Rescue system of electric vehicle
By combining machine learning models with data on the electric vehicle's three-electric system and external environment, the problem of inaccurate human judgment in electric vehicle rescue has been solved, enabling more accurate rescue plan generation and rapid response.
Patent Information
- Application Number
- CN202411091984.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-09
- Publication Date
- 2024-11-08
AI Technical Summary
Current technologies for electric vehicle rescue rely on manual judgment, which cannot fully consider the unique problems of the electric vehicle's three-electric system and the external environment, resulting in inaccurate rescue positioning and the inability to provide the most suitable rescue plan.
By employing machine learning models combined with basic vehicle data and supplementary information, and through onboard assistance modules, central response modules, rescue demand modules, and rescue arrangement modules, the system comprehensively analyzes data from the three-electric system (battery, motor, and electronic control system), external environmental data, and descriptions from those seeking help, and generates an accurate sequence of rescue demands.
It improves the accuracy and response speed of rescue plans, reduces the bias and error of human judgment, and can more accurately reflect the actual situation of rescue needs, reducing false alarms and omissions.
Smart Images

Figure CN118917618A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of road rescue, and more particularly to an electric vehicle rescue system. Background Art
[0002] In today's fast-developing society, the need for vehicle rescue is becoming an issue that cannot be ignored. Due to their unique power structure and operating mechanism, electric vehicles encounter different problems from traditional fuel vehicles, which poses new challenges to the rescue system. Especially when electric vehicles encounter emergency situations, such as vehicle failures and accidents, how to rescue quickly and accurately has become the focus of social attention. Effective and timely rescue services can not only ensure the safety of drivers and passengers, but also reduce traffic congestion and improve road use efficiency.
[0003] In the prior art, vehicle rescue usually relies on manual judgment. Even in some rescue systems with a high degree of automation, the identification and classification of rescue needs still rely on the received help information and the experience and judgment of the rescue personnel. There are several significant problems with this approach that relies on manual experience: First, limited by manual judgment, the rescue system may not be able to fully consider all relevant factors, especially for the three-electric system problems unique to electric vehicles, and the comprehensive analysis of the external environment in which the vehicle is located, which may lead to inaccurate rescue positioning and inability to provide the most appropriate rescue plan. For example, electric vehicles may face the risk of battery thermal runaway, which requires accurate judgment based on the vehicle's real-time "three-electric" data; or, if a vehicle breaks down on a highway at night, the risks of the surrounding environment may not be fully perceived and communicated by the person seeking help. Relying on the traditional rescue model, these potential but safety-threatening factors may be ignored, resulting in insufficient rescue measures or even rescue failures. Summary of the invention
[0004] In order to overcome the problem of inaccurate rescue demand judgment in the prior art, the present invention proposes an electric vehicle rescue system to solve the above problem.
[0005] The present invention provides the following technical solutions:
[0006] An electric vehicle rescue system comprises: a vehicle-mounted rescue module, which is used to send a rescue signal, wherein the rescue signal is sent manually or automatically;
[0007] The central response module is used to receive the rescue signal and establish a communication connection with the rescue vehicle, obtain the basic data of the rescue vehicle through the communication connection, and obtain supplementary information through the communication connection;
[0008] A first rescue demand module, used to set a demand value for each of a plurality of preset rescue demands to form a first rescue demand sequence;
[0009] A second rescue demand module is used to extract a first feature vector according to basic data, extract a second feature vector according to supplementary information, combine the first feature vector and the second feature vector to obtain a rescue feature, and obtain a second rescue demand sequence according to the rescue feature using a pre-trained rescue sequence generation model;
[0010] A rescue requirement determination module, used to obtain a final rescue requirement sequence according to the first rescue requirement sequence and the second rescue requirement sequence;
[0011] The rescue arrangement module is used to set the rescue mission according to the final rescue requirement sequence.
[0012] Preferably, the basic data includes three-electric data and external data, the three-electric data includes motor data, battery data and electronic control data; the motor data includes motor speed, motor temperature and motor current; the battery data includes battery voltage, battery current and battery temperature; the electronic control data includes fault codes; the external data includes road speed limit value, visibility and traffic flow data;
[0013] The preset rescue needs are: vehicle maintenance needs, road safety needs, personnel rescue needs and fire safety needs.
[0014] Preferably, extracting the first feature vector according to the basic data includes:
[0015] Extracting external feature vectors based on external data in basic data;
[0016] Extracting three-electric characteristic vectors according to the three-electric data in the basic data;
[0017] Combining the three electrical eigenvectors and the external eigenvector to obtain a first eigenvector;
[0018] The step of obtaining the external feature vector comprises:
[0019] Obtain and normalize road speed limit, visibility and traffic flow data; and use the normalized road speed limit, visibility and traffic flow data to form an external feature vector.
