Oil consumption prediction method and device and computer storage medium

Abstract

The embodiment of the invention discloses a fuel consumption prediction method and device and a computer storage medium. The method comprises the steps that a data sequence of feature data changing along with time when a vehicle runs is obtained; the characteristic data comprises oil consumption, acceleration and other data; the linear correlation between the data sequence of the other data and the data sequence of the fuel consumption is greater than a correlation threshold; predicting predicted fuel consumption and predicted acceleration at the first future moment through a fuel consumption prediction model based on the data sequence; the first future moment is an adjacent moment after the last moment in the data sequence; under the condition that the future acceleration meets the actual acceleration condition, the future fuel consumption is determined as a fuel consumption prediction result; the actual acceleration condition comprises at least one of the following conditions: the predicted acceleration change rate at the first future moment is within a preset change rate range; the predicted acceleration at the first future moment is greater than an acceleration threshold.

Description

Technical Field

[0001] The present application relates to the field of transportation, and in particular to a fuel consumption prediction method and device, and a computer storage medium. Background Art

[0002] With the continuous development of the transportation industry, fuel consumption always accounts for a large proportion of the cost expenditure of vehicles that undertake more road transportation tasks. Since the data collected and transmitted back by on-board sensors has the characteristics of large scale and strong time series, it becomes more applicable to use sequence prediction models to predict fuel consumption. Therefore, the accuracy of the prediction results of the sequence prediction model will affect the accuracy of the predicted fuel consumption. Summary of the invention

[0003] The embodiments of the present application provide a fuel consumption prediction method and device, and a computer storage medium, which can improve the accuracy of fuel consumption prediction.

[0004] The technical solution of this application is implemented as follows:

[0005] The present application embodiment provides a fuel consumption prediction method, including:

[0006] Acquire a data sequence of characteristic data that changes with time when the vehicle is running; the characteristic data includes fuel consumption, acceleration and other data; the linear correlation between the data sequence of the other data and the data sequence of the fuel consumption is greater than a correlation threshold; based on the data sequence, predict the predicted fuel consumption and predicted acceleration at a first future moment through a fuel consumption prediction model; the first future moment is an adjacent moment after the last moment in the data sequence; if the future acceleration meets an actual acceleration condition, determine that the future fuel consumption is a fuel consumption prediction result; the actual acceleration condition includes at least one of the following: the predicted acceleration change rate at the first future moment is within a preset change rate range; the predicted acceleration at the first future moment is greater than an acceleration threshold.

[0007] The embodiment of the present application provides a fuel consumption prediction device, comprising:

[0008] An acquisition module, used for acquiring a data sequence of characteristic data of the vehicle during operation that varies with time; the characteristic data includes fuel consumption, acceleration and other data; the linear correlation between the data sequence of the other data and the data sequence of the fuel consumption is greater than a correlation threshold;

[0009] A prediction module, configured to predict the predicted fuel consumption and predicted acceleration at a first future moment based on the data sequence by using a fuel consumption prediction model; the first future moment is an adjacent moment after the last moment in the data sequence;

[0010] A determination module is used to determine that the future fuel consumption is a fuel consumption prediction result when the future acceleration meets an actual acceleration condition; the actual acceleration condition includes at least one of the following: the predicted acceleration change rate at the first future moment is within a preset change rate range; the predicted acceleration at the first future moment is greater than an acceleration threshold.

[0011] The embodiment of the present application provides a fuel consumption prediction device, comprising:

[0012] Memory for storing computer programs;

[0013] The processor is used to implement the steps of the above-mentioned fuel consumption prediction method when executing the computer program stored in the memory.

[0014] An embodiment of the present application provides a computer storage medium, which stores one or more computer programs. The one or more computer programs can be executed by one or more processors to implement the steps of the above-mentioned fuel consumption prediction method.

[0015] The fuel consumption prediction method and device and computer storage medium provided in the embodiment of the present application can predict fuel consumption based on a data sequence of characteristic data with a high linear correlation with fuel consumption, and also predict acceleration. Only when the predicted acceleration meets the actual acceleration condition, the predicted fuel consumption is used as the fuel consumption prediction result. In this way, the accuracy of the prediction can be improved, and the predicted future fuel consumption can be further screened by using the actual acceleration condition, which can improve the accuracy of the fuel consumption prediction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic flow chart of an optional fuel consumption prediction method provided in an embodiment of the present application;

[0017] Figure 2 A schematic flow chart of an optional fuel consumption prediction method provided in an embodiment of the present application;

[0018] Figure 3 A schematic flow chart of an optional fuel consumption prediction method provided in an embodiment of the present application;

[0019] Figure 4 A schematic flow chart of an optional fuel consumption prediction method provided in an embodiment of the present application;

[0020] Figure 5 A schematic diagram of the structure of an optional prediction device provided in an embodiment of the present application;

[0021] Figure 6 A schematic diagram of the structure of an optional prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0023] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it can be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0024] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described here can be implemented in an order other than that illustrated or described here.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0026] In order to illustrate the technical solution of the embodiment of the present application, the technical background is first described below:

