Charging station load prediction method and device based on BiConvLSTM

Abstract

The invention discloses a BiConvLSTM-based charging station load prediction method and device, and relates to the field of energy demand load prediction of an electric vehicle charging station network, and the method comprises the steps: constructing a training data set, and screening the features of data as input data; an Encoder module with a BiConvLSTM neural network layer as a basic framework is constructed; the method comprises the following steps of: constructing a Decoder module which takes BiLSTM as a basic framework; and establishing a deep learning electric vehicle charging station prediction model based on the Encoder module and the Decoder module, training the established model by adopting the training data set, and realizing charging station load real-time prediction based on the trained deep learning electric vehicle charging station prediction model. According to the method, convolution and loop operation can be effectively utilized to enhance the capturing capability of the spatial-temporal characteristics, so that the prediction accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of energy demand load forecasting of an electric vehicle charging station network, and specifically to a charging station load forecasting method and device based on BiConvLSTM. Background Art

[0002] Electric vehicle charging stations are sites for charging electric vehicles. With the popularity of electric vehicles, electric vehicle charging stations will inevitably become the focus of the development of the automotive industry and the energy industry. Electric vehicle charging stations can better solve the problem of fast charging while saving energy and reducing emissions.

[0003] With the popularity of electric vehicles, load forecasting of charging station networks has become critical. Traditional linear models perform poorly in handling the high dynamics and uncertainty of electric vehicle charging demand, while existing deep learning models have improved but still fall short in capturing spatiotemporal correlations. Summary of the invention

[0004] The present application provides a charging station load prediction method and device based on BiConvLSTM, which can effectively utilize convolution and loop operations to enhance the ability to capture spatiotemporal features, thereby improving prediction accuracy.

[0005] In a first aspect, an embodiment of the present application provides a charging station load prediction method based on BiConvLSTM, and the charging station load prediction method based on BiConvLSTM includes:

[0006] Based on the electric vehicle charging station network, historical charging data and corresponding meteorological data are collected, and the collected data are preprocessed to construct a training data set, and the characteristics of the data are screened as input data;

[0007] Construct an Encoder module based on the BiConvLSTM neural network layer, and extract information about the input data in spatial and temporal dimensions through the constructed Encoder module;

[0008] Construct a Decoder module based on BiLSTM architecture, decode the encoder output through the constructed Decoder module, and input the decoded output into the time series distribution fully connected layer to obtain the final load forecast value;

[0009] A deep learning electric vehicle charging station prediction model is established based on the Encoder module and the Decoder module, and the established model is trained using a training data set. Based on the trained deep learning electric vehicle charging station prediction model, real-time prediction of the charging station load is achieved.

[0010] In combination with the first aspect, in one embodiment, the historical charging data and corresponding meteorological data are collected based on the electric vehicle charging station network, and the collected data are preprocessed to construct a training data set, and the characteristics of the data are screened as input data, specifically including:

[0011] Collect historical charging data and related meteorological data from the communication infrastructure of each charging station, integrate the characteristic data of the collected data through the data warehouse, and form the original data set after pre-processing the characteristic data;

[0012] Resample the data in the constructed original data set to obtain the daily total load of the electric vehicle charging station, and use the daily total load as the data label;

[0013] Perform data normalization on the data in the original data set to adjust the numerical differences between data points and finally obtain the training data set;

[0014] For the obtained training data set, the total daily load of the electric vehicle charging station is used as the data label, and the EVCS attributes, EV attributes and charging event data of the screened data are used as the input data.

[0015] In combination with the first aspect, in one implementation, the data in the original data set is subjected to a data normalization operation, and the specific operation method is as follows:

[0016] z= x-μ

[0017] σ

[0018] Among them, x represents the original value, μ represents the mean of x, σ represents the standard deviation of the original data set, and z represents the normalized value.

[0019] In combination with the first aspect, in one implementation,

[0020] The EVCS attributes include site ID, site name, site location address, postal code of the area where the site is located, site MAC address, site longitude and latitude, and charging level of the site charging port;

[0021] The EV attribute is information about the electric vehicle, and the information about the electric vehicle includes an electric vehicle ID and a postal code of a registration location of the electric vehicle;

[0022] The charging event data includes the start date and time of the charging event, the end date and time of the charging event, the charging duration, the average residence time, the initial state of charge, the final SOC, the energy transferred, the greenhouse gas savings, and the charging event end information;

[0023] The charging event end information includes an active charging stop event and a server-terminated charging event.

