Electric vehicle charging pile load prediction method, system and storage medium

Abstract

The present invention relates to the technical field of electric vehicles, and discloses a method, system and storage medium for predicting the load of an electric vehicle charging pile, comprising: inputting a pre-constructed first neural network model into a charging pile to be analyzed, and adjusting the first neural network model according to target charging information of the charging pile to be analyzed to obtain a second neural network model; wherein the first neural network model is constructed by a first amount of charging information, and the difference between the amount of target charging information and the first amount is greater than a preset difference; collecting real-time charging information from the charging pile to be analyzed, and inputting the real-time charging information into the second neural network model, and obtaining the load state information of the charging pile to be analyzed output by the second neural network model; the present invention solves the problem that the charging pile has little data obtained and it is difficult to predict the load by migrating the neural network model.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicles, and in particular to a method, system and storage medium for predicting the load of an electric vehicle charging pile. Background Art

[0002] With the introduction of the concept of low-carbon environmental protection and the development of electric vehicle related technologies, hybrid vehicles and pure electric vehicles are rapidly becoming popular. In this context, in order to meet the power replenishment of the rapidly popularized hybrid vehicles and pure electric vehicles, the popularity of car charging piles will also be greatly improved. When using car charging piles to charge electric vehicles, electric vehicles will be connected to the AC power grid through car charging piles. At this time, the charging electric vehicles will generate power fluctuations on the power grid, thereby affecting the AC power grid that is not regulated. In severe cases, it will affect the power quality. Therefore, when using charging piles to charge electric vehicles, the charging piles need to be regulated as necessary to cope with the power fluctuations generated when the electric vehicles are charged. However, the regulation of existing charging piles is based on the load generated on the charging piles during the charging process of the electric vehicles, and the load generated by the electric vehicles on the charging piles is related to many factors. The existing charging piles mainly predict the load through prediction models, but the prediction models require a large amount of data to establish and verify. When the charging piles are used for a short time, the charging piles obtain less data and it is difficult to predict the load. Summary of the invention

[0003] The present invention provides a method, system and storage medium for predicting the load of an electric vehicle charging pile, so as to solve the above-mentioned defects and shortcomings in the prior art.

[0004] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0005] In a first aspect, the present invention provides a method for predicting the load of an electric vehicle charging pile, comprising:

[0006] Inputting a pre-constructed first neural network model into the charging pile to be analyzed, and adjusting the first neural network model according to target charging information of the charging pile to be analyzed to obtain a second neural network model;

[0007] The first neural network model is constructed by a first amount of charging information, the amount of target charging information of the charging pile to be analyzed is less than the first amount, and the difference between the amount of target charging information and the first amount is greater than a preset difference;

[0008] Collect real-time charging information from the charging pile to be analyzed, input the real-time charging information into the second neural network model, and obtain the load state information of the charging pile to be analyzed output by the second neural network model.

[0009] Optionally, before inputting the pre-built first neural network model into the charging pile to be analyzed, the method further includes:

[0010] Collecting a first amount of charging information from an original charging pile, and processing the first amount of charging information to obtain original training information;

[0011] Inputting the original training information into a pre-built original neural network model to train the original neural network model to obtain a first neural network model;

[0012] Wherein, the training of the original neural network model satisfies the following relationship:

[0013]

[0014] In the formula, f t is the forget gate vector, i t is the input gate vector, is the candidate state vector, C t is the cell state value at the current moment, C t-1 is the cell state value at the previous moment, o t is the output gate vector, is the output value or predicted value at the current moment, is the output value or predicted value of the previous moment, σ is the sigmoid activation function, is the hyperbolic tangent activation function, W f is the weight matrix of the forget gate, W i is the weight matrix of the input gate, W C is the weight matrix of the candidate state, W o is the weight matrix of the output gate, x t is the input at the current moment, b f is the bias of the forget gate, b i is the bias of the input gate, b C is the bias of the candidate state, b o is the bias of the output gate, · is the bitwise multiplication of the elements in the vector, and ⊙ is the Hadamard product symbol.

[0015] Optionally, the first amount of charging information includes original charging load data and original travel information, and the processing of the first amount of charging information to obtain original training information includes:

[0016] Obtaining original charging load data and original travel information from the first amount of charging information, and obtaining the rated charging power of the original charging pile;

[0017] The ratio of the original charging load data to the rated charging power is calculated, and the ratio and the original travel information are used as original training information.

