CN115293055B - Training method and device for oil consumption prediction model of mining vehicle and electronic equipment - Google Patents
Training method and device for oil consumption prediction model of mining vehicle and electronic equipment Download PDFInfo
- Publication number
- CN115293055B CN115293055B CN202211220194.7A CN202211220194A CN115293055B CN 115293055 B CN115293055 B CN 115293055B CN 202211220194 A CN202211220194 A CN 202211220194A CN 115293055 B CN115293055 B CN 115293055B
- Authority
- CN
- China
- Prior art keywords
- fuel consumption
- target period
- time sequence
- vehicle
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
Abstract
The disclosure relates to a training method and device for a fuel consumption prediction model of a mining area vehicle and electronic equipment, and relates to the technical field of engineering machinery. The training method of the oil consumption prediction model of the mining vehicle comprises the following steps: acquiring historical time sequence data, task data of a target period, a true value of historical time sequence oil consumption and a true value of oil consumption of the target period; according to the historical time sequence data, a predicted value of the historical time sequence fuel consumption is generated by utilizing a fuel consumption prediction model of the mining area vehicle; according to the historical time sequence data and the task data of the target period, a predicted value of the oil consumption of the target period is generated by utilizing an oil consumption prediction model of the mining area vehicle; and jointly training a fuel consumption prediction model according to the actual value of the historical time sequence fuel consumption, the predicted value of the historical time sequence fuel consumption, the actual value of the fuel consumption in the target period and the predicted value of the fuel consumption in the target period. According to the method and the device, accuracy of oil consumption prediction of the mining vehicles is improved.
Description
Technical Field
The disclosure relates to the technical field of engineering machinery, in particular to a training method and device for a fuel consumption prediction model of a mining area vehicle.
Background
In mining operations in mining areas, fuel consumption is closely related to the cost and profitability of the mining area. The engineering amount of the mine is large, so that the load and the oil consumption of the vehicle are large, and the oil consumption is large. If the oil reserve is not timely replenished, the transportation task of the vehicle can be influenced, and the yield of a mining area can be influenced.
Operation data of mining vehicles, such as distance of transportation, actual load of vehicles, may affect fuel consumption. In the related art, the fuel consumption of a mining vehicle is predicted by manually labeling operation data of the mining vehicle to generate a label.
Disclosure of Invention
According to a first aspect of the present disclosure, there is provided a training method of a fuel consumption prediction model of a mining vehicle, including:
acquiring historical time sequence data, task data of a target period, a true value of historical time sequence oil consumption and a true value of oil consumption of the target period, wherein the historical time sequence data comprises a sequence of historical data of a plurality of continuous periods before the target period, and the true value of the historical time sequence oil consumption comprises a sequence of true values of oil consumption of a vehicle in a plurality of continuous periods before the target period;
according to the historical time sequence data, a predicted value of the historical time sequence fuel consumption is generated by utilizing a fuel consumption prediction model of the mining area vehicle;
according to the historical time sequence data and the task data of the target period, a predicted value of the oil consumption of the target period is generated by utilizing an oil consumption prediction model of the mining area vehicle;
and jointly training a fuel consumption prediction model according to the actual value of the historical time sequence fuel consumption, the predicted value of the historical time sequence fuel consumption, the actual value of the fuel consumption in the target period and the predicted value of the fuel consumption in the target period.
In some embodiments, the jointly training the fuel consumption prediction model according to the actual value of the historical time-series fuel consumption, the predicted value of the historical time-series fuel consumption, the actual value of the fuel consumption in the target period, and the predicted value of the fuel consumption in the target period includes:
calculating a first loss function according to the actual value of the historical time sequence oil consumption and the predicted value of the historical time sequence oil consumption;
calculating a second loss function according to the actual value of the oil consumption in the target period and the predicted value of the oil consumption in the target period;
and jointly training a fuel consumption prediction model according to the first loss function and the second loss function.
In some embodiments, the jointly training the fuel consumption prediction model according to the first loss function and the second loss function includes:
and jointly training the fuel consumption prediction model according to the weighted results of the first loss function and the second loss function.
In some embodiments, the generating the predicted value of the historical time series fuel consumption according to the historical time series data by using a fuel consumption prediction model of the mining vehicle includes:
generating time sequence characteristics by utilizing a multi-head self-attention network of a fuel consumption prediction model of the mining area vehicle according to the historical time sequence data;
and generating a predicted value of the historical time sequence oil consumption according to the time sequence characteristics.
In some embodiments, the generating a timing feature from historical timing data using a multi-headed self-attention network of a fuel consumption prediction model of a mining vehicle includes:
masking the missing values in the historical time sequence data;
generating a position code for the history time sequence data after mask processing;
and generating time sequence characteristics by utilizing a multi-head self-attention network, a feedforward neural network and a residual error network of the oil consumption prediction model of the mining vehicle according to the historical time sequence data and the position codes.
In some embodiments, the task data of the target period includes a planned load and a driving mileage of the vehicle in the target period, and the generating the predicted value of the fuel consumption of the target period according to the historical time sequence data and the task data of the target period by using the fuel consumption prediction model of the mining vehicle includes:
generating task characteristics according to the planned load and the driving mileage of the vehicle in the target period;
and generating a predicted value of the fuel consumption of the target period by using a fuel consumption prediction model of the mining vehicle according to the time sequence characteristics and the task characteristics.
In some embodiments, the generating the mission feature according to the planned load and the driving range of the vehicle in the target period includes:
acquiring the empty load and the full load of the vehicle;
generating a plurality of load ranges according to the empty load and the full load of the vehicle:
and generating task characteristics corresponding to the load ranges according to the driving mileage of the vehicle with the planned load in each load range.
In some embodiments, the generating a plurality of load ranges from the empty load and the full load of the vehicle includes:
the plurality of load ranges are non-linearly divided based on an average value of the empty loads of the plurality of vehicles at the specified time and a maximum value of the full loads of the plurality of vehicles.
