A charging pile usage information prediction method based on multi-information coupling

Abstract

The present invention relates to a method for predicting the use of charging piles based on multi-information coupling, comprising the following steps: Step 1: Using Python crawler technology to obtain the charging station information of the prediction area; Step 2: Performing correlation analysis on the obtained charging station information to obtain factors with high correlation with the busy and idle states of the charging piles, and combining these highly correlated factors with historical operation data to determine input variables; Step 3: Based on the obtained input variables, using a deep belief network prediction model to predict the use of charging piles in the charging stations in the prediction area; Step 4: Outputting the prediction results of the use of charging piles in a certain period of time at the charging stations in the prediction area. The present invention has the advantages of high prediction accuracy, the ability to plan the charging of electric vehicles, and convenience for users, and is a highly innovative method for predicting the use of charging piles based on multi-information coupling.

Description

Technical Field

[0001] The invention belongs to the technical field of electric vehicles and relates to a charging pile usage information prediction method based on multi-information coupling. Background Art

[0002] With the rapid growth of the global economy and the increasing consumption of fossil fuels, electric vehicles (EVs) have emerged as a new type of transportation. However, the current growth scale of charging piles is far from meeting the rapidly growing demand for electric vehicles. Timely and fast charging is still an important factor affecting the development of electric vehicles.

[0003] Users have an increasing demand for smart travel. Unlike traditional fuel vehicles, electric vehicles take a long time to charge. In addition, due to the low coverage of charging stations and limited charging resources, charging stations cannot promptly inform electric vehicle owners of the busy and idle status of charging stations, which may prevent electric vehicles from charging immediately after arriving at charging stations. In order to reduce the waiting time of users at charging stations, the real-time operating status of charging stations must be considered. However, the existing charging navigation APP does not predict the number of charging piles used at charging stations, which can easily cause electric vehicle users to "have piles when leaving, but congestion when arriving at the station". Therefore, the existing prediction technology for electric vehicle charging station information still needs to be improved and developed. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art, comprehensively consider the usage of charging piles in a charging station and the temporal and spatial factors that affect the charging of electric vehicles at the station, predict the usage of charging piles in a certain period of time in the future, and provide a charging pile usage information prediction method based on multi-information coupling.

[0005] The present invention solves the technical problem by the following technical solutions:

[0006] A charging pile usage information prediction method based on multi-information coupling includes the following steps:

[0007] Step 1: Use Python crawler technology to obtain charging station information in the predicted area;

[0008] Step 2: Perform correlation analysis on the acquired charging station information to obtain factors with high correlation with the busy and idle status of the charging piles, and combine these highly correlated factors with historical operation data to determine input variables;

[0009] Step 3: Based on the input variables obtained, the deep belief network prediction model is used to predict the usage of charging piles in the charging stations in the prediction area;

[0010] Step 4: Output the prediction results of the usage of charging piles in a certain period of time at the charging stations in the prediction area.

[0011] Furthermore, the charging station information includes but is not limited to any of the following: charging station status information, road condition information, historical operation data and weather information.

[0012] Moreover, the charging station status information includes but is not limited to any one of the following: location information of the charging station, and status information of the charging piles already in use at the charging station.

[0013] Furthermore, the road condition information includes but is not limited to any one of the following: road topology information, and traffic flow information of a road section near a charging station.

[0014] Furthermore, the factors with higher correlation include but are not limited to any one of the following: day type, weather conditions and traffic, and geographical location.

[0015] Moreover, the input variables include the number of uses before the time to be measured, the weather conditions of the predicted day, the day type, the ambient temperature, the surrounding traffic index and the most relevant charging station usage.

