Taxi fault prediction model training method, vehicle fault prediction method and device

Abstract

The invention relates to a taxi fault prediction model training method and a vehicle fault prediction method and device, and the method comprises the steps: obtaining first data of a taxi to form learning sample data, and carrying out the normalization processing of the learning sample data, and obtaining a Bayesian logic regression model; acquiring second data of the taxi to form test sample data, inputting the test sample data into the model to obtain a fault of a prediction result, comparing the fault with an actual fault, and if a fault error is within a first set value, determining that the prediction result is accurate; and if the error is greater than the first set value, calculating the multiple of the actual error and the first set value, and adding the same number of corresponding test sample data into the learning sample data according to the multiple, so that the model is continuously trained and optimized. And judging whether the model is successfully trained or adjusting the interface until the training is successful according to the comparison of the error and the set value within the set time, or carrying out model learning again until the model is successful. Therefore, model training is more accurate, follow-up taxi fault prediction is facilitated, and prediction is more accurate.

Description

Technical Field

[0001] The present invention relates to the field of taxis, and in particular to a taxi fault prediction model training method, a taxi fault prediction method, a taxi fault prediction model training device, and a taxi fault prediction device. Background Art

[0002] With the convenience of life, taxis have become an indispensable means of transportation. Modern taxis usually include an on-board diagnostic system, which is a system for self-diagnosis and reporting problems on the vehicle. Reporting problems can follow standards such as On-Board Diagnostics II (OBD-II). The vehicle can output messages indicating problems through a physical port. The OBD-II standard specifies a 16-pin D-shaped female connector as a port for outputting messages. Through OBD, the vehicle's driving status data can be obtained in real time, as well as vehicle user data, environmental data, etc. saved in the database. The above data is closely related to the occurrence of vehicle failures. Therefore, a model is needed to analyze the above data and predict the occurrence of failures. The above data of existing taxis is often only used to record information and cannot be used to predict failures. It is also a waste of additional information provided by the data, but it can actually provide great help for taxi failure prediction. Summary of the invention

[0003] The technical problem to be solved by the present invention is that the above data of existing taxis is often only used to record information and cannot be used to predict faults, which is also a waste of the additional information provided by the data. However, it can actually provide great help for taxi fault prediction. In view of the above defects of the prior art, a taxi fault prediction model training method, a taxi fault prediction method and a device are provided.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0005] In a first aspect of an embodiment of the present invention, a taxi fault prediction model training method is constructed, and the training method comprises the following steps:

[0006] Acquire driving data of a plurality of taxis at the first time to form learning sample data, wherein the driving data includes vehicle driving state data, user state data, environmental information data and taxi fault data;

[0007] Use learning sample data to learn and predict the severity of faults and establish a regression model;

[0008] Acquire driving data of a plurality of taxis at a second time after the first time to form test sample data, wherein the driving data includes vehicle driving state data, user state data, environmental information data and taxi fault data;

[0009] The test sample data is used as the input of the regression model to verify the prediction results and cyclically optimize the regression model.

[0010] Preferably, in the process of obtaining the driving data of the plurality of taxis at the first time to form the learning sample data, the process further includes:

[0011] The driving data of multiple taxis at the first time are sorted out to form learning sample data, and the learning sample data are normalized;

[0012] The process of obtaining the driving data of the plurality of taxis at a second time after the first time to form the test sample data further includes:

[0013] The driving data of multiple taxis in the second room are sorted to form test sample data, and the test sample data are normalized.

[0014] Preferably, in the process of using the test sample data as the input of the regression model to verify the prediction result and cyclically optimize the regression model, it also includes:

[0015] The vehicle driving status data, user status data and environmental information data in the test sample data are input into the regression model to obtain the predicted failure result, and the predicted failure result is compared with the actual failure. If the error value is within the first set value range, the regression model is available. If the error value is greater than the first set value range, the regression model is adjusted.

[0016] Preferably, in the process of comparing the predicted failure result with the actual failure, if the error value is greater than the first value setting range, then adjusting the regression model further includes:

[0017] If the predicted result shows a fault and the actual fault is greater than the first set value, the actual error is compared with the first set value, and the multiple of the actual error value and the first set value is calculated. The corresponding prediction sample data weights are accumulated according to the calculated multiples, and the same number of prediction sample data as the multiples are added to the learning sample data to obtain new learning sample data, and the regression model is re-established.