In some embodiments, the generating task features corresponding to the load ranges according to the driving mileage of the vehicle with the planned load in each load range includes:
for each load range, a mission feature corresponding to the load range is generated from a sum of mileage of all vehicles of the plurality of vehicles for which the planned load is within the load range, an average value of planned loads of all vehicles of the plurality of vehicles for which the planned load is within the load range, and an average value of empty loads of the plurality of vehicles at a specified time.
In some embodiments, the generating the predicted value of the fuel consumption of the target period according to the time sequence characteristic and the task characteristic includes:
processing task characteristics by using a residual error network of a fuel consumption prediction model of the mining area vehicle;
and generating a predicted value of the oil consumption of the target period according to the sum of the task characteristics and the time sequence characteristics processed by the residual error network.
In some embodiments, the actual value of historical time series fuel consumption comprises a sequence of actual values of total fuel consumption of a plurality of vehicles over a plurality of consecutive time periods prior to the target time period;
the actual value of fuel consumption for the target period includes an actual value of total fuel consumption for a plurality of vehicles within the target period.
In some embodiments, the historical data includes at least one of historical weather data for the mine, historical empty mileage and historical full mileage for the vehicle.
In some embodiments, the acquiring task data for the target period includes:
obtaining a planned production of different types of ore within a target period;
the planned load of the vehicle in the target period is determined based on the planned production of different types of ore in the target period.
According to a second aspect of the present disclosure, there is provided a method of predicting fuel consumption of a mining vehicle, comprising
Acquiring historical time sequence data and task data of a target period, wherein the historical time sequence data comprises a sequence of historical data of a plurality of continuous periods before the target period;
and predicting the oil consumption of the target period by using an oil consumption prediction model of the mining area vehicle according to the historical time sequence data and the task data of the target period.
In some embodiments, the fuel consumption prediction model of the mining vehicle is trained according to the training method of the fuel consumption prediction model of any embodiment of the disclosure.
According to a third aspect of the present disclosure, a training apparatus for a fuel consumption prediction model of a mining vehicle includes:
the system comprises an acquisition module, a control module and a control module, wherein the acquisition module is configured to acquire historical time sequence data, task data of a target period, a true value of historical time sequence oil consumption and a true value of oil consumption of the target period, wherein the historical time sequence data comprises a sequence of historical data of a plurality of continuous periods before the target period, the true value of the historical time sequence oil consumption comprises a sequence of true values of oil consumption of the vehicle in a plurality of continuous periods before the target period, and the task data of the target period comprises a planned load and a driving mileage of the vehicle in the target period;
the historical oil consumption generation module is configured to generate a predicted value of historical time sequence oil consumption by utilizing an oil consumption prediction model of the mining vehicle according to the historical time sequence data;
the target oil consumption generation module is configured to generate a predicted value of oil consumption of a target period by utilizing an oil consumption prediction model of a mining area vehicle according to the historical time sequence data and the task data of the target period;
the combined training module is configured to combine and train the oil consumption prediction model according to the actual value of the oil consumption of the historical time sequence, the predicted value of the oil consumption of the historical time sequence, the actual value of the oil consumption of the target period and the predicted value of the oil consumption of the target period.
According to a fourth aspect of the present disclosure, a fuel consumption prediction apparatus for a mining vehicle includes
An acquisition module configured to acquire historical time series data and task data of a target period, wherein the historical time series data comprises a sequence of historical data of a plurality of continuous periods before the target period, and the task data of the target period comprises a planned load and a driving mileage of the vehicle in the target period;
the prediction module is configured to predict the fuel consumption of the target period by using a fuel consumption prediction model of the mining vehicle according to the historical time sequence data and the task data of the target period.
According to a fifth aspect of the present disclosure, an electronic device, comprises:
a memory; and
a processor coupled to the memory, the processor configured to perform a training method of a fuel consumption prediction model of a mining vehicle according to any embodiment of the present disclosure, or to perform a prediction method of fuel consumption of a mining vehicle according to any embodiment of the present disclosure, based on instructions stored in the memory.
According to a sixth aspect of the present disclosure, there is provided a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement a method of training a fuel consumption prediction model of a mining vehicle according to any embodiment of the present disclosure, or a method of predicting fuel consumption of a mining vehicle according to any embodiment of the present disclosure.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the description, serve to explain the principles of the disclosure.
The disclosure may be more clearly understood from the following detailed description taken in conjunction with the accompanying drawings in which:
FIG. 1 illustrates a flow chart of a training method of a fuel consumption prediction model of a mining vehicle according to some embodiments of the present disclosure;
FIG. 2 illustrates a schematic diagram of planned production and haul distances, according to some embodiments of the present disclosure;
FIG. 3 illustrates a schematic diagram of a fuel consumption prediction model of a mining vehicle according to some embodiments of the present disclosure;
FIG. 4 illustrates a flow chart of a method of predicting fuel consumption of a mining vehicle according to some embodiments of the present disclosure;
FIG. 5 illustrates a block diagram of a training apparatus for a fuel consumption prediction model of a mining vehicle, according to some embodiments of the present disclosure;
FIG. 6 illustrates a block diagram of a predictive device for fuel consumption of a mining vehicle according to some embodiments of the present disclosure;
FIG. 7 illustrates a block diagram of an electronic device according to further embodiments of the present disclosure;
FIG. 8 illustrates a block diagram of a computer system for implementing some embodiments of the present disclosure.
Detailed Description
Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that: the relative arrangement of the components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present disclosure unless it is specifically stated otherwise.
Meanwhile, it should be understood that the sizes of the respective parts shown in the drawings are not drawn in actual scale for convenience of description.
The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application, or uses.