[0024] In combination with the first aspect, in one implementation, constructing an Encoder module based on a BiConvLSTM neural network layer, and extracting information of input data in spatial and temporal dimensions through the constructed Encoder module, specifically includes:

[0025] A fully connected layer and a BiConvLSTM layer are stacked to construct an encoder module. The input layer of the encoder module is specifically used to receive a five-dimensional tuple of shape [n, T, r, c, ch], where n represents the number of samples, T represents the time step, r represents the number of feature rows, c represents the number of feature map columns, and ch represents the number of channels. The input matrix of the BiConvLSTM layer is a one-dimensional sequence of total power consumption of the charging station. The input matrix of each time step is regarded as an image in a tabular format, and a spatiotemporal convolution operation is performed to obtain a two-dimensional feature map output;

[0026] The ConvLSTM unit of the backward layer of BiConvLSTM is set to capture the spatiotemporal features in two directions based on the forward layer, so that the ConvLSTM unit of the forward layer can exchange sequence knowledge with the backward layer;

[0027] Based on the data flow operation of bidirectional spatiotemporal information and the final output, the information of the input data in spatial and temporal dimensions is extracted.

[0028] In combination with the first aspect, in one implementation, the construction of a Decoder module based on the BiLSTM architecture, decoding the encoder output through the constructed Decoder module, and inputting the decoded output into a time series distribution fully connected layer to obtain a final load prediction value, specifically includes:

[0029] A flattened fully connected layer, a repeated vector layer, and three BiLSTM layers are cascaded to construct a decoder module. The flattened layer of the decoder module is used to receive the feature map output of the encoding layer and flatten the feature map output into a one-dimensional vector with a length of 1536. The repeated vector layer copies and expands the one-dimensional vector into a 1536×7 matrix. The three BiLSTM layers are used to receive the 1536×7 matrix and output the final daily charging load forecast value.

[0030] The data flow operation of the BiLSTM layer is expressed as:

[0031]

[0032] Among them, the hidden state H of BiLSTM at time t t include and represents the hidden state of the forward ConvLSTM unit, represents the hidden state of the backward ConvLSTM unit, LSTM represents the long short-term memory neural network, H t+1 represents the output of the next unit, X t represents the input of the current unit, C t+1 Indicates the state of the next unit, H t-1 represents the output of the previous unit, C t-1 Indicates the status of the previous unit.

[0033] In combination with the first aspect, in one implementation, a deep learning electric vehicle charging station prediction model is established based on the Encoder module and the Decoder module, and the established model is trained using a training data set, and real-time prediction of charging station load is achieved based on the trained deep learning electric vehicle charging station prediction model, specifically including:

[0034] The encoder module constructed by stacking a fully connected layer and a BiConvLSTM layer is combined with the decoder module constructed by cascading a flattened fully connected layer, a repeated vector layer, and three BiLSTM layers to obtain a deep learning electric vehicle charging station prediction model;

[0035] The constructed training dataset is divided into a training set and a test set to train the deep learning electric vehicle charging station prediction model;

[0036] The real-time data of the target charging station to be predicted is input into the trained deep learning electric vehicle charging station prediction model to predict the future charging load of the charging station.

[0037] In a second aspect, an embodiment of the present application provides a charging station load prediction device based on BiConvLSTM, and the charging station load prediction device based on BiConvLSTM includes:

[0038] The acquisition module is used to collect historical charging data and corresponding meteorological data based on the electric vehicle charging station network, and construct a training data set after preprocessing the collected data, and at the same time screen the characteristics of the data as input data;

[0039] A construction module is used to construct an Encoder module based on the BiConvLSTM neural network layer, and extract information about the input data in the spatial and temporal dimensions through the constructed Encoder module;

[0040] Create a module for building a Decoder module based on BiLSTM architecture, decode the encoder output through the built Decoder module, and input the decoded output into the time series distribution fully connected layer to obtain the final load forecast value;

[0041] The prediction module is used to establish a deep learning electric vehicle charging station prediction model based on the Encoder module and the Decoder module, and use the training data set to train the established model, and realize the real-time prediction of the charging station load based on the trained deep learning electric vehicle charging station prediction model.