[0018] Preferably, in the process of comparing the predicted failure result with the actual failure, if the error value is greater than the first set value range, then adjusting the regression model further includes:

[0019] If the error between the predicted failure and the actual failure within the set time is within the first set value range, the model training is successful;

[0020] If the error between the predicted fault and the actual fault occurs intermittently within the set time and exceeds the first set value range, but does not exceed the second set value range, the regression model interface data is adjusted to adjust the regression model and training is performed until the model training is successful;

[0021] If the error between the predicted failure and the actual failure exceeds the second set value range within the set time, the regression model is re-learned until the model training is successful.

[0022] In a second aspect of an embodiment of the present invention, a taxi fault prediction method is constructed, the method comprising the following steps:

[0023] Acquiring real-time driving data of multiple taxis during driving to form test data, wherein the driving data includes driving state data, user state data and environment data;

[0024] The test data is input into the regression model trained by the above-mentioned taxi fault prediction model training method to perform fault prediction.

[0025] In a third aspect of an embodiment of the present invention, a taxi fault prediction model training device is constructed, and the training device comprises:

[0026] A data acquisition module, used to obtain driving data of multiple taxis at a first time to form learning sample data, and to obtain driving data of multiple taxis at a second time after the first time to form test sample data;

[0027] A data sorting module sorts the acquired learning sample data and test sample data so that the driving data of each vehicle corresponds;

[0028] The learning prediction module uses the learning sample data to learn the fault severity and establish a regression model for prediction, and uses the test sample data to verify the regression model and cyclically optimize the regression model.

[0029] In a fourth aspect of the embodiment of the present invention, a taxi fault prediction device is constructed, the prediction device comprising:

[0030] A data acquisition module, used to obtain the driving data of multiple taxis at the first time to form learning sample data, obtain the driving data of multiple taxis at the second time after the first time to form test sample data, and obtain the real-time driving information of multiple taxis to form test data;

[0031] A data sorting module sorts the acquired learning sample data, test sample data and test data so that the driving data of each vehicle corresponds;

[0032] The learning prediction module uses the learning sample data to learn the fault severity and establish a regression model for prediction, and uses the test sample data to verify the regression model and cyclically optimize the regression model, and uses the test data to input into the optimized regression model for fault prediction.

[0033] In a fifth aspect of an embodiment of the present invention, an intelligent terminal is constructed, characterized in that the intelligent terminal includes at least one memory, at least one processor, and a taxi fault prediction model training program stored in the memory and executable on the processor, and the taxi fault prediction model training program implements the steps of the taxi fault prediction model training method as described above when executed by the processor.

[0034] In a sixth aspect of an embodiment of the present invention, a computer-readable storage medium is constructed, characterized in that a taxi fault prediction model training program is stored on the computer-readable storage medium, and when the taxi fault prediction model training program is executed by a processor, the steps of the taxi fault prediction model training method as described above are implemented.

[0035] The beneficial effect of the present invention is that: learning sample data is formed by acquiring existing driving data of taxis, and the Bayesian logistic regression model is obtained after normalizing the learning sample data. Then, taxi driving data is acquired to form test sample data, and the test sample data is input into the model to obtain the predicted result of failure, and compared with the actual failure. If the error between the predicted result of failure and the actual failure is within the first set value, the predicted result is considered accurate; if the error is greater than the first set value, the multiple of the actual error and the first set value is calculated, and the corresponding test sample data is added to the learning sample data in the same amount according to the multiple, so that the model is continuously trained and optimized. And according to the prediction error within the set time being the first set value, the model is successful. If it is intermittently greater than the first set value but less than the second set value, the model interface is adjusted to adjust the model and repeat the prediction until the model is successful. If it is greater than the second set value, the model is re-learned and the sample data is repeatedly learned until the model is successful. This method uses the existing fault data of the taxi to learn and establish a model, and trains the model based on the existing data to make the model more accurate, which is convenient for subsequent taxi fault prediction and more accurate prediction. Then, accurate fault prediction is achieved based on the daily real-time data interface. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. The drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work:

[0037] Figure 1 A flowchart of a taxi fault prediction model training method according to a preferred embodiment of the present invention;

[0038] Figure 2 A flowchart of a taxi fault prediction method according to a preferred embodiment of the present invention;

[0039] Figure 3 It is a structural schematic diagram of a taxi fault prediction model training device according to a preferred embodiment of the present invention;

[0040] Figure 4 A schematic diagram of the structure of a taxi fault prediction device according to a preferred embodiment of the present invention;

[0041] Figure 5 Schematic diagram of the structure of a smart terminal according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the following will be described clearly and completely in combination with the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are partial embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0043] A taxi fault prediction model training method according to a preferred embodiment of the present invention; Figure 1 As shown, it is a flowchart of a taxi fault prediction model training method provided by an embodiment of the present invention. The method can be executed by a device, and the device can be implemented by software and / or hardware.

