Mine Truck Engine Cylinder Temperature Early Warning Method Based on Random Forest Algorithm

Abstract

The present invention relates to a method for warning the cylinder temperature of a mining truck engine based on the random forest algorithm. Since the cylinder temperature of a mining truck engine has a lag effect, that is, the cylinder temperature is affected by the past values of itself and other explanatory variables, it is an urgent problem to be solved to perform time lag processing on the variables. The steps and methods of the present invention are as follows: First, the cylinder temperature is predicted based on the eigenvalue related to the cylinder temperature, its rolling window value, and the statistical eigenvalue using the random forest algorithm. Then, according to the difference between the predicted value and the actual value and the threshold range obtained by statistically analyzing a large amount of cylinder temperature data, the calculation flow chart is as follows: (1) Collect data, process missing values and outliers; (2) Process eigenvalues and construct a data set; (3) Build a random forest algorithm model and add new variables; (4) Train the model, adjust parameters, and predict the cylinder temperature. The present invention is used for the method for warning the cylinder temperature of a mining truck engine based on the random forest algorithm.

Description

Technical Field

[0001] The present invention relates to a method for warning the cylinder temperature of a mining truck engine based on a random forest algorithm. Background Art

[0002] The random forest algorithm is a supervised learning algorithm that integrates multiple trees through the idea of ensemble learning. It has the characteristics of high flexibility, strong adaptability, and not being easily trapped in overfitting. Since the cylinder temperature of a mining truck engine has a lag effect, that is, the cylinder temperature is affected by the past values of itself and other explanatory variables, it is necessary to perform time lag processing on the variables, which is an urgent problem to be solved. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for warning the cylinder temperature of a mining truck engine based on a random forest algorithm. This method mainly predicts the real-time cylinder temperature based on relevant parameters and compares it with the actual cylinder temperature to detect abnormal cylinder temperature in advance and reduce the maintenance cost of mining truck equipment.

[0004] The above purpose is achieved through the following technical solutions:

[0005] A method for warning the cylinder temperature of a mining truck engine based on a random forest algorithm, the method includes the following steps: First, predict the cylinder temperature based on the eigenvalue related to the cylinder temperature, its rolling window value, and statistical eigenvalue based on the random forest algorithm, and then calculate according to the difference between the predicted value and the actual value and the threshold range obtained by statistically analyzing a large amount of cylinder temperature data. The flow chart is as follows:

[0006] (1) Collect data and process missing values and outliers:

[0007] First, construct a data set. It is necessary to perform measurement point sampling on ambient temperature, ambient pressure, wind speed, vehicle speed, horsepower, engine speed, accelerator pedal percentage, brake pedal percentage, k# cylinder temperature, and k + 1# cylinder temperature, and the sampling frequency is 2s;

[0008] (2) Process eigenvalues and construct a data set:

[0009] Use a box plot to statistically analyze each eigenvalue one by one to determine the maximum and minimum ranges of each eigenvalue. Values outside the range can be identified as outliers. If the number of missing values and outliers is small, they can be replaced by the average value of the front and back data or deleted;

[0010] (3) Build a random forest algorithm model and add new variables:

[0011] A. Add time lag variables, perform time lag processing on independent variables x1 - x9 respectively, and the lag intervals are -1, -2, -3, -4. The time intervals of the corresponding new variables are -2s, -4s, -6s, -8s. For example, the ambient temperature is , when the lag interval is -1, , and similar processing is done for the other variables;

[0012] B. Increase the rolling window statistical values. Respectively increase the rolling window statistical values for the independent variables x1 - x9, namely the rolling standard deviation, rolling gradient, rolling mean, rolling maximum value, and rolling minimum value. The window size is set to 5. For example, the rolling standard deviation of the ambient temperature at this time is:

[0013] ;

[0014] The rolling gradient of the ambient temperature at this time is:

[0015] ;

[0016] The rolling mean of the ambient temperature at this time is:

[0017] ;

[0018] The rolling maximum value of the ambient temperature at this time is:

[0019] ;

[0020] The rolling minimum value of the ambient temperature at this time is:

[0021] ;

[0022] Similar processing is done for the other variables;

[0023] C. Combine the data obtained in steps A, B, and step (2) into a dataset, and divide it into a training set and a test set according to a ratio of 7:3;

[0024] (4) Train the model, adjust the parameters, and predict the cylinder temperature:

[0025] Use the random forest algorithm to predict the dependent variable x10. The parameter configuration is: the number of decision trees n_estimators = 18, the maximum depth of the decision tree max_depth = 18, the number of features considered when restricting branching max_features = 'auto', that is, max_features = sqrt(n_features), and the function for measuring the quality of tree splitting selects the default mean squared error criterion ='mse'. The mean squared error in the decision tree can be expressed as:

[0026] ;

[0027] where Xm is the sample set of the current node, y is the sample target variable of the current node, is the average value of the sample target variable of the current node, and the other parameters are set to the default values. Beneficial effects

[0028] 1. The present invention is a method for predicting the cylinder temperature of a mining truck engine based on the random forest algorithm. This method predicts the real-time cylinder temperature according to relevant parameters and compares it with the actual cylinder temperature to detect abnormal cylinder temperature in advance, reducing the maintenance cost of mining truck equipment. In addition, rolling window statistical variables such as rolling variance, rolling gradient, rolling average, rolling maximum, and rolling minimum are added, which can make the prediction more accurate.

[0029] The present invention predicts the cylinder temperature based on the random forest algorithm according to the eigenvalue related to the cylinder temperature, its rolling window value, and statistical eigenvalue. According to the difference between the predicted value and the actual value and the threshold range obtained by statistical analysis of a large amount of cylinder temperature data, early warning is carried out when the mining truck engine fails, ensuring the safe operation of the equipment.

[0030] When the mining truck engine fails, it will first be reflected in the cylinder temperature. According to the method in the present invention, early warning can be carried out in advance, reducing equipment maintenance costs, saving costs, and improving work efficiency. Brief description of the drawings

[0031] Appendix Figure 1 is the calculation flow chart of the present invention.

[0032] Appendix Figure 2 is the statistical chart of the cylinder temperature prediction error of the present invention. Specific implementation manners Example 1:

[0033] A method for predicting the cylinder temperature of a mining truck engine based on the random forest algorithm, the method comprising the following steps: First, predict the cylinder temperature based on the random forest algorithm according to the eigenvalue related to the cylinder temperature, its rolling window value, and statistical eigenvalue, and then calculate the flow chart according to the difference between the predicted value and the actual value and the threshold range obtained by statistical analysis of a large amount of cylinder temperature data as follows:

[0034] (1) Collect data and process missing values and outliers;

[0035] First, construct a data set. It is necessary to perform measurement point sampling on environmental temperature, environmental pressure, wind speed, vehicle speed, horsepower, engine speed, accelerator pedal percentage, brake pedal percentage, k# cylinder temperature, and k+1# cylinder temperature, and the sampling frequency is 2s;

[0036] (2) Process eigenvalues and construct a data set;

[0037] Use box plots to perform statistical analysis on each eigenvalue one by one to determine the maximum and minimum ranges of each eigenvalue. Values outside the range can be identified as outliers. If the number of missing values and outliers is small, they can be replaced with the average value of the front and back data or deleted;

[0038] (3) Build a random forest algorithm model and add new variables:

[0039] A. Add time lag variables. Perform time lag processing on independent variables x1 - x9 respectively. The lag intervals are -1, -2, -3, -4, and the corresponding time intervals of the new variables are -2s, -4s, -6s, -8s. For example, for the ambient temperature , when the lag interval is -1, , and similar processing is done for the other variables;

[0040] B. Add rolling window statistical values. Add rolling window statistical values to independent variables x1 - x9 respectively, namely rolling standard deviation, rolling gradient, rolling mean, rolling maximum, and rolling minimum. The window size is set to 5. At this time, the rolling standard deviation of the ambient temperature is:

[0041] ;

[0042] At this time, the rolling gradient of the ambient temperature is:

[0043] ;

[0044] At this time, the rolling mean of the ambient temperature is:

[0045] ;

[0046] At this time, the rolling maximum of the ambient temperature is:

[0047] ;

[0048] At this time, the rolling minimum of the ambient temperature is:

[0049] ;

[0050] Similar processing is done for the other variables;

[0051] C. Combine the data obtained in steps A, B, and step (2) into a dataset and divide it into a training set and a test set according to a ratio of 7:3;

[0052] (4)Train the model, adjust the parameters, and predict the cylinder temperature:

[0053] Predict the dependent variable x10 using the random forest algorithm with the following parameter settings: the number of decision trees n_estimators = 18, the maximum depth of the decision tree max_depth = 18, the number of features to consider when restricting branching max_features = 'auto', i.e., max_features = sqrt(n_features), and the function for measuring the quality of tree splitting is set to the default mean squared error criterion ='mse'. In a decision tree, the mean squared error can be expressed as:

[0054] ;

[0055] where Xm is the sample set of the current node, y is the target variable of the current node sample, is the average value of the target variable of the current node sample, and the remaining parameters are set to their default values.

[0056] The following is an example:

[0057] In this experiment, taking the cylinder temperature warning of a mining truck engine as an example, sampling is carried out at the measurement points in Table 1 with a sampling frequency of 5 s. The total training data provides 384 hours of sample data (i.e., the number of training sample points is 276,480), and the total test data provides 48 hours of data (i.e., the number of samples in the test set is 34,560). After box plot analysis, it is determined that when the cylinder temperature exceeds 650 °C or is less than 130 °C (at this time, it is considered that the truck has stalled and data upload stops), it is abnormal data. If it exceeds the range for 1 minute continuously, an alarm is issued. If it is less than 1 minute, it is considered a system jump and can be ignored;

[0058] Based on the above data, the method of the present invention is used to predict the cylinder temperature of the mining truck engine, and the test results shown in Table 2 and Appendix Figure 2 are obtained.

[0059] Through the analysis of Appendix Figure 2 it can be found that for more than 90% of the absolute value differences between the predicted values and the actual values of the present invention, they are less than 15 °C. To reduce the false alarm rate, the absolute error can be set to 25 °C. When the absolute value of the difference between the actual value and the predicted value exceeds 25 °C and lasts for 1 minute, it can be considered that the engine has a fault and an alarm is issued;

[0060] Table 1: Measurement Point Data

[0061] ;

[0062] Table 2: Test Results

[0063] 。

Claims

1. A method for warning the temperature of a mining truck engine cylinder based on the random forest algorithm, characterized in that: The method includes the following steps: First, the cylinder temperature is predicted based on the eigenvalue related to the cylinder temperature, its rolling window value, and the statistical eigenvalue using the random forest algorithm. Then, according to the difference between its predicted value and the actual value and the threshold range obtained from the statistical analysis of a large number of cylinder temperature data, the flowchart for calculation is as follows: (1) Collect data and process missing values and outliers: First, construct a data set. It is necessary to sample the measurement points of environmental temperature, environmental pressure, wind speed, vehicle speed, horsepower, engine speed, accelerator pedal percentage, brake pedal percentage, cylinder temperature of k, and cylinder temperature of k + 1. The sampling frequency is 2s; (2) Process eigenvalues and construct a data set: Use box plots to conduct statistical analysis on each eigenvalue one by one to determine the maximum and minimum ranges of each eigenvalue. Values outside the range can be identified as outliers. If the number of missing values and outliers is small, they can be replaced by the average value of the previous and subsequent data or deleted; (3) Build a random forest algorithm model and add new variables: A. Add time lag variables, and perform time lag processing on independent variables x1 - x9 respectively. The lag intervals are -1, -2, -3, -4, and the time intervals of the corresponding new variables are -2s, -4s, -6s, -8s. The environmental temperature is , when the lag interval is -1, , and similar processing is performed on the other variables; B. Add rolling window statistical values. Add rolling window statistical values to the independent variables x1 - x9 respectively, namely rolling standard deviation, rolling gradient, rolling mean, rolling maximum value, and rolling minimum value. The window size is set to 5. At this time, the rolling standard deviation of the environmental temperature is: ， At this time, the rolling gradient of the environmental temperature is: ， At this time, the rolling mean of the environmental temperature is: ， At this time, the rolling maximum value of the environmental temperature is: ， At this time, the rolling minimum value of the environmental temperature is: ， Similar processing is also done for the other variables; C. Combine the data obtained in steps A, B, and step (2) into a data set and divide it into a training set and a test set according to a ratio of 7:3; (4) Train the model, adjust the parameters, and predict the cylinder temperature: Use the random forest algorithm to predict the dependent variable x10. The parameter configuration is: the number of decision trees n_estimators = 18, the maximum depth of the decision tree max_depth = 18, the number of features considered when restricting branching max_features = 'auto', that is, max_features = sqrt(n_features), and the function for measuring the quality of tree splitting selects the default mean squared error criterion ='mse'. The mean squared error in the decision tree can be expressed as: ， where Xm is the sample set of the current node, and y is the target variable of the current node sample, is the average value of the target variable of the current node sample, and the remaining parameters are set to default values.

Images