCU sends it to the host computer via the communication module. The host computer processes the received electromagnetic environment data and uses the processed data to train the electromagnetic environment alarm model. During normal vehicle operation, electromagnetic environment data is collected at various electromagnetic acquisition nodes and sent to the host computer via the onboard router. The host computer analyzes the electromagnetic data collected by each node in real time using the trained electromagnetic environment alarm model. If the current test data is similar to electromagnetic data collected in a past fault environment, or if the current test data deviates significantly from the dataset included in the model, the electromagnetic environment alarm model uses resources from the host computer to issue an alarm to the vehicle operator.

[0006] Furthermore, the collected data includes vehicle electromagnetic environment parameters, timestamps, and geographical location information. Data collection needs to cover different time periods, different road conditions, and different scenarios.

[0007] Furthermore, the host computer processes the received electromagnetic environment data, specifically including:

[0008] Interpolation and noise reduction processing:

[0009] Interpolation can be used to handle missing values ​​or delete invalid data segments.

[0010] Use filters or smoothing algorithms to reduce data noise;

[0011] Category labeling:

[0012] If the collected data is from when the vehicle is driving normally, the collected data will be recorded in the first dataset. If electromagnetic interference or equipment failure occurs during the data collection process, the collected data will be recorded in the second dataset and the type of failure will be marked.

[0013] Enhancement:

[0014] The `find_peaks` function is used to extract peak values ​​from the labeled data as input features. Fixed-length data sequences are generated by truncating and expanding the data near each peak, and the category labels are converted into one-hot encoded form.

[0015] Furthermore, training the electromagnetic environment alarm model using the processed data specifically includes:

[0016] For the labeled data, a 1D-CNN model is constructed. The 1D-CNN model includes an input layer, a convolutional layer, a pooling layer, two fully connected layers, and an output layer. The hidden layer uses a convolutional layer with 32 filters of size 3, a stride of 2, a ReLU activation function, and a same padding method to extract local features from the data and introduce non-linearity. The pooling layer downsamples through a pooling window of size 2. The first fully connected layer contains 100 neurons, and the second layer contains 50 neurons. The activation function for both is ReLU. The output layer has multiple neurons, corresponding to various electromagnetic environment states of the vehicle, and the Softmax function is used to convert the output into a probability distribution.

[0017] The 1D-CNN model is trained using labeled data. The Adam optimizer is used to accelerate the convergence speed of the model. The classification cross-entropy is used as the loss function to measure the difference between the model's prediction results and the actual labels. The 1D-CNN model outputs anomaly alerts.

[0018] Furthermore, training the electromagnetic environment alarm model using the processed data specifically includes:

[0019] For unlabeled data, the HTM (Hierarchical Temporal Memory) model is used. The input encoding stage of the HTM model is responsible for converting the input data into a binary string composed of 0s and 1s, which serves as the input to the spatial pool. The spatial pool of the HTM model converts the binary string into a sparse binary pattern, generating a sparse distributed representation of the input data. The temporal pool of the HTM model represents the contextual pattern of the input data as the active state of the neurons, and uses the predicted state of the neurons to represent the input data pattern to be passed in at the next time step. The decoding stage of the HTM model uses a sparse distributed representation classifier as a decoder, which uses the relationship between the input data bits and the neurons to decode, and is responsible for converting the predicted neuron pattern output by the temporal pool into an output with the same format as the input data.

[0020] The HTM model is trained using unlabeled data. The HTM model calculates an anomaly score between the sparse code P(x) of the current input x and the HTM model's prediction of the sparse code P(x). The anomaly score is a value between 0 and 1 used to quantify the degree of difference between the current state and the state predicted by the HTM model. The degree of difference is positively correlated with the probability of an anomaly.

[0021] Furthermore, training the electromagnetic environment alarm model using the processed data specifically includes:

[0022] By processing abnormal alarms from the outputs of 1D-CNN and HTM models using a moving window, abnormal electromagnetic environments of vehicles at different time periods can be captured, as follows:

[0023]

[0024]

[0025] In the formula, S k S represents the weighted sum of outlier scores at all times within a predetermined time window. i W represents the outlier score at time i. i Let represent the weight corresponding to the abnormal score at time i, and l represent the length of the moving window.

[0026] Finally, according to S k The value determines the current abnormal electromagnetic environment of the vehicle.