[0044] Specifically, in this embodiment, the taxi fault prediction model training method includes:

[0045] S101: Acquire first information of a taxi to form learning sample data;

[0046] In a preferred embodiment of the present invention, the real-time status of the vehicle is first uploaded to MongoBD in real time through the TBOX of the taxi, and the data collection module reads all the vehicle status of the vehicle one day ago and the real-time status of all vehicles from MongoDB, where the real-time status of each vehicle can be obtained from multiple dimensions, such as obtaining the vehicle's form time, speed, acceleration, longitude and latitude, engine speed, driving direction, high beam, low beam and other driving data related to driving behavior during a complete journey of the vehicle, and learning is performed based on these driving data. It is also possible to collect the driving data of the vehicle within a specific time period, or the driving data of a whole day, and the collection can be set according to the needs.

[0047] Furthermore, the driving data is the driving status data of the vehicle, and other driving data of the vehicle can also be obtained from MySQL through the data acquisition module, such as user status data, environmental data and taxi fault data, etc. The user status data includes driver data, such as the driver's gender, age and city where the driver is located, etc. The environmental data includes the weather change conditions of the current city or place, the road conditions, the brightness outside the car, etc., and can also include the road conditions of the driving section, such as the data of traffic lights, the time and position of traffic lights, the position of speed bumps, speed limit sections and speed limit values, and lane distribution, etc., as well as the recorded taxi fault data.

[0048] Furthermore, the driving data can be matched through the vehicle data stored in MySQL. For example, the driving status data, user status data, environmental data and taxi fault data can be associated with the same frame number, license plate number, etc., so that the driving status data, user status data and environmental data of each vehicle are matched, and the fault data of the vehicle is matched, thereby forming the driving data into learning sample data.

[0049] S102: normalizing the learning sample data;

[0050] In the preferred embodiment of the present invention, since the driving data of each learning sample data is multi-dimensional, the units are not unified. For example, the driving speed of the vehicle in the driving status data is 60 yards and the mileage is 10 kilometers; the driver in the user status data is male and 45 years old; the weather in the environmental data is sunny or moderate rain, the number of traffic lights is 8, and the traffic light time is 150 seconds for red light and 30 seconds for green light; the taxi fault data has no fault or minor fault, serious fault light. This is not conducive to AI learning, so all or part of the driving data can be normalized, and the data of each dimension can be adjusted to between [0,1].

[0051] Furthermore, during the normalization process, each set of data is digitally converted according to its situation in the current data. For example, weather data can be digitally converted according to weather registration, such as 0 for sunny days, 10 for light rain, and 100 for heavy rain or snow. Fault data is digitally converted according to severity, such as 0 for no fault, and the larger the fault, the larger the value, up to 100. Other related data can also be normalized using similar methods to adjust the driving data to between [0,1].

[0052] S103: learning and predicting the fault severity of the normalized learning sample data to establish a Bayesian logistic regression model;

[0053] In a preferred embodiment of the present invention, the AI ​​learning module adopts a Bayesian logistic regression model (BLRM) to learn and predict the severity of the fault through the driving data normalized by the BLRM.

[0054] Furthermore, in the process of model building, the Bayesian logistic regression model is established using python's pymc3. First, the prior weight is set according to the normalized learning sample data. For example, the prior weight of the sample without fault is 0, and the weight of the sample with fault is between [0.5, 1]. The prior weight is set according to the severity of the fault. The more serious the fault, the greater the fault severity. Then the logistic regression coefficient is calculated, which is calculated using pymc3.Determin i st ic (interface f), where the parameter f is the interface, and different parameters can be designed. Then, the logistic function is predicted through pymc3.Determin i st ic (interface g), where the parameter g is the interface, and different parameters can be designed. The data process is observed through pymc3.Bernou lli to obtain the model unit. The MCMC method is resampled, the parameters are calculated to run the model unit, and finally the parameter unit is determined, such as taking the sample mean as the final parameter to obtain the Bayesian logistic regression model.

