An electric vehicle instantaneous energy consumption estimation method and device

Abstract

The application provides an electric vehicle instantaneous energy consumption estimation method and device, comprising the following steps: A, obtaining electric vehicle driving data and historical weather data; B, adding a new column as a subsequent target feature in the driving data; C, splicing the driving data and the historical weather data after uniforming the time format and precision to form merged data; D, constructing a neural network; E, dividing the merged data into a training set and a test set, and training the neural network constructed in step D by using the training set to obtain an instantaneous energy consumption estimation model. The application has high calculation precision, fast calculation speed and strong adaptability.

Description

TECHNICAL FIELD

[0001] The application relates to an electric vehicle instantaneous energy consumption estimation method and device. BACKGROUND

[0002] The driving range of an electric vehicle has always been an important indicator of attention, and reliable and accurate estimation of the instantaneous energy consumption of an electric vehicle is crucial for the driver's trip planning. Since the energy consumption of an electric vehicle is not a simple linear model, but is coupled by multiple factors, the energy consumption prediction based on the establishment of a physical model of the vehicle is difficult to achieve high accuracy.

[0003] At present, the methods widely used for estimating the energy consumption of an electric vehicle can be divided into two categories. The first category of method is to take the battery as the research object, study the relationship between the SOC, temperature and other characteristics of the battery and the energy consumption, and establish an energy model to predict the instantaneous energy consumption through the state parameters of the battery. However, since only the battery is considered and the driver's behavior and the environmental conditions are ignored, the error of the instantaneous energy consumption estimation in a complex dynamic traffic environment is large. The second category of method is to take the driver's behavior factors, vehicle factors (speed, acceleration, etc.), environmental temperature and the like as the research object, perform feature recognition, and use a machine learning method for modeling analysis. This method is the current mainstream method. However, the machine learning method for electric vehicle energy consumption prediction has some shortcomings. The first is the neglect of the diversity of weather information, and only the environmental temperature and wind speed are considered. The second is that the traditional machine learning regression algorithm is not good enough for classification feature processing. The third is that the model parameter optimization of the traditional machine learning regression algorithm cannot quickly adapt to real-time data updates and needs to be continuously retrained according to the recent data. With the passage of time, the real data distribution may be quite different from the training data distribution, thereby affecting the accuracy. SUMMARY

[0004] The application provides an electric vehicle instantaneous energy consumption estimation method and device, which has high calculation accuracy, fast calculation speed and strong adaptability.

[0005] The application is implemented by the following technical solutions:

[0006] An electric vehicle instantaneous energy consumption estimation method comprises the following steps:

[0007] A, obtaining electric vehicle driving data, the driving data comprising time and location information, and obtaining historical weather data according to the time and location information, the historical weather data comprising atmospheric temperature, dew point temperature, atmospheric pressure, wind direction, wind speed, humidity, cloud cover, horizontal visibility and special weather;

[0008] B、the driving data further comprises instantaneous current data and instantaneous voltage data, a new column is added in the driving data as a subsequent target feature, and the column data is the product of the instantaneous current data and the corresponding instantaneous voltage data;

[0009] C、unify the time format of the driving data and the time format of the historical weather data, unify the precision of the driving data and the precision of the historical weather data, and splice the driving data and the historical weather data after the time format and the precision are unified to form merged data;

[0010] D、construct a neural network: the neural network comprises an input layer with a classification feature input part and a digital feature input part, a dropout neural network layer associated with the classification feature input part and the digital feature input part respectively, an FM neural network layer associated with the classification feature input part and the corresponding dropout neural network layer, and a local connection layer associated with the digital feature input part and the corresponding dropout neural network layer;

[0011] E、divide the merged data into a training set and a test set, and train the neural network constructed in step D using the training set to obtain an instantaneous energy consumption estimation model.

[0012] Further, in step A, the driving data further comprises speed, acceleration pedal and brake pedal state, and the position information is latitude and longitude data.

[0013] Further, in step A, the sampling period of the historical weather data is half an hour.

[0014] Further, in step B, before the new column is added in the driving data, the instantaneous current data less than or equal to 0 in the driving data and the instantaneous voltage data corresponding to the instantaneous current data are deleted, and after the new column is added in the driving data, the instantaneous current data column and the instantaneous voltage data column are deleted.

[0015] Further, step C specifically comprises:

[0016] C1、the time format of the driving data is "year-month-day-hour-minute-second", the time format of the historical weather data is "day-month-year-hour-minute", and the time format of the historical weather data is modified to "year-month-day-hour-minute-second";

[0017] C2、the time precision of the driving data is "second", the time precision of the historical weather data is the sampling period, and the time format of the driving data and the historical weather data is modified to "year-month-day-hour", so that the time precision of the driving data and the historical weather data is the same;

[0018] C3, according to the time characteristics, the historical weather data is matched to the driving data to form the combined data based on the driving data processed through steps C1 and C2.