[0027] The beneficial effects of this invention are as follows:

[0028] This invention employs multiple electromagnetic acquisition nodes distributed throughout the vehicle to collect vehicle electromagnetic environment data, thereby improving acquisition efficiency. The collected electromagnetic environment data includes vehicle electromagnetic environment parameters, timestamps, and geographical location information. During data collection, different time periods, road conditions, and scenarios need to be covered, thus improving the accuracy of the collected data.

[0029] This invention uses a 1D-CNN model to classify specific data containing abnormal state types, thereby improving the efficiency and accuracy of classification.

[0030] To improve the generalization ability of 1D-CNN models, this invention performs data preprocessing and enhancement, including using the find_peaks function to extract peaks from the original data as input features, generating fixed-length data sequences by truncating and expanding the data near each peak, and converting the class labels into one-hot encoded form.

[0031] This invention processes data that does not contain anomalous state types using an Hidden Trace Meter (HTM) model. The HTM model calculates an anomaly score St between the sparse code P(x) of the current input x and the HTM network model's prediction of P(x). The anomaly score St is a value between 0 and 1, used to quantify the degree of difference between the current state and the model's predicted state. The higher the value of St, the greater the difference between the current data and the model's prediction, i.e., the greater the probability of an anomaly. This anomaly detection mechanism is not only simple and intuitive but also effectively captures subtle changes in the input data, making it particularly effective for scenarios such as vehicle electromagnetic environment monitoring.

[0032] The sliding window technology of this invention combines the detection results of these two models, greatly improving the accuracy of anomaly detection. Attached Figure Description

[0033] Figure 1 This is an architecture diagram of a distributed monitoring and alarm system for vehicle electromagnetic environment provided in an embodiment of the present invention;

[0034] Figure 2 This is a block diagram of the electromagnetic acquisition node structure provided in an embodiment of the present invention;

[0035] Figure 3 This is a block diagram of the 1D-CNN model structure provided in the embodiments of the present invention;

[0036] Figure 4 This is a block diagram of the HTM model structure provided in the embodiments of the present invention. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0038] This invention provides a distributed monitoring and alarm system for vehicle electromagnetic environment, such as... Figure 1 As shown, it specifically includes: a host computer, a router, and multiple electromagnetic acquisition nodes, which are distributed throughout the vehicle.

[0039] The host computer and router are connected via wired or wireless means. The router and electromagnetic acquisition nodes that collect electromagnetic interference signals are also connected via wired or wireless means. The host computer is mainly used for comprehensive data processing. Its display function can be integrated with the vehicle's control panel for convenient observation and operation by passengers. Alternatively, it can be connected to an external screen for maintenance and research personnel to observe and operate from different locations inside or outside the vehicle. The host computer's data storage and comprehensive processing functions refer to storing the electromagnetic emission data obtained from each electromagnetic acquisition node on the host computer and analyzing and calculating the collected data using algorithms. The host computer can be placed in the vehicle as an independent unit or integrated with the vehicle's onboard computer system.

[0040] like Figure 2 As shown, the electromagnetic interference (EMI) acquisition node includes a wireless or wired communication module, a microcomputer or microprocessor, a data acquisition module, and an acquisition probe or antenna. The wireless or wired communication module is responsible for transmitting control commands from the host computer to the EMI acquisition node via wired or wireless means, or for periodically uploading test data from the EMI acquisition node to the host computer via wired or wireless means according to the control commands. The microcomputer or microprocessor is responsible for periodically acquiring EMI data from the data acquisition module according to the host computer's instructions and processing it into a frequency domain data file of a unified format. The data acquisition module is responsible for acquiring EMI data from the microcomputer or microprocessor according to the requirements of the microcomputer or microprocessor and transmitting it back to the microcomputer or microprocessor. Depending on the type of signal being acquired, different sensors are connected to the data acquisition module. For conducted signals, the data acquisition module is connected to an acquisition probe, current clamp, etc.; for radiated interference signals, the spectrum acquisition module is connected to an antenna.

[0041] This invention allows for distributed monitoring to be operated via a software interface on a host computer. The software mainly includes three functions: initial setup of sampling nodes, display of the on-vehicle electromagnetic environment at specific frequencies, and display of the electromagnetic spectrum and spectrum waterfall plot at each node. These will be described in detail below.