[0055] S104: Acquire the second information of the taxi to form test sample data;

[0056] In a preferred embodiment of the present invention, the taxi driving data of the day is obtained, such as driving status data, user status data, environmental data and taxi fault data, etc. The specific data type is the same as the learning sample data, and no repetition is made here, and the obtained data is sorted to obtain the test sample data, and the test sample data can also be normalized to obtain the test sample data that is more conducive to calculation. It should be noted that on the time axis, the test sample data is after the learning sample data. For example, if the test sample data is October 1, the learning sample data is before October 1 (excluding October 1), and then it is sorted and normalized in the same way as the learning sample data to facilitate subsequent prediction.

[0057] S105: Use test sample data to verify the model prediction results and cyclically optimize the model;

[0058] In a preferred embodiment of the present invention, the test sample data is input into the Bayesian logistic regression model for prediction, and the prediction result is obtained, and the prediction result is compared with the actual result. If the error between the predicted result and the actual fault is within a first set value, then the prediction result of the fault can be considered accurate; but if the error between the predicted result and the actual fault exceeds the first set value, the test sample data is added to the learning sample data for recalculation and modeling to obtain a new Bayesian logistic regression model.

[0059] Furthermore, when the error between the predicted result and the actual fault exceeds the set value, the weight of the test sample data will be accumulated according to the percentage of the error, and then multiple identical test sample data will be added to the learning sample data according to the different weights to make the test result more accurate. Assuming that the first set value of the error between the predicted result and the actual fault is 10%, when the error between the predicted result and the actual fault of test sample data A is 55%, the weight of test sample data A is increased by 5, and 5 identical test sample data A are added to the learning sample data to obtain new learning sample data. In this way, the model is trained to make the prediction result more close to the actual fault situation, thereby improving the prediction accuracy of the model.

[0060] Furthermore, in order to further improve the accuracy of model prediction, daily driving data can be used as prediction sample data to input into the model for prediction within a set time, so as to continuously optimize model parameters. If the error value between the predicted result failure and the actual failure is within the first set value within a set time period (such as a month), it can be considered that the Bayesian logistic regression model has been trained and can be used for subsequent taxi failure prediction. However, if the error value between the predicted result failure and the actual failure is intermittent within a continuous set time and does not exceed the second set value, the interface data in the model establishment process is adjusted to adjust the Bayesian logistic regression model. However, if the error value between the predicted result failure and the actual failure exceeds the second set value during the prediction process, it can be considered that the model needs to be re-learned. It should be noted that the first set value is smaller than the second set value. For example, if the first set value is 10%, the second set value may be 50%, that is, if the error between the predicted failure and the actual failure is within 10% within the set time, the model is successful and can be used for taxi failure prediction; if the predicted failure and the actual failure occasionally exceed 10% but do not exceed 50% within the set time, the interface in the model is adjusted to adjust the model and continue to use the adjusted model for fault prediction; but if the error between the predicted failure and the actual failure reaches more than 50% within the set time, it indicates that the model needs to be re-learned, and the same steps as above are used to regenerate the learning sample data and modeling process, and obtain test sample data until the model is successful.

[0061] Through the above method, taxi driving data is obtained to form learning sample data, and the Bayesian logistic regression model is obtained after the learning sample data is normalized. After obtaining the taxi driving data to form the test sample data, the test sample data is input into the model to obtain the predicted result failure, and compared with the actual failure. If the error between the predicted result failure and the actual failure is within the first set value, the prediction result is considered accurate; if the error is greater than the first set value, the multiple of the actual error and the first set value is calculated, and the corresponding test sample data is added to the learning sample data in the same amount according to the multiple, so that the model is continuously trained and optimized. And according to the prediction error within the set time is the first set value, the model is successful. If it is greater than the first set value but less than the second set value intermittently, the model interface is adjusted to adjust the model and repeat the prediction. If it is greater than the second set value, the model learning is repeated, and the learning sample data is repeated until the model is successful.

[0062] A taxi fault prediction method according to a preferred embodiment of the present invention; Figure 2 As shown, it is a flow chart of a taxi fault prediction method provided by an embodiment of the present invention. The method can be executed by a device, and the device can be implemented by software and / or hardware.