[0016] Moreover, the method of using the deep belief network prediction model to predict the usage of charging piles in the charging station in the prediction area according to the obtained input variables is:

[0017] (1) First, a part of the historical operation data of the charging station and its associated weather conditions, day type, ambient temperature, surrounding traffic index and the most relevant charging station usage are used as the training set of the model, which is standardized as the input variable of the model, and the output variable is the known number of historical charging piles used. Through this step, the prediction model trains each layer of the RBM network separately and unsupervised to obtain the weight matrix and weight of each layer of the RBM between the input variable of the number of charging piles used and the number of charging piles used at the prediction time;

[0018] (2) A BP network is set up at the last layer of the DBN to receive the output feature vector of the RBM as its input feature vector. This layer feeds back the error information to each RBM layer from top to bottom, and fine-tunes the weight matrix and weight parameters of the entire DBN model in a supervised manner.

[0019] (3) Input the number of times the charging station was used before the time to be tested, the weather conditions on the predicted day, the day type, the ambient temperature, the surrounding traffic index, and the number of times the charging station with the highest correlation was used for standardization processing;

[0020] (4) Output the number of charging piles that will be used by the charging station in the short term in the future.

[0021] The advantages and beneficial effects of the present invention are:

[0022] 1. This charging pile usage information prediction method based on multi-information coupling takes into account the nonlinear and random changes in the number of charging piles used in electric vehicle charging stations under a multi-information coupling environment. It fully considers the influence of external environment such as weather conditions, meteorological conditions, traffic conditions and regional charging stations, improves the accuracy of model prediction, and lays the foundation for electric vehicle charging planning.

[0023] 2. This charging pile usage information prediction method based on multi-information coupling has an accuracy of 88.17% and an average absolute percentage error of 9.14% when the traffic conditions around the charging station are not considered. After taking the traffic factors into account, the accuracy of the charging station prediction model is increased by 6.34%, and the average absolute percentage error is reduced by 4.71%, and its prediction accuracy is higher.

[0024] 3. The present invention has the advantages of high prediction accuracy, ability to plan charging for electric vehicles, and convenience for users. It is a highly innovative charging pile usage information prediction method based on multi-information coupling. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a schematic diagram of DBN prediction effect analysis of the present invention;

[0026] Figure 2 This is a schematic diagram for comparing the prediction effects when the present invention takes traffic factors into consideration;

[0027] Figure 3 Schematic diagram of the DBN model structure of the present invention. DETAILED DESCRIPTION

[0028] The present invention is further described in detail below through specific examples. The following examples are only illustrative and not restrictive, and the protection scope of the present invention cannot be limited thereto.

[0029] The present invention proposes a method for predicting the operating status of an electric vehicle charging station under the interconnection and coupling of multiple information. Firstly, the traffic road condition information and charging station information in the prediction area are collected by using Amap and e-charging network based on Python crawler technology, and the correlation between the local meteorological conditions and the surrounding traffic conditions and the busy and idle status of the charging piles is analyzed. Finally, a deep belief network prediction model is used to predict the usage of the charging piles in the charging station, and the future operation data of the charging station is predicted, so as to provide electric vehicle users with an effective charging guidance strategy.

[0030] The deep belief network is a probabilistic generative model. Compared with the traditional discriminative neural network, the generative model is to establish a joint distribution between observed data and labels. The deep belief network DBN consists of multiple restricted Boltzmann machine RBM (Restricted Boltzmann Machines) layers. A typical network structure is as follows Figure 3 shown.

[0031] Its innovation lies in:

[0032] Step 1: Use Python crawler technology to obtain charging station status information and road condition information;

[0033] Step 2: Perform Pearson correlation analysis on the data obtained in step 1 to determine the input variables of the prediction method;

[0034] Step 3: Based on the input variables obtained, the deep belief network prediction model is used to predict the usage of charging piles in the charging stations in the prediction area;

[0035] The DBN model training process is mainly divided into two steps:

[0036] Step 1: Train each layer of the RBM network separately and unsupervised to ensure that as much feature information as possible is retained when the feature vector is mapped to different feature spaces, aiming to hierarchically characterize the complex nonlinear relationship between the factors affecting the number of charging piles in use and the number of charging piles in use at the prediction time;

[0037] Step 2: Set up a BP network at the last layer of DBN, receive the output feature vector of RBM as its input feature vector, and train the entity relationship classifier in a supervised manner. Moreover, each layer of RBM network can only ensure that the weights in its own layer are optimally mapped to the feature vector of that layer, not the feature vector of the entire DBN. Therefore, the back propagation network also propagates the error information from top to bottom to each layer of RBM to fine-tune the entire DBN network. The process of RBM network training model can be regarded as the initialization of the weight parameters of a deep BP network, which enables DBN to overcome the shortcomings of BP network that is prone to local optimality and long training time due to random initialization of weight parameters.