[0018] Optionally, the method further includes:

[0019] Respectively obtaining an actual value and a predicted value output by the first neural network model and a number of cases for training the first amount of charging information;

[0020] The squares of the differences between the actual value and the predicted value are summed with the number of cases as the upper bound, and the ratio of the summed value to the number of cases is calculated, and then the square root of the ratio is taken to obtain the root mean square error index;

[0021] Inputting the root mean square error index into a target neural network model to verify the target neural network model;

[0022] Wherein, the target neural network model includes a first neural network model and a second neural network.

[0023] Optionally, the target charging information includes target charging load data and target travel information;

[0024] The step of adjusting the first neural network model according to the target charging information of the charging pile to be analyzed to obtain the second neural network model includes:

[0025] Removing the output layer in the first neural network model and freezing the training parameters in the first neural network model to obtain a relay neural network model;

[0026] A new output layer is added to the relay neural network model, and the target charging information is input into the relay neural network model to train the relay neural network model to obtain a second neural network model.

[0027] Optionally, the method further includes:

[0028] After unfreezing the training parameters in the relay neural network model and adjusting the learning rate parameters in the relay neural network model, the target charging information is input into the relay neural network model to train the relay neural network model.

[0029] In a second aspect, an embodiment of the present application provides an electric vehicle charging pile load prediction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the methods described in the first aspect when executing the computer program.

[0030] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the method steps described in the first aspect.

[0031] Beneficial effects:

[0032] The electric vehicle charging pile load prediction method provided by the present invention inputs a pre-constructed first neural network model into the charging pile to be analyzed, and uses the target charging information in the charging pile to be analyzed to adjust the first neural network model to obtain a second neural network model. When the charging pile to be analyzed is used to charge the electric vehicle, real-time charging information in the charging pile to be analyzed can be obtained, and the real-time charging information is input into the second neural network model to predict the load that the charging pile to be analyzed may receive during the charging process, thereby helping the charging pile to be analyzed to regulate the fluctuation of the power grid when charging the electric vehicle, and solving the problem that the charging pile to be analyzed with a short use time is difficult to predict the load. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a flow chart of a method for predicting the load of an electric vehicle charging pile according to a preferred embodiment of the present invention;

[0034] Figure 2 A schematic diagram of a simple structure of a charging pile and a power grid in a preferred embodiment of the present invention;

[0035] Figure 3 A diagram showing the fitting results of the load curve of the migration model provided in the preferred embodiment of the present invention;

[0036] Figure 4 It is a real-time flow chart of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0037] The technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the usual meanings understood by persons with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "one" do not indicate quantity restrictions, but indicate the existence of at least one. Words such as "connect" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship also changes accordingly.

[0039] See also Figure 1 The present application embodiment provides a method for predicting the load of an electric vehicle charging pile, comprising:

[0040] Inputting a pre-constructed first neural network model into the charging pile to be analyzed, and adjusting the first neural network model according to target charging information of the charging pile to be analyzed to obtain a second neural network model;

[0041] The first neural network model is constructed by a first amount of charging information, the amount of target charging information of the charging pile to be analyzed is less than the first amount, and the difference between the amount of target charging information and the first amount is greater than a preset difference;

[0042] Collect real-time charging information from the charging pile to be analyzed, input the real-time charging information into the second neural network model, and obtain the load state information of the charging pile to be analyzed output by the second neural network model.

[0043] In the above embodiment, a first neural network model that is easy to migrate is constructed based on a first amount of charging information, and the first neural network model is input into the charging pile to be analyzed. The first neural network model is adjusted using the target charging information collected from the charging pile to be analyzed. After the adjustment, a second neural network model that is suitable for the charging pile to be analyzed is obtained. Then, real-time charging information collected in real time during the operation of the charging pile to be analyzed is input into the second neural network model. The second neural network model can be used to predict the load state information of the charging pile to be analyzed during the operation. The first amount of charging information in this process is collected from charging piles that have been in use for a long time, and the first amount is much larger than the amount of target charging information.