[0063] Specifically, in this embodiment, the taxi fault prediction method includes:

[0064] S201: Acquire third information of the taxi to form prediction data;

[0065] In a preferred embodiment of the present invention, the real-time driving data of the taxi driver, such as driving condition data, user status data and environmental data, is used to form prediction data, and the prediction data can also be normalized or normalized in the model, and the prediction data is included in the trained and optimized model.

[0066] S202: inputting the prediction data into the success model for calculation to output the probability of failure;

[0067] In a preferred embodiment of the present invention, the prediction data is simultaneously input into the success model, the driving data transmitted from the interface is calculated, the failure probability of the taxi driver's driving behavior is obtained, and the probability of failure is output and returned to the caller.

[0068] Corresponding to a taxi fault prediction model training method, the present invention also provides a taxi fault prediction model training device, specifically, as Figure 3 As shown, the taxi fault prediction model training device includes: a data acquisition module 301, a data sorting module 302 and a learning prediction module 303.

[0069] The data collection module 301 is used to obtain first information of the taxi to form learning sample data, and obtain second information of the taxi to form test sample data;

[0070] Specifically, the driving information of the taxi at different times is obtained. The driving information is obtained from multiple different dimensions, including driving status data, user status data, environmental data, and taxi fault data. The specific data information has been described in detail above and will not be repeated here. It should be noted that on the time axis, the test sample data is later than the learning sample data. For example, if the test sample data is on October 1st, the learning sample data is before October 1st (excluding October 1st), so as to facilitate subsequent predictions.

[0071] A data sorting module 302 is used to sort and normalize the acquired first information and second information;

[0072] Specifically, the sample data of multiple dimensions are sorted so that the information corresponds to each other, such as the driving status data of a certain car corresponds to its user status data, environmental data and taxi fault data. Then the corresponding data is normalized to facilitate machine language learning.

[0073] The learning prediction module 303 uses the normalized learning sample data to establish a Bayesian logistic regression model, and uses the normalized test sample data to train the model to obtain a successful model;

[0074] Specifically, after the model is established using the learning sample data, the model is trained using the test sample data, the test sample data is input into the Bayesian logistic regression model for prediction, and the prediction result is obtained, and the prediction result is compared with the actual result. If the error between the predicted result and the actual fault is within the first set value, then the prediction result of the fault can be considered accurate; but if the error between the predicted result and the actual fault exceeds the first set value, the test sample data is added to the learning sample data to recalculate the model and re-obtain a new Bayesian logistic regression model. When the error between the predicted result and the actual fault exceeds the set value, the weight of the test sample data is accumulated according to the percentage of the error, and then multiple identical test sample data are added to the learning sample data according to different weights to make the test result more accurate. Assuming that the first set value of the error between the predicted result and the actual fault is 10%, when the error between the predicted result and the actual fault of the test sample data A is 55%, the weight of the test sample data A is increased by 5, and 5 identical test sample data A are added to the learning sample data to obtain new learning sample data. In this way, the model is trained to make the prediction result more inclined to the actual fault situation, thereby improving the prediction accuracy of the model. In order to further improve the accuracy of model prediction, daily driving data can be used as prediction sample data to input into the model for prediction within a set time, thereby realizing continuous optimization of model parameters. If the error value between the predicted result and the actual fault is within the first set value within a set time period (such as a month), it can be considered that the Bayesian logistic regression model has been trained and can be used for subsequent taxi fault prediction. However, if the error value between the predicted result and the actual fault is intermittent within a continuous set time and does not exceed the second set value, the interface data in the model establishment process is adjusted to adjust the Bayesian logistic regression model, and it is continued to be used for prediction until the prediction error is within the first set value within the set time, and the model is considered to be successful. However, if the error between the predicted result and the actual fault exceeds the second set value during the prediction process, it can be considered that the model needs to be re-learned. It should be noted that the first set value is less than the second set value, such as the first set value is 10%, and the second set value can be 50%, that is, if the error between the predicted result and the actual fault is within 10% within the set time, the model is successful and can be used for taxi fault prediction; if the predicted result and the actual fault occasionally exceed 10% but not more than 50% within the set time, the interface in the model is adjusted to adjust the model and continue to use the adjusted model for fault prediction; but if the error between the predicted result and the actual fault reaches more than 50% within the set time, it indicates that the model needs to be re-learned, and the same steps as above are used to regenerate the learning sample data and modeling process, and obtain test sample data until the model is successful.