[0042] (1) Initial settings for sampling nodes

[0043] The start frequency, end frequency, frequency step, and time interval of the acquisition system can be set through the software interface in the host computer. By clicking to generate the configuration file and start acquisition, the set parameters can be uniformly transmitted to all electromagnetic acquisition nodes controlled by the system.

[0044] (2) Demonstrate the electromagnetic environment on the vehicle at a specific frequency.

[0045] The host computer displays the electromagnetic environment at a specific frequency point of the vehicle at a given moment. By dragging the frequency bar at the bottom, the electromagnetic environment observed in the graph can be switched at different frequency points.

[0046] (3) Display the electromagnetic spectrum and spectrum waterfall diagram at each node.

[0047] The host computer software interface can display the spectrum of each node location. By selecting a node, users can switch between the electromagnetic spectra of different nodes. It can also draw a spectrum waterfall plot by reading historical spectrum data files.

[0048] The following section provides a detailed explanation of the data processing and electromagnetic alarm model training for the host computer.

[0049] The electromagnetic environment data received by the host computer includes vehicle electromagnetic environment parameters, timestamps, geographical location and other information, covering different time periods (day and night, seasons), different road conditions (mountainous, plains, indoors) and scenarios (normal, abnormal events).

[0050] The host computer processes the received electromagnetic environment data, and the processing methods include:

[0051] Preprocessing:

[0052] Missing values ​​can be handled or invalid data segments can be deleted using interpolation (linear / spline);

[0053] Use filters (such as Kalman filtering, wavelet transform) or smoothing algorithms (moving average) to reduce data noise;

[0054] Feature extraction:

[0055] This study primarily extracts time-domain features, frequency-domain features, time-series features, and event-related features from electromagnetic environment data. Time-domain features mainly focus on parameters such as mean, variance, kurtosis, skewness, zero-crossing rate, and autocorrelation coefficient. Frequency-domain features mainly focus on parameters such as FFT spectrum, frequency band energy proportion, and dominant frequency components. Time-series features mainly focus on parameters such as sliding window statistics (maximum / minimum / slope within the window) and difference sequences.

[0056] Category labeling:

[0057] For datasets extracted from electromagnetic environment data, it is necessary to label the datasets based on the vehicle's operating conditions in different electromagnetic environments. If the data is recorded when the vehicle is driving without abnormalities, it should be recorded in dataset I. If the vehicle experiences electromagnetic interference or equipment failure during the creation of the electromagnetic environment dataset, it should be recorded in dataset II, and the type of abnormality should be labeled.

[0058] Suppose that in the dataset I obtained from vehicle testing, the spectral intensity data measured by the nth node (out of N) in the m-th test (out of M tests) is Amp. mn =(Amp 1 mnAmp 2 mn ...Amp k mn ...Amp K mn Define the operator Min(Amp) n To find the minimum value of the M results for the k-th amplitude data at the n-th node, the minimum value of all spectral intensity data tested from the first node can be expressed as Min(Amp1) = (min(Amp1) / (k-th amplitude data)). 1 11 Amp 1 21 ...Amp 1 m1 ...Amp 1 M1 ), min(Amp 2 11 Amp 2 21 ...Amp 2 m1 ...Amp 2 M1 ), …,min(Amp k 11 Amp k 21 ...Amp k m1 ...Amp k M1 ), …,min(Amp K 11 Amp K 21 ...Amp K m1 ...Amp K M1 And so on, the results of the calculations performed on the data from N nodes are denoted as Min(Amp1), Min(Amp2), ..., Min(Amp3). n ),…,Min(Amp N Define the operator Max(Amp) n To find the maximum value of the M results for the k-th amplitude data at the n-th node, the maximum value of all spectral intensity data measured at the first node can be expressed as Max(Amp1) = (max(Amp1) / (k-th amplitude data)). 1 11 Amp 1 21 ...Amp 1 m1 ...Amp 1 M1), max(Amp 2 11 Amp 2 21 ...Amp 2 m1 ...Amp 2 M1 ), ...,max(Amp k 11 Amp k 21 ...Amp k m1 ...Amp k M1 ), …,min(Amp K 11 Amp K 21 ...Amp K m1 ...Amp K M1 And so on, the results of the calculations performed on the data from N nodes are denoted as Max(Amp1), Max(Amp2), ..., Max(Amp... n ),…,Max(Amp N ).