Techniques, methods, and apparatus known to one of ordinary skill in the relevant art may not be discussed in detail, but are intended to be part of the specification where appropriate.
In all examples shown and discussed herein, any specific values should be construed as merely illustrative, and not a limitation. Thus, other examples of the exemplary embodiments may have different values.
It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further discussion thereof is necessary in subsequent figures.
In the related art, only the influence of the current day mission plan on the fuel consumption is considered, and the influence of time sequence data of a plurality of continuous days in the past on the fuel consumption is ignored, so that the prediction accuracy is not high. In addition, in the related art, the labels of the training data need to be manually marked, which results in high training cost.
The invention provides a training method and device for a fuel consumption prediction model of a mining vehicle and electronic equipment, which improve the efficiency and accuracy of fuel consumption prediction of the mining vehicle and reduce the cost.
FIG. 1 illustrates a flow chart of a training method of a fuel consumption prediction model of a mining vehicle according to some embodiments of the present disclosure.
As shown in fig. 1, the training method of the fuel consumption prediction model of the mining vehicle includes steps S11 to S14.
In step S11, historical time series data, task data of a target period, a true value of historical time series fuel consumption, and a true value of fuel consumption of the target period are obtained, wherein the historical time series data includes a sequence of historical data of a plurality of continuous periods before the target period, and the true value of historical time series fuel consumption includes a sequence of true values of fuel consumption of the vehicle of a plurality of continuous periods before the target period.
For example, in units of days, the target period is 4 months 1 day, and then the plurality of consecutive periods before the target period may be 3 months 1 day to 3 months 31 days. The sequence of the history data of the plurality of consecutive periods preceding the target period is:
history data of day 3 month 1, history data of day 3 month 2, history data of day … … month 31.
Similarly, the true values of fuel consumption for a number of consecutive periods preceding the target period are:
oil consumption of 3 months 1 day, oil consumption of 3 months 2 days, and oil consumption of … … months 31 days.
In some embodiments, the actual value of the historical time-series fuel consumption comprises a sequence of actual values of the total fuel consumption of the plurality of vehicles over a plurality of consecutive time periods prior to the target time period; the actual value of fuel consumption for the target period includes an actual value of total fuel consumption for a plurality of vehicles within the target period.
For example, the total oil consumption of all vehicles in the mining area is integrally considered, so that the overall oil consumption prediction of all vehicles in the mining area is realized, guidance can be provided for the overall oil quantity reserve supplement and mining task optimization in the mining area, the efficiency is improved, and the cost is saved.
In some embodiments, the historical data includes at least one of historical weather data for the mine, historical empty mileage and historical full mileage for the vehicle.
For example, weather data, the empty mileage and the full mileage of the vehicle are all time series data, that is, the weather data today and the past today, the empty mileage or the full mileage of the vehicle, affect the weather data, the empty mileage and the full mileage of the vehicle, respectively.
Further effects on fuel consumption can be caused under different extreme climates and construction conditions. The influence of weather time sequence data is considered, so that the oil consumption prediction of the strip mine in the complex climate environment is realized.
In some embodiments, the mission data of the target period includes a planned load and a driving range of the vehicle in the target period, and the acquiring the mission data of the target period includes: obtaining a planned production of different types of ore within a target period; the planned load of the vehicle in the target period is determined based on the planned production of different types of ore in the target period.
For example, the full driving mileage and the no-load driving mileage are uploaded in real time by the unmanned vehicle, and the daily full driving mileage and the no-load driving mileage are counted at regular time by the background big data management center. The ore types are different, the total weight of the vehicle is different when the vehicle is fully loaded, the vehicle load under different ore types is the actual load of the vehicle in one-time transportation, the vehicle has no load, partial loading, full loading and other conditions, and the load under the full loading is also different to a certain extent. Thus, the planned production of different types of ore may be established first, from which the planned load of the vehicle in the target period is generated.
In some embodiments, the planned production is calculated from a mine allocation plan for the mine, wherein the allocation plan is obtained from a data source.
For example, a mine allocation plan is obtained from a dispatch center module.
FIG. 2 illustrates a schematic diagram of planned production and wheelbase according to some embodiments of the present disclosure.
As shown in fig. 2, the production of crushing stations 1 and 2 is related to the production of quarries 1, 2 and the distance of the transportation from quarry 1 to crushing station 1 needs to be calculated. Similarly, the output of the dump is related to the output of the collection sites 1, 2.
In step S12, a predicted value of the historical time series fuel consumption is generated from the historical time series data by using the vehicle fuel consumption prediction model.
For example, the vehicle fuel consumption prediction model includes a neural network, and the time series of the historical fuel consumption corresponding to the model can be predicted from the time series of the historical data by using the neural network.
In some embodiments, generating a predicted value of historical time series fuel consumption using a vehicle fuel consumption prediction model from historical time series data includes: generating a time sequence characteristic by utilizing a multi-head self-attention network of a vehicle fuel consumption prediction model according to the historical time sequence data; and generating a predicted value of the historical time sequence oil consumption according to the time sequence characteristics.
For example, using a multi-head self-attention network, the characteristics of the sequence of weather data, empty mileage and full mileage of the vehicle over the past thirty days are extracted, and the sequence of fuel consumption over the past thirty days is calculated from the extracted sequence characteristics.
The method and the device perform feature extraction on the multi-source time sequence data such as the vehicle condition, weather and the like through a self-attention mechanism, and enhance the capturing capability of long-sequence features.
In some embodiments, generating a timing feature from historical timing data using a multi-headed self-attention network of a vehicle fuel consumption prediction model includes: masking the missing values in the historical time sequence data; generating a position code for the history time sequence data after mask processing; and generating time sequence characteristics by utilizing a multi-head self-attention network, a feedforward neural network and a residual error network of the vehicle fuel consumption prediction model according to the historical time sequence data and the position codes.