[0075] Corresponding to a taxi fault prediction method, the present invention also provides a taxi fault prediction device, specifically, as Figure 4 As shown, the taxi fault prediction device includes: a data acquisition module 401, a data sorting module 402 and a learning prediction module 403.

[0076] The data collection module 401 is used to obtain the first information of the taxi to form the learning sample data, obtain the second information of the taxi to form the test sample data, and obtain the third information of the taxi to form the prediction data;

[0077] Specifically, the driving information of the taxi at different times is obtained. The driving information is obtained from multiple different dimensions, including driving status data, user status data, environmental data, and taxi fault data. The specific data information has been explained in detail above and will not be repeated here. It should be noted that on the time axis, the test sample data is later than the learning sample data. For example, if the test sample data is on October 1, the learning sample data is before October 1 (excluding October 1), so as to facilitate subsequent predictions. The predicted data is real-time driving data, and there is no taxi fault data in the predicted data at this time.

[0078] A data sorting module 402 is used to sort and normalize the acquired first information, second information and third information;

[0079] Specifically, the sample data of multiple dimensions of the first information and the second information are sorted so that the information corresponds to each other, such as the driving status data of a certain car corresponds to its user status data, environmental data and taxi fault data. Then the corresponding data is normalized to facilitate machine language learning. The third information only has driving status data, user status data and environmental data for normalization.

[0080] The learning prediction module 403 uses the normalized learning sample data to establish a Bayesian logistic regression model, uses the normalized test sample data to train the model to obtain a successful model, and inputs the normalized prediction data into the successful model for fault prediction;

[0081] Specifically, after a successfully trained model is obtained through the aforementioned method, the prediction data is input into the trained model to obtain a fault prediction result, which is then returned to the caller to implement the fault prediction result of the taxi.

[0082] Those skilled in the art will understand that Figure 5The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the smart terminal to which the scheme of the present invention is applied. The specific smart terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0083] In an embodiment of the present invention, a smart terminal is provided. The smart terminal includes a memory, a processor, and a taxi fault prediction model training program stored in the memory and executable on the processor. When the taxi fault prediction model training program is executed by the processor, the following operation instructions are performed:

[0084] Acquire first information of the taxi, the first information including driving status data, user status data, environmental data and taxi fault data;

[0085] Arrange the driving status data, user status data, environmental data and taxi fault data of the first information to form learning sample data;

[0086] Normalizing the learning sample data to obtain normalized learning sample data;

[0087] Use the normalized learning sample data to learn and predict the severity of the fault and establish a Bayesian logistic regression model;

[0088] Acquire second information of the taxi, the second information including driving status data, user status data, environmental data and taxi fault data;

[0089] Arrange the driving status data, user status data, environmental data and taxi fault data of the second information to form prediction sample data;

[0090] Normalizing the predicted sample data to obtain normalized predicted sample data;

[0091] The normalized prediction sample data is input into the Bayesian logistic regression model to predict the failure results;

[0092] The predicted fault is compared with the actual fault. If the error is within the first set value, the predicted result is considered to be accurate. If the error exceeds the first set value, the corresponding predicted sample data weight is accumulated according to the percentage of the error to the first set value, and the predicted sample data with the same weight is added to the learning sample data.

[0093] If the error is within the first set value within the set time, the model is considered successful; if the error is between the first set value and the second set value within the set time, the model interface is adjusted to adjust the model and continue to predict; if the error exceeds the second set value, the model learning is repeated;

[0094] Acquire the third information of the taxi, the third information including driving status data, user status data and environment data;

[0095] Arrange the driving state data, user state data and environment data of the third information to form prediction sample data;

[0096] Normalizing the predicted data to obtain normalized predicted data;

[0097] The normalized prediction data is input into the successfully trained Bayesian logistic regression model to predict the fault results.

[0098] An embodiment of the present invention also provides a computer-readable storage medium, on which a taxi fault prediction model training program is stored. When the taxi fault prediction model training program is executed by a processor, the steps of any taxi fault prediction model training method provided by an embodiment of the present invention are implemented.

[0099] It should be understood that the present invention is described by some embodiments, and those skilled in the art will appreciate that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.