[0019] Further, in step E, the loss function is set as MAPE: MAPE = sum(|y'-y|*100% / y) / n, and the dropout value and the number of neurons in each hidden layer are continuously adjusted to minimize MAPE, so as to obtain the instantaneous energy consumption estimation model, wherein y is the actual value, y' is the predicted value, and n is the sample size.

[0020] Further, in step E, the combined data is first subjected to z-score standardization, and then the standardized combined data is divided into a training set and a test set.

[0021] Further, in step E, the initial value range of the dropout value is [0.1, 0.5].

[0022] The application also realizes the following technical solutions:

[0023] An instantaneous energy consumption estimation device for an electric vehicle comprises:

[0024] A data acquisition module is configured to acquire driving data of an electric vehicle, the driving data comprising time, location information, instantaneous current data and instantaneous voltage data, and to acquire historical weather data according to the time and location information, the historical weather data comprising atmospheric temperature, dew point temperature, atmospheric pressure, wind direction, wind speed, humidity, cloud cover, horizontal visibility and special weather, and adding a new column to the driving data as a subsequent target feature, the column data being the product of the instantaneous current data and the corresponding instantaneous voltage data;

[0025] A data splicing module is configured to unify the time format of the acquired driving data and the time format of the acquired historical weather data, unify the precision of the driving data and the precision of the historical weather data, and splice the driving data and the historical weather data after the time format and the precision are unified to form combined data;

[0026] A neural network construction module is configured to construct a neural network comprising an input layer having a classification feature input part and a digital feature input part, a dropout neural network layer associated with the classification feature input part and the digital feature input part, an FM neural network layer associated with the classification feature input part and the corresponding dropout neural network layer, and a local connection layer associated with the digital feature input part and the corresponding dropout neural network layer;

[0027] A training module is configured to divide the combined data into a training set and a test set, and train the neural network constructed in step D using the training set to obtain an instantaneous energy consumption estimation model.

[0028] The present application has the following beneficial effects:

[0029] 1、The present application obtains historical weather data according to time and location information in driving data, and the historical weather data includes atmospheric temperature, dew point temperature, atmospheric pressure, wind direction, wind speed, humidity, cloud cover, horizontal visibility and special weather, then a power column is added to the driving data, the column is used as a subsequent target feature, the driving data and the historical weather data are spliced, so as to train and estimate the neural network subsequently, when the neural network is constructed, the dropout neural network layer and the FM neural network layer are combined, the running speed is improved while the accuracy is ensured, the FM neural network layer can well solve the parameter estimation and feature combination problem of the classification feature by introducing the cross feature and the hidden vector, the dropout neural network layer reduces the overfitting problem by repeatedly randomly selecting neurons and taking the average value, the use of the local connection layer can reduce the parameter amount and speed up the running speed, therefore, under the condition of fully considering the influence of weather diversity on the energy consumption of the electric vehicle and the neural network constructed by the present application, the operation accuracy and operation speed of the instantaneous energy consumption estimation can be greatly improved, the model parameter optimization can quickly adapt to the real-time data update, and the situation that the real data distribution and the training data distribution are greatly different with the passage of time, and then the accuracy is affected, is avoided.

[0030] 2、Before a new column is added to the driving data, the instantaneous current data and the corresponding instantaneous voltage data less than or equal to 0 in the driving data are deleted, which can prevent the two data from interfering with the estimation of the feature pair target feature in the subsequent calculation process, thereby further improving the accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0031] The present application will be further described in detail below with reference to the accompanying drawings.

[0032] Figure 1 The flowchart of the present application.

[0033] Figure 2 The structure diagram of the neural network of the present application. DETAILED DESCRIPTION

[0034] As shown in the figure, the electric vehicle instantaneous energy consumption estimation method comprises the following steps: Figure 1

[0035] ​A, obtaining electric vehicle driving data from the new energy vehicle national detection and management platform, the driving data including time, location information, instantaneous current data, instantaneous voltage data, speed, acceleration pedal and brake pedal state, the location information being longitude and latitude data, and logging into the global historical weather network, obtaining corresponding historical weather data of the electric vehicle during driving according to the time and location information, the historical weather data including atmospheric temperature, dew point temperature, atmospheric pressure, wind direction, wind speed, humidity, cloud cover, horizontal visibility and special weather, and the sampling period of the historical weather data being half an hour;