For example, the acquired data is time-sequentially segmented (weather is divided into daily weather) according to time continuity, and then feature completion is performed. Specifically, time series data is extracted in steps. Because mining works in severe weather such as partial holidays and some storm snow, etc., the mining can be stopped for a few days, therefore, the missing value in the continuous time sequence characteristic is completed first, and the characteristic data of date interruption is set to zero. Then, a timing feature mask is generated based on the timing data interrupt condition.
The self-attention transformation process is as follows, which combines residual connection to generate time sequence characteristics by using a multi-head self-attention module and a feedforward neural network. Wherein, query vector q, key vector k and value vector v, attention vector is attention:
the transformation process of the multi-headed self-attention combined residual is as follows, wherein,
representing model parameters, LN representing layer standardization, concat representing splicing, inputting x in Output is x multi ：
The transformation process of the feedforward neural network combined with the residual module is as follows, wherein W 1 ，W 2 ，b 1 ，b 2 For model parameters, relu represents the activation function, and the corresponding output is x out ：
x out ＝LN((W 2 ·relu(W 1 ·x multi +b 1 )+b 2 )+x multi )
In step S13, a predicted value of fuel consumption in the target period is generated using the vehicle fuel consumption prediction model based on the history time series data and the task data in the target period.
In some embodiments, generating a predicted value of fuel consumption for the target time period using a vehicle fuel consumption prediction model based on the historical time series data and the task data for the target time period includes: generating task characteristics according to the planned load and the driving mileage of the vehicle in the target period; and generating a predicted value of the oil consumption of the target period according to the time sequence characteristics and the task characteristics.
For example, task features are extracted from task data, and the fuel consumption of tomorrow is predicted together by considering the task features and the time sequence features.
In some embodiments, the planned load and range of the vehicle over the target period, generating the mission feature, includes: acquiring the empty load and the full load of the vehicle; generating a plurality of load ranges according to the empty load and the full load of the vehicle: and generating task characteristics corresponding to the load ranges according to the driving mileage of the vehicle with the planned load in each load range.
For example, the load zone is a weight zone between the empty weight of the vehicle and the maximum load of the vehicle obtained from the history data. The ore types of different quarries have different ratios, the average load of the vehicles corresponding to the ore types is obtained, and the load of the vehicles under different ore types is obtained, wherein the proportion of the same ore, soil and ore is different, so that the load of the vehicles is also different under the condition of carrying the same volume of the ores of different types. The load can affect the fuel consumption, so the load conditions of different vehicles can be evaluated by dividing the load section, thereby generating task characteristics for predicting the fuel consumption.
In some embodiments, generating a plurality of load ranges from an empty load and a full load of the vehicle includes: the plurality of load ranges are non-linearly divided based on an average value of the empty loads of the plurality of vehicles at the specified time and a maximum value of the full loads of the plurality of vehicles.
For example, an average load of the empty loads of all vehicles over a specified time (e.g., the past year) empty Maximum load of a single vehicle among all vehicles up to a specified time (e.g., the past year) max Within the range of (1), the loading zone is non-linearly divided into n cells, the width w of the ith zone i The method comprises the following steps:
wherein p is i Is an intermediate variable.
For example, the fuel consumption is counted for one day, and the vehicle is mainly in the idle state and the loading state in one day, and the larger the loading amount is, the larger the fuel consumption is. The method and the device realize finer division of large loading capacity by non-linear division of the loading section. The load is divided into n cells with different widths, the larger the load is, the smaller the interval is, the finer the division is, the discretization of data is realized, and the difference characteristics among different loading capacities under the loading condition can be more accurately obtained, so that the prediction accuracy is improved.
In some embodiments, generating a mission feature corresponding to a load range from a range of vehicles for which the load is planned, includes: for each load range, a mission feature corresponding to the load range is generated from a sum of mileage of all vehicles of the plurality of vehicles for which the planned load is within the load range, an average value of planned loads of all vehicles of the plurality of vehicles for which the planned load is within the load range, and an average value of empty loads of the plurality of vehicles at a specified time.
For example, for each load zone w i Selecting a planned load within a load interval w within a target period i All vehicles within. Find the load interval w i Average of planned loads for all vehicles within: average load i . And calculate the load interval w i Sum of the mileage of all vehicles in: mileage dis i Load section w i And mileage dis i Composing feature pairs, constructing combined features f from feature pairs i Obtaining n task features f corresponding to n load zones i I.e.
Wherein alpha is i And beta i For the parameters of the fuel consumption prediction model of the vehicle, the model is thatUpdating is carried out in the training process.
According to the method, the load is divided into n cells with different widths, the larger the load is, the smaller the interval is, the finer the division is, the discretization of data is realized, n statistical characteristics are obtained, and the method is used for inputting data of a vehicle fuel consumption prediction model.
In some embodiments, generating a predicted value of fuel consumption for a target period from the timing characteristics and the mission characteristics includes: processing task characteristics by using a residual error network of a vehicle fuel consumption prediction model; and generating a predicted value of the oil consumption of the target period according to the sum of the task characteristics and the time sequence characteristics processed by the residual error network.
For example, further feature extraction is performed on the task features by using a residual network, then the feature extraction result is added with the time sequence features, and the added result passes through at least one full connection layer, so that the predicted value of the oil consumption is obtained.
In step S14, the vehicle fuel consumption prediction model is jointly trained according to the actual value of the historical time-series fuel consumption, the predicted value of the historical time-series fuel consumption, the actual value of the fuel consumption in the target period, and the predicted value of the fuel consumption in the target period.
For example, the model is jointly trained using the predictions generated from the task data and the time series data from different data sources.
