Training method and system of oil consumption estimation model

Abstract

The invention provides a training method and system of a fuel consumption estimation model. The training method and system of the fuel consumption estimation model comprise a data collection method, a data preprocessing method, a model training method, a model iterative training method and a model iterative training convergence judgment method. The training method and system for the fuel consumption estimation model have the advantage that the fuel consumption of the vehicle can be estimated without installing a sensor in the fuel tank of the vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent transportation systems, and in particular to a training method and system for a fuel consumption estimation model. Background Art

[0002] The current fuel consumption estimation system mainly installs sensors in the fuel tank of the vehicle to sense the fuel level, but the fuel consumption estimation system requires the cost of installing the sensors, and the sensor installation process may cause damage to the vehicle body.

[0003] Therefore, it is necessary to provide a new training method and system for a fuel consumption estimation model to solve the above technical problems. Summary of the invention

[0004] The technical problem solved by the present invention is to provide a training method and system for a fuel consumption estimation model that can estimate the fuel consumption of a vehicle without installing a sensor in the vehicle fuel tank.

[0005] In order to solve the above technical problems, the training method of the fuel consumption estimation model provided by the present invention includes: a data collection method: using a plurality of vehicle-mounted devices to collect driving data, the driving data at least includes the vehicle-mounted device number, time point information, positioning information, and vehicle speed information, and the vehicle-mounted device can report a refueling data to the intelligent transportation system server after each refueling, the refueling data includes the vehicle-mounted device number, time point information, and refueling amount information;

[0006] Data pre-processing method: Use the intelligent transportation system server to analyze driving data and refueling data, count multiple driving data and multiple refueling data of the same vehicle equipment number in the same month, calculate the number of data in each speed range as the input variable of the fuel consumption estimation model, calculate the total amount of refueling in that month as the output variable, and obtain a model data after the information of a vehicle equipment in the same month is counted, and calculate the information of each vehicle equipment in each month to obtain multiple model data as the initial data set, and initialize the fuel consumption estimation model;

[0007] Model training method: using the intelligent transportation system server to apply model data and adopt an optimization algorithm to train the fuel consumption estimation model, and after the fuel consumption estimation model training is completed, calculating the average fuel consumption estimation error value;

[0008] Model iterative training method: Using a clustering method on the intelligent transportation system server, the data set is divided into two groups according to the total amount of fuel refueled, and an initial data set is divided into two clustered data sets. The clustered data sets are then substituted into the model training method to train two fuel consumption estimation models separately. After the fuel consumption estimation model training is completed, the average fuel consumption estimation error value after clustering is calculated.

[0009] Model iterative training convergence judgment method: If the average fuel consumption estimation error value of the i+1th model iterative training method is greater than the average fuel consumption estimation error value of the i-th model iterative training method, the fuel consumption estimation model trained by the i-th model iterative training method is adopted; if the average fuel consumption estimation error value of the i+1th model iterative training method is less than the average fuel consumption estimation error value of the i-th model iterative training method, the model iterative training method is executed again.

[0010] Preferably, the fuel consumption estimation model is a linear model, and each input variable in the fuel consumption estimation model has a weight variable.

[0011] Preferably, the optimization algorithm is a gradient descent algorithm.

[0012] Preferably, the optimization algorithm is a genetic algorithm.

[0013] Preferably, the optimization algorithm is a particle swarm optimization algorithm.

[0014] Preferably, the clustering method is K-Means algorithm.

[0015] Preferably, when the clustering method calculates similarity, the similarity method is based on the chi-square value.

[0016] Preferably, the clustering method is the DBSCAN algorithm.

[0017] Preferably, the average fuel consumption estimation error value is based on the mean square error.

[0018] The present invention also provides a training system for a fuel consumption estimation model, comprising: a plurality of vehicle-mounted devices for executing a data collection method;

[0019] An intelligent transportation system server is used to execute a data pre-processing method, a model training method, a model iterative training method, and a model iterative training convergence judgment method.