[0038] (1) DBN unsupervised pre-training stage

[0039] RBM consists of visible layers and hidden layers. The unsupervised pre-training phase can be completed by training the connection weights and weight parameters between each RBM layer to achieve the ideal state of the model. The parameter initialization model between n RBM layers is

[0040]

[0041] In the formula, parameter θ = (b, c, W); v is the visible layer, that is, the input normalized data, b is the weight on the visible layer; h is the hidden layer, which obeys the corresponding distribution and is obtained through Gibbs sampling, c is the weight on the hidden layer; W is the connection weight matrix between the visible layer v and the hidden layer h, v T is the transpose of the visual layer v.

[0042] If the parameter θ is known, according to the input v(i), the hidden layer h(i) is:

[0043]

[0044] In the formula, σ(·) is the activation function, h(i) is the output value of the i-th node in the hidden layer h, and v T is the transpose of the visual layer v, W :,i is the i-th column of the weight matrix W, c i is the i-th component in the hidden layer weight c;

[0045] Similarly, if the hidden layer h(j) is known, the visible layer v(j) is:

[0046]

[0047] Where v(j) is the output value of the jth node in the visual layer v, W j,: is the jth row of the weight matrix W, b j is the jth component in the visual layer weight v;

[0048] RBM uses an energy model to represent a measure of the steady state of a system. The more ordered the system is or the more concentrated the probability distribution is, the smaller the energy of the system is. Conversely, the more disordered the system is and the more divergent the probability distribution is, the greater the energy of the system is. The minimum value of the energy function corresponds to the most stable state of the entire system.

[0049] The energy function of the network is defined as:

[0050]

[0051] Where v(j) is the output value of the jth node in the visual layer v, W ij is the weight between the visible layer v and the hidden layer h, h i is the output value of the i-th node in the hidden layer h, b j is the jth component in the visual layer weight v, c i is the i-th component in the hidden layer weight c.

[0052] Through the energy function, the probability distribution of the system random variables can be obtained:

[0053]

[0054] Where Z is the regularization factor, which is the sum of the energy functions associated with all visible units and hidden units.

[0055] This probability distribution can be considered as the result of normalizing the energy function with the exponent e. A state of a sample corresponds to an energy, and the corresponding probability distribution corresponds to the probability distribution of the energy. The maximum likelihood estimation method is used to train the data and obtain the parameter optimization objective function when the model is in the ideal state:

[0056]

[0057] By adjusting the internal parameter θ of RBM, RBM can fit the training samples well, that is, the probability distribution of RBM is consistent with the probability distribution of samples. For probability The logarithm of .

[0058] For a sample data x, the contrast divergence algorithm is used to train it:

[0059] ■ Assign x to the visible layer v 1 , using formula (2) to calculate the probability P(h) of each neuron being activated in the visible layer 1 |v 1 );

[0060] ■ Take a sample from the calculated probability distribution using Gibbs sampling: h 1 ~P(h 1 |v 1 );

[0061] ■Use h 1 Reconstruct the visible layer, that is, infer the visible layer through the hidden layer, and use formula (3) to calculate the probability P(v 2 |h 1 );

[0062] ■Similarly, take Gibbs sampling to draw a sample from the calculated probability distribution: v 2 ~P(v 2 |h 1 );

[0063] ■Through v 2 Calculate the probability of each neuron in the hidden layer being activated again and get the probability distribution P(v 2 |h 1 );

[0064] ■Update weights:

[0065] W←W+λ(P(h 1 |v 1 )v 1-P(h 2 |v 2 )v 2 )

[0066] b←b+λ(v 1 -v 2 )

[0067] c←c+λ(h 1 -h 2 )

[0068] The weight matrix and weight of each layer of RBM are obtained, the pre-training process of parameter initialization is completed, and the deep features of the number of charging piles in use are obtained.