[0044] Optionally, before inputting the pre-built first neural network model into the charging pile to be analyzed, the method further includes:

[0045] Collecting a first amount of charging information from an original charging pile, and processing the first amount of charging information to obtain original training information;

[0046] Inputting the original training information into a pre-built original neural network model to train the original neural network model to obtain a first neural network model;

[0047] Wherein, the training of the original neural network model satisfies the following relationship:

[0048]

[0049] In the formula, f t is the forget gate vector, i t is the input gate vector, is the candidate state vector, C t is the cell state value at the current moment, C t-1 is the cell state value at the previous moment, o t is the output gate vector, is the output value or predicted value at the current moment, is the output value or predicted value of the previous moment, σ is the sigmoid activation function, is the hyperbolic tangent activation function, W f is the weight matrix of the forget gate, W i is the weight matrix of the input gate, W C is the weight matrix of the candidate state, W o is the weight matrix of the output gate, x t is the input at the current moment, b f is the bias of the forget gate, b i is the bias of the input gate, b C is the bias of the candidate state, b o is the bias of the output gate, · is the bitwise multiplication of the elements in the vector, and ⊙ is the Hadamard product symbol.

[0050] In the above embodiment, the original neural network model is trained by using the original charging information so that the original neural network model meets the migration conditions, and the trained first neural network model is input into the charging pile to be analyzed through transfer learning;

[0051] The process of transfer learning satisfies the following relationship:

[0052]

[0053] Where D s and D t Represent the source domain data and target domain data respectively, X s and T srepresents the source domain samples and corresponding labels, X t and T t Represents the target domain samples and corresponding labels, Y s is the source domain prediction value, Y t is the predicted value of the target domain, f s Represents the sample X s To the source domain predicted value Y s Function mapping, f t Represents the sample X t To the target domain predicted value Y t Function mapping, δ s represents the model parameters of the source domain, δ t represents the model parameters of the target domain; transfer learning is achieved by s ,T s} to find its features and obtain the corresponding function mapping f s , and then map the function f s After transfer learning to the target domain, the target domain data {X t ,T t} to learn the function mapping f of the target domain t ; The relationship here is only an example and not a limitation.

[0054] Optionally, the first amount of charging information includes original charging load data and original travel information, and the processing of the first amount of charging information to obtain original training information includes:

[0055] Obtaining original charging load data and original travel information from the first amount of charging information, and obtaining the rated charging power of the original charging pile;

[0056] The ratio of the original charging load data to the rated charging power is calculated, and the ratio and the original travel information are used as original training information.

[0057] In the above embodiment, the original charging load data in the first amount of charging information is normalized to convert the original charging load data into training data suitable for training the original neural network model, and then the original neural network model is trained with the training data together with the original travel data to obtain the first neural network model;

[0058] Among them, the normalization process satisfies the following relationship:

[0059]

[0060] Where P nor is the normalized value of the original charging load data, P charge is the original charging load data, Pmax is the maximum rated charging power of the charging pile; the relationship here is only for example and not for limitation;

[0061] The neural network model obtained after normalization processing needs to be denormalized to obtain the normal load state information prediction value;

[0062] Among them, the anti-normalization process satisfies the following relationship:

[0063] P perd =P norpred ·P max ;

[0064] Where P perd is the predicted value of load status information, P norpred is the normalized prediction value output by the neural network model, P max is the maximum rated charging power of the charging pile; the relationship here is only for example and not for limitation.

[0065] Optionally, the method further includes:

[0066] Respectively obtaining an actual value and a predicted value output by the first neural network model and a number of cases for training the first amount of charging information;

[0067] The squares of the differences between the actual value and the predicted value are summed with the number of cases as the upper bound, and the ratio of the summed value to the number of cases is calculated, and then the square root of the ratio is taken to obtain the root mean square error index;

[0068] Inputting the root mean square error index into a target neural network model to verify the target neural network model;

[0069] Wherein, the target neural network model includes a first neural network model and a second neural network.

[0070] In the above embodiment, the first neural network model and the second neural network model can be verified respectively using the root mean square error index, thereby improving the accuracy of the trained first neural network model and the accuracy of the migrated second neural network model;

[0071] Among them, the root mean square error index satisfies the following relationship:

[0072]

[0073] In the formula, RMSE represents the root mean square error index, m represents the number of cases, and y i Indicates the actual value, Represents the predicted value; the relationship here is only an example and is not limiting.