[0059] For the data in dataset II, for a certain type of anomaly, suppose that in the j-th test (out of J tests), the spectral intensity data measured by the n-th node (out of N tests) is Amp. jn =(Amp 1 jn Amp 2 jn ...Amp k jn ...Amp K jn ), by Amp jn With Max(Amp) n Compare each item to obtain Amp. k jn Higher than Max(Amp) n The corresponding preset threshold item number Num jn =(num 1 jn ,…,num p jn It should be noted that this invention does not limit the value of the preset threshold; those skilled in the art can select it according to actual needs. Statistical analysis is performed on the results of J tests, and those serial numbers with a frequency greater than 80% are selected and denoted as Num. n =(num1 n ,…,num p1 n ). Calculate the Num values ​​of all N nodes. n Find the union of the sets to obtain the sequence number Num = (num_1, ..., num_q), then find the spectral intensity data Amp corresponding to this sequence number. jn (Num) = (Amp) num_1 jn Amp num_2 jn ,…,Amp num_q jn When a vehicle malfunctions, the main spectral data and corresponding amplitudes that need to be monitored are as follows.

[0060] To improve the generalization ability of the model, this invention uses the find_peaks function to extract peaks from the preprocessed data as input features, generates a fixed-length data sequence by truncating and expanding the data near each peak, and converts the labeled category labels into one-hot encoded form.

[0061] The electromagnetic alarm model of this invention adopts a structure of 1D-CNN model plus HTM model. The 1D-CNN model is used to process alarm models of specific datasets containing abnormal state types, and the HTM model is used to solve sudden electromagnetic abnormal alarms without standard alarm datasets. Finally, the sliding window technique is used to process alarms under both models.

[0062] For labeled datasets, this invention constructs a 1D-CNN model, which is also designed to handle a specific dataset classification task containing anomalous state types. The core of the 1D-CNN model design lies in accurately capturing key features in the data to achieve efficient classification.

[0063] The structure of a 1D-CNN model is as follows: Figure 3As shown, the input layer is designed as a tensor of shape (batch_size, 100, 1), where 100 represents the length of the data sequence and 1 represents the feature dimension of the sequence. This design ensures that the model can receive and process data of a specified shape. In the hidden layer design, a convolutional layer is used, containing 32 filters of size 3 with a stride of 2, using the ReLU activation function and the same padding method to extract local features from the data and introduce non-linearity. Subsequently, a pooling layer downsamples through a pooling window of size 2, reducing the number of model parameters and computational complexity while preserving key features. To map the extracted features to the class space, two fully connected layers are designed: the first layer contains 100 neurons, and the second layer contains 50 neurons, both using the ReLU activation function. Finally, the output layer consists of multiple neurons, corresponding to various electromagnetic environment states of the vehicle, and the Softmax function is used to convert the output into a probability distribution for subsequent classification decisions.

[0064] The 1D-CNN model was trained using a labeled dataset. For training and optimization, the Adam optimizer was employed to accelerate convergence, and classification cross-entropy was used as the loss function to measure the difference between the model's predictions and the actual labels. Accuracy was used as the evaluation metric to assess the model's performance on the test set.

[0065] In terms of 1D-CNN model evaluation and visualization, a confusion matrix was plotted to visually demonstrate the classification performance of the 1D-CNN model on the test set, and accuracy and loss curves were plotted during the training process to monitor the training progress and performance changes of the 1D-CNN model.

[0066] Since most data acquired during vehicle operation is anomaly-free data, i.e., unlabeled datasets, this invention also provides an unsupervised learning method. Unsupervised learning, by deeply exploring the intrinsic relationships within the data, can reveal latent features in the absence of explicit labels. This capability is particularly important in distributed electromagnetic environment detection and alarm systems, as electromagnetic signals are often difficult to capture and may be hidden within complex background noise. Traditional statistical methods face numerous challenges in processing this type of data, while clustering-based anomaly detection methods are limited by the characteristics of one-dimensional time-series data. Although deep learning models possess powerful feature learning capabilities, their performance is significantly affected in situations lacking labeled data.