[0042] In conjunction with the second aspect, in one embodiment, the historical charging data and corresponding meteorological data are collected based on the electric vehicle charging station network, and the collected data are preprocessed to construct a training data set, and the characteristics of the data are screened as input data, specifically including:

[0043] Collect historical charging data and related meteorological data from the communication infrastructure of each charging station, integrate the characteristic data of the collected data through the data warehouse, and form the original data set after pre-processing the characteristic data;

[0044] Resample the data in the constructed original data set to obtain the daily total load of the electric vehicle charging station, and use the daily total load as the data label;

[0045] Perform data normalization on the data in the original data set to adjust the numerical differences between data points and finally obtain the training data set;

[0046] For the obtained training data set, the total daily load of the electric vehicle charging station is used as the data label, and the EVCS attributes, EV attributes and charging event data of the screened data are used as the input data.

[0047] In conjunction with the second aspect, in one implementation, the data in the original data set is subjected to a data normalization operation, and a specific operation method is as follows:

[0048]

[0049] Among them, x represents the original value, μ represents the mean of x, σ represents the standard deviation of the original data set, and z represents the normalized value.

[0050] The beneficial effects brought by the technical solution provided by the embodiments of the present application include:

[0051] By constructing an encoder network containing multiple ConvLSTM layers to extract spatiotemporal features, and then using multiple BiLSTM layers as a decoder network to generate prediction results, convolution and recurrent operations are used to enhance the ability to capture spatiotemporal features, thereby improving prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of the charging station load prediction method based on BICONVLSTM in this application;

[0053] Figure 2 This is a functional module diagram of a charging station load prediction device based on BICONVLSTM in this application;

[0054] Figure 3 This is a schematic diagram of the hardware structure of the charging station load prediction device based on BICONVLSTM in this application. DETAILED DESCRIPTION

[0055] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0056] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0057] On the first aspect, the embodiment of the present application provides a charging station load forecasting method based on BICONVLSTM, which can effectively utilize convolution and loop operations to enhance the ability to capture spatiotemporal features, thereby improving the prediction accuracy.

[0058] In one embodiment, referring to Figure 1 , Figure 1 This is a flow chart of the charging station load forecasting method based on BICONVLSTM in this application. Figure 1 As shown in the figure, the charging station load prediction method based on BICONVLSTM includes:

[0059] S1: Collect historical charging data and corresponding meteorological data based on the electric vehicle charging station network, construct a training data set after preprocessing the collected data, and filter the data features as input data;

[0060] Furthermore, in one embodiment, historical charging data and corresponding meteorological data are collected based on the electric vehicle charging station network, and a training data set is constructed after preprocessing the collected data, and the characteristics of the data are screened as input data, specifically including:

[0061] S101: Collect historical charging data and related meteorological data from the communication infrastructure of each charging station, integrate the characteristic data of the collected data through the data warehouse, and pre-process the characteristic data to form an original data set; wherein the pre-processing is to delete data noise points and fill in data missing items using the average method;

[0062] S102: resampling the data in the constructed original data set to obtain the daily total load of the electric vehicle charging station, and using the daily total load as a data label;

[0063] S103: performing a data normalization operation on the data in the original data set to adjust the numerical differences between the data points, and finally obtaining a training data set;

[0064] Specifically, the data in the original data set is normalized, and the specific operation method is as follows:

[0065]

[0066] Among them, x represents the original value, μ represents the mean value of x, σ represents the standard deviation of the original data set, and z represents the normalized value. The distribution is adjusted by subtracting the mean of each input variable and dividing by the standard deviation so that the mean value is zero and the standard deviation is one, thereby improving the model training performance and speeding up the model training.

[0067] S104: For the obtained training data set, the total daily load of the electric vehicle charging station is used as a data label, and the EVCS (Electric Vehicle Charging Station) attribute, EV (Electric Vehicle) attribute and charging event data of the data are screened as input data.

[0068] Specifically, the EVCS attributes include site ID, site name, site location address, postal code of the site area, site MAC address, site longitude and latitude, and charging level of the site charging port; the EV attributes are the information of the electric vehicle, and the information of the electric vehicle includes the electric vehicle ID and the postal code of the electric vehicle registration place; the charging event data includes the start date and time of the charging event, the end date and time of the charging event, the charging duration, the average residence time, the initial state of charge, the final SOC (State of Charge), the transferred energy, the greenhouse gas savings, and the charging event end information; the charging event end information includes the active stop charging event and the server terminated charging event.