[0027] With the continuous development of the transportation industry, for vehicles that undertake more road transportation tasks, fuel consumption expenses always account for a large proportion of the operator's cost expenditure. At the same time, the increase in fuel consumption has also brought about problems such as energy shortages and environmental pollution. At present, through the Internet of Vehicles technology, a large amount of vehicle operation status data collected by on-board sensors can be centrally transmitted back to the Internet of Vehicles big data analysis platform. These data have the characteristics of large scale, strong time series and rich attributes. Factors related to vehicle driving conditions, external environment, driving behavior, etc. that will affect vehicle fuel consumption can all be reflected in these data. Based on the above characteristics, more and more fuel consumption predictions often use transformers and other models that predict the data at the next moment based on sequence data. However, even if the prediction accuracy of the model is high, there will still be problems with insufficient accuracy of the prediction results.

[0028] The embodiments of the present application provide a fuel consumption prediction method and device, and a computer storage medium. Since the fuel consumption prediction model predicts acceleration while predicting fuel consumption, the accuracy of the predicted fuel consumption is further screened by the predicted acceleration, thereby improving the accuracy of the predicted fuel consumption. The following describes an exemplary application of the fuel consumption prediction device (hereinafter described as a prediction device) provided in the embodiments of the present application. The above-mentioned fuel consumption prediction device provided in the embodiments of the present application can be implemented as various types of electronic devices such as laptop computers, tablet computers, desktop computers, mobile devices (e.g., mobile phones, portable music players, personal digital assistants, dedicated messaging devices, portable gaming devices, smart watches), etc.

[0029] Figure 1 An exemplary fuel consumption prediction method provided in an embodiment of the present application is shown, and the method may include: S101-S103.

[0030] S101, obtaining a data sequence of characteristic data that changes over time when the vehicle is running; the characteristic data includes fuel consumption, acceleration and other data; the linear correlation between the data sequence of other data and the data sequence of fuel consumption is greater than a correlation threshold.

[0031] In an embodiment of the present application, a data sequence of characteristic data of vehicle operation is used to characterize characteristic data of vehicle operation acquired at a plurality of consecutive sampling moments, and the time intervals between two adjacent moments in the plurality of sampling moments are the same. The data sequence is a sequence of real characteristic data that changes over time. The characteristic data of vehicle operation include fuel consumption, acceleration and other data. Among them, other data may be data with a high degree of correlation with fuel consumption that is screened out from the original driving data. The original driving data may include at least one of the following: speed, torque, engine speed, air intake, DPF pressure difference, water temperature and engine fuel consumption rate, etc., which may be set according to actual needs, and the embodiment of the present application does not impose any limitation.

[0032] In some embodiments, the server collects and stores the original driving data collected at each collection time, and the prediction device can directly obtain the original driving data from the server. Here, the data collection field names of the original driving data may include: vehicle identification, driving data identification, collection time and collection location. Exemplarily, Table 1 shows the original driving data collected at any collection time and the corresponding data collection field names. As shown in Table 1, the original driving data at each collection time can be distinguished by the data collection field name. Among them, at least one of the VIN model number and the terminal receiving device ID represents the vehicle identification; the longitude and latitude represent the collection location; the remaining data collection field names, except for the collection time, are all driving data identification. Each driving data identification corresponds to a type of original driving data.

[0033] Table 1

[0034]

[0035] Among them, SCR stands for Selective Catalytic Reduction, which is a treatment process for nitrogen oxides NOx in diesel vehicle exhaust emissions; DPF stands for Diesel Particulate Filter.

[0036] It should be noted that the formats of various raw driving data stored in the server are the same, that is, the units of the same raw driving data at different collection times are all preset unified units. For example, the raw driving data includes speed, and the unit of speed is km / h.

[0037] In some embodiments, the prediction device can collect the original driving data of the vehicle at a first time interval to obtain a data sequence of the original driving data. For example, the first time interval is 2s, and the data sequence includes the original driving data of 20 moments from t1 to t20, wherein the interval between two adjacent moments is 2s; that is, t1 is moment 0, t2 is 2s, t3 is 4s, and so on, and the original driving data is collected every 2s. In some embodiments, the prediction device can directly filter the data sequences of other data, the data sequences of fuel consumption, and the data sequences of acceleration from the data sequence of the original driving data, and use them together as the data sequence of the feature data. In some embodiments, the prediction device can extract the original driving data at the first collection moment from the data sequence of the original driving data, and form the original driving data at the first collection moment into a data sequence to be filtered; wherein the second time interval between any two adjacent first collection moments is the same. For example, the original driving data at moments t1, t3, t5, ..., t17 and t19 are selected from a total of 20 moments t1-t20 as the data sequence to be filtered. After obtaining the data sequence to be filtered, data sequences of other data, data sequences of fuel consumption and data sequences of acceleration can be filtered from the data sequence to be filtered and together used as the historical data sequence of feature data.