In some embodiments, jointly training the vehicle fuel consumption prediction model based on the actual value of the historical time-series fuel consumption, the predicted value of the historical time-series fuel consumption, the actual value of the fuel consumption for the target period, and the predicted value of the fuel consumption for the target period includes: calculating a first loss function according to the actual value of the historical time sequence oil consumption and the predicted value of the historical time sequence oil consumption; calculating a second loss function according to the actual value of the oil consumption in the target period and the predicted value of the oil consumption in the target period; and jointly training a vehicle fuel consumption prediction model according to the first loss function and the second loss function.
For example, the first and second loss functions are calculated based on the historical fuel consumption sequence predicted from the historical time series data, and the fuel consumption of the target period predicted from the task data of the target period, respectively, and then the vehicle fuel consumption prediction model is jointly trained based on the first loss function and the second loss function, i.e., the parameters of the model are updated. In the process of updating the model parameters, each time the parameters are updated, the parameters are simultaneously influenced by a first loss function and a second loss function.
That is, in each round of (epoch) training, two types of labels, that is, the actual value of the historical fuel consumption sequence and the actual value of the fuel consumption of the target period, may be used simultaneously, and the parameters of the model may be corrected, instead of using only one label, that is, the actual value of the fuel consumption of the target period. Although the predicted value of the target period is the final required result, the predicted value and the true value of the historical oil consumption sequence are additionally introduced, so that more references can be provided for model parameter updating, the training efficiency is improved, and the accuracy of model prediction is improved.
In some embodiments, training the vehicle fuel consumption prediction model according to the first and second loss functions includes: and training a vehicle fuel consumption prediction model according to the weighted results of the first loss function and the second loss function.
For example, a corresponding weight is set for each loss function, and the parameters of the model are updated according to the weighted sum of the loss functions.
According to the method and the device, different types of data are respectively processed, the time sequence of the historical oil consumption is predicted according to the sequence of the historical data, and the oil consumption of a target period is predicted according to comprehensive consideration of the historical time sequence data and the task data. And then, according to the predicted historical oil consumption and the predicted oil consumption of the target period, the model is trained in a combined mode. When the method and the device are used for predicting the oil consumption of the target period, the influence of time sequence data on the oil consumption is additionally considered, the capturing capacity of time sequence features is enhanced, and the accuracy of prediction is improved. In addition, when the model is trained, the real value of the historical time sequence oil consumption and the predicted value of the historical time sequence oil consumption are additionally used for updating parameters of the model, so that training efficiency of the model and prediction accuracy of the model are improved.
In addition, the tags (the true value of the historical time sequence oil consumption and the true value of the oil consumption in the target period) of the training data used in the training model are all true oil consumption values, manual labeling is not needed, and cost is reduced.
FIG. 3 illustrates a schematic diagram of a fuel consumption prediction model of a mining vehicle according to some embodiments of the present disclosure.
As shown in fig. 3, for the multidimensional time sequence features composed of weather, full load, no-load mileage and the like, position encoding is performed first, then the time sequence features are generated by utilizing a multi-head self-attention module, a residual error module and a feedforward neural network, and the time sequence features are input into a full-connection layer to obtain a sequence of the historical fuel consumption 1.
For task data, firstly extracting task features, then constructing a deep network through multi-layer residual connection, acquiring deep task features, and further extracting the task features. The extraction result is spliced (e.g., added) with the timing characteristics, and then the spliced result is input to another full-connection layer, so as to generate the fuel consumption 2 of the period to be predicted.
Based on fuel consumption 1 and 2, parameters of the model (including parameter α of task data characteristics i And beta i ) And updating. And circularly acquiring daily uploaded time sequence data and task data, predicting oil consumption and updating parameters according to a prediction result. Through model cyclic training, the model can be continuously optimized.
Fig. 4 illustrates a flow chart of a method of predicting fuel consumption of a mining vehicle according to some embodiments of the present disclosure.
As shown in fig. 4, the method for predicting fuel consumption of a mining vehicle includes steps S21 to S22.
In step S21, history time series data and task data of a target period are acquired, wherein the history time series data includes a sequence of history data of a plurality of continuous periods before the target period;
in step S22, fuel consumption in the target period is predicted using a fuel consumption prediction model of the mining vehicle based on the historical time series data and the task data in the target period.
For example, when it is necessary to predict the fuel consumption in the open day, history data such as weather of a plurality of days today and before the present day, and planned task data in the open day are acquired, and prediction is performed by using a trained fuel consumption prediction model based on the acquired data. And taking the oil consumption of the target period generated by the model as a prediction result.
In some embodiments, a fuel consumption prediction model of a mining vehicle is trained according to the method of any of the embodiments of the present disclosure.
In some embodiments, the mission data of the target period includes a planned load and a driving range of the vehicle in the target period, and generating a predicted value of fuel consumption of the target period according to the historical time sequence data and the mission data of the target period by using a fuel consumption prediction model of the mining vehicle includes:
generating time sequence characteristics by utilizing a multi-head self-attention network of a fuel consumption prediction model of the mining area vehicle according to the historical time sequence data;
generating task characteristics according to the planned load and the driving mileage of the vehicle in the target period;
and generating a predicted value of the fuel consumption of the target period by using a fuel consumption prediction model of the mining vehicle according to the time sequence characteristics and the task characteristics.
For example, when predicting using a model, a time series feature and a task feature are generated from time series data and task data, respectively, and a predicted value of fuel consumption in a target period is generated from the time series feature and the task feature. Different from the training method, the prediction value of the historical time sequence oil consumption can not be generated during prediction, so that the prediction efficiency of the model is improved.
Fig. 5 illustrates a block diagram of a training apparatus for a fuel consumption prediction model of a mining vehicle, according to some embodiments of the present disclosure.
As shown in fig. 5, the training device 5 for the fuel consumption prediction model of the mining vehicle includes an acquisition module 51, a historical fuel consumption generation module 52, a target fuel consumption generation module 53, and a joint training module 54.