[0069] S2: Construct an Encoder module based on the BiConvLSTM neural network layer to extract information about the input data in spatial and temporal dimensions. BiConvLSTM, or bi-directional long short-term memory, uses a temporal convolution bidirectional long short-term memory model.

[0070] Furthermore, in one embodiment, an Encoder module based on the BiConvLSTM neural network layer is constructed, and the information of the input data in the spatial and temporal dimensions is extracted through the constructed Encoder module, specifically including:

[0071] S201: stacking a fully connected layer and a BiConvLSTM layer to construct an encoder module, wherein the input layer of the encoder module is specifically used to receive a five-dimensional tuple of shape [n, T, r, c, ch], wherein n represents the number of samples, T represents the time step, r represents the number of feature rows, c represents the number of feature map columns, ch represents the number of channels, and the input matrix of the BiConvLSTM layer is a one-dimensional sequence of total power consumption of the charging station. The input matrix of each time step is regarded as an image in a tabular format, and a spatiotemporal convolution operation is performed to obtain a two-dimensional feature map output;

[0072] S202: Setting the ConvLSTM (an architecture combining convolutional neural network and long short-term memory network) unit of the backward layer of BiConvLSTM to capture the spatiotemporal features in two directions based on the forward layer, so that the ConvLSTM unit of the forward layer can exchange sequence knowledge with the backward layer;

[0073] It should be noted that the difference between the input of BiConvLSTM and ConvLSTM is that it can flow in two directions and has the potential to utilize information on both sides. When ConvLSTM starts training with the arrival of input, if the input gate is activated, the new data will be accumulated into the memory cell. Similarly, if the forget gate is opened, the past cell state can be forgotten. Finally, whether the latest memory cell value of the past cell state will be passed to the final state is further controlled by the output gate. Due to this unique structure, ConvLSTM can extract more useful feature representations than CNN. In addition, compared with LSTM, the implementation of the three-gate mechanism uses a three-dimensional tensor to expand the one-dimensional to multi-dimensional convolution operation.

[0074] S203: Based on the data flow operation of the bidirectional spatiotemporal information and the final output, the information of the input data in the spatial and temporal dimensions is extracted.

[0075] The ConvLSTM unit of the forward layer receives the spatial feature map as input and outputs the feature map in the forward order according to the standard data flow operation of the ConvLSTM layer. The deeper layer consists of the backward unit, which receives the output features from the forward layer as input, and then combines the forward features and the backward features into the final output using the equation. In this way, information is encouraged to flow between the forward and backward ConvLSTM units, and deeper spatiotemporal features can be extracted through the backward unit.

[0076] S3: Construct a Decoder module based on BiLSTM architecture, decode the encoder output through the constructed Decoder module, and input the decoded output into the time series distribution fully connected layer to obtain the final load forecast value;

[0077] Furthermore, in one embodiment, a Decoder module based on BiLSTM is constructed, the encoder output is decoded by the constructed Decoder module, and the decoded output is input into a time series distribution full connection layer to obtain the final load prediction value, specifically including:

[0078] S301: cascade a flattened fully connected layer, a repeated vector layer, and three BiLSTM layers to construct a decoder module, wherein the flattening layer of the decoder module is used to receive the feature map output of the encoding layer and flatten the feature map output into a one-dimensional vector with a length of 1536, the repeated item vector layer copies and expands the one-dimensional vector into a 1536×7 matrix, and the three BiLSTM layers are used to receive the 1536×7 matrix and output the final daily charging load forecast value;

[0079] S302: The data flow operation for the BiLSTM layer is expressed as:

[0080]

[0081] Among them, the hidden state H of BiLSTM at time t t include and represents the hidden state of the forward ConvLSTM unit, represents the hidden state of the backward ConvLSTM unit, LSTM represents the long short-term memory neural network, H t+1 represents the output of the next unit, X t represents the input of the current unit, C t+1 Indicates the state of the next unit, H t-1 represents the output of the previous unit, C t-1 Indicates the status of the previous unit.

[0082] Bidirectional Long Short-Term Memory (BiLSTM) is a variant of the regular LSTM that can process both past and future information, whereas the conventional LSTM can only pass information in one direction. The working principle of BiLSTM is that the same output connects two LSTM networks, but in opposite directions. The forward LSTM can receive information about the past data of the input sequence, while the backward LSTM can obtain information about the future data of the input sequence. Each layer of the bidirectional LSTM network passes the output of the forward and backward layers to the activation layer (neural network), and considers the output of this activation layer, which also includes information about or the relationship between past and future data points.