[0067] To address these challenges, a more suitable alarm model is needed for scenarios with non-standard alarm data. The Hidden Time Metric (HTM) model, with its unique neuroscience principles and online learning capabilities, demonstrates significant advantages in processing time-series data and identifying anomalous behavior. The HTM model compresses high-dimensional input data into a low-dimensional sparse representation through a sparse coding mechanism, while preserving key data features, providing a solid foundation for subsequent anomaly detection. Furthermore, the HTM model's online learning capability allows it to continuously adapt to new data patterns, which is particularly important in the dynamically changing vehicle electromagnetic detection environment.

[0068] HTM model structure as follows Figure 4 As shown, the input encoding layer of the HTM model is responsible for converting the input data into a binary string composed of 0s and 1s, which serves as the input to the space pool. The space pool then further converts these binary strings into sparse binary patterns, generating a sparse distributed representation of the input data.

[0069] Spatial pooling plays a crucial role in Hidden Memory (HTM) models. It not only transforms the input data into a sparse representation but also significantly impacts the learning of the temporal pool and the overall performance of the HTM model.

[0070] The time pool is another key component of the Hidden Matrix Theory (HTM) model. Its main task is to represent the contextual patterns of the input data as the active states of neurons, and to characterize the input data patterns at the next time step using the predicted states of the neurons. The time pool focuses more on the temporal dependencies between data, generating predictions for the next time step based on the current context. Even with the same input and the same active micropillars, the time pool will select different sets of neurons for representation depending on the context of the input.

[0071] The time pooling algorithm comprises four stages: neuron initialization, active neuron selection, learning, and predictive neuron selection. However, the decoding stage of the HTM model is responsible for converting the predicted neuron patterns from the time pooling output into an output format identical to the input data. It employs a sparse distribution representation classifier as the decoder, utilizing the relationship between input data bits and neurons for decoding. In this way, the HTM model can transform input sequence data into meaningful output, fulfilling the needs of various tasks.

[0072] The HTM anomaly detection process is as follows:

[0073] The Hidden Motion Matrix (HTM) model employs an innovative sparse coding mechanism to transform input data into a sparse distributed representation, which is then pooled in a sparse space to form an output prediction in the form of a sparse binary vector P(x) representing the current input and a sparse vector F(xt-1) representing the previous time step (t-1). The core of this mechanism lies in HTM's ability to effectively compress high-dimensional input data into a low-dimensional sparse representation while preserving key data features, thus providing a foundation for subsequent anomaly detection.

[0074] Specifically, the HTM model calculates an anomaly score St between the sparse code P(x) of the current input x and the HTM network model's prediction F(xt-1) of P(x). This anomaly score St is a value between 0 and 1, used to quantify the degree of difference between the current state and the model's predicted state. The higher the value of St, the greater the difference between the current data and the model's prediction, i.e., the greater the probability of an anomaly. This anomaly detection mechanism is not only simple and intuitive, but also effectively captures subtle changes in the input data, making it particularly effective for scenarios such as vehicle electromagnetic environment monitoring.

[0075] At the application level, the Hidden Time Matrix (HTM) model, with its unique neuroscience principles, demonstrates significant advantages in processing time series data. Compared to traditional feature extraction methods, the HTM model can directly process raw time series data without the need for complex preprocessing and feature selection. This characteristic not only preserves the integrity and contextual relevance of the data but also enables a more accurate simulation and understanding of data dynamics, thereby improving the accuracy and robustness of anomaly detection.

[0076] Furthermore, the HTM model supports online learning, which is another major advantage. As it continuously receives new data, the HTM model can constantly update its internal state to adapt to changes in data patterns. This characteristic makes the HTM model particularly suitable for dynamically changing environments, such as real-time anomaly detection in vehicle electromagnetic environment monitoring. The online learning capability allows the HTM model to continuously learn and adapt to new data patterns without relying on a predefined training set. This feature overcomes the dependence of traditional methods on existing risk data in vehicle electromagnetic environment monitoring, enabling the HTM model to effectively identify electromagnetic environments that may place vehicles at risk.

[0077] The Hidden Time Metric (HTM) model, with its unique sparse coding mechanism and online learning capabilities, demonstrates significant advantages in processing time-series data and identifying anomalous behavior. In the field of vehicle electromagnetic environment monitoring, the HTM model not only overcomes the limitations of traditional methods but also provides a more accurate and reliable anomaly detection solution.