[0074] Optionally, the target charging information includes target charging load data and target travel information;

[0075] The step of adjusting the first neural network model according to the target charging information of the charging pile to be analyzed to obtain the second neural network model includes:

[0076] Removing the output layer in the first neural network model and freezing the training parameters in the first neural network model to obtain a relay neural network model;

[0077] A new output layer is added to the relay neural network model, and the target charging information is input into the relay neural network model to train the relay neural network model to obtain a second neural network model.

[0078] In the above embodiment, a multi-layer long short-term memory network (LSTM) is a commonly used time recurrent neural network model. When the first neural network model is trained using target charging information with a small amount of data, only all LSTM layer structures in the first neural network model are retained, and the original output layer is removed. Then, a new output layer is added based on the original LSTM layer structure, and the target charging information is used to train it. At the same time, during the training process, the training parameters of all LSTM layers are frozen, and only the output layer parameters are kept in a trainable and adjustable state. Through this training method, the integrity of the first neural network model can be guaranteed during the training of the first neural network model using the target charging information, and at the same time, it can be trained into a second neural network model suitable for the actual charging situation of the charging pile to be analyzed.

[0079] Optionally, the method further includes:

[0080] After unfreezing the training parameters in the relay neural network model and adjusting the learning rate parameters in the relay neural network model, the target charging information is input into the relay neural network model to train the relay neural network model.

[0081] In the above embodiment, after the training of the first neural network model is completed, fine-tuning is still required, that is, unfreezing the top LSTM layer and adjusting the learning rate parameter of the neural network model to one tenth of the current value, and then training the first neural network model according to the adjusted learning rate parameter of the neural network model until the data reaches the minimum error. At this time, an accurate second neural network model can be obtained.

[0082] See also Figure 2-4In another embodiment, loads A and B are old charging piles with smart terminals, and load C is a new charging pile with a smart terminal to supplement the charging gap of residents' electric vehicles. All three charging piles are 37.5kw DC charging piles. The operation time of piles A and B has reached more than one year, and pile C has only been in operation for half a month. In this context, the load prediction model on piles A and B is migrated to the newly installed pile C to realize the charging load prediction of the charging pile C;

[0083] First, the multi-layer LSTM load prediction model of piles A and B is obtained. The model training and testing of piles A and B use more than 120,000 load data from May 2020 to July 2021 and more than 120,000 corresponding travel information (date nature, daily maximum temperature, minimum temperature and average temperature). The specific parameters of the two LSTM models are shown in Table 1:

[0084] Table 1

[0085] Parameter name parameter Number of layers 4 (3 LSTM layers, 1 Dense output layer) Number of hidden units in LSTM layer (128,64,32,16) Optimizer Adam Loss Function Mean Absolute Error (MAE) batch_size 50 epochs 200 Learning_rate 0.01

[0086] The root mean square error of the two models for the load value prediction for the next five minutes is about 2kw, and the root mean square error of the load value prediction for the next ten minutes is about 3.2kw. Then the two models are used as LSTM-1 candidate models A and B, and migrated to the C pile for training and testing. The data set used for training and testing is more than 5,000 load data and corresponding travel information (date nature, daily maximum temperature, minimum temperature and average temperature) collected by the C pile in more than half a month. During the training process, only the LSTM layer structure of the A and B models is migrated, and a new output layer is rebuilt. The trainable parameters of the LSTM layer are frozen, and the training parameters of the output layer are kept in a trainable state. More than 5,000 charging load data and corresponding travel information of the C pile are put into the A and B migration models for training. The error value results for the load value prediction for the next five minutes are shown in the first two columns of Table 2:

[0087] Table 2

[0088] Model code A Migration Model (5min) B Migration Model (5min) C Fine-tuning model (5min) Root mean square error (kw) 5.3465 4.6835 2.6753