[0020] Compared with the related art, the training method and system of the fuel consumption estimation model provided by the present invention have the following beneficial effects:

[0021] The present invention provides a training method and system for a fuel consumption estimation model, which does not require the installation of a sensor in a fuel tank of a vehicle, thereby being able to avoid damage to the vehicle body during the installation or replacement of the sensor. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 An architecture diagram of a training system for a fuel consumption estimation model provided by an embodiment of the present invention;

[0023] Figure 2A flowchart of a method for training a fuel consumption estimation model provided by an embodiment of the present invention;

[0024] Figure 3 A schematic diagram of a model iterative training method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0026] Figure 1 An architecture diagram of a fuel consumption estimation model training system provided in an embodiment of the present invention. The present invention provides a fuel consumption estimation model training system, and the system includes: n vehicle-mounted devices and an intelligent transportation system server.

[0027] The vehicle-mounted devices can execute the data collection method, and each has its own vehicle-mounted device number, global positioning system, and mobile network module. The global positioning system can use time point information, positioning information, and vehicle speed information, and the vehicle-mounted devices can report a piece of driving data to an intelligent transportation system server every second through their mobile network modules, and the vehicle-mounted devices can report a piece of refueling data to the intelligent transportation system server after each refueling.

[0028] The driving data at least includes the vehicle equipment number, time point information, positioning information, and vehicle speed information:

[0029]

[0030] The refueling data includes the vehicle equipment number, time point information, and refueling amount information:

[0031]

[0032]

[0033] The intelligent transportation system server can execute the data pre-processing method, the model training method, the model iterative training method, and the model iterative training convergence judgment method.

[0034] Figure 2 A flowchart of a fuel consumption estimation model training method provided in an embodiment of the present invention; the present invention provides a fuel consumption estimation model training method, comprising the following steps: a data collection method, a data pre-processing method, a model training method, a model iterative training method, and a model iterative training convergence judgment method.

[0035] The data collection method is that a plurality of vehicle-mounted devices are equipped with a global positioning system and a mobile network module and can report a piece of driving data every second. The driving data at least includes the vehicle-mounted device number, time point information, positioning information, and vehicle speed information to an intelligent transportation system server, and the vehicle-mounted devices can report a piece of refueling data after each refueling. The refueling data includes the vehicle-mounted device number, time point information, and refueling amount information to the intelligent transportation system server.

[0036] The data pre-processing method is that the intelligent transportation system server can analyze the driving data and the refueling data, count multiple driving data and multiple refueling data of the same vehicle-mounted equipment number in the same month, calculate the number of data in each speed range as the input variable of the fuel consumption estimation model, calculate the total amount of fuel refueled in the month as the output variable, obtain a model data after the information of a vehicle-mounted equipment in the same month is counted, and calculate the information of each vehicle equipment in each month to obtain multiple model data as the initial data set, and initialize the fuel consumption estimation model.

[0037] The speed interval is composed of 10 kilometers per hour as a speed interval, including speed interval x0: 0 kilometers per hour, speed interval x1: 0-10 kilometers per hour, speed interval x2: 10-20 kilometers per hour, speed interval x3: 20-30 kilometers per hour, speed interval x4: 30-40 kilometers per hour, speed interval x5: 40-50 kilometers per hour, speed interval x6: 50-60 kilometers per hour, speed interval x7: 60-70 kilometers per hour, speed interval x8: 70-80 kilometers per hour, speed interval x9: 80-90 kilometers per hour, speed interval x10: 90-100 kilometers per hour, speed interval x11: 100-110 kilometers per hour, speed interval x12: 110-120 kilometers per hour, speed interval x13: greater than 120 kilometers per hour.

[0038] The data set contains the information of each vehicle in each month. The number of data in each speed range is used as the input variables x0~x13 of the fuel consumption estimation model, and the total amount of fuel (liters) y in that month is used as the output variable:

[0039]

[0040] The fuel consumption estimation model is y=w0*x0+w1*x1+w2*x2+w3*x3+w4*x4+w5*x5+w6*x6+w7*x7+w8*x8+w9*x9+w10*x10+w11*x11+w12*x12+w13*x13, and the weight variables are w0~w13.