[0083] S4: A deep learning electric vehicle charging station prediction model is established based on the Encoder module and the Decoder module, and the established model is trained using a training data set. Based on the trained deep learning electric vehicle charging station prediction model, real-time prediction of the charging station load is achieved.

[0084] Furthermore, in one embodiment, a deep learning electric vehicle charging station prediction model is established based on the Encoder module and the Decoder module, and the established model is trained using a training data set. Based on the trained deep learning electric vehicle charging station prediction model, real-time prediction of charging station load is achieved, specifically including:

[0085] S401: combining an encoder module constructed by stacking a fully connected layer and a BiConvLSTM layer with a decoder module constructed by cascading a flattened fully connected layer, a repeated vector layer, and three BiLSTM layers to obtain a deep learning electric vehicle charging station prediction model;

[0086] S402: Divide the constructed training data set into a training set and a test set, and train the deep learning electric vehicle charging station prediction model;

[0087] S403: Input the real-time data of the target charging station to be predicted into the trained deep learning electric vehicle charging station prediction model to predict the future charging load of the charging station.

[0088] The BiConvLSTM-based charging station load prediction method of the embodiment of the present application constructs an encoder network including multiple ConvLSTM layers to extract spatiotemporal features, and then uses multiple BiLSTM layers as a decoder network to generate prediction results, and uses convolution and loop operations to enhance the ability to capture spatiotemporal features, thereby improving prediction accuracy.

[0089] In a second aspect, an embodiment of the present application also provides a charging station load prediction device based on BICONVLSTM.

[0090] In one embodiment, referring to Figure 2 , Figure 2 This is a functional module diagram of the charging station load prediction device based on BICONVLSTM in this application. Figure 2 As shown, the charging station load prediction device based on BICONVLSTM includes: an acquisition module, a construction module, a creation module, and a prediction module.

[0091] The acquisition module is used to collect historical charging data and corresponding meteorological data based on the electric vehicle charging station network, and construct a training data set after preprocessing the collected data, and at the same time screen the characteristics of the data as input data; the construction module is used to construct an Encoder module based on the BiConvLSTM neural network layer, and extract the information of the input data in the spatial and temporal dimensions through the constructed Encoder module; the creation module is used to construct a Decoder module based on the BiLSTM architecture, and decode the encoder output through the constructed Decoder module, and input the decoded output into the time series distribution fully connected layer to obtain the final load prediction value; the prediction module is used to establish a deep learning electric vehicle charging station prediction model based on the Encoder module and the Decoder module, and use the training data set to train the established model, and realize the real-time prediction of the charging station load based on the trained deep learning electric vehicle charging station prediction model.

[0092] In this application, historical charging data and corresponding meteorological data are collected based on the electric vehicle charging station network, and the collected data are preprocessed to construct a training data set, and the characteristics of the data are screened as input data, specifically including:

[0093] Collect historical charging data and related meteorological data from the communication infrastructure of each charging station, integrate the characteristic data of the collected data through the data warehouse, and form the original data set after pre-processing the characteristic data;

[0094] Resample the data in the constructed original data set to obtain the daily total load of the electric vehicle charging station, and use the daily total load as the data label;

[0095] Perform data normalization on the data in the original data set to adjust the numerical differences between data points and finally obtain the training data set;

[0096] For the obtained training data set, the total daily load of the electric vehicle charging station is used as the data label, and the EVCS attributes, EV attributes and charging event data of the screened data are used as the input data.

[0097] In this application, data normalization is performed on the data in the original data set. The specific operation method is as follows:

[0098]

[0099] Among them, x represents the original value, μ represents the mean of x, σ represents the standard deviation of the original data set, and z represents the normalized value.

[0100] In a third aspect, an embodiment of the present application provides a charging station load prediction device based on BICONVLSTM. The charging station load prediction device based on BICONVLSTM can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0101] Reference Figure 3 , Figure 3 The hardware structure diagram of the charging station load prediction device based on BICONVLSTM involved in the embodiment of the present application is shown in FIG. In the embodiment of the present application, the charging station load prediction device based on BICONVLSTM may include a processor, a memory, a communication interface and a communication bus.

[0102] The communication bus may be of any type and is used to interconnect the processor, the memory, and the communication interface.