[0069] (2) DBN supervised fine-tuning stage

[0070] A BP network is set up at the last layer of DBN, which receives the output feature vector of RBM as its input feature vector, and uses a back propagation network to propagate the error information from top to bottom to each layer of RBM, and fine-tune the entire DBN network. The process of RBM network training model can be regarded as the initialization of the weight parameters of a deep BP network, which enables DBN to overcome the shortcomings of BP network that is prone to fall into local optimality and long training time due to random initialization of weight parameters.

[0071] In order to train the DBN model and minimize the prediction error, the structural parameters of the RBM model are updated, and the optimization objective function is constructed according to the energy loss criterion:

[0072]

[0073] The task of learning is to continuously reduce J by adjusting the value of W,b. is the loss function, and λ is the learning rate. The loss function is constructed using cross entropy, that is:

[0074]

[0075] In the formula, δ(·) means that it is 1 when y(k)=1, otherwise it is 0. To prevent the training model from being dependent on the data set and causing overfitting, the regularization term constraint is:

[0076]

[0077] The optimization objective function (7) is a convex optimization problem. The parameter solution depends on the setting value of the initial parameters. The error back propagation algorithm based on gradient descent is used to iteratively solve the error of each layer, minimize the objective function (7), and update the parameter value on each layer of RBM, as shown in formula (10):

[0078]

[0079] In the formula, α is the learning rate, Indicates the parameter update of the kth iteration of the lth layer RBM. This method is used to obtain the model parameters and complete the training process of the charging pile usage prediction model.

[0080] Through the above analysis of the factors affecting the number of charging piles in the charging station, it can be seen that the selection of input variables in the deep belief network model should include the number of uses before the test time, the weather conditions on the predicted day, the day type, the ambient temperature, the surrounding traffic index and the most relevant charging station usage. The input and output variables of the DBN prediction model constructed by the present invention are as follows:

[0081] Table 3. DBN prediction model input and output

[0082]

[0083] Since the dimensions of the input variables of various influencing factors are different, the values ​​will be very different, so the sample data needs to be normalized first. The present invention divides Monday to Friday into working days, Saturday and Sunday as rest days, and the corresponding influencing factors are 1 and 0.8 respectively.

[0084] Weather types are divided into three categories: sunny, cloudy, and rainy and snowy. Considering that cloudy weather conditions have little impact on user travel behavior, sunny and cloudy weather are classified into one category, and the corresponding impact factors of the two categories are 1 and 0.8 respectively.

[0085] Since the fluctuation range of traffic index and temperature is relatively large, they are normalized so that their values ​​are between (0, 1). The formula is shown in formula (11):

[0086]

[0087] In the formula is the normalized sample data at time i, x i is the sample data before normalization, x max is the maximum value of the variable before normalization, x min is the minimum value of the variable before normalization.

[0088] When evaluating the accuracy of the prediction model, the accuracy rate P and the mean absolute percentage error MAPE are selected as evaluation indicators to reflect the accuracy of the model in predicting the number of charging piles in the charging station. The indicators can be expressed as:

[0089]

[0090]

[0091] In the formula, is the predicted value of the charging pile at that moment, e i It is the actual value of the number of charging piles in use at that moment.

[0092] Step 4: Output the prediction results of the charging pile usage at the charging station during a certain period of time.

[0093] The actual operation data of a charging station in Beijing from November to December 2019, as well as the meteorological data, surrounding road condition data, and daily type data of the corresponding dates are used as an example. The original data is processed according to the invention and divided into 4320 segments of data, of which 4248 segments are training set data and 72 segments are test set data. The DBN model parameter settings are shown in Table 1.

[0094] Table 1 DBN model parameter selection

[0095]

[0096]

[0097] After setting the above parameters, a complete electric vehicle charging station information prediction model based on deep belief network can be formed. The prediction effect of the model is as follows: Figure 1 As shown, the accuracy and mean absolute percentage error of the model are 94.51% and 4.43% respectively. Compared with the BP neural network prediction method, the comparison between the two methods can further illustrate the practical effect of the method proposed in the present invention.