[0020] Preferably, the step of acquiring the three electrical characteristic vectors includes:
[0021] A motor vector, a battery vector and a fault vector are set and initialized to be empty, wherein the motor vector includes three dimensions corresponding to the motor speed, the motor temperature and the motor current, the battery vector includes three dimensions corresponding to the battery voltage, the battery current and the battery temperature, the dimension of the fault vector corresponds to the total number of fault codes, and each dimension corresponds to one fault code;
[0022] Obtain motor data, battery data and electronic control data among the three electric data;
[0023] Use each item of motor data to compare with the preset normal range. If the data is within the normal range, the value of the corresponding dimension in the motor vector is set to 1, otherwise the value of the corresponding dimension is set to 0;
[0024] Use each item of battery data to compare with the preset normal range. If the data is within the normal range, the value of the corresponding dimension in the battery vector is set to 1, otherwise the value of the corresponding dimension is set to 0;
[0025] Obtain all fault codes in the electronic control data. In the fault vector, set the dimension value corresponding to each acquired fault code to 1, and set the values of the remaining dimensions to 0;
[0026] The motor vector, battery vector and fault vector are combined to obtain the three-electric characteristic vectors.
[0027] Preferably, the step of extracting the second feature vector according to the supplementary information comprises:
[0028] Obtain supplementary information and determine whether the supplementary information is empty; if it is empty, use a preset vector as the second eigenvector;
[0029] If it is not empty, the second feature vector is obtained using text processing technology.
[0030] Preferably, the step of obtaining the second feature vector using text processing technology includes:
[0031] Preset an event word list, wherein the event word list contains a plurality of preset event words;
[0032] Construct an event vector, wherein the dimension of the event vector is the same as the total number of event words, and each dimension corresponds to an event word;
[0033] Traverse each event word in the event word list, and for each event word, search in the supplementary information to determine whether the event word appears in the supplementary information; if not, set the dimension value corresponding to the event word in the event vector to 0; if it appears, extract the judgment word set corresponding to the event word from the supplementary information, and set the dimension value corresponding to the event word according to the judgment word set;
[0034] A character word set is preset, the character word set includes a first word group and a second word group, and the first word group and the second word group respectively contain a plurality of keywords;
[0035] Constructing a character vector, wherein the character vector includes two dimensions, corresponding to the first phrase and the second phrase respectively;
[0036] Traverse each keyword in the first phrase and the second phrase, and for each keyword, search in the supplementary information to determine whether the keyword appears in the supplementary information; if not, set the measurement value of the keyword to 1; if it appears, extract the measurement word corresponding to the keyword from the supplementary information, and obtain the measurement value corresponding to the keyword according to the measurement word;
[0037] The dimension value corresponding to the first phrase in the character vector is set to the sum of the measurement values corresponding to all the keywords in the first phrase; the dimension value corresponding to the second phrase in the character vector is set to the sum of the measurement values corresponding to all the keywords in the second phrase;
[0038] The event vector and the character vector are combined to obtain the second feature vector.
[0039] Preferably, the step of setting the dimension value corresponding to the event word according to the judgment word set includes:
[0040] Get n characters before and after the event word, respectively, as a judgment word, and all judgment words constitute the judgment word set corresponding to the event word;
[0041] Use the judgment words to match the preset negative word library. If the proportion of judgment words in the judgment word set that successfully match the negative word library exceeds the preset proportion value, the dimension value corresponding to the event word in the event vector is set to 0, otherwise the dimension value corresponding to the event word is set to 1;
[0042] The step of obtaining the measurement value corresponding to the keyword according to the measurement word includes:
[0043] Get the first n characters of the keyword as a quantifier, and determine whether there is a numeral in the quantifier; if not, set the quantifier value corresponding to the keyword to 1; if so, set the quantifier value corresponding to the keyword to the value corresponding to the numeral, where n is a preset value.
[0044] Preferably, the training process of the rescue sequence generation model includes:
[0045] Obtain historical rescue data, obtain a learning data set based on the historical rescue data, and divide the learning data set into a training set and a validation set according to a predetermined ratio;
[0046] Using a fully connected neural network as the basic framework, a machine learning model with an output layer of 4 output neurons was constructed as a rescue sequence generation model;
[0047] Obtain some learning data from the training set and input them into the rescue sequence generation model, obtain the output of the model, and calculate the loss of the model according to the loss function; calculate the gradient of the model through the back propagation algorithm according to the loss, and update the weight and bias of the model according to the gradient of the model;
[0048] The training of the rescue sequence generation model is completed until the performance of the rescue sequence generation model on the validation set reaches the preset standard.