[0036] B, deleting the instantaneous current data less than or equal to 0 in the driving data and the instantaneous voltage data corresponding to the instantaneous current data, because when the instantaneous current is equal to 0, it represents that the electric vehicle is in a stop state, which is meaningless for research, and when the instantaneous current is less than 0, it represents that the electric vehicle is in a charging state caused by braking regeneration, which is not within the research scope;

[0037] Then a new column is added to the driving data as a subsequent target feature, the column data being the product of the instantaneous current data and the corresponding instantaneous voltage data, that is, the instantaneous total power, and compared with the traditional energy consumption indicator ECR, the instantaneous total power is more accurate in expressing the specific moment energy consumption; the instantaneous total power as the subsequent target feature is specifically: taking the instantaneous total power value as the true value, inputting other features except the instantaneous total power into the neural network at each instantaneous moment for training, and the training control target is the instantaneous total power at this moment, and the closer the estimated value of the instantaneous total power after training to the true value, the more successful the neural network training is;

[0038] After adding a new column to the driving data, the instantaneous current data column and the instantaneous voltage data column are deleted to avoid interference with subsequent calculations;

[0039] C, unifying the time format of obtaining the driving data and the time format of obtaining the historical weather data, unifying the precision of the driving data and the precision of the historical weather data, and splicing the driving data and the historical weather data after unifying the time format and the precision to form merged data; specifically including the following steps:

[0040] C1, setting the driving data as df1 and the historical weather data as df2, the time format of the driving data being “year-month-day-hour-minute-second”, and the time format of the historical weather data being “day-month-year-hour-minute”, modifying the time format of the historical weather data to “year-month-day-hour-minute-second” by using the function pd.to_datetime(df2[‘time’], format‘%d.%m.%Y%H:%M’);

[0041] C2, the time precision of the driving data is "second", and the time precision of the historical weather data is a sampling period, the time format of the driving data is modified to "year-month-day-hour" by using a function df1['time'].astype('datetime64[h]'), the time format of the historical weather data is modified to "year-month-day-hour" by using a function df2['time'].astype('datetime64[h]'), so that the time precisions of the driving data and the historical weather data are the same;

[0042] C3, based on the driving data processed by steps C1 and C2, the historical weather data is matched to the driving data according to the time characteristics by using a function df3=pd.merge(df1,df2,how='left',on='time'), to form the merged data df3;

[0043] D, constructing a neural network: as shown in Figure 2 The neural network includes an input layer with a classification feature input part and a digital feature input part, a dropout neural network layer associated with the classification feature input part and the digital feature input part, an FM neural network layer associated with the classification feature input part and the corresponding dropout neural network layer, and a local connection layer associated with the digital feature input part and the corresponding dropout neural network layer; wherein the classification feature input part includes wind direction, cloud cover, horizontal visibility and special weather, and the rest are digital feature input parts;

[0044] E, the merged data is z-score standardized, and the standardized merged data is divided into a training set and a test set, the neural network constructed in step D is trained using the training set to obtain an instantaneous energy consumption estimation model, the loss function is set as MAPE: MAPE=sum(|y′-y|*100% / y) / n, the dropout value and the number of neurons in the dropout neural network layer, the FM neural network layer and the local connection layer are adjusted to minimize the MAPE, and the instantaneous energy consumption estimation model is obtained, wherein the initial value of the dropout value is 0.1, y is the actual value, y′ is the predicted value, and n is the sample size.

[0045] An electric vehicle instantaneous energy consumption estimation device, comprising the following modules:

[0046] The data acquisition module is used for acquiring the electric vehicle driving data, the driving data including time, position information, instantaneous current data and instantaneous voltage data, and acquiring historical weather data according to the time and position information, the historical weather data including atmospheric temperature, dew point temperature, atmospheric pressure, wind direction, wind speed, humidity, cloud cover, horizontal visibility and special weather, and adding a new column as a subsequent target feature in the driving data, the column data being the product of the instantaneous current data and the corresponding instantaneous voltage data;

[0047] The data splicing module is used for unifying the time format of the acquired driving data and the time format of the acquired historical weather data, unifying the precision of the driving data and the precision of the historical weather data, and splicing the driving data and the historical weather data after the time format and the precision are unified, to form the merged data;

[0048] The neural network construction module is used for constructing the neural network, the neural network including an input layer with a classification feature input part and a digital feature input part, a dropout neural network layer respectively associated with the classification feature input part and the digital feature input part, an FM neural network layer associated with the classification feature input part and the corresponding dropout neural network layer, and a local connection layer associated with the digital feature input part and the corresponding dropout neural network layer;

[0049] The training module is used for dividing the merged data into a training set and a test set, training the constructed neural network by using the training set, to obtain the instantaneous energy consumption estimation model.