The obtaining module 51 is configured to obtain historical time series data, task data of a target period, a true value of historical time series fuel consumption, and a true value of fuel consumption of the target period, where the historical time series data includes a sequence of historical data of a plurality of consecutive periods before the target period, the true value of historical time series fuel consumption includes a sequence of true values of fuel consumption of the vehicle of a plurality of consecutive periods before the target period, and the task data of the target period includes a planned load and a driving range of the vehicle in the target period, for example, S11 as shown in fig. 1 is performed.
The historical fuel consumption generation module 52 is configured to generate a predicted value of historical time-series fuel consumption using a fuel consumption prediction model of the mining vehicle according to the historical time-series data, for example, to perform S12 shown in fig. 1.
The target fuel consumption generation module 53 is configured to generate a predicted value of fuel consumption for the target period using a fuel consumption prediction model of the mining vehicle according to the historical time series data and the task data for the target period, for example, to perform S13 shown in fig. 1.
The joint training module 54 is configured to joint train the fuel consumption prediction model according to the actual value of the historical time series fuel consumption, the predicted value of the historical time series fuel consumption, the actual value of the fuel consumption of the target period, and the predicted value of the fuel consumption of the target period, for example, to perform S14 as shown in fig. 1.
Fig. 6 illustrates a block diagram of a fuel consumption prediction apparatus for a mining vehicle according to some embodiments of the present disclosure.
As shown in fig. 6, the prediction apparatus 6 for fuel consumption of a mining vehicle includes an acquisition module 61 and a prediction module 62.
An acquisition module 61 configured to acquire historical time series data and task data of a target period, wherein the historical time series data includes a sequence of historical data of a plurality of consecutive periods preceding the target period, the task data of the target period including a planned load and a driving range of the vehicle within the target period, for example, performing step S21 as shown in fig. 2.
The prediction module 62 is configured to predict the fuel consumption of the target period using a fuel consumption prediction model of the mining vehicle according to the historical time series data and the task data of the target period, for example, to perform step S22 shown in fig. 2.
Fig. 7 illustrates a block diagram of an electronic device according to further embodiments of the present disclosure.
As shown in fig. 7, the electronic device 7 includes a memory 71; and a processor 72 coupled to the memory 71, the memory 71 for storing instructions for performing a training method of a fuel consumption prediction model of a mining vehicle according to any of the embodiments of the present disclosure, or for performing a prediction method of fuel consumption of a mining vehicle according to any of the embodiments of the present disclosure. The processor 72 is configured to execute a training method of a fuel consumption prediction model of the mining vehicle or a prediction method of fuel consumption of the mining vehicle of any of the embodiments of the present disclosure based on instructions stored in the memory 71.
FIG. 8 illustrates a block diagram of a computer system for implementing some embodiments of the present disclosure.
As shown in FIG. 8, computer system 80 may be in the form of a general purpose computing device. Computer system 80 includes a memory 810, a processor 820, and a bus 800 that connects the various system components.
It should be noted that: the relative arrangement of the components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present disclosure unless it is specifically stated otherwise. These instructions to perform the training method of the fuel consumption prediction model of the mining vehicle or the prediction method of fuel consumption of the mining vehicle in any of the embodiments of the present disclosure are also stored in a readable storage medium.
Through the training method and device for the oil consumption prediction model of the mining vehicle and the electronic equipment, the oil consumption prediction efficiency and accuracy of the mining vehicle can be improved.
Thus far, a training method of a fuel consumption prediction model of a mining vehicle, a prediction method and apparatus of fuel consumption of a mining vehicle, an electronic device, and a computer-readable storage medium according to the present disclosure have been described in detail. Techniques, methods, and apparatus known to one of ordinary skill in the relevant art may not be discussed in detail. The techniques, methods and apparatus should be considered part of the description where appropriate.
Claims (15)
1. A method of training a fuel consumption predictive model of a mining vehicle, comprising:
acquiring historical time sequence data, task data of a target period, a true value of historical time sequence oil consumption and a true value of oil consumption of the target period, wherein the historical time sequence data comprises a sequence of historical data of a plurality of continuous periods before the target period, the true value of the historical time sequence oil consumption comprises a sequence of the true value of oil consumption of the vehicle in the plurality of continuous periods before the target period, and the task data of the target period comprises a planned load and a driving mileage of the vehicle in the target period;
according to the historical time sequence data, a predicted value of the historical time sequence fuel consumption is generated by utilizing a fuel consumption prediction model of the mining area vehicle, and the method comprises the following steps of
Generating time sequence characteristics by utilizing a multi-head self-attention network of a fuel consumption prediction model of the mining area vehicle according to the historical time sequence data;
generating a predicted value of historical time sequence oil consumption according to the time sequence characteristics;
according to the historical time sequence data and the task data of the target period, generating a predicted value of the oil consumption of the target period by utilizing an oil consumption prediction model of the mining area vehicle, wherein the predicted value comprises
Acquiring the empty load and the full load of the vehicle;
according to the empty load and the full load of the vehicle, generating a plurality of load ranges in a nonlinear manner, wherein the larger the load corresponding to the load range is, the smaller the interval width is;
generating task features according to the planned load and the driving mileage of the vehicle in a target period, wherein the task features comprise generating task features corresponding to the load ranges according to the driving mileage of the vehicle with the planned load in each load range;
according to the time sequence characteristics and the task characteristics, a predicted value of the oil consumption of the target period is generated by utilizing an oil consumption prediction model of the mining area vehicle;
and jointly training a fuel consumption prediction model according to the actual value of the historical time sequence fuel consumption, the predicted value of the historical time sequence fuel consumption, the actual value of the fuel consumption in the target period and the predicted value of the fuel consumption in the target period.
