Multi-vehicle state intelligent monitoring and alarming method and system based on mobile communication

Abstract

The invention discloses a multi-vehicle state intelligent monitoring alarm method and system based on mobile communication, and relates to the technical field of intelligent alarm and safety monitoring, and the method comprises the steps: building a vehicle abnormal anchor point state vector library based on a historical vehicle state monitoring alarm log set; monitoring a plurality of vehicles according to a preset monitoring period through a state monitoring branch array connected with the mobile communication center to obtain a plurality of vehicle state monitoring data sequences; obtaining a plurality of vehicle low-dimensional dense state vectors through long-time and short-time low-dimensional dense compression; and transmitting to a mobile communication center and matching with a vehicle abnormal anchor point state vector library to determine alarm information. According to the invention, the technical problem that the existing alarm device is difficult to efficiently process multi-vehicle complex state data and cannot accurately match abnormity to realize accurate alarm requirements in multi-vehicle state monitoring is solved, and the purposes that the alarm device efficiently processes the multi-vehicle complex state data, and the alarm efficiency is improved are achieved. And the technical effects of multi-vehicle state intelligent monitoring and accurate alarm requirements are met.

Description

Technical Field

[0001] This invention relates to the field of intelligent alarm and security monitoring technology, and in particular to a method and system for intelligent monitoring and alarming of multiple vehicle statuses based on mobile communication. Background Technology

[0002] Real-time monitoring and anomaly warning of multiple vehicle statuses are crucial for traffic safety and fleet operation efficiency, with alarm devices being a key component in ensuring vehicle safety. Current technologies for vehicle status monitoring and alarms largely rely on traditional single-sensor or simple communication alarm devices, playing a role in monitoring the localized status of a single vehicle. However, with the increasing demand for collaborative management of multiple vehicles, traditional alarm devices have limitations in real-time data acquisition, large-scale status analysis, and accurate anomaly matching. They struggle to efficiently process complex status data from multiple vehicles, leading to alarm delays or false alarms, and ultimately failing to meet the needs for intelligent monitoring and accurate alarms for multiple vehicle statuses. Summary of the Invention

[0003] This application provides a method and system for intelligent monitoring and alarming of multiple vehicle statuses based on mobile communication, which solves the technical problem that existing alarm devices are unable to efficiently process complex status data of multiple vehicles and accurately match anomalies to achieve the need for precise alarms in multi-vehicle status monitoring.

[0004] The first aspect of this application provides a multi-vehicle status intelligent monitoring and alarm method based on mobile communication. The method includes: constructing a vehicle anomaly anchor point status vector library based on a historical vehicle status monitoring alarm log set; performing status monitoring on multiple vehicles separately according to a preset monitoring cycle through a status monitoring branch array connected to a mobile communication center to obtain multiple vehicle status monitoring data sequences; traversing the multiple vehicle status monitoring data sequences and performing long-short-time low-dimensional dense compression to obtain multiple vehicle low-dimensional dense status vectors; transmitting the multiple vehicle low-dimensional dense status vectors to the mobile communication center respectively, matching them with the vehicle anomaly anchor point status vector library, and determining alarm information based on the matching results.

[0005] A second aspect of this application provides a multi-vehicle status intelligent monitoring and alarm system based on mobile communication. The system includes: a vehicle anomaly anchor point status vector library construction module, which constructs a vehicle anomaly anchor point status vector library based on a set of historical vehicle status monitoring alarm logs; a vehicle status monitoring data sequence acquisition module, used to perform status monitoring on multiple vehicles according to a preset monitoring cycle through a status monitoring branch array connected to a mobile communication center, thereby obtaining multiple vehicle status monitoring data sequences; a vehicle low-dimensional dense status vector acquisition module, used to traverse the multiple vehicle status monitoring data sequences and perform long-short-time low-dimensional dense compression to obtain multiple vehicle low-dimensional dense status vectors; and an alarm information acquisition module, used to transmit the multiple vehicle low-dimensional dense status vectors to the mobile communication center respectively, match them with the vehicle anomaly anchor point status vector library, and determine alarm information based on the matching results.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application constructs a vehicle anomaly anchor point state vector library based on a set of historical vehicle status monitoring alarm logs. A status monitoring branch array, connected to a mobile communication center, collects multi-vehicle status monitoring data sequences at preset intervals. These data are then compressed into low-dimensional dense state vectors using a long-short-time low-dimensional compression process. These vectors are transmitted to the mobile communication center and matched with the anomaly anchor point state vector library to determine alarm information. Furthermore, the application allows for resource reallocation of the monitoring branch array based on the alarm information, thereby achieving intelligent monitoring and accurate alarms for multiple vehicle states. This makes the multi-vehicle status monitoring alarm results more accurate and reliable, achieving the technical effect of enabling the alarm device to efficiently process complex multi-vehicle status data and meeting the technical requirements for intelligent monitoring and accurate alarms for multiple vehicle states. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart illustrating the intelligent monitoring and alarm method for multiple vehicle statuses based on mobile communication provided in this application embodiment.

[0009] Figure 2 This is a schematic diagram of the structure of a multi-vehicle status intelligent monitoring and alarm system based on mobile communication provided in an embodiment of this application.

[0010] Figure labeling: Module 1 for constructing vehicle abnormal anchor point state vector library, Module 2 for acquiring vehicle state monitoring data sequence, Module 3 for acquiring vehicle low-dimensional dense state vector, and Module 4 for acquiring alarm information. Detailed Implementation

[0011] This application provides a method and system for intelligent monitoring and alarming of multiple vehicle statuses based on mobile communication, which solves the technical problem that existing alarm devices are unable to efficiently process complex status data of multiple vehicles and accurately match anomalies to achieve the need for precise alarms in multi-vehicle status monitoring.

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0013] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0014] Example 1, as Figure 1 As shown, a multi-vehicle status intelligent monitoring and alarm method based on mobile communication is provided, wherein the method includes: Step A100: Based on the historical vehicle status monitoring alarm log set, construct a vehicle anomaly anchor point status vector library.

[0015] In this embodiment, the historical vehicle status monitoring alarm log is a collection of logs recording past vehicle status monitoring and alarm situations, including information related to abnormal vehicle states. The vehicle abnormal anchor point state vector is a representative abnormal state vector obtained from the historical vehicle abnormal state vector collection after semantic relevance analysis and clustering filtering.

[0016] Specifically, a vehicle anomaly anchor point state vector library is constructed based on the historical vehicle status monitoring alarm log set. This includes extracting the historical vehicle anomaly state vector set, obtaining the set of semantic correlation coefficient mean values ​​through pairwise semantic correlation analysis, taking the first m corresponding vectors as the historical vehicle anomaly cluster state vector set, and then using this anchor point to filter and construct the vehicle anomaly anchor point state vector library. The specific steps are explained in detail in A110-A140.

[0017] Step A200: The status monitoring branch array, which is connected to the mobile communication center, performs status monitoring on multiple vehicles according to a preset monitoring cycle to obtain multiple vehicle status monitoring data sequences.

[0018] In this embodiment, the mobile communication center is the core processing node that maintains a communication connection with the condition monitoring branch array and is responsible for receiving vehicle status data transmitted by the condition monitoring branch array.

[0019] Optionally, firstly, a status monitoring branch array connected to the mobile communication center is deployed on the vehicle. This array, through hardware modules such as the OBD interface and sensor nodes, covers monitoring dimensions such as basic vehicle driving, powertrain, tires, and suspension systems. The preset monitoring cycle is set according to the vehicle type and application scenario; for example, it is set to 1 second / time for commercial vehicles due to their complex operating conditions, and 5 seconds / time for passenger vehicles.