[0103] The communication interface includes input / output (I / O) interface, physical interface and logical interface, etc., which are used to realize the interconnection of devices inside the charging station load forecasting device based on BICONVLSTM, and the interface used to realize the interconnection of the charging station load forecasting device based on BICONVLSTM with other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, a fiber optic interface, an ATM interface, etc.; the user device can be a display (Display), a keyboard (Keyboard), etc.

[0104] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0105] The processor may be a general-purpose processor, which may call the BICONVLSTM-based charging station load prediction program stored in the memory and execute the BICONVLSTM-based charging station load prediction method provided in the embodiment of the present application. For example, the general-purpose processor may be a central processing unit (CPU). The method executed when the BICONVLSTM-based charging station load prediction program is called may refer to the various embodiments of the BICONVLSTM-based charging station load prediction method of the present application, which will not be repeated here.

[0106] Those skilled in the art will understand that Figure 3 The hardware structure shown in the figure does not constitute a limitation on the present application, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0107] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0108] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0109] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0110] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0111] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0112] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A charging station load prediction method based on BiConvLSTM, characterized in that: The BiConvLSTM-based charging station load prediction method includes: Based on the electric vehicle charging station network, historical charging data and corresponding meteorological data are collected, and the collected data are preprocessed to construct a training data set, and the characteristics of the data are screened as input data; Construct an Encoder module based on the BiConvLSTM neural network layer, and extract information about the input data in spatial and temporal dimensions through the constructed Encoder module; Construct a Decoder module based on BiLSTM architecture, decode the encoder output through the constructed Decoder module, and input the decoded output into the time series distribution fully connected layer to obtain the final load forecast value; A deep learning electric vehicle charging station prediction model is established based on the Encoder module and the Decoder module, and the established model is trained using a training data set. Based on the trained deep learning electric vehicle charging station prediction model, real-time prediction of the charging station load is achieved.

2. A charging station load prediction method based on BiConvLSTM as claimed in claim 1, characterized in that: The method collects historical charging data and corresponding meteorological data based on the electric vehicle charging station network, constructs a training data set after preprocessing the collected data, and screens the characteristics of the data as input data, specifically including: Collect historical charging data and related meteorological data from the communication infrastructure of each charging station, integrate the characteristic data of the collected data through the data warehouse, and form the original data set after pre-processing the characteristic data; Resample the data in the constructed original data set to obtain the daily total load of the electric vehicle charging station, and use the daily total load as the data label; Perform data normalization on the data in the original data set to adjust the numerical differences between data points and finally obtain the training data set; For the obtained training data set, the total daily load of the electric vehicle charging station is used as the data label, and the EVCS attributes, EV attributes and charging event data of the screened data are used as the input data.

3. A charging station load prediction method based on BiConvLSTM as claimed in claim 2, characterized in that: The data normalization operation is performed on the data in the original data set, and the specific operation method is as follows: z＝ x-μ σ Among them, x represents the original value, μ represents the mean of x, σ represents the standard deviation of the original data set, and z represents the normalized value.

4. A charging station load prediction method based on BiConvLSTM as claimed in claim 2, characterized in that: The EVCS attributes include site ID, site name, site location address, postal code of the area where the site is located, site MAC address, site longitude and latitude, and charging level of the site charging port; The EV attribute is information about the electric vehicle, and the information about the electric vehicle includes an electric vehicle ID and a postal code of a registration location of the electric vehicle; The charging event data includes the start date and time of the charging event, the end date and time of the charging event, the charging duration, the average residence time, the initial state of charge, the final SOC, the energy transferred, the greenhouse gas savings, and the charging event end information; The charging event end information includes an active charging stop event and a server-terminated charging event.

5. A charging station load prediction method based on BiConvLSTM as claimed in claim 1, characterized in that: The encoder module based on the BiConvLSTM neural network layer is constructed to extract information of the input data in the spatial and temporal dimensions, specifically including: A fully connected layer and a BiConvLSTM layer are stacked to construct an encoder module. The input layer of the encoder module is specifically used to receive a five-dimensional tuple of shape [n, T, r, c, ch], where n represents the number of samples, T represents the time step, r represents the number of feature rows, c represents the number of feature map columns, and ch represents the number of channels. The input matrix of the BiConvLSTM layer is a one-dimensional sequence of total power consumption of the charging station. The input matrix of each time step is regarded as an image in a tabular format, and a spatiotemporal convolution operation is performed to obtain a two-dimensional feature map output; The ConvLSTM unit of the backward layer of BiConvLSTM is set to capture the spatiotemporal features in two directions based on the forward layer, so that the ConvLSTM unit of the forward layer can exchange sequence knowledge with the backward layer; Based on the data flow operation of bidirectional spatiotemporal information and the final output, the information of the input data in spatial and temporal dimensions is extracted.