[0038] In an embodiment of the present application, the prediction device can determine the original driving data as other data when the linear correlation between the data sequence of the original driving data and the data sequence of the fuel consumption is greater than a correlation threshold. The correlation threshold can be set according to actual needs and is not limited by the embodiment of the present application. In some embodiments, the linear correlation can be characterized by the Pearson coefficient.

[0039] In some embodiments, all data in the data sequence of the feature data may be dimensionless data. In some embodiments, the dimensionless processing method is normalization processing; the normalization processing method may be any of the following: deviation normalization (Max-Min) method, mean normalization method, and Z-score normalization method, etc., which are not limited in the embodiments of the present application.

[0040] S102. Based on the data sequence, predict the predicted fuel consumption and predicted acceleration at a first future moment by using a fuel consumption prediction model; the first future moment is an adjacent moment after the last moment in the data sequence.

[0041] In the embodiment of the present application, after the data sequence is input into the fuel consumption prediction model, the fuel consumption prediction model can output the predicted fuel consumption and predicted acceleration at the first future moment. The first future moment and the last moment in the data sequence are continuous, and the time interval between the two is the same as the time interval between any two adjacent moments in the data sequence.

[0042] For example, the data sequence includes 6 moments, namely t1-t6, and the interval between any two moments is 2s. The data sequence is input into the fuel consumption prediction model, and the fuel consumption and acceleration at the moment t7 can be predicted. Among them, the interval between t7 and t6 is 2s.

[0043] In the embodiment of the present application, the prediction device can use the training data sequence to train the original prediction model to obtain the fuel consumption prediction model. The data types in the training data sequence include all types in the feature data.

[0044] Exemplarily, the training data sequence includes A data sequence, B data sequence, C data sequence, fuel consumption sequence and acceleration sequence; wherein, a continuous part of A data sequence, B data sequence and C data sequence is taken as the input data of the model, and the output data of the model is the predicted fuel consumption at the first future moment and the predicted acceleration at the first future moment. The actual fuel consumption at the first future moment in the fuel consumption sequence is taken as the fuel consumption target value and compared with the predicted fuel consumption at the first future moment, and the actual acceleration at the first future moment in the acceleration sequence is taken as the acceleration target value and compared with the predicted acceleration at the first future moment, and the model parameters are adjusted according to the comparison result, thereby obtaining a fuel consumption prediction model.

[0045] The model may be a model for predicting a sequence, such as a transformer model, an informer, or a long short-term memory network (LSTM) model, etc. This embodiment of the present application does not limit this.

[0046] S103. When the predicted acceleration at the first future moment satisfies the actual acceleration condition, the predicted fuel consumption is determined to be the predicted fuel consumption result, and the actual acceleration condition includes at least one of the following: the predicted acceleration change rate at the first future moment is within a preset change rate range; the predicted acceleration at the first future moment is greater than the acceleration threshold.

[0047] In the embodiment of the present application, due to the limitations of the vehicle's own conditions such as the vehicle's power, the acceleration of the vehicle during operation should meet the actual acceleration conditions. Here, the actual acceleration conditions include at least one of the following: the acceleration change rate corresponding to the predicted acceleration at the first future moment is within the preset change rate range; the absolute value of the predicted acceleration at the first future moment is less than the acceleration threshold. In some embodiments, the preset change rate range and acceleration threshold are different for different types of vehicles. Here, the preset change rate range and acceleration threshold can be set according to actual needs, and the embodiment of the present application does not limit them.

[0048] In the embodiment of the present application, the acceleration change rate corresponding to the acceleration at the first future moment is used to characterize the rate of change of the acceleration from the moment before the first future moment to the acceleration at the first future moment. The difference between the accelerations at two moments is obtained by dividing the difference between the two moments. For example, the acceleration at the current moment is 55m / s. 2 / s, the acceleration at the next moment is 89m 2 / s, the interval between two adjacent moments is 5s, then the acceleration change rate is 6.8m / s 3 If the preset change rate range is 0-5m / s 3 , the prediction device can determine that the acceleration at the next moment does not meet the actual acceleration condition.

[0049] After obtaining the predicted fuel consumption at the first future moment and the predicted acceleration at the first future moment, the prediction device can first determine whether the predicted acceleration at the first future moment meets the actual acceleration condition. If so, the predicted fuel consumption at the first future moment is determined as the fuel consumption prediction result; otherwise, it is determined that the predicted fuel consumption at the first future moment is not accurate enough and is not used as the fuel consumption prediction result.

[0050] For example, the acceleration threshold may be 4 m / s. When the absolute value of the predicted acceleration at the first future moment is greater than or equal to 4 m / s, the prediction device may determine that the prediction result of the predicted fuel consumption at the first future moment is inaccurate.

[0051] It is understandable that since the fuel consumption can be predicted based on the data sequence of characteristic data with high linear correlation with the fuel consumption, the acceleration is also predicted at the same time, and the predicted fuel consumption is used as the fuel consumption prediction result only when the predicted acceleration meets the actual acceleration condition. In this way, the accuracy of the prediction can be improved, and the predicted future fuel consumption can be further screened by using the actual acceleration condition, which can improve the accuracy of the fuel consumption prediction result.