Claims

1. A taxi fault prediction model training method, characterized in that: The training method includes the following steps: Acquire driving data of a plurality of taxis at the first time to form learning sample data, wherein the driving data includes vehicle driving state data, user state data, environmental information data and taxi fault data; Use learning sample data to learn and predict the severity of faults and establish a regression model; Acquire driving data of a plurality of taxis at a second time after the first time to form test sample data, wherein the driving data includes vehicle driving state data, user state data, environmental information data and taxi fault data; The test sample data is used as the input of the regression model to verify the prediction results and cyclically optimize the regression model.

2. The training method according to claim 1, characterized in that: In the process of obtaining the driving data of the plurality of taxis at the first time to form the learning sample data, the process further includes: The driving data of multiple taxis at the first time are sorted out to form learning sample data, and the learning sample data are normalized; The process of obtaining the driving data of the plurality of taxis at a second time after the first time to form the test sample data further includes: The driving data of multiple taxis in the second room are sorted to form test sample data, and the test sample data are normalized.

3. The training method according to claim 1, characterized in that: In the process of using the test sample data as the input of the regression model to verify the prediction result and cyclically optimize the regression model, it also includes: The vehicle driving status data, user status data and environmental information data in the test sample data are input into the regression model to obtain the predicted failure result, and the predicted failure result is compared with the actual failure. If the error value is within the first set value range, the regression model is available. If the error value is greater than the first set value range, the regression model is adjusted.

4. The training method according to claim 3, characterized in that: In the process of comparing the predicted failure result with the actual failure, if the error value is greater than the first value setting range, then adjusting the regression model also includes: If the predicted result shows a fault and the actual fault is greater than the first set value, the actual error is compared with the first set value, and the multiple of the actual error value and the first set value is calculated. The corresponding prediction sample data weights are accumulated according to the calculated multiples, and the same number of prediction sample data as the multiples are added to the learning sample data to obtain new learning sample data, and the regression model is re-established.

5. The training method according to claim 3, characterized in that: In the process of comparing the predicted result of the fault with the actual fault, if the error value is greater than the first set value range, then adjusting the regression model, the process also includes: If the error between the predicted failure and the actual failure is within the first set value range within the set time, the model training is successful; If the error between the predicted fault and the actual fault occurs intermittently within the set time and exceeds the first set value range, but does not exceed the second set value range, the regression model interface data is adjusted to adjust the regression model and training is performed until the model training is successful; If the error between the predicted failure and the actual failure exceeds the second set value range within the set time, the regression model is re-learned until the model training is successful.

6. A taxi fault prediction method, characterized in that: The method comprises the following steps: Acquiring real-time driving data of multiple taxis during driving to form test data, wherein the driving data includes driving state data, user state data and environment data; The test data is input into the regression model trained by the taxi fault prediction model training method described in any one of claims 1 to 5 to perform fault prediction.

7. A taxi fault prediction model training device, characterized in that: The training device includes: A data acquisition module, used to obtain driving data of multiple taxis at a first time to form learning sample data, and to obtain driving data of multiple taxis at a second time after the first time to form test sample data; A data sorting module sorts the acquired learning sample data and test sample data so that the driving data of each vehicle corresponds; The learning prediction module uses the learning sample data to learn the fault severity and establish a regression model for prediction, and uses the test sample data to verify the regression model and cyclically optimize the regression model.

8. A taxi fault prediction device, characterized in that: The prediction device includes: A data acquisition module, used to obtain the driving data of multiple taxis at the first time to form learning sample data, obtain the driving data of multiple taxis at the second time after the first time to form test sample data, and obtain the real-time driving information of multiple taxis to form test data; A data sorting module sorts the acquired learning sample data, test sample data and test data so that the driving data of each vehicle corresponds; The learning prediction module uses the learning sample data to learn the fault severity and establish a regression model for prediction, and uses the test sample data to verify the regression model and cyclically optimize the regression model, and uses the test data to input into the optimized regression model for fault prediction.

9. An intelligent terminal, characterized in that: The intelligent terminal includes at least one memory, at least one processor, and a taxi fault prediction model training program stored in the memory and executable on the processor. When the taxi fault prediction model training program is executed by the processor, the steps of the taxi fault prediction model training method as described in any one of claims 1 to 5 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a taxi fault prediction model training program, and when the taxi fault prediction model training program is executed by the processor, the steps of the taxi fault prediction model training method according to any one of claims 1 to 5 are implemented.

Images