[0089] From Table 2, we can see that the root mean square error of the predicted value of the migration model of pile B after training with the data samples of pile C is smaller. Therefore, it can be determined that the migration model of pile B is better and more suitable for migration to pile C. Finally, by unfreezing the trainable parameters of the top LSTM of the migration model of pile B and adjusting the learning rate to one tenth of the original, the model is trained and fine-tuned to obtain the C fine-tuning model. At this time, the root mean square error of the load value predicted by the C fine-tuning model for the next five minutes is shown in the last column of Table 2, and the fitting of its load curve is shown in the attached figure. Figure 3As shown in the figure, at this time, the load prediction accuracy of the C pile has been significantly improved, and with the accumulation of historical load information of the C pile, the model can continue to iterate and optimize, and will become more and more accurate. Therefore, the method of this embodiment can use the load prediction model of the similar charging piles in the area to migrate to the new charging pile, and use a small amount of data from the new charging pile to meet the requirements of charging load prediction. The implementation flow chart of the transfer learning solution is shown in the attached figure. Figure 4 As shown, this migration method can efficiently establish a load forecasting model for the new charging pile and accurately predict the charging load of the new charging pile.

[0090] An embodiment of the present application also provides an electric vehicle charging pile load prediction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above-described methods when executing the computer program.

[0091] An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0092] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for predicting the load of an electric vehicle charging pile, characterized in that: include: Inputting a pre-constructed first neural network model into the charging pile to be analyzed, and adjusting the first neural network model according to target charging information of the charging pile to be analyzed to obtain a second neural network model; The first neural network model is constructed by a first amount of charging information, the amount of target charging information of the charging pile to be analyzed is less than the first amount, and the difference between the amount of target charging information and the first amount is greater than a preset difference; Collecting real-time charging information from the charging pile to be analyzed, inputting the real-time charging information into the second neural network model, and obtaining load state information of the charging pile to be analyzed output by the second neural network model; Before inputting the pre-built first neural network model into the charging pile to be analyzed, the method further includes: Collecting a first amount of charging information from an original charging pile, and processing the first amount of charging information to obtain original training information; Inputting the original training information into a pre-built original neural network model to train the original neural network model to obtain a first neural network model; Wherein, the training of the original neural network model satisfies the following relationship: In the formula, f t is the forget gate vector, i t is the input gate vector, is the candidate state vector, C t is the cell state value at the current moment, C t-1 is the cell state value at the previous moment, o t is the output gate vector, h t is the output value or predicted value at the current moment, h t-1 is the output value or predicted value of the previous moment, σ is the sigmoid activation function, tanh is the hyperbolic tangent activation function, W f is the weight matrix of the forget gate, W i is the weight matrix of the input gate, W C is the weight matrix of the candidate state, W o is the weight matrix of the output gate, x t is the input at the current moment, b f is the bias of the forget gate, b i is the bias of the input gate, b C is the bias of the candidate state, b o is the bias of the output gate, · is the bitwise multiplication of the elements in the vector, and ⊙ is the Hadamard product symbol; The method further comprises: Respectively obtaining an actual value and a predicted value output by the first neural network model and a number of cases for training the first amount of charging information; The squares of the differences between the actual value and the predicted value are summed with the number of cases as the upper bound, and the ratio of the summed value to the number of cases is calculated, and then the square root of the ratio is taken to obtain the root mean square error index; Inputting the root mean square error index into a target neural network model to verify the target neural network model; Wherein, the target neural network model includes a first neural network model and a second neural network; The target charging information includes target charging load data and target travel information; The step of adjusting the first neural network model according to the target charging information of the charging pile to be analyzed to obtain the second neural network model includes: Removing the output layer in the first neural network model and freezing the training parameters in the first neural network model to obtain a relay neural network model; A new output layer is added to the relay neural network model, and the target charging information is input into the relay neural network model to train the relay neural network model to obtain a second neural network model.

2. The electric vehicle charging pile load prediction method according to claim 1, characterized in that: The first amount of charging information includes original charging load data and original travel information, and the processing of the first amount of charging information to obtain original training information includes: Obtaining original charging load data and original travel information from the first amount of charging information, and obtaining the rated charging power of the original charging pile; The ratio of the original charging load data to the rated charging power is calculated, and the ratio and the original travel information are used as original training information.

3. The electric vehicle charging pile load prediction method according to claim 1, characterized in that: The method further comprises: After unfreezing the training parameters in the relay neural network model and adjusting the learning rate parameters in the relay neural network model, the target charging information is input into the relay neural network model to train the relay neural network model.

4. A system for predicting the load of an electric vehicle charging pile, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method described in any one of claims 1 to 3 are implemented.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method steps described in any one of claims 1 to 3 are implemented.

Images