[0052] In some embodiments of the present application, the historical data sequence of characteristic data that changes over time when the vehicle is running is obtained in S101, such as Figure 2 As shown, it may include: S201-S203.

[0053] S201, obtaining a standard data sequence of a vehicle when it is traveling; the data in the standard data sequence is dimensionless data; the characteristic data in the standard data sequence includes fuel consumption, acceleration and other data.

[0054] In the embodiments of the present application, the standard data sequence is dimensionless data. In some embodiments, after obtaining the data sequence of the original driving data, the prediction device can perform standardization processing on the original driving data, eliminate the dimensions of all data, and classify the data into the same order of magnitude for processing. In this way, it is possible to reduce the errors caused by the difference in the order of magnitude of the values ​​due to different units, and improve the prediction accuracy of the fuel consumption prediction model.

[0055] In some embodiments of the present application, the implementation of obtaining the standard data sequence when the vehicle is traveling in S201 is as follows: Figure 3 As shown, it may include: S301-S302.

[0056] S301 , preprocessing abnormal data in an original data sequence of a vehicle when the vehicle is traveling to obtain a preprocessed data sequence.

[0057] In the embodiment of the present application, the original data sequence is the data sequence of the original driving data. After the prediction device obtains the original data sequence, it needs to check the abnormal data therein, pre-process the abnormal data to obtain a pre-processed data sequence; and then standardize the data in the pre-processed data sequence to eliminate the dimension.

[0058] It should be noted that during the data collection stage, it may be interfered by various factors, such as signal interruption, bad weather, and abnormal vehicle driving, resulting in abnormal data in the original data sequence. Abnormal data may include at least one of the following: missing data, abnormal communication data, and data that does not conform to actual working conditions. Among them, abnormal communication data refers to the process of vehicle driving. At a certain moment, due to reasons such as communication signal interruption, all data is not collected, and the data collected at the previous moment is directly used as the data of the current moment. In the original data sequence, all data are repeated. In some embodiments, when the prediction device detects that all data in the original data sequence are repeated in units of multiple collection moments, and the number of multiple collection moments is greater than the number of communication anomalies, these repeated data can be determined as communication anomaly data. The number of communication anomalies can be set according to actual needs, and the embodiments of the present application are not limited. In some embodiments, the prediction device can delete the data of other collection moments except the first collection moment in multiple collection moments to obtain a preprocessed data sequence. The data not conforming to the actual working condition may be data exceeding the design upper limit of the vehicle; for example, the design upper limit of the vehicle includes a speed threshold, and the data not conforming to the actual working condition may include speed data having a speed greater than the speed threshold; or acceleration data having an acceleration greater than the acceleration threshold.

[0059] In some embodiments of the present application, the abnormal data includes missing data; in S301, the abnormal data in the original data sequence of vehicle travel is preprocessed to obtain the implementation method of the preprocessed data sequence, such as Figure 4 As shown, it may include: S401-S402.

[0060] S401. Use a prediction filling method to predict the missing data to obtain a missing data prediction result; the prediction filling method includes any one of the following: Lagrange interpolation method, simple average method, weighted moving average method and exponential smoothing method.

[0061] In the embodiment of the present application, for missing data at any time, the prediction device can use a prediction filling method, that is, to make a prediction based on data at other times near the time of the missing data to obtain a data prediction result.

[0062] In some embodiments, when the acquisition frequency is greater than the frequency threshold, the prediction device can use the simple average method as the prediction filling method; in this way, a good filling effect can be achieved with less calculation. The frequency threshold can be set according to actual needs and is not limited in the embodiments of the present application.

[0063] S402: Use the data prediction results to fill in the missing data to obtain a preprocessed data sequence.

[0064] In the embodiment of the present application, after determining the data prediction result, the prediction device can use the data prediction result to replace the missing data, thereby completing the data of the original data sequence and obtaining a preprocessed data sequence.

[0065] For example, the speed at time t1 is 76 km / h, the speed at time t2 is missing data, and the speed at time t3 is 79 km / h. The prediction device can use a simple average method, using the speed at time t1 of 76 km / h plus the fill difference, the fill difference is the data difference between t3 and t1 of 3 km / h divided by the time difference between t3 and t1 of 2 to obtain the result of 1.5 km / h, so that the data prediction result at time t2 is 74.5 km / h.

[0066] In some embodiments of the present application, abnormal data includes data that does not conform to actual working conditions; the prediction device can delete the data that does not conform to actual working conditions in the original data sequence to obtain a preprocessed data sequence, thereby reducing the impact of abnormal data on future fuel consumption and future acceleration, and improving the accuracy of future fuel consumption and future acceleration.

[0067] In some embodiments, the prediction device may also delete data that does not conform to the actual working conditions to obtain missing data, and then use S401-S402 to predict and fill in the missing data to obtain a preprocessed data sequence.