[0098] Table 2 Comparison of evaluation performance of each algorithm under different training set sizes

[0099]

[0100] As shown in Table 2, when the training set size is 2880, the average absolute percentage error of the BP neural network is smaller than that of the DBN, and the accuracy is the highest. As the training set size increases, the prediction accuracy shows a trend of gradual attenuation; when the training set size becomes larger, the average absolute percentage error of the DBN model gradually decreases, and the accuracy steadily increases. It can be seen that the DBN prediction method has good deep feature extraction capabilities and good prediction stability, and is suitable for scenarios with large-scale operation data of charging stations.

[0101] In addition, in order to verify the advantages of considering multiple information coupling factors proposed by the present invention, the present invention compares the prediction effects of considering the traffic conditions near the charging station and considering the surrounding traffic conditions. Figure 2The results show that when the traffic conditions around the charging station are not considered, the prediction accuracy of the DBN model is 88.17%, and the average absolute percentage error is 9.14%. After the traffic factors are taken into account, the accuracy of the charging station prediction model is increased by 6.34%, and the average absolute percentage error is reduced by 4.71%, and its prediction accuracy is higher.

[0102] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0103] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0104] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A charging pile usage information prediction method based on multi-information coupling, characterized in that: The steps include: Step 1: Use Python crawler technology to obtain charging station information in the predicted area; Step 2: Perform correlation analysis on the acquired charging station information to obtain factors with high correlation with the busy and idle status of the charging piles, and combine these highly correlated factors with historical operation data to determine input variables; Step 3: Based on the input variables obtained, the deep belief network prediction model is used to predict the usage of charging piles in the charging stations in the prediction area; Step 4: Output the prediction results of the usage of charging piles in a certain period of time at the charging stations in the prediction area; The method of using the deep belief network prediction model to predict the usage of charging piles in the charging station in the prediction area according to the obtained input variables is: (1) First, a part of the historical operation data of the charging station is used as the training set of the model, which is standardized and used as the input variable of the model. The output variable is the known number of historical charging piles in use. Through this step, the prediction model trains each layer of the RBM network separately and unsupervised to obtain the weight matrix and weight of each layer of the RBM between the input variable of the number of charging piles in use and the number of charging piles in use at the prediction time; (2) A BP network is set up at the last layer of the DBN to receive the output feature vector of the RBM as its input feature vector. This layer feeds back the error information to each RBM layer from top to bottom, and fine-tunes the weight matrix and weight parameters of the entire DBN model in a supervised manner. (3) Input the number of times the charging station was used before the time to be tested, the weather conditions on the predicted day, the day type, the ambient temperature, the surrounding traffic index, and the number of times the charging station with the highest correlation was used for standardization processing; (4) Output the number of charging piles that will be used by the charging station in the short term in the future.

2. The method for predicting charging pile usage information based on multi-information coupling according to claim 1 is characterized in that: The charging station information includes but is not limited to any of the following: charging station status information, road condition information, historical operation data and weather information.

3. The method for predicting charging pile usage information based on multi-information coupling according to claim 2 is characterized in that: The charging station status information includes but is not limited to any of the following: location information of the charging station, and status information of charging piles already in use at the charging station.

4. The method for predicting charging pile usage information based on multi-information coupling according to claim 2 is characterized in that: The road condition information includes but is not limited to any of the following: road topology information, and traffic flow information of a section of road near a charging station.

5. The method for predicting charging pile usage information based on multi-information coupling according to claim 1, characterized in that: The highly relevant factors include but are not limited to any one of the following: day type, weather conditions and traffic, and geographical location.

6. The method for predicting charging pile usage information based on multi-information coupling according to claim 1, characterized in that: The input variables include the number of uses before the time to be tested, the weather conditions of the predicted day, the day type, the ambient temperature, the surrounding traffic index and the most relevant charging station usage.

Images