[0078] HTM (Hypertext Transfer Model) is an advanced anomaly detection technique with a powerful ability to assign anomaly scores at each time step in continuous time series data. However, relying solely on these scores to determine the existence of anomalous behavior within a specific time period requires more in-depth analysis and judgment. In previous HTM anomaly detection practices, setting anomaly score thresholds was commonly used to define anomalies. However, using anomaly score thresholds for anomaly judgment has another significant drawback. Because the behavioral patterns and characteristics of electromagnetic risks vary, the distribution of anomaly scores for each sample will also exhibit significant differences. Therefore, for each sample to be detected, it is not easy to accurately define the anomaly score threshold, let alone find a unified standard applicable to all samples.

[0079] In view of the above problems, a more scientific, unified, and effective standard is needed to improve the application of HTM in vehicle electromagnetic environment monitoring and alarm. Such a standard should be able to comprehensively consider various characteristics of vehicle electromagnetic risks, as well as the differences between different samples, so as to achieve more accurate and reliable detection of electromagnetic risks.

[0080] The sliding window technique is widely used in time series data processing, especially in anomaly detection. It achieves goals such as feature extraction, pattern recognition, or prediction by moving a fixed-size window across the data sequence and performing specific operations on each window. The flexibility of the sliding window lies in its ability to adjust the window size and step size to adapt to different data characteristics and anomaly patterns. This flexibility allows for data analysis at different time scales, thereby detecting anomalous behavior at different levels.

[0081] In vehicle electromagnetic environment monitoring and alarm scenarios, the sliding window technique can be used to extract features from time-series data to identify potential anomaly patterns. By moving the window and calculating anomaly scores within each window, anomalies at different time periods can be captured, and the characteristics and trends of these anomalous time periods can be further analyzed. This method provides a more comprehensive understanding of potential electromagnetic risks and a basis for developing effective detection and response strategies.

[0082] Anomalies in the electromagnetic environment of vehicles can manifest in two ways: either they are very noticeable at a specific point in time, or the electromagnetic environment remains at a relatively high level over a considerable period. To integrate the anomaly detection results from the 1D-CNN and HTM models, improve the accuracy of anomaly detection, and address the shortcomings of traditional threshold detection methods, this invention proposes a novel monitoring strategy. This strategy not only focuses on obvious anomalies but also on the average height of anomaly scores over a specific time period. A new metric, S, is defined here. k S kThis represents the weighted sum of outlier scores at all times within a predetermined time window. In this way, even if two time windows have the same average outlier score, if a significant outlier occurs within one window, its S... k The value will also be significantly higher than the other window. This design allows the algorithm to more sensitively capture significant anomalies, thereby improving the effectiveness of electromagnetic environment anomaly detection.

[0083] The calculation method is as follows:

[0084]

[0085]

[0086] In the formula, S k S represents the weighted sum of outlier scores at all times within a predetermined time window. i W represents the outlier score at time i. i Let represent the weight corresponding to the abnormal score at time i, and l represent the length of the moving window.

[0087] This method also considers cases where there are no obvious single anomalies but the overall anomaly level is high. By calculating S... k The value reflects the relatively high average anomaly score during periods of electromagnetic environmental anomalies. This comprehensive index setting aims to achieve efficient monitoring and accurate identification of electromagnetic environment monitoring through precise calculation and analysis of anomaly scores.

[0088] During normal operation, the vehicle collects electromagnetic environment data through various acquisition nodes and aggregates it to the host computer via a router. The host computer calls the trained electromagnetic environment alarm model in real time to analyze the electromagnetic data collected by each node. If the current test data is similar to the electromagnetic data collected in a past fault environment or if the current test data deviates significantly from the dataset contained in the model, the electromagnetic environment alarm model calls the resources of the host computer to issue an alarm to the vehicle operator.