[0041] The fuel consumption estimation model is a linear model, and each input variable in the model has a weight variable.

[0042] The model training method, the intelligent transportation system server uses the model data and an optimization algorithm to train the fuel consumption estimation model, the training process mainly optimizes the weight variables, and after the fuel consumption estimation model training is completed, the average fuel consumption estimation error value is calculated, and the average fuel consumption estimation error value is based on the mean square error (MSE).

[0043] The optimization algorithm is the gradient descent algorithm.

[0044] The optimization algorithm is a genetic algorithm.

[0045] The optimization algorithm is the particle swarm optimization algorithm.

[0046] The model iterative training method, the intelligent transportation system server can use a clustering method to divide the data set into two groups according to the total amount of fuel refueled, divide an initial data set into two clustered data sets, and then substitute these clustered data sets into the model training method to train them separately to generate two fuel consumption estimation models, and after the training of these fuel consumption estimation models is completed, calculate the average fuel consumption estimation error value after clustering.

[0047] The model iterative training convergence judgment method is as follows: if the average fuel consumption estimation error value of executing the i+1th model iterative training method is greater than the average fuel consumption estimation error value of the ith model iterative training method, the fuel consumption estimation model trained by the ith model iterative training method is adopted; if the average fuel consumption estimation error value of executing the i+1th model iterative training method is less than the average fuel consumption estimation error value of the ith model iterative training method, the model iterative training method is executed again.

[0048] The clustering method is K-Means algorithm.

[0049] When the clustering method calculates similarity, the similarity method is based on the chi-square value.

[0050] The clustering method is the DBSCAN algorithm.

[0051] Figure 3 A schematic diagram of a model iteration training method provided by an embodiment of the present invention, wherein the initial data set is the first data set, including data of vehicle-mounted equipment numbered 1 to vehicle-mounted equipment numbered n, and after the fuel consumption estimation model training is completed, the calculated average fuel consumption estimation error value is 0.06;

[0052] Using the iterative training method of the model and the clustering method, clustering is performed according to the total amount of fuel refueled to generate two categories, and two data sets are generated according to the clustering results, namely the second data set and the third data set. In this embodiment, the second data set includes the vehicle-mounted device number 1 and the vehicle-mounted device number 2, and the third data set includes the vehicle-mounted device number n;

[0053] Using the model training method, the fuel consumption estimation model is trained using the second data set and the third data set respectively, wherein the average fuel consumption estimation error value of the fuel consumption estimation model trained using the second data set is 0.05, and the average fuel consumption estimation error value of the fuel consumption estimation model trained using the third data set is 0.07;

[0054] Using the iterative training convergence judgment method of the model, the average fuel consumption estimation error value of 0.07 of the fuel consumption estimation model trained by the third data set is greater than the average fuel consumption estimation error value of 0.06 of the fuel consumption estimation model trained by the first data set. Therefore, the vehicle-mounted equipment in the third data set is suitable for the fuel consumption estimation model trained by the first data set.

[0055] The average fuel consumption estimation error value of 0.05 of the fuel consumption estimation model trained by the second data set is smaller than the average fuel consumption estimation error value of 0.06 of the fuel consumption estimation model trained by the first data set, so the second data set continues to execute the model iterative training method;

[0056] The second data set uses clustering method to cluster according to the total amount of fuel refueled, generating two categories. Based on the clustering results, two data sets are generated, namely the fourth data set and the fifth data set;

[0057] Using the iterative training convergence judgment method of the model, the average fuel consumption estimation error value of 0.055 of the fuel consumption estimation model trained by the fourth data set is greater than the average fuel consumption estimation error value of 0.05 of the fuel consumption estimation model trained by the second data set. Therefore, the vehicle-mounted equipment in the fourth data set is suitable for the fuel consumption estimation model trained by the second data set.