[0050] The above is only a preferred embodiment of the present application, and therefore cannot limit the scope of the present application, that is, equivalent changes and modifications made according to the patent application scope and the content of the specification should still be within the scope of the present application.

Claims

1. A method for estimating instantaneous energy consumption of an electric vehicle, the method comprising: It comprises the following steps: ​ A. Obtain electric vehicle driving data, which includes time and location information, and obtain historical weather data based on the time and location information, which includes atmospheric temperature, dew point temperature, atmospheric pressure, wind direction, wind speed, humidity, cloud cover, horizontal visibility and special weather; B. The driving data also includes instantaneous current data and instantaneous voltage data, and a new column is added to the driving data as a subsequent target feature, which is the product of the instantaneous current data and the corresponding instantaneous voltage data; C. The time format of the driving data and the time format of the historical weather data are unified, the precision of the driving data and the precision of the historical weather data are unified, and the driving data and the historical weather data after the time format and the precision are unified are spliced to form merged data; D. Construct a neural network: the neural network includes an input layer with a classification feature input part and a digital feature input part, a dropout neural network layer associated with the classification feature input part and the digital feature input part, an FM neural network layer associated with the classification feature input part and the corresponding dropout neural network layer, and a local connection layer associated with the digital feature input part and the corresponding dropout neural network layer; E. Divide the merged data into a training set and a test set, and use the training set to train the neural network constructed in step D to obtain an instantaneous energy consumption estimation model; In step A, the driving data also includes speed, acceleration pedal and brake pedal state, and the location information is latitude and longitude data; In step B, before adding the new column to the driving data, delete the instantaneous current data less than or equal to 0 and the corresponding instantaneous voltage data in the driving data, and after adding the new column to the driving data, delete the instantaneous current data column and the instantaneous voltage data column.

2. The method of claim 1, wherein: In step A, the sampling period of the historical weather data is half an hour.

3. The method of claim 2, wherein: Step C specifically includes: C1, the time format of the driving data is "year-month-day-hour-minute-second", and the time format of the historical weather data is "day-month-year-hour-minute", the time format of the historical weather data is modified to "year-month-day-hour-minute-second"; C2, the time precision of the driving data is "second", and the time precision of the historical weather data is the sampling period, the time format of the driving data and the historical weather data is modified to "year-month-day-hour", so that the time precision of the driving data and the historical weather data is the same; C3, based on the driving data processed by steps C1 and C2, the historical weather data is matched to the driving data according to the time characteristics to form the merged data.

4. The method of claim 1 or 2 or 3, wherein: The neural network further includes a hidden layer, and in step E, the loss function is set as MAPE: MAPE = sum(|y'-y|*100% / y) / n, the objective is to minimize MAPE, and the dropout value and the number of neurons in each hidden layer are adjusted to obtain the instantaneous energy consumption estimation model, wherein y is the actual value, y' is the predicted value, and n is the sample size.

5. The method of claim 1 or 2 or 3, wherein: In step E, the combined data is first subjected to z-score standardization, and then the standardized combined data is divided into a training set and a test set.

6. The method of claim 4, wherein: In step E, the initial value range of the dropout value is [0.1, 0.5].

7. An estimation device of the estimation method of the instantaneous energy consumption of an electric vehicle according to any one of claims 1 to 6, characterized in that: The method comprises the following steps: a data acquisition module is used to acquire electric vehicle driving data, the driving data comprising time, position information, instantaneous current data, instantaneous voltage data, speed, acceleration pedal and brake pedal state, the position information being latitude and longitude data, and historical weather data is acquired according to the time and position information, the historical weather data comprising atmospheric temperature, dew point temperature, atmospheric pressure, wind direction, wind speed, humidity, cloud cover, horizontal visibility and special weather; after the instantaneous current data less than or equal to 0 in the driving data and the instantaneous voltage data corresponding to the instantaneous current data are deleted, a new column is added to the driving data as a subsequent target feature, the column data being the product of the instantaneous current data and the corresponding instantaneous voltage data, and the instantaneous current data column and the instantaneous voltage data column are deleted; a data splicing module is used to unify the time format of the acquired driving data and the time format of the acquired historical weather data, unify the precision of the driving data and the precision of the historical weather data, and splice the driving data and the historical weather data after the time format and the precision are unified to form combined data; a neural network construction module is used to construct a neural network comprising an input layer having a classification feature input part and a digital feature input part, a dropout neural network layer associated with the classification feature input part and the digital feature input part respectively, an FM neural network layer associated with the classification feature input part and the corresponding dropout neural network layer, and a local connection layer associated with the digital feature input part and the corresponding dropout neural network layer; a training module is used to divide the combined data into a training set and a test set, and the neural network constructed in step D is trained by using the training set to obtain an instantaneous energy consumption estimation model.

Images