[0049] Preferably, the step of obtaining historical rescue data and obtaining a learning data set based on the historical rescue data includes:
[0050] Obtain any rescue event in history, obtain basic data and supplementary information corresponding to the rescue event, extract a corresponding first feature vector according to the basic data, and extract a corresponding second feature vector according to the supplementary information;
[0051] Obtain the rescue force mobilized for the rescue incident, the rescue force includes maintenance manpower corresponding to vehicle maintenance needs, traffic police manpower corresponding to road safety needs, medical manpower corresponding to personnel rescue needs, and firefighting manpower corresponding to fire safety needs;
[0052] According to the rescue force, the demand values corresponding to the four rescue demands are marked to obtain the historical rescue demand sequence;
[0053] The first eigenvector, the second eigenvector and the historical rescue demand sequence are used as one piece of learning data, and multiple pieces of learning data are obtained to form a learning data set.
[0054] Preferably, the step of obtaining the final rescue requirement sequence according to the first rescue requirement sequence and the second rescue requirement sequence comprises:
[0055] The first rescue sequence and the second rescue sequence are obtained, and the phase difference of each corresponding element in the first rescue sequence and the second rescue sequence is compared. The calculation formula of the phase difference is as follows:
[0056]
[0057] Where Df i represents the phase difference between the i-th corresponding element in the first rescue sequence and the second rescue sequence, st i represents the i-th element in the first rescue sequence, rd i represents the i-th element in the second rescue sequence;
[0058] The element value in the final rescue sequence is obtained according to the phase difference, and the formula for obtaining the element value of the final rescue sequence is as follows:
[0059]
[0060] In the formula, Fn i represents the i-th element in the final rescue sequence, and Dfy represents the preset phase difference threshold.
[0061] The present invention provides an electric vehicle rescue system, which has the following beneficial effects:
[0062] By using machine learning models and comprehensively analyzing the basic data and supplementary information of the vehicle, the rescue needs are identified and combined with the results of manual judgment, the deviation and error of manual judgment are reduced; by integrating the three-electric system data of the vehicle, the external environment data and the description of the rescue personnel, a more suitable rescue plan can be provided; by constructing an event word list and a judgment word set, the specific events described in the text can be identified, and the authenticity and urgency of the events can be accurately judged according to the judgment words in the context. At the same time, for key information such as people and vehicles, the present invention not only pays attention to whether they appear, but also quantifies their quantity by extracting measurement words, thereby providing a more detailed and specific basis for rescue decisions. This text processing technology significantly improves the accuracy and response speed of the prediction. It can more accurately reflect the actual situation of the rescue needs and reduce false alarms and omissions. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] Figure 1 A schematic diagram of a module of an electric vehicle rescue system of the present invention;
[0064] Figure 2 The present invention is a schematic diagram of a model training process of an electric vehicle rescue system. DETAILED DESCRIPTION
[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0066] Example 1
[0067] See also Figure 1 In this embodiment, an electric vehicle rescue system includes:
[0068] The vehicle-mounted assistance module is used to send an assistance signal, and the assistance signal can be sent manually or automatically;
[0069] In this embodiment, the active sending method of the distress signal can be: directly triggered by the driver or passenger through button operation, voice control or APP operation, etc.; and the automatic sending method can be: through the vehicle's built-in sensors and intelligent systems to monitor emergency situations such as collisions, airbag deployment, abnormal conditions, etc.
[0070] The central response module is used to receive the rescue signal and establish a communication connection with the rescue vehicle, obtain the basic data of the rescue vehicle through the communication connection, and obtain supplementary information through the communication connection;
[0071] A first rescue demand module, used to set a demand value for each of a plurality of preset rescue demands to form a first rescue demand sequence;
[0072] The basic data includes three-electric data and external data, the three-electric data includes motor data, battery data and electronic control data; the motor data includes motor speed, motor temperature and motor current; the battery data includes battery voltage, battery current and battery temperature; the electronic control data includes fault codes; the external data includes road speed limit, visibility and traffic flow data;
[0073] The preset rescue needs are: vehicle maintenance needs, road safety needs, personnel rescue needs and fire safety needs.
[0074] In this embodiment, the three electric data in the basic data of the vehicle can be collected in real time through the on-board sensors. For example, the motor speed, motor temperature and motor current are obtained through the sensors on the motor controller; the battery voltage, battery current and battery temperature are obtained through the battery management system (BMS); the electronic control fault code is provided by the vehicle electronic control unit (ECU). The road speed limit value can be obtained through the vehicle navigation system or in combination with GPS positioning and electronic map data. Visibility can be analyzed by the on-board camera combined with image processing technology, or by calling real-time meteorological data. Traffic flow data can be obtained through the on-board camera combined with visual processing technology, or by accessing the real-time road condition data platform of the traffic management department. Supplementary information can be in the form of recording or text. For example, the rescuer can directly describe the situation on the scene through voice recording, or the rescue center staff can enter the key information into the system in text form based on the communication content with the rescuer.
[0075] The first rescue demand sequence is usually set by the rescue center staff based on the communication with the rescue personnel and the basic data obtained. Each rescue demand corresponds to a demand value, and the higher the value, the higher the demand. For example, if the staff learned through communication with the rescue personnel that only one car had a flat tire and no one was injured, and the car was in a safe area, the first rescue demand sequence can be set to (1, 0, 0, 0), indicating that only the vehicle needs to be repaired.
[0076] A second rescue demand module is used to extract a first feature vector according to basic data, extract a second feature vector according to supplementary information, combine the first feature vector and the second feature vector to obtain a rescue feature, and obtain a second rescue demand sequence according to the rescue feature using a pre-trained rescue sequence generation model;
[0077] The extracting the first feature vector according to the basic data comprises:
[0078] Extracting external feature vectors based on external data in basic data;
[0079] Extracting three-electric characteristic vectors according to the three-electric data in the basic data;
[0080] Combining the three electrical eigenvectors and the external eigenvector to obtain a first eigenvector;
[0081] The step of obtaining the external feature vector comprises:
[0082] Obtain and normalize road speed limit, visibility and traffic flow data; and use the normalized road speed limit, visibility and traffic flow data to form an external feature vector.
[0083] The steps of obtaining the three electrical characteristic vectors include:
[0084] A motor vector, a battery vector and a fault vector are set and initialized to be empty, wherein the motor vector includes three dimensions corresponding to the motor speed, the motor temperature and the motor current, the battery vector includes three dimensions corresponding to the battery voltage, the battery current and the battery temperature, the dimension of the fault vector corresponds to the total number of fault codes, and each dimension corresponds to one fault code;
[0085] Obtain motor data, battery data and electronic control data among the three electric data;
[0086] Use each item of motor data to compare with the preset normal range. If the data is within the normal range, the value of the corresponding dimension in the motor vector is set to 1, otherwise the value of the corresponding dimension is set to 0;
[0087] Use each item of battery data to compare with the preset normal range. If the data is within the normal range, the value of the corresponding dimension in the battery vector is set to 1, otherwise the value of the corresponding dimension is set to 0;
[0088] Obtain all fault codes in the electronic control data. In the fault vector, set the dimension value corresponding to each acquired fault code to 1, and set the values of the remaining dimensions to 0;
[0089] The motor vector, battery vector and fault vector are combined to obtain the three-electric characteristic vectors.
[0090] The step of extracting the second feature vector according to the supplementary information comprises:
[0091] Obtain supplementary information and determine whether the supplementary information is empty; if it is empty, use a preset vector as the second eigenvector;
[0092] If it is not empty, the second feature vector is obtained using text processing technology.
[0093] The step of using text processing technology to obtain the second feature vector includes:
[0094] Preset an event word list, wherein the event word list contains a plurality of preset event words;
[0095] Construct an event vector, wherein the dimension of the event vector is the same as the total number of event words, and each dimension corresponds to an event word;
[0096] Traverse each event word in the event word list, and for each event word, search in the supplementary information to determine whether the event word appears in the supplementary information; if not, set the dimension value corresponding to the event word in the event vector to 0; if it appears, extract the judgment word set corresponding to the event word from the supplementary information, and set the dimension value corresponding to the event word according to the judgment word set;
[0097] A character word set is preset, the character word set includes a first word group and a second word group, and the first word group and the second word group respectively contain a plurality of keywords;
[0098] Constructing a character vector, wherein the character vector includes two dimensions, corresponding to the first phrase and the second phrase respectively;
[0099] Traverse each keyword in the first phrase and the second phrase, and for each keyword, search in the supplementary information to determine whether the keyword appears in the supplementary information; if not, set the measurement value of the keyword to 1; if it appears, extract the measurement word corresponding to the keyword from the supplementary information, and obtain the measurement value corresponding to the keyword according to the measurement word;
[0100] The dimension value corresponding to the first phrase in the character vector is set to the sum of the measurement values corresponding to all the keywords in the first phrase; the dimension value corresponding to the second phrase in the character vector is set to the sum of the measurement values corresponding to all the keywords in the second phrase;
[0101] The event vector and the character vector are combined to obtain the second feature vector.
[0102] The step of setting the dimension value corresponding to the event word according to the judgment word set includes:
[0103] Get n characters before and after the event word, respectively, as a judgment word, and all judgment words constitute the judgment word set corresponding to the event word;
[0104] Use the judgment words to match the preset negative word library. If the proportion of judgment words in the judgment word set that successfully match the negative word library exceeds the preset proportion value, the dimension value corresponding to the event word in the event vector is set to 0, otherwise the dimension value corresponding to the event word is set to 1;
[0105] The step of obtaining the measurement value corresponding to the keyword according to the measurement word includes:
[0106] Get the first n characters of the keyword as a quantifier, and determine whether there is a numeral in the quantifier; if not, set the quantifier value corresponding to the keyword to 1; if so, set the quantifier value corresponding to the keyword to the value corresponding to the numeral, where n is a preset value.
[0107] In this embodiment, first, the first eigenvector is extracted based on the basic data. The basic data includes external data such as road speed limit, visibility and traffic flow and three-electric data involving motors, batteries and electronic control systems. The external data is normalized to obtain an external eigenvector, for example, the road speed limit of 60km / h, visibility of 100m and traffic flow of 1000 vehicles / h are normalized to (0.6, 0.8, 0.5). Next, the motor vector, battery vector and fault vector are initialized, and these vectors are filled according to the actual three-electric data (such as motor speed, temperature, current, battery voltage, current, temperature, and fault code of the electronic control system). For example, if the motor data are all within the normal range, the motor vector is (1, 1, 1); similarly, if the battery data are all normal, the battery vector is (1, 1, 1). The fault vector is marked according to the fault code in the electronic control data, for example, the corresponding dimension of "motor failure" is set to 1, and the rest are 0. Finally, the external eigenvector and the three-electric eigenvector are combined to form the first eigenvector.
[0108] Secondly, extract the second feature vector according to the supplementary information. If the supplementary information is not empty, the text processing technology is used for feature extraction. The preset event word list ("rollover", "smoke", "fire", etc.) and character word set (including the first phrase such as "adult", "woman", "child", etc., and the second phrase such as "car", "truck", "motorcycle", etc.) are used to construct event vectors and character vectors. By traversing the event word list and the character word set, the corresponding words are searched in the supplementary information, and the values of the corresponding dimensions in the vector are set according to the appearance or absence of the words and their context, that is, the judgment word set and the measurement word. For example: part of the rescue information is "On the XX highway, a car suddenly smoked, and adults and children were trapped in the car. The situation is urgent. Request immediate rescue." The event word "rollover" did not appear, and the corresponding dimension value was set to 0. The event word "smoke" appeared, and no obvious negative words (such as "not" and "not") were found in its context, so it is considered that the event did occur, and the corresponding dimension value is set to 1. The event word "fire" did not appear, and the corresponding dimension value was set to 0. However, for the sake of clarity, it is set to 0 here, and is set only based on the direct mention of "smoke". Therefore, the event vector is (..., 0, 1, 0, ...) where ... represents the vector dimension corresponding to the unlisted event words. The first phrase (character type): "Adult" appears and is not accompanied by a specific quantifier, and the default quantifier is 1. "Woman" is not explicitly mentioned, and the quantifier is set to 0. "Child" appears, also without an accompanying quantifier, and the quantifier is 1. The dimension value corresponding to the first phrase is the sum of the two, that is, (1+0+1)=2. The second phrase (vehicle type): "Sedan" is explicitly mentioned and is a single vehicle, and the quantifier is set to 1. "Truck" and "motorcycle" are not mentioned, and the quantifiers are both set to 0. The dimension value corresponding to the second phrase is 1. Therefore, the character vector is (2, 1), where the first dimension corresponds to the first phrase and the second dimension corresponds to the second phrase. Subsequently, the event vector and the character vector are combined to obtain the second feature vector.
[0109] Finally, the first feature vector and the second feature vector are combined to form a complete rescue feature vector. This vector integrates information from multiple aspects such as the external environment, vehicle status, and text description, and can more comprehensively reflect the actual needs of road rescue.