[0058] Using the iterative training convergence judgment method of the model, the average fuel consumption estimation error value of 0.055 of the fuel consumption estimation model trained by the 5th data set is greater than the average fuel consumption estimation error value of 0.05 of the fuel consumption estimation model trained by the 2nd data set. Therefore, the vehicle-mounted equipment in the 5th data set is suitable for the fuel consumption estimation model trained by the 2nd data set.

[0059] Compared with the related art, the training method and system of the fuel consumption estimation model provided by the present invention have the following beneficial effects:

[0060] The present invention provides a training method and system for a fuel consumption estimation model, which does not require the installation of a sensor in a fuel tank of a vehicle, thereby being able to avoid damage to the vehicle body during the installation or replacement of the sensor.

[0061] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for training a fuel consumption estimation model, characterized in that: The following steps are involved: Data collection method: using multiple vehicle-mounted devices to collect driving data, the driving data at least includes the vehicle-mounted device number, time point information, location information, and vehicle speed information, and the vehicle-mounted device can report a refueling data to the intelligent transportation system server after each refueling, the refueling data includes the vehicle-mounted device number, time point information, and refueling amount information; Data pre-processing method: Use the intelligent transportation system server to analyze driving data and refueling data, count multiple driving data and multiple refueling data of the same vehicle equipment number in the same month, calculate the number of data in each speed range as the input variable of the fuel consumption estimation model, calculate the total amount of refueling in that month as the output variable, obtain a model data after the information of a vehicle equipment in the same month is counted, and calculate the information of each vehicle equipment in each month, obtain multiple model data as the initial data set, and initialize the fuel consumption estimation model; Model training method: using the intelligent transportation system server to apply model data and adopt an optimization algorithm to train the fuel consumption estimation model, and after the fuel consumption estimation model training is completed, calculating the average fuel consumption estimation error value; Model iterative training method: Using a clustering method on the intelligent transportation system server, the data set is divided into two groups according to the total amount of fuel refueled, and an initial data set is divided into two clustered data sets. The clustered data sets are then substituted into the model training method to train two fuel consumption estimation models separately. After the fuel consumption estimation model training is completed, the average fuel consumption estimation error value after clustering is calculated. Model iterative training convergence judgment method: If the average fuel consumption estimation error value of the i+1th model iterative training method is greater than the average fuel consumption estimation error value of the i-th model iterative training method, then the fuel consumption estimation model trained by the i-th model iterative training method is used; If the average fuel consumption estimation error value of the i+1th model iteration training method is less than the average fuel consumption estimation error value of the i-th model iteration training method, the model iteration training method is executed again.

2. The training method of the fuel consumption estimation model according to claim 1, characterized in that: The fuel consumption estimation model is a linear model, and each input variable in the fuel consumption estimation model has a weight variable.

3. The training method of the fuel consumption estimation model according to claim 2, characterized in that: The optimization algorithm is a gradient descent algorithm.

4. The method for training a fuel consumption estimation model according to claim 2, characterized in that: The optimization algorithm is a genetic algorithm.

5. The method for training a fuel consumption estimation model according to claim 2, characterized in that: The optimization algorithm is a particle swarm optimization algorithm.

6. The method for training a fuel consumption estimation model according to claim 2, characterized in that: The clustering method is the K-Means algorithm.

7. The method for training a fuel consumption estimation model according to claim 6, characterized in that: When the clustering method calculates similarity, the similarity method is based on the chi-square value.

8. The method for training a fuel consumption estimation model according to claim 1, characterized in that: The clustering method is the DBSCAN algorithm.

9. The method for training a fuel consumption estimation model according to claim 1, characterized in that: The average fuel consumption estimation error value is based on the mean square error.

10. A training system for a fuel consumption estimation model, characterized in that: include: A plurality of vehicle-mounted devices are used to perform the data collection method; An intelligent transportation system server is used to execute a data pre-processing method, a model training method, a model iterative training method, and a model iterative training convergence judgment method.

Images