[0020] Once monitoring is initiated, the branch array collects vehicle status data periodically: for basic driving conditions, it simultaneously acquires vehicle speed, engine speed, accelerator pedal position, brake pedal position, and driving mode; for the powertrain, it collects engine operating temperature, fuel consumption rate, and engine load for gasoline vehicles, and current battery level and charging status for electric vehicles; and for the tire and suspension systems, it collects tire pressure, tire temperature, tire wear, suspension pressure, and vibration frequency.

[0021] Each data collection from each vehicle is accompanied by a timestamp. The branch array integrates the continuous collection results from the same vehicle in chronological order to form a vehicle condition monitoring data sequence, which contains full-dimensional condition parameters at each time point. Multiple vehicles execute this process synchronously, ultimately obtaining multiple vehicle condition monitoring data sequences, enabling continuous and multi-dimensional characterization of the conditions of multiple vehicles.

[0022] By periodically collecting and constructing time series data through the condition monitoring branch array, dynamic data covering the core operating dimensions of the vehicle is fully covered, providing continuous and accurate condition baseline support for subsequent anomaly identification.

[0023] Step A300: Traverse the multiple vehicle state monitoring data sequences and perform long-short-time low-dimensional dense compression to obtain multiple vehicle low-dimensional dense state vectors.

[0024] In one embodiment of this application, multiple vehicle condition monitoring data sequences are traversed and subjected to long-term and short-term low-dimensional dense compression to obtain multiple vehicle low-dimensional dense state vectors. The first vehicle condition monitoring data sequence needs to be extracted, compressed according to the long-term and short-term low-dimensional dense compression bandwidths respectively, and the state analyzer is called to identify and obtain the long-term and short-term low-dimensional dense compressed state vectors. The first vehicle low-dimensional dense state vector is determined interactively, and other sequences are processed in the same way to obtain multiple vehicle low-dimensional dense state vectors. The specific steps are described in detail in A310-A360.

[0025] Step A400: Transmit the multiple low-dimensional dense state vectors of vehicles to the mobile communication center respectively, match them with the vehicle abnormal anchor point state vector library, and determine the alarm information based on the matching result.

[0026] Specifically, multiple low-dimensional dense state vectors of vehicles are transmitted to the mobile communication center via a mobile communication link. Encryption protocols are used during transmission to ensure data integrity and security, preventing leakage or tampering of state information during transmission. Upon receiving these vectors, the mobile communication center activates a matching engine to compare them with a database of abnormal vehicle anchor point state vectors.

[0027] During matching, for each vehicle's low-dimensional dense state vector, the similarity between it and all anchor vectors in the vehicle anomaly anchor state vector library is calculated. The similarity calculation can use Euclidean distance as the metric, the same as in step A122, to generate a similarity set for each vector by quantifying the correlation between the current vehicle state and the known anomaly states.

[0028] The system determines whether to trigger an alarm based on the similarity set. If the similarity between a vehicle vector and at least one abnormal anchor vector in the database reaches or exceeds a preset alarm threshold, it indicates that the vehicle's current state is close to a known abnormal pattern, and the mobile communication center will generate an alarm message containing the vehicle identifier, abnormal type, and suspected abnormal parameters. If all similarities are below the threshold, the vehicle is considered to be in a normal state, and no alarm is generated.

[0029] By transmitting the low-dimensional dense state vector of vehicles to the mobile communication center and matching it with the abnormal anchor point database, the abnormal state of vehicles can be accurately identified based on the similarity results, thus realizing intelligent monitoring and timely alarm of multiple vehicle states.

[0030] Furthermore, step A100 in the method provided in this application embodiment includes: A110: Extract the vehicle abnormal status vector from the historical vehicle status monitoring alarm log set to obtain the historical vehicle abnormal status vector set.

[0031] A120: Perform pairwise semantic correlation analysis on the set of historical vehicle abnormal state vectors to obtain the set of mean semantic correlation coefficients of historical vehicle abnormal state vectors.

[0032] A130: Extract the mean semantic correlation coefficient of the top m historical vehicle abnormal state vectors from the set of mean semantic correlation coefficients of the historical vehicle abnormal state vectors, and use the corresponding historical vehicle abnormal state vectors as the set of historical vehicle abnormal clustering state vectors.

[0033] A140: Based on the set of historical vehicle anomaly clustering state vectors, anchor points are filtered in the set of historical vehicle anomaly state vectors to construct the vehicle anomaly anchor point state vector library.

[0034] Specifically, the first step is to collect a sufficient number of historical vehicle status monitoring alarm logs. For example, a logistics fleet's vehicle monitoring system recorded approximately 100,000 logs over the past three years. These logs cover vehicle status data for different vehicle models and driving environments, including specific status information at the time of each alarm, such as parameters like engine speed, coolant temperature, and brake system pressure, as well as the alarm trigger time and vehicle identifier. This data together constitutes the historical vehicle status monitoring alarm log set.

[0035] From these logs, it's necessary to filter out vehicle status data directly related to the alarms. For example, in the logs, when the vehicle triggers a high-temperature alarm due to coolant temperature exceeding 110°C, the corresponding status data is engine speed 2500 rpm, coolant temperature 115°C, and brake system pressure 8 MPa; when the tire pressure is below 1.8 bar, triggering a low-pressure alarm, the corresponding status data is engine speed 1800 rpm, coolant temperature 85°C, and brake system pressure 6 MPa. These specific status parameters recorded when alarms are triggered constitute the raw data of the vehicle's abnormal state.

[0036] Next, these raw vehicle abnormal state data are converted into vector form, i.e., raw historical vehicle abnormal state vectors. During vectorization, different state parameters are used as the dimensions of the vector, and parameter values ​​are used as the components of the corresponding dimensions. For example, the state data of the high-temperature alarm can be converted into vector [2500, 115, 8], and the state data of the low-pressure alarm can be converted into vector [1800, 85, 6]. By performing the same vectorization process on the state data of all alarm records in the historical vehicle status monitoring alarm log set, all valid abnormal state data are extracted, and after conversion, the corresponding vehicle abnormal state vectors are obtained. The set of these vectors is the historical vehicle abnormal state vector set.

[0037] Then, pairwise semantic correlation analysis is performed on the set of historical vehicle abnormal state vectors to obtain the set of their semantic correlation coefficient mean values. First, the first historical vehicle abnormal state vector needs to be extracted. Pairwise semantic correlation analysis is performed on it and the vectors in the set using Euclidean distance to obtain the set of semantic correlation coefficients of the first historical vehicle abnormal state. The mean of this set is processed to obtain the mean of the semantic correlation coefficients of the first historical vehicle abnormal state vector. The same analysis is then performed on the set to obtain the set of the mean of the semantic correlation coefficients of the historical vehicle abnormal state vector. The specific steps are explained in detail in A121-A124.

[0038] After obtaining the set of mean semantic correlation coefficients for historical vehicle anomaly state vectors, the first step is to sort all the means in this set in descending order. Assume this set contains 8000 means, ranging from 0.0012 to 0.0045. These means correspond to the average correlation between the 8000 historical vehicle anomaly state vectors and other vectors. After descending sorting, the means are distributed from high to low. The larger mean values ​​at the beginning indicate a stronger overall correlation between the corresponding vector and other vectors in the set, better reflecting the common characteristics of a certain type of anomaly state.

[0039] Next, based on the requirements for anomaly type coverage and clustering accuracy in practical applications, the specific value of m is determined. If it is necessary to cover most common anomaly types and ensure the representativeness of the clustering, m=200 can be selected. In the mean set after descending order, the mean of the top 200 values ​​is extracted. The numerical range of these means is significantly higher than that of other means in the set, indicating that the corresponding vectors have stronger correlation and representativeness in the overall anomaly state.

[0040] Subsequently, by establishing the correspondence between the mean and the vector, the original historical vehicle anomaly state vectors corresponding to these 200 means were found. For example, the first-ranked mean of 0.0045 corresponds to the vector [2400, 118, 7.5], which records the key parameters when a certain type of truck engine experiences a high-temperature alarm; the 50th-ranked mean of 0.0042 corresponds to the vector [1700, 82, 5.8], reflecting the state characteristics of a small passenger car with low tire pressure. These vectors were extracted and summarized one by one to form a historical vehicle anomaly cluster state vector set containing 200 vectors, providing a highly representative core reference vector for subsequent anchor point selection based on cluster vectors.

[0041] Finally, based on the historical vehicle anomaly clustering state vector set, anchor point selection is performed on the historical vehicle anomaly state vector set to construct a vehicle anomaly anchor point state vector library. The historical vehicle anomaly state vector set needs to be added to the neighborhood of the historical vehicle anomaly clustering state vector corresponding to the maximum semantic relevance to obtain the historical vehicle anomaly clustering state vector neighborhood set. Then, the mean drift algorithm is used to select anchor points for this neighborhood set, thereby constructing the vehicle anomaly anchor point state vector library. The specific steps are explained in detail in A141-A142.

[0042] Furthermore, step A120 in the method provided in this application embodiment includes: A121: Extract the first historical vehicle abnormal state vector from the set of historical vehicle abnormal state vectors.

[0043] A122: By using Euclidean distance, perform pairwise semantic correlation analysis on the first historical vehicle abnormal state vector and the historical vehicle abnormal state vector to obtain the set of semantic correlation coefficients of the first historical vehicle abnormal state.

[0044] A123: The semantic correlation coefficient set of the first historical vehicle abnormal state is averaged to obtain the average semantic correlation coefficient of the first historical vehicle abnormal state vector.

[0045] A124: Perform pairwise semantic correlation analysis on the set of historical vehicle abnormal state vectors to obtain the set of mean semantic correlation coefficients of the historical vehicle abnormal state vectors.

[0046] In this embodiment of the application, Euclidean distance is a distance metric used to analyze the semantic correlation between historical vehicle abnormal state vectors. The smaller the distance value, the more similar the vehicle abnormal states represented by the two vectors are, and the higher the semantic correlation; the larger the distance value, the lower the semantic correlation.

[0047] Optionally, when processing the set of historical vehicle abnormal state vectors, a vector is first selected from the set as the first historical vehicle abnormal state vector. Assuming that the set contains 8000 vectors, for example, vector V1[2500,115,8] is selected as the first vector, which corresponds to the state parameters when a vehicle experiences a high temperature alarm.

[0048] Next, the semantic relevance of the first historical vehicle abnormal state vector to all other vectors in the set is calculated using Euclidean distance. Taking V1 as an example, the Euclidean distance between it and vector V2[1800,85,6] is calculated, and the result is √[(2500-1800)]. 2 +(115-85) 2 +(8-6) 2The correlation coefficient is approximately 1 / 700.65 ≈ 0.0014. The Euclidean distance between V1 and vector V3[2200,100,7] is approximately 300.38, with a correlation coefficient of approximately 0.0033. This process is repeated to obtain the correlation coefficients between V1 and the remaining 7999 vectors, forming the first historical vehicle anomaly state semantic correlation coefficient set containing 7999 coefficients.

[0049] Subsequently, the set of coefficients is averaged. If the sum of the 7999 coefficients is 16, then the mean is 16 ÷ 7999 ≈ 0.002, which is the mean of the semantic correlation coefficient of the first historical vehicle abnormal state vector.

[0050] Then, the same process is followed to iterate through each vector in the set. Taking vector V2 as the first vector, its Euclidean distance to the other 7999 vectors is calculated to obtain the set of correlation coefficients, and then the mean is calculated; the mean is obtained after processing vector V3, until all 8000 vectors have been calculated, and finally a set of semantic correlation coefficient mean values ​​of historical vehicle abnormal state vectors containing 8000 mean values ​​is formed.

[0051] By extracting a single vector, calculating its Euclidean distance correlation coefficient with other vectors, taking the mean, and traversing all vectors, a set of mean semantic correlation coefficients of historical vehicle abnormal state vectors was obtained, providing a quantitative basis for subsequent selection of representative clustered state vectors.

[0052] Furthermore, step A140 in the method provided in this application embodiment includes: A141: Add the set of historical vehicle abnormal state vectors to the neighborhood of the historical vehicle abnormal clustering state vector corresponding to the maximum semantic relevance, and obtain the neighborhood set of historical vehicle abnormal clustering state vectors.

[0053] A142: The mean shift algorithm is used to filter the neighborhood set of the historical vehicle anomaly clustering state vectors to construct the vehicle anomaly anchor point state vector library.

[0054] In this embodiment, the mean drift algorithm is an algorithm used to filter anchor points for the neighborhood set of historical vehicle anomaly clustering state vectors.

[0055] Specifically, when processing the set of historical vehicle anomaly state vectors, it is necessary to find the corresponding historical vehicle anomaly cluster state vector for each vector in the set. Specifically, for any vector in the set of historical vehicle anomaly state vectors, its semantic relevance coefficient with all vectors in the set of historical vehicle anomaly cluster state vectors must be calculated, and the cluster vector corresponding to the maximum value must be determined. Assuming the set of historical vehicle anomaly state vectors contains 8000 vectors and the set of historical vehicle anomaly cluster state vectors contains 200 vectors, then each vector needs to have its relevance calculated with these 200 cluster vectors.

[0056] For example, if the correlation coefficients between a vector Vx and 200 cluster vectors range from 0.0015 to 0.0042, with the highest coefficient (0.0042) being with cluster vector Vc3, then Vx is added to the neighborhood of Vc3. This process is repeated for 8000 vectors. Ultimately, each cluster vector Vc1-Vc200 will have a certain number of correlated vectors clustered in its neighborhood, forming a set of 200 historical vehicle anomaly cluster state vector neighborhoods.

[0057] Next, the mean-shift algorithm is applied to the neighborhood of each historical vehicle anomaly clustering state vector to select anchor points. The mean-shift algorithm iteratively calculates the density center of vectors within a neighborhood, gradually moving towards the region with the highest density, ultimately finding the most representative vector in each neighborhood as the anchor point. Taking the neighborhood of Vc1 as an example, this neighborhood contains multiple related vectors. The algorithm determines that vector Va1 is the density center by calculating the distance and distribution density between vectors. This vector has the most similar anomaly state vectors clustered around it, thus becoming the anchor point for this neighborhood. After processing each of the 200 neighborhoods, the corresponding anchor point vectors are selected, and these anchor point vectors together constitute the vehicle anomaly anchor point state vector library.

[0058] By assigning historical vehicle anomaly state vectors to the neighborhood of corresponding cluster vectors and using the mean shift algorithm to filter anchor points, a vehicle anomaly anchor point state vector library was constructed, providing an accurate set of benchmark vectors for subsequent real-time vehicle state vector anomaly matching.

[0059] Furthermore, step A300 in the method provided in this application embodiment includes: A310: Extract the first vehicle condition monitoring data sequence from the plurality of vehicle condition monitoring data sequences.

[0060] A320: Compress the first vehicle status monitoring data sequence according to the long-term low-dimensional dense compression bandwidth to obtain the first long-term vehicle low-dimensional dense compressed status monitoring data sequence.

[0061] A330: Compress the first vehicle status monitoring data sequence according to the short-time low-dimensional dense compression bandwidth to obtain the first short-time vehicle low-dimensional dense compressed status monitoring data sequence.

[0062] A340: Call the state analyzer to perform state identification on the first long-term vehicle low-dimensional dense compressed state monitoring data sequence and the first short-term vehicle low-dimensional dense compressed state monitoring data sequence respectively, and obtain the first long-term vehicle low-dimensional dense compressed state vector and the first short-term vehicle low-dimensional dense compressed state vector.

[0063] A350: Interact with the first long-term vehicle low-dimensional dense compressed state vector and the first short-term vehicle low-dimensional dense compressed state vector to determine the first vehicle low-dimensional dense state vector.

[0064] A360: Similarly, the multiple vehicle status monitoring data sequences are subjected to long-short-time low-dimensional dense compression to obtain the multiple vehicle low-dimensional dense state vectors.

[0065] Specifically, after acquiring multiple vehicle condition monitoring data sequences, these sequences are first sorted and labeled. Each sequence is associated with a unique vehicle identifier, such as the vehicle VIN (Vehicle Identification Number, a 17-character code consisting of letters and numbers used to uniquely identify a vehicle, containing important information such as the vehicle's manufacturer, place of manufacture, year of manufacture, model, and engine code), and also contains continuous condition data of the vehicle within a preset monitoring period. Assuming that a total of 30 vehicles are monitored through the condition monitoring branch array, this forms 30 vehicle condition monitoring data sequences. Each sequence records the vehicle's condition information every 5 minutes in chronological order, covering multiple parameters such as vehicle speed, engine temperature, and tire pressure, as shown in Table 1.

[0066] Table 1: Example Table of Vehicle Condition Monitoring Data Sequences

[0067] Subsequently, the first vehicle condition monitoring data sequence is determined from these sequences according to preset extraction rules. The extraction rules can be based on the sorting of vehicle identifiers, such as ascending order of alphanumeric VIN codes, with the sequence corresponding to the first vehicle in the sorted sequence being taken as the first sequence. For consistency, a fixed sorting rule is usually adopted. For example, after sorting 30 sequences by VIN code, the VIN code of the first vehicle is LHGCM82633A000001, and its corresponding sequence is the first vehicle condition monitoring data sequence.

[0068] The extracted first vehicle status monitoring data sequence needs to undergo integrity verification to ensure that it contains all preset data points within the monitoring period without any missing or abnormal interruptions. If some data in the sequence is missing, it needs to be supplemented by interpolation or based on adjacent valid data. For example, if the engine temperature data at a certain time point is missing, it can be supplemented by fitting the temperature values ​​of the preceding and following 10 minutes. After verification and supplementation, the first sequence completely preserves the vehicle's status change trajectory within the monitoring period, including timestamps for all monitoring time periods, as well as parameters such as vehicle speed and engine speed at each time point.

[0069] Next, the first vehicle status monitoring data sequence is compressed according to the long-term low-dimensional dense compression bandwidth to obtain the first long-term vehicle low-dimensional dense compressed status monitoring data sequence. The first vehicle status monitoring data needs to be extracted, and the first search data is retrieved based on it and the long-term bandwidth. The cosine similarity between the two is calculated. If it is greater than or equal to the preset threshold, the search data is removed and compression continues to obtain the sequence. The specific steps are explained in detail in A321-A323.

[0070] When processing the first vehicle status monitoring data sequence, compression is first performed using a short-time, low-dimensional, dense compression bandwidth. Assume the first vehicle status monitoring data sequence records the continuous operation status of a ride-hailing vehicle for two hours, collected every minute, totaling 120 data points, including parameters such as real-time vehicle speed, instantaneous engine fuel consumption, and battery output power. The short-time, low-dimensional, dense compression bandwidth is set to 2 minutes, meaning the next data point is retrieved every 2 minutes from the current data point.

[0071] The first vehicle status monitoring data in the sequence is extracted as [50km / h, 8L / h, 30kW] at time t0. Based on this data and a 2-minute bandwidth, the first data retrieved at time t2 (t0+2 minutes) is [55km / h, 8.2L / h, 32kW]. The cosine similarity between the two is calculated, and the result is 0.82. If the preset similarity threshold is 0.8, then since 0.82≥0.8, the first data is discarded, and the data at time t4 is retrieved again using t0 as the baseline. If the calculation result is 0.78<0.8, then the data at t0 and t2 are retained, and the retrieval continues using t2 as the baseline. By traversing the sequence according to this logic, the original 120 data points are compressed into dozens of data points after a short time, forming the first short-time low-dimensional dense compressed vehicle status monitoring data sequence. This sequence can reflect the details of the vehicle's status fluctuations within every 2 minutes.

[0072] Subsequently, the state analyzer is invoked to process the first long-term and first short-term compressed sequences respectively. For the first long-term vehicle low-dimensional dense compressed state monitoring data sequence, assuming it consists of multiple data points compressed with a 5-minute bandwidth, the state analyzer extracts trend features from the data, such as the average vehicle speed over 2 hours, the frequency of peak fuel consumption, and the power stability range, and transforms them into the first long-term vehicle low-dimensional dense compressed state vector, for example, [48km / h, 2 times / hour, 28-32kW]. For the first short-term compressed sequence, the state analyzer focuses on short-term abrupt change features, such as the maximum increase in vehicle speed within 2 minutes, the instantaneous peak fuel consumption, and the number of sudden power changes, and transforms them into the first short-term vehicle low-dimensional dense compressed state vector, for example, [5km / h / 2min, 8.5L / h, 1 time].

[0073] Furthermore, the state analyzer, as a neural network model, is constructed with the core objective of processing temporally compressed data. The model employs a CNN-LSTM hybrid architecture. The input layer dimension is set according to the number of features in the compressed sequence. If each data point in a long-term compressed sequence contains three parameters—vehicle speed, fuel consumption, and power—then the input layer dimension is (x, 3), corresponding to the temporal features of x data points. The input layer dimension for short-term compressed sequences is (y, 3), adapting to the structure of y data points. After the input layer, a convolutional layer is connected, using 16 3×3 convolutional kernels to extract local features, with ReLU as the activation function. The output of the convolutional layer is connected to two LSTM layers, each with 32 neurons, capturing temporal dependencies. Following the LSTM layers are two fully connected layers: the first layer has 16 neurons, and the second layer has the same number of neurons as the output vector dimension—3 neurons for the three features in the long-term vector and 3 neurons for the three features in the short-term vector. The output layer uses a linear activation function to ensure continuous numerical values ​​in the output.

[0074] The training process requires historical compressed sequences and corresponding labeled data. The input data consists of 100,000 long-term and short-term compressed sequences from the past three months; the labels are manually labeled state vectors, including trend features (long-term) and mutation features (short-term). During data preprocessing, all parameters are normalized, mapping the values ​​to the [0,1] interval. 80% of the data is used as the training set, and 20% as the validation set. The loss function is mean squared error, the optimizer is Adam, the initial learning rate is 0.001, and it decays to 0.8 every 5 epochs. Training iterations are performed for 50 epochs, with a batch size of 32. Model performance is monitored using the validation set loss; training stops when the validation loss does not decrease for three consecutive epochs. The final mean squared error of the model on the validation set is controlled to be within 0.01.

[0075] The model takes as input a tensor of a long-term or short-term compressed sequence. The long-term input has a shape of (x, 3), and the short-term input has a shape of (y, 3), both containing time-series data of parameters such as vehicle speed, fuel consumption, and power. The output is a low-dimensional dense state vector. The long-term output is such as [48km / h, 2 times / hour, 28-32kW], corresponding to the average vehicle speed, peak fuel consumption frequency, and stable power range. The short-term output is such as [5km / h / 2min, 8.5L / h, 1 time], corresponding to the maximum increase in vehicle speed, instantaneous peak fuel consumption, and number of power fluctuations. Both are numerical vectors reflecting the core features of the sequence.

[0076] Among them, short-term low-dimensional dense compression captures the subtle changes in the vehicle's state in the short term, while long-term low-dimensional dense compression preserves the overall trend in the long term. The state analyzer then transforms the two types of compressed sequences into corresponding vectors, providing basic data that includes both macroscopic features and microscopic details for subsequent vector interaction.

[0077] Then, the low-dimensional dense compressed state vectors of the first long-time and the first short-time vehicles are interacted to determine the low-dimensional dense state vector of the first vehicle. It is necessary to perform inner product mapping on the same type of state components to obtain the set of inner product mapping results for long and short time. The set is normalized to construct a low-dimensional dense correlation matrix. Then, the matrix is ​​used to interact with the first long-time vector to obtain the low-dimensional dense state vector of the first vehicle. The specific steps are explained in detail in A351-A353.

[0078] Subsequently, when processing multiple vehicle condition monitoring data sequences, the processing flow for each sequence remains consistent with that for the first vehicle condition monitoring data sequence. That is, for each sequence, it is first compressed using a long-term low-dimensional dense compression bandwidth to obtain the corresponding long-term vehicle low-dimensional dense compressed condition monitoring data sequence; then, it is compressed using a short-term low-dimensional dense compression bandwidth to obtain the corresponding short-term vehicle low-dimensional dense compressed condition monitoring data sequence. The state identification steps of the state analyzer are repeated to obtain multiple long-term vehicle low-dimensional dense compressed state vectors and multiple short-term vehicle low-dimensional dense compressed state vectors.

[0079] The remaining 29 sequences are processed using the same logic, with each sequence undergoing long-term compression, short-term compression, state recognition, and vector interaction. Ultimately, after processing, each of the 30 vehicle state monitoring data sequences corresponds to a low-dimensional dense state vector, forming a set containing 30 vectors.

[0080] By performing long-term and short-term low-dimensional dense compression and vector interaction on multiple vehicle status monitoring data sequences one by one, multiple low-dimensional dense status vectors that can comprehensively reflect the long-term and short-term status characteristics of each vehicle were obtained, laying the data foundation for subsequent transmission to the mobile communication center for anomaly matching.

[0081] Furthermore, step A320 in the method provided in this application embodiment includes: A321: Extract the first vehicle status monitoring data from the first vehicle status monitoring data sequence.

[0082] A322: Based on the first vehicle status monitoring data and the long-term low-dimensional dense compression bandwidth, the first vehicle status monitoring data sequence is retrieved to obtain the first retrieved vehicle status monitoring data.

[0083] A323: Calculate the cosine similarity between the first vehicle status monitoring data and the first retrieved vehicle status monitoring data. If the calculation result is greater than or equal to a preset similarity threshold, then remove the first retrieved vehicle status monitoring data and continue to compress the first vehicle status monitoring data sequence based on the first vehicle status monitoring data and the long-term low-dimensional dense compression bandwidth to obtain the first long-term vehicle low-dimensional dense compressed status monitoring data sequence.

[0084] Specifically, when performing long-term low-dimensional dense compression on the first vehicle condition monitoring data sequence, the first vehicle condition monitoring data is extracted from the sequence. For example, suppose the first vehicle condition monitoring data sequence records the condition of a truck for 2 consecutive hours, collected once every 30 seconds, for a total of 240 data points. The first data point is at time t0 [vehicle speed 55km / h, engine speed 1800r / min, fuel consumption 7.2L / h, coolant temperature 82℃].

[0085] Subsequently, a retrieval is performed based on the first vehicle status monitoring data and a long-term low-dimensional dense compression bandwidth. The long-term low-dimensional dense compression bandwidth is set to 5 minutes (300 seconds), meaning that starting from the first data t0, the retrieval proceeds to the second data point at t0+300 seconds, i.e., t... 10 The first vehicle status monitoring data obtained from the time data is [vehicle speed 57km / h, engine speed 1850r / min, fuel consumption 7.3L / h, coolant temperature 83℃].

[0086] Next, the cosine similarity between the first data point and the first retrieved data point is calculated. Using the cosine similarity formula, the similarity is 0.96. If the preset similarity threshold is 0.85, since 0.96 ≥ 0.85, it indicates that the vehicle states at these two times are highly similar; therefore, the first retrieved data point is discarded. Then, using the first data point t0 as a baseline, the data point t0+600 seconds is retrieved using a long-term bandwidth of 5 minutes. 20 Repeat the above calculation and filtering process for the time data.

[0087] After traversing and compressing the entire first vehicle condition monitoring data sequence, the original 240 data entries were reduced to a few dozen, forming the first long-term low-dimensional dense compressed vehicle condition monitoring data sequence. This retained data not only covers key condition information from different time periods but also avoids redundant and similar data.

[0088] By extracting the first data, searching by long-term bandwidth, calculating cosine similarity, and filtering out similar data, long-term low-dimensional dense compression of the first vehicle state monitoring data sequence was achieved, resulting in a concise compressed sequence that reflects long-term state characteristics, providing an efficient data foundation for subsequent state recognition.

[0089] Furthermore, step A324 in the method provided in this application embodiment includes: A324-1: If the calculation result is less than the preset similarity threshold, the first vehicle status monitoring data and the first retrieved vehicle status monitoring data are retained, and the first vehicle status monitoring data sequence is compressed based on the first retrieved vehicle status monitoring data and the long-term low-dimensional dense compression bandwidth to obtain the first long-term vehicle low-dimensional dense compressed status monitoring data sequence.

[0090] In one embodiment, when performing long-term low-dimensional dense compression on the first vehicle status monitoring data sequence, if the cosine similarity calculation result between the first vehicle status monitoring data and the first retrieved vehicle status monitoring data is less than a preset threshold, it needs to be processed according to a specific procedure. For example, suppose the first vehicle status monitoring data sequence records the continuous 3-hour operating status of a city bus, collected every 2 minutes, totaling 90 data points, including parameters such as vehicle speed, engine load, and remaining battery power. The first vehicle status monitoring data is [30km / h, 40%, 95%] at time t0, and the long-term low-dimensional dense compression bandwidth is set to 10 minutes, meaning the next data point is retrieved every 10 minutes from the current data point.

[0091] Based on the first data and a 10-minute bandwidth, the first retrieved vehicle status monitoring data at time t5 (t0+10 minutes) is [50km / h, 60%, 90%]. The cosine similarity between the two is calculated, yielding a result of 0.72, while the preset similarity threshold is 0.85. Since 0.72 < 0.85, it indicates a significant difference in vehicle status between the two times. Therefore, both the first data t0 and the first retrieved data t5 are retained in the compressed sequence.

[0092] Subsequently, using the first retrieved data t5 as the new baseline, the sequence was further searched based on a long-term low-dimensional dense compression bandwidth of 10 minutes to obtain t. 10 The second retrieval data at time (t5 + 10 minutes) [45km / h, 55%, 88%]. Calculate t5 and t...10 The cosine similarity is 0.78, which is still less than 0.85, so t is retained. 10 Data. Following this logic, the entire sequence is continuously traversed. Each time, the latest retained data is used as the basis to retrieve data for the next bandwidth interval, calculate the similarity, and determine whether to retain it.

[0093] After a complete compression process, the original 90 data points of the first vehicle status monitoring data sequence were reduced to a few dozen, forming the first long-term low-dimensional dense compressed vehicle status monitoring data sequence. These retained data points fully cover the 3-hour monitoring period, and each data point reflects significant changes in vehicle status, avoiding the loss of critical status information due to the removal of similar data.

[0094] By retaining the first and second digits and the retrieved data when the cosine similarity is less than a preset threshold, and continuing to compress based on the retrieved data, a low-dimensional dense compressed sequence that can accurately reflect the long-term state change characteristics of the vehicle was obtained, providing a high-quality data foundation for the subsequent extraction of long-term state vectors.

[0095] Furthermore, step A350 in the method provided in this application embodiment includes: A351: Perform inner product mapping on the same type of state components in the first long-term vehicle low-dimensional dense compressed state vector and the first short-term vehicle low-dimensional dense compressed state vector to determine the set of long-term and short-term inner product mapping results.

[0096] A352: Normalize the set of long and short time inner product mapping results to construct a low-dimensional dense correlation matrix.

[0097] A353: The first long-term vehicle low-dimensional dense compressed state vector is obtained by interacting with the low-dimensional dense correlation matrix.

[0098] Optionally, when interacting with the first long-term and first short-term low-dimensional dense compressed state vectors of the vehicle, an inner product mapping is first performed on the state components of the same type in both vectors. Assume the first long-term vector is [average vehicle speed 40 km / h, average engine temperature 85℃, average fuel consumption 6 L / h], containing components reflecting the long-term stable state of the vehicle; the first short-term vector is [instantaneous vehicle speed 55 km / h, instantaneous engine temperature 90℃, instantaneous fuel consumption 8 L / h], containing components reflecting the short-term dynamic changes of the vehicle. The state components of the same type are vehicle speed, engine temperature, and fuel consumption. An inner product mapping is performed on these components, i.e., 40×55=2200, 85×90=7650, 6×8=48, yielding the set of long-term and short-term inner product mapping results [2200, 7650, 48]. This set quantifies the correlation between the long-term and short-term state components.

[0099] Next, the set of long-time and short-time inner product mapping results is normalized to construct a low-dimensional dense correlation matrix. The normalization process maps the values ​​to the interval between 0 and 1, eliminating the influence of different orders of magnitude. For example, the maximum value in the above set is 7650 and the minimum value is 48. After calculation, (2200-48) / (7650-48)≈0.28, (7650-48) / (7650-48)=1, (48-48) / (7650-48)=0, resulting in the normalized result [0.28,1,0]. Based on this, a 3×3 low-dimensional dense correlation matrix is ​​constructed, with the diagonal elements being the normalized result and the off-diagonal elements being 0. That is, the low-dimensional dense correlation matrix is ​​[[0.28,0,0];[0,1,0];[0,0,0]], which represents the correlation weights of long-time and short-time state components.

[0100] Finally, the low-dimensional dense compressed state vector of the first long-term vehicle is interacted with using a low-dimensional dense correlation matrix. The interaction process is achieved through matrix-vector multiplication, i.e., [40×0.28, 85×1, 6×0]=[11.2, 85, 0], and the result is the low-dimensional dense state vector of the first vehicle. This vector integrates the stable characteristics of the long-term state and the dynamic characteristics of the short-term state, preserving the overall trend of vehicle operation while incorporating key information about short-term fluctuations.

[0101] By performing inner product mapping on state components of the same type, normalizing them to construct an association matrix, and interacting with long-term vectors, a first low-dimensional dense state vector of the vehicle that integrates long-term and short-term features was obtained, providing an accurate and comprehensive state representation for subsequent matching with the vehicle anomaly anchor point state vector library.

[0102] Furthermore, step A500 in the method provided in this application embodiment includes: A510: Retrieves the set of alarm information within the preset monitoring window.

[0103] A520: Based on the alarm information set, perform a status monitoring requirement analysis on the status monitoring branch array to obtain a set of status monitoring branch requirement coefficients.

[0104] A530: Based on the set of condition monitoring branch demand coefficients, the condition monitoring branch array is reassigned to obtain an updated condition monitoring branch array, and the updated condition monitoring branch array is used to perform condition monitoring on the multiple vehicles again according to the preset monitoring cycle.

[0105] In one embodiment, firstly, a preset monitoring window time range is determined, and all generated alarm information is collected within this range to form an alarm information set. This alarm information includes the vehicle identifier that triggered the alarm, the anomaly type, the corresponding status monitoring branch array number, etc., and fully records the anomaly monitoring situation within this time period.

[0106] Next, a condition monitoring demand analysis was performed on the condition monitoring branch arrays based on the alarm information set. The alarm frequency generated by each branch array within the monitoring window was counted, and then the alarm frequency of each branch was divided by the total alarm frequency to obtain the condition monitoring branch demand coefficient for each branch, forming a demand coefficient set. Branches with higher demand coefficients indicate that the vehicle areas or types they are responsible for monitoring are more prone to anomalies, and their demand for monitoring resources is more urgent.

[0107] Next, resources are reallocated to the condition monitoring branch array based on the set of branch demand coefficients. More monitoring resources are allocated to branches with high demand coefficients, such as shortening their monitoring cycle and increasing the number of monitoring nodes; resource investment is appropriately reduced for branches with low demand coefficients, maintaining only basic monitoring capabilities. This results in an updated condition monitoring branch array, which is then used to monitor multiple vehicles again according to a preset monitoring cycle, achieving dynamic optimization of resource allocation.

[0108] By acquiring alarm information, analyzing demand coefficients, and reallocating resources, the resource investment of the condition monitoring branch array is matched with the actual monitoring needs, thereby improving the targeting and efficiency of multi-vehicle condition monitoring.

[0109] In summary, the multi-vehicle status intelligent monitoring and alarm method based on mobile communication provided in this application has the following technical effects: This application constructs a vehicle anomaly anchor point state vector library based on a historical vehicle status monitoring alarm log set. It utilizes a status monitoring branch array connected to a mobile communication center to monitor multiple vehicles at preset intervals, obtaining status monitoring data sequences. These sequences are then compressed using a low-dimensional, dense compression process to obtain low-dimensional, dense vehicle state vectors. These vectors are transmitted to the mobile communication center and matched with the anomaly anchor point state vector library to determine alarm information. Combined with alarm information within a preset monitoring window, resources of the status monitoring branch array are reallocated. This allows for accurate monitoring of multiple vehicle states and intelligent alarm generation, making the multi-vehicle status monitoring alarm results more precise and reliable. The application achieves the technical effect of enabling the alarm device to efficiently process complex multi-vehicle status data, meeting the technical requirements for intelligent multi-vehicle status monitoring and accurate alarm generation.

[0110] Example 2, as Figure 2 As shown, based on the same inventive concept as the aforementioned Embodiment 1, this application provides a multi-vehicle status intelligent monitoring and alarm system based on mobile communication, the system comprising: Vehicle Anomaly Anchor Point State Vector Library Construction Module 1: The vehicle anomaly anchor point state vector library construction module 1 constructs a vehicle anomaly anchor point state vector library based on a collection of historical vehicle status monitoring alarm logs.

[0111] The vehicle status monitoring data sequence acquisition module 2 is used to perform status monitoring on multiple vehicles according to a preset monitoring cycle through a status monitoring branch array that is connected to the mobile communication center, thereby obtaining multiple vehicle status monitoring data sequences.

[0112] The vehicle low-dimensional dense state vector acquisition module 3 is used to traverse the multiple vehicle state monitoring data sequences and perform long and short time low-dimensional dense compression to obtain multiple vehicle low-dimensional dense state vectors.

[0113] Alarm information acquisition module 4 is used to transmit the multiple low-dimensional dense state vectors of vehicles to the mobile communication center respectively, match them with the vehicle abnormal anchor point state vector library, and determine alarm information based on the matching result.

[0114] Furthermore, the vehicle anomaly anchor point state vector library construction module 1 is used to perform the following steps: Extract vehicle abnormal state vectors from the historical vehicle status monitoring alarm log set to obtain a historical vehicle abnormal state vector set; perform pairwise semantic correlation analysis on the historical vehicle abnormal state vector set to obtain a set of mean semantic correlation coefficients of historical vehicle abnormal state vectors; extract the mean semantic correlation coefficients of the top m historical vehicle abnormal state vectors in the set of mean semantic correlation coefficients of historical vehicle abnormal state vectors, and use their corresponding historical vehicle abnormal state vectors as a set of historical vehicle abnormal cluster state vectors; based on the set of historical vehicle abnormal cluster state vectors, perform anchor point screening on the set of historical vehicle abnormal state vectors to construct the vehicle abnormal anchor point state vector library.

[0115] Furthermore, the vehicle anomaly anchor point state vector library construction module 1 is used to perform the following steps: Extract the first historical vehicle abnormal state vector from the set of historical vehicle abnormal state vectors; perform pairwise semantic correlation analysis on the first historical vehicle abnormal state vector and the historical vehicle abnormal state vector using Euclidean distance to obtain a set of semantic correlation coefficients for the first historical vehicle abnormal state; perform mean processing on the set of semantic correlation coefficients for the first historical vehicle abnormal state to obtain the mean of the semantic correlation coefficients for the first historical vehicle abnormal state vector; traverse the set of historical vehicle abnormal state vectors and perform pairwise semantic correlation analysis to obtain a set of the mean of the semantic correlation coefficients for the historical vehicle abnormal state vectors.

[0116] Furthermore, the vehicle anomaly anchor point state vector library construction module 1 is used to perform the following steps: The set of historical vehicle abnormal state vectors is added to the neighborhood of the historical vehicle abnormal cluster state vector corresponding to the maximum semantic relevance, to obtain the neighborhood set of historical vehicle abnormal cluster state vectors; the mean shift algorithm is used to filter anchor points in the neighborhood set of historical vehicle abnormal cluster state vectors to construct the vehicle abnormal anchor point state vector library.

[0117] Furthermore, the vehicle low-dimensional dense state vector acquisition module 3 is used to perform the following steps: Extract a first vehicle condition monitoring data sequence from the plurality of vehicle condition monitoring data sequences; compress the first vehicle condition monitoring data sequence according to the long-term low-dimensional dense compression bandwidth to obtain a first long-term vehicle low-dimensional dense compressed condition monitoring data sequence; compress the first vehicle condition monitoring data sequence according to the short-term low-dimensional dense compression bandwidth to obtain a first short-term vehicle low-dimensional dense compressed condition monitoring data sequence; call the state analyzer to perform state identification on the first long-term vehicle low-dimensional dense compressed condition monitoring data sequence and the first short-term vehicle low-dimensional dense compressed condition monitoring data sequence respectively to obtain a first long-term vehicle low-dimensional dense compressed condition vector and a first short-term vehicle low-dimensional dense compressed condition vector; interact the first long-term vehicle low-dimensional dense compressed condition vector and the first short-term vehicle low-dimensional dense compressed condition vector to determine the first vehicle low-dimensional dense condition vector; and so on, perform long-term and short-term low-dimensional dense compression on the plurality of vehicle condition monitoring data sequences respectively to obtain the plurality of vehicle low-dimensional dense condition vectors.

[0118] Furthermore, the vehicle low-dimensional dense state vector acquisition module 3 is used to perform the following steps: Extract the first vehicle status monitoring data from the first vehicle status monitoring data sequence; based on the first vehicle status monitoring data and the long-term low-dimensional dense compression bandwidth, search the first vehicle status monitoring data sequence to obtain the first retrieved vehicle status monitoring data; calculate the cosine similarity between the first vehicle status monitoring data and the first retrieved vehicle status monitoring data; if the calculation result is greater than or equal to a preset similarity threshold, remove the first retrieved vehicle status monitoring data, and continue to compress the first vehicle status monitoring data sequence based on the first vehicle status monitoring data and the long-term low-dimensional dense compression bandwidth to obtain a first long-term vehicle low-dimensional dense compressed status monitoring data sequence.

[0119] Furthermore, the vehicle low-dimensional dense state vector acquisition module 3 is used to perform the following steps: If the calculation result is less than the preset similarity threshold, the first vehicle status monitoring data and the first retrieved vehicle status monitoring data are retained, and the first vehicle status monitoring data sequence is compressed based on the first retrieved vehicle status monitoring data and the long-term low-dimensional dense compression bandwidth to obtain the first long-term vehicle low-dimensional dense compressed status monitoring data sequence.

[0120] Furthermore, the vehicle low-dimensional dense state vector acquisition module 3 is used to perform the following steps: The inner product mapping is performed on the same type of state components in the first long-term vehicle low-dimensional dense compressed state vector and the first short-term vehicle low-dimensional dense compressed state vector to determine the long-term and short-term inner product mapping result set; the long-term and short-term inner product mapping result set is normalized to construct a low-dimensional dense correlation matrix; the low-dimensional dense correlation matrix is ​​used to interact with the first long-term vehicle low-dimensional dense compressed state vector to obtain the first vehicle low-dimensional dense state vector.

[0121] Furthermore, the alarm information acquisition module 4 is used to perform the following steps: Obtain a set of alarm information within a preset monitoring window; perform a status monitoring requirement analysis on the status monitoring branch array based on the alarm information set to obtain a set of status monitoring branch requirement coefficients; reallocate resources of the status monitoring branch array based on the set of status monitoring branch requirement coefficients to obtain an updated status monitoring branch array after reallocation, and use the updated status monitoring branch array to perform status monitoring on the multiple vehicles again according to a preset monitoring cycle.

[0122] The multi-vehicle status intelligent monitoring and alarm system based on mobile communication provided in the embodiments of the present invention can execute the multi-vehicle status intelligent monitoring and alarm method based on mobile communication provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method.

[0123] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0124] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A multi-vehicle status intelligent monitoring and alarm method based on mobile communication, characterized in that, The method includes: Based on the historical vehicle status monitoring alarm log set, construct a vehicle anomaly anchor point status vector library; The status monitoring branch array, which is connected to the mobile communication center, performs status monitoring on multiple vehicles according to a preset monitoring cycle, and obtains multiple vehicle status monitoring data sequences. The multiple vehicle state monitoring data sequences are traversed and subjected to long-short-time low-dimensional dense compression to obtain multiple low-dimensional dense state vectors of vehicles. The multiple low-dimensional dense state vectors of the vehicles are transmitted to the mobile communication center, matched with the vehicle abnormal anchor point state vector library, and alarm information is determined based on the matching results.

2. The intelligent monitoring and alarm method for multiple vehicle statuses based on mobile communication as described in claim 1, characterized in that, Based on historical vehicle status monitoring alarm logs, a vehicle anomaly anchor point status vector library is constructed, including: Extract the vehicle abnormal state vector from the historical vehicle status monitoring alarm log set to obtain the historical vehicle abnormal state vector set; Perform pairwise semantic correlation analysis on the set of historical vehicle abnormal state vectors to obtain the set of mean semantic correlation coefficients of historical vehicle abnormal state vectors. Extract the mean semantic correlation coefficient of the top m historical vehicle abnormal state vectors from the set of mean semantic correlation coefficients of the historical vehicle abnormal state vectors, and use the corresponding historical vehicle abnormal state vectors as the set of historical vehicle abnormal clustering state vectors. Based on the set of historical vehicle anomaly clustering state vectors, anchor points are filtered from the set of historical vehicle anomaly state vectors to construct the vehicle anomaly anchor point state vector library.

3. The intelligent monitoring and alarm method for multiple vehicle statuses based on mobile communication as described in claim 2, characterized in that, Perform pairwise semantic correlation analysis on the set of historical vehicle abnormal state vectors to obtain the set of mean semantic correlation coefficients of historical vehicle abnormal state vectors, including: Extract the first historical vehicle abnormal state vector from the set of historical vehicle abnormal state vectors; By using Euclidean distance, pairwise semantic correlation analysis is performed on the first historical vehicle abnormal state vector and the historical vehicle abnormal state vector to obtain the set of semantic correlation coefficients of the first historical vehicle abnormal state. The semantic correlation coefficient set of the first historical vehicle abnormal state is averaged to obtain the average semantic correlation coefficient of the first historical vehicle abnormal state vector. The set of historical vehicle abnormal state vectors is traversed and subjected to pairwise semantic correlation analysis to obtain the set of mean semantic correlation coefficients of the historical vehicle abnormal state vectors.

4. The intelligent monitoring and alarm method for multiple vehicle statuses based on mobile communication as described in claim 2, characterized in that, Based on the historical vehicle anomaly clustering state vector set, anchor points are filtered from the historical vehicle anomaly state vector set to construct the vehicle anomaly anchor point state vector library, including: The set of historical vehicle abnormal state vectors is added to the neighborhood of the historical vehicle abnormal cluster state vector corresponding to the maximum semantic relevance, to obtain the neighborhood set of historical vehicle abnormal cluster state vectors. The mean-shift algorithm is used to filter anchor points in the neighborhood set of the historical vehicle anomaly clustering state vectors, and the vehicle anomaly anchor point state vector library is constructed.

5. The intelligent monitoring and alarm method for multiple vehicle statuses based on mobile communication as described in claim 1, characterized in that, By traversing the multiple vehicle state monitoring data sequences and performing long-short-time low-dimensional dense compression, multiple low-dimensional dense state vectors of vehicles are obtained, including: Extract the first vehicle condition monitoring data sequence from the plurality of vehicle condition monitoring data sequences; The first vehicle status monitoring data sequence is compressed according to the long-term low-dimensional dense compression bandwidth to obtain the first long-term vehicle low-dimensional dense compressed status monitoring data sequence. The first vehicle status monitoring data sequence is compressed according to the short-time low-dimensional dense compression bandwidth to obtain the first short-time vehicle low-dimensional dense compressed status monitoring data sequence. The state analyzer is invoked to perform state identification on the first long-term vehicle low-dimensional dense compressed state monitoring data sequence and the first short-term vehicle low-dimensional dense compressed state monitoring data sequence respectively, so as to obtain the first long-term vehicle low-dimensional dense compressed state vector and the first short-term vehicle low-dimensional dense compressed state vector. Interact with the first long-term vehicle low-dimensional dense compressed state vector and the first short-term vehicle low-dimensional dense compressed state vector to determine the first vehicle low-dimensional dense state vector; Similarly, the multiple vehicle status monitoring data sequences are subjected to long-short-time low-dimensional dense compression to obtain the multiple vehicle low-dimensional dense state vectors.

6. The intelligent monitoring and alarm method for multiple vehicle statuses based on mobile communication as described in claim 5, characterized in that, The first vehicle condition monitoring data sequence is compressed according to the long-term low-dimensional dense compression bandwidth to obtain the first long-term vehicle low-dimensional dense compressed condition monitoring data sequence, including: Extract the first vehicle status monitoring data from the first vehicle status monitoring data sequence; Based on the first vehicle status monitoring data and the long-term low-dimensional dense compression bandwidth, the first vehicle status monitoring data sequence is retrieved to obtain the first retrieved vehicle status monitoring data. Calculate the cosine similarity between the first vehicle status monitoring data and the first retrieved vehicle status monitoring data. If the calculation result is greater than or equal to a preset similarity threshold, then remove the first retrieved vehicle status monitoring data and continue to compress the first vehicle status monitoring data sequence based on the first vehicle status monitoring data and the long-term low-dimensional dense compression bandwidth to obtain the first long-term vehicle low-dimensional dense compressed status monitoring data sequence.

7. The intelligent monitoring and alarm method for multiple vehicle statuses based on mobile communication as described in claim 6, characterized in that, If the calculation result is less than the preset similarity threshold, the first vehicle status monitoring data and the first retrieved vehicle status monitoring data are retained, and the first vehicle status monitoring data sequence is compressed based on the first retrieved vehicle status monitoring data and the long-term low-dimensional dense compression bandwidth to obtain the first long-term vehicle low-dimensional dense compressed status monitoring data sequence.

8. The intelligent monitoring and alarm method for multiple vehicle statuses based on mobile communication as described in claim 5, characterized in that, Interacting with the first long-term vehicle low-dimensional dense compressed state vector and the first short-term vehicle low-dimensional dense compressed state vector, the first vehicle low-dimensional dense state vector is determined, including: Perform inner product mapping on the same type of state components in the first long-term vehicle low-dimensional dense compressed state vector and the first short-term vehicle low-dimensional dense compressed state vector to determine the set of long-term and short-term inner product mapping results. The set of long and short time inner product mapping results is normalized to construct a low-dimensional dense correlation matrix; The first long-term vehicle low-dimensional dense compressed state vector is obtained by interacting with the low-dimensional dense correlation matrix.

9. The intelligent monitoring and alarm method for multiple vehicle statuses based on mobile communication as described in claim 8, characterized in that, include: Obtain the set of alarm information within the preset monitoring window; Based on the alarm information set, a status monitoring requirement analysis is performed on the status monitoring branch array to obtain a set of status monitoring branch requirement coefficients. Based on the set of condition monitoring branch demand coefficients, the condition monitoring branch array is reassigned to obtain an updated condition monitoring branch array. The updated condition monitoring branch array is then used to perform condition monitoring on the multiple vehicles again according to a preset monitoring cycle.

10. A multi-vehicle status intelligent monitoring and alarm system based on mobile communication, characterized in that, The system is used to implement the intelligent multi-vehicle status monitoring and alarm method based on mobile communication as described in any one of claims 1-9, the system comprising: The vehicle anomaly anchor point state vector library construction module constructs a vehicle anomaly anchor point state vector library based on the historical vehicle status monitoring alarm log collection. The vehicle condition monitoring data sequence acquisition module is used to perform condition monitoring on multiple vehicles according to a preset monitoring cycle through a condition monitoring branch array that is connected to the mobile communication center, and obtain multiple vehicle condition monitoring data sequences. The vehicle low-dimensional dense state vector acquisition module is used to traverse the multiple vehicle state monitoring data sequences and perform long and short time low-dimensional dense compression to obtain multiple vehicle low-dimensional dense state vectors. The alarm information acquisition module is used to transmit the multiple low-dimensional dense state vectors of the vehicles to the mobile communication center respectively, match them with the vehicle abnormal anchor point state vector library, and determine the alarm information based on the matching result.

Images