[0110] By constructing an event word list and a set of judgment words, it is possible to identify specific events described in the text, and accurately judge the authenticity and urgency of the event based on the judgment words in the context. At the same time, for key information such as people and vehicles, the present invention not only focuses on whether they appear, but also quantifies their quantity by extracting measurement words, thereby providing a more detailed and specific basis for rescue decisions. This text processing technology significantly improves the accuracy and response speed of predictions. It can more accurately reflect the actual situation of rescue needs and reduce false positives and false negatives.
[0111] A rescue requirement determination module, used to obtain a final rescue requirement sequence according to the first rescue requirement sequence and the second rescue requirement sequence;
[0112] The step of obtaining the final rescue requirement sequence according to the first rescue requirement sequence and the second rescue requirement sequence comprises:
[0113] The first rescue sequence and the second rescue sequence are obtained, and the phase difference of each corresponding element in the first rescue sequence and the second rescue sequence is compared. The calculation formula of the phase difference is as follows:
[0114]
[0115] Where Df i represents the phase difference between the i-th corresponding element in the first rescue sequence and the second rescue sequence, st i represents the i-th element in the first rescue sequence, rd i represents the i-th element in the second rescue sequence;
[0116] The element value in the final rescue sequence is obtained according to the phase difference, and the formula for obtaining the element value of the final rescue sequence is as follows:
[0117]
[0118] In the formula, Fn i represents the i-th element in the final rescue sequence, and Dfy represents the preset phase difference threshold.
[0119] The rescue arrangement module is used to set the rescue mission according to the final rescue requirement sequence.
[0120] In this embodiment, the first rescue demand sequence and the second rescue demand sequence are obtained. The first rescue demand sequence represents the rescue demand based on the staff's experience, and the second rescue demand sequence represents the rescue demand obtained by the machine learning model based on data analysis. The difference between the two is calculated. If it does not exceed a threshold value such as 0.1, it is considered that the two judgments are the same. The judgment result of the machine learning model can be taken as the final value. If it exceeds the threshold, the average value between the two can be taken as the final value. Then the final rescue sequence is obtained, and the various rescue forces required can be determined according to the value of the final sequence, and the corresponding human resources can be dispatched or coordinated according to the required rescue forces.
[0121] Example 2
[0122] See also Figure 1 and Figure 2 Based on Example 1, the training process of the rescue sequence generation model includes:
[0123] S1. Obtain historical rescue data, obtain a learning data set based on the historical rescue data, and divide the learning data set into a training set and a verification set according to a predetermined ratio;
[0124] S2. Using a fully connected neural network as the basic framework, a machine learning model with an output layer of 4 output neurons was constructed as a rescue sequence generation model;
[0125] S3. Obtain some learning data from the training set and input them into the rescue sequence generation model, obtain the output result of the model, and calculate the loss of the model according to the loss function; calculate the gradient of the model through the back propagation algorithm according to the loss, and update the weight and bias of the model according to the gradient of the model;
[0126] S4. Until the performance of the rescue sequence generation model on the validation set reaches a preset standard, the training of the rescue sequence generation model is completed.
[0127] The steps of obtaining historical rescue data and obtaining a learning data set based on the historical rescue data include:
[0128] Obtain any rescue event in history, obtain basic data and supplementary information corresponding to the rescue event, extract a corresponding first feature vector according to the basic data, and extract a corresponding second feature vector according to the supplementary information;
[0129] Obtain the rescue force mobilized for the rescue incident, the rescue force includes maintenance manpower corresponding to vehicle maintenance needs, traffic police manpower corresponding to road safety needs, medical manpower corresponding to personnel rescue needs, and firefighting manpower corresponding to fire safety needs;
[0130] According to the rescue force, the demand values corresponding to the four rescue demands are marked to obtain the historical rescue demand sequence;
[0131] The first eigenvector, the second eigenvector and the historical rescue demand sequence are used as one piece of learning data, and multiple pieces of learning data are obtained to form a learning data set.
[0132] In this embodiment, the training process of the model can be as follows: First, 1,000 pieces of data on historical rescue events are obtained, and these data come from various sources, such as traffic accident reports, emergency rescue dispatch systems, etc. Each piece of data includes basic data and supplementary information of the rescue event. Two eigenvectors, the first eigenvector and the second eigenvector, are extracted based on the basic data and supplementary information. For each piece of data, the rescue forces mobilized for the rescue event are also obtained, including maintenance manpower corresponding to vehicle maintenance needs, traffic police manpower corresponding to road safety needs, medical manpower corresponding to personnel rescue needs, and firefighting manpower corresponding to fire safety needs; according to the actual number of manpower mobilized, the demand values corresponding to the four rescue needs are marked to form a historical rescue demand sequence. Using the first eigenvector, the second eigenvector and the historical rescue demand sequence as a learning data, 1,000 pieces of learning data are obtained to form a learning data set.
[0133] Using a fully connected neural network as the basic framework, a machine learning model with an output layer of 4 output neurons is constructed as a rescue sequence generation model. The learning data set is divided into a training set and a validation set according to a predetermined ratio, such as 8:2.
[0134] Use the training set to train the rescue sequence generation model, use the mean square error loss function to calculate the model's loss, and use the back propagation algorithm to update the model's weights and biases. Repeat this process until the model's performance on the validation set reaches the preset standard (for example, the accuracy rate reaches 90%). After training, the rescue sequence generation model can generate the corresponding rescue demand sequence based on the input feature vector.
[0135] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0136] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
[0137] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. An electric vehicle rescue system, characterized in that: include: The vehicle-mounted assistance module is used to send an assistance signal, and the assistance signal can be sent manually or automatically; The central response module is used to receive the rescue signal and establish a communication connection with the rescue vehicle, obtain the basic data of the rescue vehicle through the communication connection, and obtain supplementary information through the communication connection; A first rescue demand module, used to set a demand value for each of a plurality of preset rescue demands to form a first rescue demand sequence; A second rescue demand module is used to extract a first feature vector according to basic data, extract a second feature vector according to supplementary information, combine the first feature vector and the second feature vector to obtain a rescue feature, and obtain a second rescue demand sequence according to the rescue feature using a pre-trained rescue sequence generation model; A rescue requirement determination module, used to obtain a final rescue requirement sequence according to the first rescue requirement sequence and the second rescue requirement sequence; The rescue arrangement module is used to set the rescue mission according to the final rescue requirement sequence.
2. An electric vehicle rescue system according to claim 1, characterized in that: The basic data includes three-electric data and external data, the three-electric data includes motor data, battery data and electronic control data; the motor data includes motor speed, motor temperature and motor current; the battery data includes battery voltage, battery current and battery temperature; the electronic control data includes fault codes; the external data includes road speed limit, visibility and traffic flow data; The preset rescue needs are: vehicle maintenance needs, road safety needs, personnel rescue needs and fire safety needs.
3. An electric vehicle rescue system according to claim 2, characterized in that: The extracting the first feature vector according to the basic data comprises: Extracting external feature vectors based on external data in basic data; Extracting three-electric characteristic vectors according to the three-electric data in the basic data; Combining the three electrical eigenvectors and the external eigenvector to obtain a first eigenvector; The step of obtaining the external feature vector comprises: Obtain and normalize road speed limit, visibility and traffic flow data; and use the normalized road speed limit, visibility and traffic flow data to form an external feature vector.
4. An electric vehicle rescue system according to claim 3, characterized in that: The steps of obtaining the three electrical characteristic vectors include: A motor vector, a battery vector and a fault vector are set and initialized to be empty, wherein the motor vector includes three dimensions corresponding to the motor speed, the motor temperature and the motor current, the battery vector includes three dimensions corresponding to the battery voltage, the battery current and the battery temperature, the dimension of the fault vector corresponds to the total number of fault codes, and each dimension corresponds to one fault code; Obtain motor data, battery data and electronic control data among the three electric data; Use each item of motor data to compare with the preset normal range. If the data is within the normal range, the value of the corresponding dimension in the motor vector is set to 1, otherwise the value of the corresponding dimension is set to 0; Use each item of battery data to compare with the preset normal range. If the data is within the normal range, the value of the corresponding dimension in the battery vector is set to 1, otherwise the value of the corresponding dimension is set to 0; Obtain all fault codes in the electronic control data. In the fault vector, set the dimension value corresponding to each acquired fault code to 1, and set the values of the remaining dimensions to 0; The motor vector, battery vector and fault vector are combined to obtain the three-electric characteristic vectors.
5. An electric vehicle rescue system according to claim 4, characterized in that: The step of extracting the second feature vector according to the supplementary information comprises: Obtain supplementary information and determine whether the supplementary information is empty; if it is empty, use a preset vector as the second eigenvector; If it is not empty, the second feature vector is obtained using text processing technology.
6. An electric vehicle rescue system according to claim 5, characterized in that: The step of using text processing technology to obtain the second feature vector includes: Preset an event word list, wherein the event word list contains a plurality of preset event words; Construct an event vector, wherein the dimension of the event vector is the same as the total number of event words, and each dimension corresponds to an event word; Traverse each event word in the event word list, and for each event word, search in the supplementary information to determine whether the event word appears in the supplementary information; if not, set the dimension value corresponding to the event word in the event vector to 0; if it appears, extract the judgment word set corresponding to the event word from the supplementary information, and set the dimension value corresponding to the event word according to the judgment word set; A character word set is preset, the character word set includes a first word group and a second word group, and the first word group and the second word group respectively contain a plurality of keywords; Constructing a character vector, wherein the character vector includes two dimensions, corresponding to the first phrase and the second phrase respectively; Traverse each keyword in the first phrase and the second phrase, and for each keyword, search in the supplementary information to determine whether the keyword appears in the supplementary information; if not, set the measurement value of the keyword to 1; if it appears, extract the measurement word corresponding to the keyword from the supplementary information, and obtain the measurement value corresponding to the keyword according to the measurement word; The dimension value corresponding to the first phrase in the character vector is set to the sum of the measurement values corresponding to all the keywords in the first phrase; the dimension value corresponding to the second phrase in the character vector is set to the sum of the measurement values corresponding to all the keywords in the second phrase; The event vector and the character vector are combined to obtain the second feature vector.
7. An electric vehicle rescue system according to claim 6, characterized in that: The step of setting the dimension value corresponding to the event word according to the judgment word set includes: Get n characters before and after the event word, respectively, as a judgment word, and all judgment words constitute the judgment word set corresponding to the event word; Use the judgment words to match the preset negative word library. If the proportion of judgment words in the judgment word set that successfully match the negative word library exceeds the preset proportion value, the dimension value corresponding to the event word in the event vector is set to 0, otherwise the dimension value corresponding to the event word is set to 1; The step of obtaining the measurement value corresponding to the keyword according to the measurement word includes: Get the first n characters of the keyword as a quantifier, and determine whether there is a numeral in the quantifier; if not, set the quantifier value corresponding to the keyword to 1; if so, set the quantifier value corresponding to the keyword to the value corresponding to the numeral, where n is a preset value.
8. An electric vehicle rescue system according to claim 7, characterized in that: The training process of the rescue sequence generation model includes: Obtain historical rescue data, obtain a learning data set based on the historical rescue data, and divide the learning data set into a training set and a validation set according to a predetermined ratio; Using a fully connected neural network as the basic framework, a machine learning model with an output layer of 4 output neurons was constructed as a rescue sequence generation model; Obtain some learning data from the training set and input them into the rescue sequence generation model, obtain the output of the model, and calculate the loss of the model according to the loss function; calculate the gradient of the model through the back propagation algorithm according to the loss, and update the weight and bias of the model according to the gradient of the model; The training of the rescue sequence generation model is completed until the performance of the rescue sequence generation model on the validation set reaches the preset standard.
9. An electric vehicle rescue system according to claim 8, characterized in that: The steps of obtaining historical rescue data and obtaining a learning data set based on the historical rescue data include: Obtain any rescue event in history, obtain basic data and supplementary information corresponding to the rescue event, extract a corresponding first feature vector according to the basic data, and extract a corresponding second feature vector according to the supplementary information; Obtain the rescue force mobilized for the rescue incident, the rescue force includes maintenance manpower corresponding to vehicle maintenance needs, traffic police manpower corresponding to road safety needs, medical manpower corresponding to personnel rescue needs, and firefighting manpower corresponding to fire safety needs; According to the rescue force, the demand values corresponding to the four rescue demands are marked to obtain the historical rescue demand sequence; The first eigenvector, the second eigenvector and the historical rescue demand sequence are used as one piece of learning data, and multiple pieces of learning data are obtained to form a learning data set.
10. The electric vehicle rescue system according to claim 7, characterized in that: The step of obtaining the final rescue requirement sequence according to the first rescue requirement sequence and the second rescue requirement sequence comprises: The first rescue sequence and the second rescue sequence are obtained, and the phase difference of each corresponding element in the first rescue sequence and the second rescue sequence is compared. The calculation formula of the phase difference is as follows: Where Df i represents the phase difference between the i-th corresponding element in the first rescue sequence and the second rescue sequence, st i represents the i-th element in the first rescue sequence, rd i represents the i-th element in the second rescue sequence; The element value in the final rescue sequence is obtained according to the phase difference, and the formula for obtaining the element value of the final rescue sequence is as follows: In the formula, Fn i represents the i-th element in the final rescue sequence, and Dfy represents the preset phase difference threshold.