[0089] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A distributed monitoring and alarm system for vehicle electromagnetic environment, characterized in that, include: The system includes a host computer, an in-vehicle router, and multiple electromagnetic data acquisition nodes distributed throughout the vehicle. Each electromagnetic data acquisition node includes a communication module, an MCU, a data acquisition module, and acquisition probes. The communication module connects to the host computer via the in-vehicle router, receives instructions from the host computer, and sends the received instructions to the MCU. The MCU controls the data acquisition module according to the instructions from the host computer to acquire electromagnetic environment data through the acquisition probes. After processing the acquired electromagnetic environment data, the MCU sends it to the host computer via the communication module. The host computer processes the received electromagnetic environment data and uses the processed data to train the electromagnetic environment alarm model. When the vehicle is running normally, it collects electromagnetic environment data through various electromagnetic acquisition nodes and sends it to the host computer via the vehicle router. The host computer calls the trained electromagnetic environment alarm model in real time to analyze the electromagnetic data collected by each node. If the current test data is similar to the electromagnetic data collected in a certain fault environment in the past, or if the current test data deviates significantly from the dataset contained in the model, the electromagnetic environment alarm model calls the resources of the host computer to issue an alarm to the vehicle operator. Training the electromagnetic environment alarm model using the processed data also includes: By processing abnormal alarms from the outputs of 1D-CNN and HTM models using a moving window, abnormal electromagnetic environments of vehicles at different time periods can be captured, as follows: ； ； In the formula, This represents the weighted sum of outlier scores at all times within a predetermined time window. This represents the outlier score at time i. This represents the weight corresponding to the outlier score at time i. Indicates the length of the moving window; Finally, according to The value determines the current abnormal electromagnetic environment of the vehicle.

2. The vehicle electromagnetic environment distributed monitoring and alarm system according to claim 1, characterized in that, The collected electromagnetic environment data includes vehicle electromagnetic environment parameters, timestamps, and geographical location information. Data collection needs to cover different time periods, different road conditions, and different scenarios.

3. The vehicle electromagnetic environment distributed monitoring and alarm system according to claim 1, characterized in that, The host computer processes the received electromagnetic environment data in the following specific ways: Interpolation and noise reduction processing: Interpolation can be used to handle missing values ​​or delete invalid data segments. Use filters or smoothing algorithms to reduce data noise; Category labeling: If the collected data is from when the vehicle is driving normally, the collected data will be recorded in the first dataset. If electromagnetic interference or equipment failure occurs during the data collection process, the collected data will be recorded in the second dataset and the type of failure will be marked. Enhancement: The `find_peaks` function is used to extract peak values ​​from the labeled data as input features. Fixed-length data sequences are generated by truncating and expanding the data near each peak, and the category labels are converted into one-hot encoded form.

4. The vehicle electromagnetic environment distributed monitoring and alarm system according to claim 3, characterized in that, Training the electromagnetic environment alarm model using the processed data specifically includes: For the labeled data, a 1D-CNN model is constructed. The 1D-CNN model includes an input layer, a convolutional layer, a pooling layer, two fully connected layers, and an output layer. The hidden layer uses a convolutional layer with 32 filters of size 3, a stride of 2, a ReLU activation function, and a same padding method to extract local features from the data and introduce non-linearity. The pooling layer downsamples through a pooling window of size 2. The first fully connected layer contains 100 neurons, and the second layer contains 50 neurons. The activation function for both is ReLU. The output layer has multiple neurons, corresponding to various electromagnetic environment states of the vehicle, and the Softmax function is used to convert the output into a probability distribution. The 1D-CNN model is trained using labeled data. The Adam optimizer is used to accelerate the convergence speed of the model. The classification cross-entropy is used as the loss function to measure the difference between the model's prediction results and the actual labels. The 1D-CNN model outputs anomaly alerts.

5. The vehicle electromagnetic environment distributed monitoring and alarm system according to claim 4, characterized in that, Training the electromagnetic environment alarm model using the processed data also includes: For unlabeled data, the Hidden Trace Meter (HTM) model is used. The input encoding stage of the HTM model is responsible for converting the input data into a binary string composed of 0s and 1s, which serves as the input to the spatial pool. The spatial pool of the HTM model converts the binary string into a sparse binary pattern, generating a sparse distributed representation of the input data. The temporal pool of the HTM model represents the contextual pattern of the input data as the active state of the neurons, and uses the predicted state of the neurons to represent the input data pattern to be passed in at the next time step. The decoding stage of the HTM model uses a sparse distributed representation classifier as a decoder, which uses the relationship between the input data bits and the neurons to decode, and is responsible for converting the predicted neuron pattern output from the temporal pool into an output with the same format as the input data. The HTM model is trained using unlabeled data. The HTM model calculates an anomaly score between the sparse code P(x) of the current input x and the HTM model's prediction of the sparse code P(x). The anomaly score is a value between 0 and 1 used to quantify the degree of difference between the current state and the state predicted by the HTM model. The degree of difference is positively correlated with the probability of an anomaly.

Images