2. The training method according to claim 1, wherein the joint training of the fuel consumption prediction model based on the actual value of the historical time-series fuel consumption, the predicted value of the historical time-series fuel consumption, the actual value of the fuel consumption for the target period, and the predicted value of the fuel consumption for the target period includes:
calculating a first loss function according to the actual value of the historical time sequence oil consumption and the predicted value of the historical time sequence oil consumption;
calculating a second loss function according to the actual value of the oil consumption in the target period and the predicted value of the oil consumption in the target period;
and jointly training a fuel consumption prediction model according to the first loss function and the second loss function.
3. The training method of claim 2, wherein jointly training the fuel consumption prediction model based on the first loss function and the second loss function comprises:
and jointly training the fuel consumption prediction model according to the weighted results of the first loss function and the second loss function.
4. The training method of claim 1, wherein generating a timing feature from historical timing data using a multi-headed self-attention network of a fuel consumption prediction model of a mining vehicle comprises:
masking the missing values in the historical time sequence data;
generating a position code for the history time sequence data after mask processing;
and generating time sequence characteristics by utilizing a multi-head self-attention network, a feedforward neural network and a residual error network of the oil consumption prediction model of the mining vehicle according to the historical time sequence data and the position codes.
5. The training method of claim 1, wherein generating the plurality of load ranges non-linearly from the empty load and the full load of the vehicle comprises:
the plurality of load ranges are non-linearly divided based on an average value of the empty loads of the plurality of vehicles at the specified time and a maximum value of the full loads of the plurality of vehicles.
6. The training method of claim 1, wherein generating mission features corresponding to load ranges from range of a vehicle for which a load is planned, comprises:
for each load range, a mission feature corresponding to the load range is generated from a sum of mileage of all vehicles of the plurality of vehicles for which the planned load is within the load range, an average value of planned loads of all vehicles of the plurality of vehicles for which the planned load is within the load range, and an average value of empty loads of the plurality of vehicles at a specified time.
7. The training method of claim 1, wherein generating a predicted value of fuel consumption for the target period from the timing characteristic and the mission characteristic comprises:
processing task characteristics by using a residual error network of a fuel consumption prediction model of the mining area vehicle;
and generating a predicted value of the oil consumption of the target period according to the sum of the task characteristics and the time sequence characteristics processed by the residual error network.
8. The training method of claim 1, wherein:
the actual value of the historical time-series oil consumption comprises a sequence of actual values of total oil consumption of a plurality of vehicles in a plurality of continuous time periods before a target time period;
the actual value of fuel consumption for the target period includes an actual value of total fuel consumption for a plurality of vehicles within the target period.
9. The training method of claim 1, wherein the historical data includes at least one of historical weather data for the mine, historical empty mileage and historical full mileage for the vehicle.
10. The training method of claim 1, wherein obtaining task data for a target period comprises:
obtaining a planned production of different types of ore within a target period;
the planned load of the vehicle in the target period is determined based on the planned production of different types of ore in the target period.
11. A method of predicting fuel consumption of a mining vehicle, comprising:
acquiring historical time sequence data and task data of a target period, wherein the historical time sequence data comprises a sequence of historical data of a plurality of continuous periods before the target period;
predicting the fuel consumption of the target period by using a fuel consumption prediction model of the mining vehicle according to the historical time sequence data and the task data of the target period, wherein the fuel consumption prediction model of the mining vehicle is trained according to the method of any one of claims 1-10.
12. A training device for a fuel consumption prediction model of a mining vehicle, comprising:
the system comprises an acquisition module, a control module and a control module, wherein the acquisition module is configured to acquire historical time sequence data, task data of a target period, a true value of historical time sequence oil consumption and a true value of oil consumption of the target period, wherein the historical time sequence data comprises a sequence of historical data of a plurality of continuous periods before the target period, the true value of the historical time sequence oil consumption comprises a sequence of true values of oil consumption of the vehicle in a plurality of continuous periods before the target period, and the task data of the target period comprises a planned load and a driving mileage of the vehicle in the target period;
a historical fuel consumption generation module configured to generate a predicted value of historical time-series fuel consumption by using a fuel consumption prediction model of the mining vehicle according to the historical time-series data, including
Generating time sequence characteristics by utilizing a multi-head self-attention network of a fuel consumption prediction model of the mining area vehicle according to the historical time sequence data;
generating a predicted value of historical time sequence oil consumption according to the time sequence characteristics;
a target fuel consumption generation module configured to generate a predicted value of fuel consumption of a target period by using a fuel consumption prediction model of the mining vehicle according to the historical time sequence data and the task data of the target period, including
Acquiring the empty load and the full load of the vehicle;
according to the empty load and the full load of the vehicle, generating a plurality of load ranges in a nonlinear manner, wherein the larger the load corresponding to the load range is, the smaller the interval width is;
generating task features according to the planned load and the driving mileage of the vehicle in a target period, wherein the task features comprise generating task features corresponding to the load ranges according to the driving mileage of the vehicle with the planned load in each load range;
according to the time sequence characteristics and the task characteristics, a predicted value of the oil consumption of the target period is generated by utilizing an oil consumption prediction model of the mining area vehicle;
the combined training module is configured to combine and train the oil consumption prediction model according to the actual value of the oil consumption of the historical time sequence, the predicted value of the oil consumption of the historical time sequence, the actual value of the oil consumption of the target period and the predicted value of the oil consumption of the target period.
13. A fuel consumption prediction device for mining vehicles comprises
An acquisition module configured to acquire historical time series data and task data of a target period, wherein the historical time series data comprises a sequence of historical data of a plurality of continuous periods before the target period, and the task data of the target period comprises a planned load and a driving mileage of the vehicle in the target period;
the prediction module is configured to predict the fuel consumption of the target period by using a fuel consumption prediction model of the mining vehicle according to the historical time sequence data and the task data of the target period, wherein the fuel consumption prediction model of the mining vehicle is trained according to the training device of claim 12.