[0068] Exemplarily, the speed threshold is 170km / h; the speed at time t1 is 76km / h, the speed at time t2 is 200km / h, and the speed at time t3 is 79km / h. When the prediction device determines that the speed at time t2 is greater than the speed threshold, the data at time t2 can be deleted to obtain a preprocessed data sequence. In some embodiments, the prediction device can also delete the data at time t2 and use a simple average method to perform prediction filling to obtain a speed of 74.5km / h at time t2.

[0069] In some embodiments, if the proportion of data that does not conform to the actual working conditions in the original data sequence is less than or equal to the deletion ratio threshold, the prediction device can delete it directly, which will not cause a decrease in data quality; if the proportion of data that does not conform to the actual working conditions in the original data sequence is greater than the deletion ratio threshold, the prediction device can delete it and continue to predict and fill. The deletion ratio threshold can be set according to actual needs, and the embodiments of the present application are not limited. It should be noted that the larger the deletion ratio threshold, the greater the impact of deleting data on data quality; the smaller the deletion ratio threshold, the higher the complexity of preprocessing.

[0070] In some embodiments, the prediction device can use the data processing library Pandas in Python to perform efficient batch processing on abnormal data and complete data preprocessing to ensure the integrity and accuracy of the data.

[0071] S302 , normalize all preprocessed data in the preprocessed data sequence to obtain a standard data sequence.

[0072] In the embodiment of the present application, after obtaining the preprocessed data sequence, the prediction device can normalize the data in the preprocessed data sequence, eliminate the dimension of all data in the preprocessed data sequence, and obtain a standard data sequence. Here, the normalization method can include any one of the following: Max-Min method, mean normalization method, and Z-score normalization method.

[0073] In some embodiments, the prediction device may use the Z-score normalization method to normalize each type of data, so that the average value of each type of data in the obtained standard data sequence is 0 and the standard deviation is 1.

[0074] S202: Determine a linear correlation characterization value between a standard data sequence of each type of data and a standard data sequence of fuel consumption in the standard data sequence.

[0075] In the embodiment of the present application, after obtaining the standard data sequence, the prediction device can calculate the linear correlation characterization value between the standard data sequence of each data and the standard sequence of fuel consumption. In some embodiments, the linear correlation characterization value can be the Pearson correlation coefficient r, see formula (1).

[0076]

[0077] Where n is the number of multiple moments in the standard data sequence, i is any moment among the n moments, and X i is the fuel consumption data at the i-th moment, is the n X in the standard data sequence i The mean of Y i is any data at the i-th moment, is the n Y in the standard data sequence i The mean of .

[0078] It should be noted that the value range of r is [-1, 1]. The closer its absolute value is to 1, the stronger the correlation between the two variables X and Y. Its positive or negative value represents whether the relationship is positively correlated or negatively correlated. When r = 0, there is no correlation between the variables; when |r|∈(0, 0.3), there is a very weak correlation between the variables; when |r|∈[0.3, 0.6), there is a moderate correlation between the variables; when |r|∈[0.6, 0.8), there is a strong correlation between the variables; when |r|∈[0.8, 1.0), there is an extremely strong correlation between the variables; when |r|=1, the two variables change completely linearly.

[0079] S203. The data corresponding to the linear correlation characterization value greater than the correlation threshold is used as other data, thereby obtaining a data sequence.

[0080] In the embodiment of the present application, the prediction device can determine whether the linear correlation characterization value of the standard data sequence of each data and the standard data sequence of fuel consumption is greater than the correlation threshold. If so, it can be determined that other characteristic data includes this type of data; otherwise, this type of data is not other characteristic data. Among them, the correlation threshold can be set according to actual needs and is not limited in the embodiment of the present application.

[0081] In some embodiments, the linear correlation characterization value is the Pearson coefficient, and the correlation threshold may be 0.3, so that other data with medium to strong correlation with the fuel consumption data can be screened out from the data sequence of the original driving data to obtain a data sequence. In this way, the original data with weak correlation or no correlation can be removed, reducing the calculation amount of the fuel consumption prediction model.

[0082] It can be understood that after obtaining the standard data sequence based on the preprocessed data sequence, calculating the linear correlation characterization value can improve the accuracy of the linear characterization value, thereby improving the accuracy of screening other feature data.

[0083] In some embodiments, the prediction device may obtain a training data sequence and train the fuel consumption prediction model based on the training data sequence. The type of feature data in the training data sequence is the same as the type of feature data in the historical data sequence.

[0084] In the embodiment of the present application, the prediction device can start from each moment in the training data sequence, intercept other data of the prediction step as the real data sequence of the first time period, input the real data sequence of the first time period into the original prediction model, and obtain the predicted fuel consumption and predicted acceleration of the corresponding second future moment. The real data sequences of multiple first time periods can obtain the corresponding predicted fuel consumption and predicted acceleration of multiple second future moments.

[0085] In some embodiments, the prediction device can obtain the predicted fuel consumption and predicted acceleration at each second future moment, compare the predicted fuel consumption at the second future moment with the actual fuel consumption at the second future moment in the training data sequence to obtain a first comparison result, and compare the predicted acceleration at the second future moment with the actual acceleration at the second future moment in the training data sequence to obtain a second comparison result, update the original prediction model based on the first comparison result and the second result, and then use the updated original prediction model to continue to predict the actual data sequence of the next first time period, perform the next update, and obtain the fuel consumption prediction model until the first update stop condition is met.

[0086] In some embodiments, the prediction device can input the training data of multiple first time periods into the original prediction model in sequence, predict the predicted fuel consumption and predicted acceleration of the corresponding multiple second future moments, compare the predicted fuel consumption of each second future moment with the actual fuel consumption of a corresponding second future moment in the training data sequence, and obtain multiple first comparison results, and compare the predicted acceleration of each second future moment with the actual acceleration of the corresponding second future moment in the training data sequence, and obtain multiple second comparison results, update the original prediction model according to the multiple first comparison results and the multiple second results, and then use the updated original prediction model to continue the next round of prediction, input the training data of multiple first time periods into the updated original prediction model in sequence, and predict the predicted fuel consumption and predicted acceleration of the corresponding multiple second future moments, and continue in this way until the second stop updating condition is met to obtain the fuel consumption prediction model.

[0087] In the embodiment of the present application, the evaluation index for the prediction result may include any of the following: mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). In this regard, it can be set according to actual needs, and the embodiment of the present application does not limit it.

[0088] In some embodiments of the present application, the fuel consumption prediction model is a sequence prediction model based on a sparse self-attention mechanism.

[0089] During the training process of the fuel consumption prediction model, after each current training sequence is input, three vectors can be obtained: query vector Q, key vector K, and value vector V. The dot product of the query vector Q and the value vector V is calculated to obtain the similarity score, which is then normalized using the softmax function to obtain the attention weight, which is used to perform weighted summation on the value vector K.

[0090] In the embodiment of the present application, the divergence formula is used to calculate the i-th query vector q i The attention weight is regarded as a probability distribution p(k j |q i ) and q(k j |q i ), as shown in formula (2) and formula (3).

[0091]

[0092]

[0093] Among them, k j represents the j-th key vector, k(q i ,k j ) represents the asymmetric exponential kernel function d is the dimension of the key vector K; L K Represents the length of vector Q.

[0094] Substituting formula (2) and (3) into the KL divergence formula, we can obtain the sparsity degree M(q i ,K), see formula (4).

[0095]

[0096] If the i-th query vector q i M(q i , K) takes a larger value, indicating that its attention weights are more diverse and may contain a more influential part of the dot product. When some value vectors V have a significant impact on the attention weights, the probability distribution will deviate from the uniform distribution.

[0097] Here, we can calculate the dot product of these value vectors V and the query vector Q, and then use softmax to get the attention weight, see formula (5). Specifically, we can choose M(q i , K) are used as the input matrix of softmax, thereby reducing the amount of calculation. The value of u is based on formula (6).

[0098]

[0099] u=c lnL Q Formula (6)

[0100] in, For q i The collection of L Q represents the time step of the Q matrix, and c is a constant sampling factor.

[0101] It should be noted that if the prediction step length is short, the prediction device can obtain the fuel consumption prediction result in time and feed it back to the driver or the driving control center to correct the driving style in time and reduce the actual fuel consumption. If the prediction step length is long, the fuel consumption prediction result determined by the prediction device can reflect the trend of fuel consumption changes and reflect the macro-change characteristics of fuel consumption.

[0102] Exemplarily, the fuel consumption prediction model is initialized based on the experimental environment shown in Table 2. The main parameters of the initialization are shown in Table 3. The regression evaluation indicators of the fuel consumption prediction model are MAE and RMSE.

[0103] Table 2

[0104]

[0105] Table 3

[0106] Parameter name Parameter meaning Parameter Value d_model Model Dimensions 512 d_ff Feedforward Neural Network Layer Dimensions 2048 batch_size The number of training samples in a batch 16 lr Learning Rate 0.0001 rate Dropout layer zero ratio 0.4 e_layers Number of Encoder stacking layers 2 d_layers Decoder stacking layers 1 activation Activation Function gelu patience Early Stopping Mechanism 10

[0107] The parameter value of patience is 10, which means that if the regression evaluation index does not improve for 10 consecutive epochs during the training process, the training is stopped and the model training is completed. The smaller the learning rate, the smaller the impact on the update frequency of the weights.

[0108] The head is set to 8, and the fuel consumption prediction model is trained using data sets 1-3. Data set 1 is the driving data of trucks, data set 2 is the driving data of tractor vehicles, and data set 3 is the driving data of dump trucks and concrete mixer trucks. The training set and test set of each data set are divided into 8:2 ratios. The prediction step size is set to 6, 12, and 24 respectively to train the fuel consumption prediction model. The evaluation results are shown in Table 4.

[0109] Table 4

[0110]

[0111] It can be seen from Table 4 that under the same prediction step, the evaluation results of different data sets are different. Among them, the RMSE of data set 1 and data set 2 is smaller than the RMSE of data set 3. At the same time, the MAE of data set 1 and data set 2 is smaller than the MAE of data set 3. In other words, the overall evaluation results of data set 1 and data set 2 are inferior to the evaluation results of data set 3. Due to the different driving conditions of the three vehicles, the fitting effects of the fuel consumption prediction model are different. However, since the RMSE and MAE of the three data sets are all less than 0.5, it shows that the fuel consumption prediction model can be applied to different data sets and has good generalization ability and adaptability.

[0112] Considering the multi-head attention mechanism in the fuel consumption prediction model, it is different from the convolution kernel in multiple convolution layers of the convolutional neural network. It splits the input feature values ​​into more fragmented small blocks, assigns a separate trainable weight parameter to each small block, and then shares a hidden layer to output the result. The experiment decomposes the head into 2, 4, 8, 16, and 32 different attentions for model training at different prediction step sizes. The evaluation results when the prediction step size is 6 are shown in Table 5, and the evaluation results when the prediction step size is 24 are shown in Table 6.

[0113] Table 5

[0114]

[0115] Table 6

[0116]

[0117] Comparing Table 5 and Table 6, it can be seen that the smaller the head value, the higher the model performance level. When the prediction step is 6, the RMSE and MAE corresponding to the head value of 2 are both minimum values. That is to say, when the prediction step is 6, the head value is set to 2 to make the fuel consumption prediction model optimal and the error is minimized. When the prediction step is 24, the RMSE and MAE corresponding to the head value of 16 are both minimum values. That is to say, when the prediction step is 24, the head value is set to 16 to make the fuel consumption prediction model optimal and the error is minimized. It can be seen that the appropriate head value should be set to be more beneficial to the Informer model.

[0118] Based on the above embodiments, the present application provides a fuel consumption prediction device, such as Figure 5 As shown, the prediction device 50 includes:

[0119] An acquisition module 501 is used to acquire a data sequence of characteristic data of a vehicle during operation that varies with time; the characteristic data includes fuel consumption, acceleration and other data; the linear correlation between the data sequence of the other data and the data sequence of the fuel consumption is greater than a correlation threshold;

[0120] A prediction module 502, configured to predict the predicted fuel consumption and predicted acceleration at a first future moment based on the data sequence by using a fuel consumption prediction model; the first future moment is an adjacent moment after the last moment in the data sequence;

[0121] The determination module 503 is used to determine that the future fuel consumption is a fuel consumption prediction result when the future acceleration meets the actual acceleration condition; the actual acceleration condition includes at least one of the following: the predicted acceleration change rate at the first future moment is within a preset change rate range; the predicted acceleration at the first future moment is greater than an acceleration threshold.

[0122] In some embodiments, the acquisition module 501 is also used to acquire a standard data sequence when the vehicle is traveling; the data in the standard data sequence is dimensionless data; the characteristic data in the standard data sequence includes the fuel consumption, the acceleration and the other characteristic data; determine the linear correlation characterization value between the sequence of each data and the sequence of the fuel consumption in the standard data sequence; use the data corresponding to the linear correlation characterization value greater than the correlation threshold as the other data, thereby obtaining the data sequence.

[0123] In some embodiments, the acquisition module 501 is further used to preprocess abnormal data in the original data sequence when the vehicle is traveling to obtain a preprocessed data sequence; and normalize all data in the preprocessed data sequence to obtain the standard data sequence.

[0124] In some embodiments, the abnormal data includes missing data; the acquisition module 501 is also used to predict the missing data using a prediction filling method to obtain a missing data prediction result; the prediction filling method includes any one of the following: Lagrange interpolation method, simple averaging method, weighted moving average method and exponential smoothing method; the missing data is filled using the data prediction result to obtain the preprocessed data sequence.

[0125] In some embodiments, the abnormal data includes data that does not conform to the actual working conditions; the acquisition module 501 is also used to delete the data that does not conform to the actual working conditions in the original data sequence to obtain the pre-processed data sequence.

[0126] In some embodiments, the prediction device 50 further includes a training module, which is used to

[0127] The real data sequence of the first time period in the training data sequence is input into the original prediction model to predict the predicted fuel consumption and predicted acceleration of the second future moment; the second future moment is a moment after the first time period and adjacent to the last moment of the first time period, based on a first comparison result of the predicted fuel consumption at the second future moment and the actual fuel consumption at the second future moment, and a second comparison result of the predicted acceleration at the second future moment and the actual acceleration at the second future moment, the original prediction model is updated until the first comparison result and the second comparison result meet the stop updating condition, thereby obtaining the fuel consumption prediction model.

[0128] In some embodiments, the fuel consumption prediction model is a sequence prediction model based on a sparse self-attention mechanism.

[0129] Figure 6 Schematic diagram of the structure of the prediction device of the embodiment of the present application. Figure 6 As shown, the prediction device 60 includes a memory 617, a processor 618, and a computer program stored in the memory 617 and executable on the processor 618; wherein the processor 618 is used to execute the driving process analysis method of the operating truck in the aforementioned embodiment when running the computer program.

[0130] It is understood that the analysis device 60 also includes a bus system 619; the various components in the terminal 9 are coupled together through the bus system 619. It is understood that the bus system 619 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 619 also includes a power bus, a control bus and a status signal bus.

[0131] It can be understood that the memory in this embodiment can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disk, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memories described in the embodiments of the present application are intended to include, but are not limited to, these and any other suitable types of memories.

[0132] The method disclosed in the above embodiment of the present application can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor may be a general-purpose processor, a DSP, or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor may be a microprocessor or any conventional processor, etc. In combination with the steps of the method disclosed in the embodiment of the present application, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0133] An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer-readable storage medium is located in a driving process analysis device for an operating truck, the computer program, when executed by a processor, implements the steps in the driving process analysis method for an operating truck in an embodiment of the present application.

[0134] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as: multiple modules or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or modules can be electrical, mechanical or other forms.

[0135] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A fuel consumption prediction method, characterized in that: include: Acquire a data sequence of characteristic data of the vehicle during operation that changes over time; the characteristic data includes fuel consumption, acceleration and other data; The linear correlation between the data sequence of the other data and the data sequence of the fuel consumption is greater than a correlation threshold; Based on the data sequence, predicting predicted fuel consumption and predicted acceleration at a first future moment by using a fuel consumption prediction model; the first future moment is an adjacent moment after the last moment in the data sequence; When the future acceleration meets the actual acceleration condition, the future fuel consumption is determined to be the fuel consumption prediction result; the actual acceleration condition includes at least one of the following: the predicted acceleration change rate at the first future moment is within a preset change rate range; the predicted acceleration at the first future moment is greater than an acceleration threshold.

2. The method according to claim 1, characterized in that The data sequence of obtaining characteristic data of the vehicle during operation that changes with time includes: Acquire a standard data sequence when the vehicle is traveling; the data in the standard data sequence is dimensionless data; the characteristic data in the standard data sequence includes the fuel consumption, the acceleration and the other characteristic data; Determine a linear correlation characterization value between a sequence of each type of data in the standard data sequence and a sequence of the fuel consumption; The data corresponding to the linear correlation characterization value greater than the correlation threshold is used as the other data to obtain the data sequence.

3. The method according to claim 2, characterized in that The step of obtaining a standard data sequence when the vehicle is traveling includes: Preprocessing abnormal data in the original data sequence when the vehicle is running to obtain a preprocessed data sequence; All data in the preprocessed data sequence are normalized to obtain the standard data sequence.

4. The method according to claim 3, characterized in that The abnormal data includes missing data; the abnormal data in the original data sequence of vehicle travel is preprocessed to obtain a preprocessed data sequence, including: The missing data are predicted using a prediction filling method to obtain a missing data prediction result; the prediction filling method includes any one of the following: Lagrange interpolation method, simple average method, weighted moving average method and exponential smoothing method; The missing data is filled using the data prediction results to obtain the preprocessed data sequence.

5. The method according to claim 3, characterized in that: The abnormal data includes data that does not conform to the actual working conditions; the abnormal data in the original data sequence of vehicle driving is preprocessed to obtain a preprocessed data sequence, including: The data that does not conform to the actual working conditions in the original data sequence is deleted to obtain the preprocessed data sequence.

6. The method according to claim 1, characterized in that Before predicting the predicted fuel consumption and predicted acceleration at the first future moment by using the fuel consumption prediction model based on the data sequence, the method further includes: Input the real data sequence of the first time period in the training data sequence into the original prediction model to predict the predicted fuel consumption and predicted acceleration at a second future moment; the second future moment is a moment after the first time period and adjacent to the last moment of the first time period; Based on a first comparison result between the predicted fuel consumption at the second future moment and the actual fuel consumption at the second future moment, and a second comparison result between the predicted acceleration at the second future moment and the actual acceleration at the second future moment, the original prediction model is updated until the first comparison result and the second comparison result meet the stop updating condition, thereby obtaining the fuel consumption prediction model.

7. The method according to any one of claims 1 to 6, characterized in that: The fuel consumption prediction model is a sequence prediction model based on a sparse self-attention mechanism.

8. A fuel consumption prediction device, characterized in that: include: An acquisition module, used to acquire a data sequence of characteristic data of the vehicle during operation that varies with time; the characteristic data includes fuel consumption, acceleration and other data; The linear correlation between the data sequence of the other data and the data sequence of the fuel consumption is greater than a correlation threshold; A prediction module, configured to predict the predicted fuel consumption and predicted acceleration at a first future moment based on the data sequence by using a fuel consumption prediction model; the first future moment is an adjacent moment after the last moment in the data sequence; A determination module is used to determine that the future fuel consumption is a fuel consumption prediction result when the future acceleration meets an actual acceleration condition; the actual acceleration condition includes at least one of the following: the predicted acceleration change rate at the first future moment is within a preset change rate range; the predicted acceleration at the first future moment is greater than an acceleration threshold.

9. A computer storage medium, characterized in that One or more computer programs are stored, and the one or more computer programs can be executed by one or more processors to implement the steps of any one of claims 1-6.

Images