6. A charging station load prediction method based on BiConvLSTM as claimed in claim 5, characterized in that: The decoder module based on BiLSTM is constructed, the encoder output is decoded by the constructed decoder module, and the decoded output is input into the time series distribution full connection layer to obtain the final load prediction value, specifically including: A flattened fully connected layer, a repeated vector layer, and three BiLSTM layers are cascaded to construct a decoder module. The flattened layer of the decoder module is used to receive the feature map output of the encoding layer and flatten the feature map output into a one-dimensional vector with a length of 1536. The repeated vector layer copies and expands the one-dimensional vector into a 1536×7 matrix. The three BiLSTM layers are used to receive the 1536×7 matrix and output the final daily charging load forecast value. The data flow operation of the BiLSTM layer is expressed as: Among them, the hidden state H of BiLSTM at time t t include and represents the hidden state of the forward ConvLSTM unit, represents the hidden state of the backward ConvLSTM unit, LSTM represents the long short-term memory neural network, H t+1 represents the output of the next unit, X t represents the input of the current unit, C t+1 Indicates the state of the next unit, H t-1 represents the output of the previous unit, C t-1 Indicates the status of the previous unit.

7. A charging station load prediction method based on BiConvLSTM as claimed in claim 6, characterized in that: The deep learning electric vehicle charging station prediction model is established based on the Encoder module and the Decoder module, and the established model is trained using a training data set. The real-time prediction of the charging station load is realized based on the trained deep learning electric vehicle charging station prediction model, which specifically includes: The encoder module constructed by stacking a fully connected layer and a BiConvLSTM layer is combined with the decoder module constructed by cascading a flattened fully connected layer, a repeated vector layer, and three BiLSTM layers to obtain a deep learning electric vehicle charging station prediction model; The constructed training dataset is divided into a training set and a test set to train the deep learning electric vehicle charging station prediction model; The real-time data of the target charging station to be predicted is input into the trained deep learning electric vehicle charging station prediction model to predict the future charging load of the charging station.

8. A charging station load prediction device based on BiConvLSTM, characterized in that: The BiConvLSTM-based charging station load prediction device includes: The acquisition module is used to collect historical charging data and corresponding meteorological data based on the electric vehicle charging station network, and construct a training data set after preprocessing the collected data, and at the same time screen the characteristics of the data as input data; A construction module is used to construct an Encoder module based on the BiConvLSTM neural network layer, and extract information about the input data in the spatial and temporal dimensions through the constructed Encoder module; Create a module for building a Decoder module based on BiLSTM architecture, decode the encoder output through the built Decoder module, and input the decoded output into the time series distribution fully connected layer to obtain the final load forecast value; The prediction module is used to establish a deep learning electric vehicle charging station prediction model based on the Encoder module and the Decoder module, and use the training data set to train the established model, and realize the real-time prediction of the charging station load based on the trained deep learning electric vehicle charging station prediction model.

9. A charging station load prediction device based on BiConvLSTM as claimed in claim 8, characterized in that: The method collects historical charging data and corresponding meteorological data based on the electric vehicle charging station network, constructs a training data set after preprocessing the collected data, and screens the characteristics of the data as input data, specifically including: Collect historical charging data and related meteorological data from the communication infrastructure of each charging station, integrate the characteristic data of the collected data through the data warehouse, and form the original data set after pre-processing the characteristic data; Resample the data in the constructed original data set to obtain the daily total load of the electric vehicle charging station, and use the daily total load as the data label; Perform data normalization on the data in the original data set to adjust the numerical differences between data points and finally obtain the training data set; For the obtained training data set, the total daily load of the electric vehicle charging station is used as the data label, and the EVCS attributes, EV attributes and charging event data of the screened data are used as the input data.

10. A charging station load prediction device based on BiConvLSTM as claimed in claim 9, characterized in that: The data normalization operation is performed on the data in the original data set, and the specific operation method is as follows: z＝ x-μ σ Among them, x represents the original value, μ represents the mean of x, σ represents the standard deviation of the original data set, and z represents the normalized value.

Images