14. An electronic device, comprising:
a memory; and
a processor coupled to the memory, the processor being configured to perform a training method of a fuel consumption prediction model of a mining vehicle according to any one of claims 1 to 10, or to perform a prediction method of fuel consumption of a mining vehicle according to claim 11, based on instructions stored in the memory.
15. A computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement a method of training a fuel consumption prediction model of a mining vehicle according to any one of claims 1 to 10, or a method of predicting fuel consumption of a mining vehicle according to claim 11.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211220194.7A CN115293055B (en) | 2022-10-08 | 2022-10-08 | Training method and device for oil consumption prediction model of mining vehicle and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211220194.7A CN115293055B (en) | 2022-10-08 | 2022-10-08 | Training method and device for oil consumption prediction model of mining vehicle and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115293055A CN115293055A (en) | 2022-11-04 |
| CN115293055B true CN115293055B (en) | 2023-05-09 |
Family
ID=83834636
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211220194.7A Active CN115293055B (en) | 2022-10-08 | 2022-10-08 | Training method and device for oil consumption prediction model of mining vehicle and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115293055B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110562096A (en) * | 2019-08-22 | 2019-12-13 | 东软睿驰汽车技术（沈阳）有限公司 | Remaining mileage prediction method and device |
| CN114202123A (en) * | 2021-12-14 | 2022-03-18 | 深圳壹账通智能科技有限公司 | Service data prediction method and device, electronic equipment and storage medium |
| CN114274780A (en) * | 2022-01-10 | 2022-04-05 | 中国第一汽车股份有限公司 | Vehicle driving mileage prediction method, device, equipment and storage medium |
| CN114435138A (en) * | 2022-01-25 | 2022-05-06 | 中国第一汽车股份有限公司 | Vehicle energy consumption prediction method and device, vehicle and storage medium |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1786413A (en) * | 2005-11-17 | 2006-06-14 | 孟江 | Method for porecasting water upwelling amount of mine well based on nerve network model |
| CN111310995A (en) * | 2020-02-12 | 2020-06-19 | 支付宝(杭州)信息技术有限公司 | Service prediction method and device based on convolutional neural network |
| CN113776610A (en) * | 2021-08-04 | 2021-12-10 | 中寰卫星导航通信有限公司 | Method and device for determining vehicle oil consumption |
| CN114781698A (en) * | 2022-03-31 | 2022-07-22 | 山东大学 | Vehicle oil consumption prediction method and system based on spatio-temporal attention mechanism chart convolution network |
-
2022
- 2022-10-08 CN CN202211220194.7A patent/CN115293055B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110562096A (en) * | 2019-08-22 | 2019-12-13 | 东软睿驰汽车技术（沈阳）有限公司 | Remaining mileage prediction method and device |
| CN114202123A (en) * | 2021-12-14 | 2022-03-18 | 深圳壹账通智能科技有限公司 | Service data prediction method and device, electronic equipment and storage medium |
| CN114274780A (en) * | 2022-01-10 | 2022-04-05 | 中国第一汽车股份有限公司 | Vehicle driving mileage prediction method, device, equipment and storage medium |
| CN114435138A (en) * | 2022-01-25 | 2022-05-06 | 中国第一汽车股份有限公司 | Vehicle energy consumption prediction method and device, vehicle and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115293055A (en) | 2022-11-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Patterson et al. | Energy efficient scheduling of open-pit coal mine trucks | |
| US20180247207A1 (en) | Online hierarchical ensemble of learners for activity time prediction in open pit mining | |
| CN112053560B (en) | Short-time traffic flow prediction method, system and storage medium based on neural network | |
| CN101620694A (en) | Logistics transportation decision-making support system and method, and transportation management system | |
| Song et al. | A match‐then‐predict method for daily traffic flow forecasting based on group method of data handling | |
| Hashemi et al. | Simulation based investigation of different fleet management paradigms in open pit mines-a case study of Sungun copper mine | |
| Kurganov et al. | Assessment of operational reliability of quarry excavator-dump truck complexes | |
| WO2016118122A1 (en) | Optimization of truck assignments in a mine using simulation | |
| Maghrebi et al. | Solving ready-mixed concrete delivery problems: Evolutionary comparison between column generation and robust genetic algorithm | |
| CN115293055B (en) | Training method and device for oil consumption prediction model of mining vehicle and electronic equipment | |
| Hsueh et al. | A short-term traffic speed prediction model based on LSTM networks | |
| Tapia et al. | An analysis of full truck versus full bucket strategies in open pit mining loading and hauling operations | |
| Zhao et al. | A novel predict-prevention quality control method of multi-stage manufacturing process towards zero defect manufacturing | |
| Medvediev et al. | Design logical linguistic models to calculate necessity in trucks during agricultural cargoes logistics using fuzzy logic | |
| Oszczypała et al. | Reliability analysis and redundancy optimization of k-out-of-n systems with random variable k using continuous time Markov chain and Monte Carlo simulation | |
| Ruta et al. | Diversified gradient boosting ensembles for prediction of the cost of forwarding contracts | |
| CN116187887A (en) | Method for constructing and tracking traffic subsection door goods communication data | |
| CN115640877A (en) | Method, device, equipment and storage medium for planning traffic flow of surface mine | |
| CN116258324A (en) | Mine car real-time dispatching system and method based on mine transportation demand prediction | |
| CN113435055B (en) | Self-adaptive migration prediction method and system in shield cutter head torque field | |
| Kuznetsov et al. | Research and Development of Hybrid Simulation Models of Mining Complexes | |
| Guseva et al. | Simulation modeling of transport flows of copper deposit | |
| Brunheroto et al. | Data analytics in fleet operations: A systematic literature review and workflow proposal | |
| Eskandari et al. | Simulation and optimization of haulage system of an open-pit mine. | |
| Borisov et al. | The System of Fuzzy Cognitive Analysis and Modeling of System Dynamics |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |