Efficient power grid data transmission method and system based on dual-prediction architecture
By using a dual prediction architecture and LSTM network in the smart grid for data preprocessing and prediction, only sending real data when the prediction error is greater than the threshold, the problem of large data transmission volume and accuracy in the smart grid is solved, and efficient Data transmission and bandwidth mitigation.
Patent Information
- Application Number
- CN202411162701.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-23
- Publication Date
- 2024-11-26
AI Technical Summary
In smart grids, as the number of applications and sensors increases, the amount of data transmission in the grid is huge. The existing technology is difficult to meet bandwidth requirements and cannot ensure the accuracy of data transmission, affecting application performance.
Using an efficient power grid data transmission method based on a dual prediction architecture, the edge gateway and cloud server deploy the same prediction model, and data preprocessing and prediction are performed through the LSTM network, and real data is sent only when the prediction error is greater than the threshold, reducing the amount of data transmission.
While ensuring data accuracy, the amount of data transmitted by the edge computing layer to the cloud computing layer is reduced, the bandwidth pressure of the power grid is alleviated, and data transmission efficiency and application performance are improved.
Smart Images

Figure CN119030886A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to communication transmission technology, and in particular to a high-efficiency power grid data transmission method and system based on a dual prediction architecture. Background Art
[0002] In smart grids, a large number of applications are deployed on cloud platforms. These applications, such as real-time monitoring of grid status and real-time equipment operation status evaluation, require a large amount of rapidly changing time series data, including current, voltage, frequency and other information, in order to understand the actual operation of the grid and issue timely alarms when abnormal conditions occur. However, with the increasing number of applications and sensors, the amount of data transmission has become very large. The limited bandwidth of the grid can hardly meet such a large amount of online data transmission. At the same time, the performance of the application is directly related to the accuracy of the acquired data. Therefore, the efficiency of data transmission must be improved to alleviate the pressure on the grid bandwidth and meet the requirements of data accuracy.
[0003] Time series data includes the specific value of the physical quantity and the timestamp that records the time when the value was collected. Through experiments, it is found that the space occupied by the timestamp is much larger than the value, and ignoring the timestamp will greatly increase the amount of data transmission.
[0004] Most of the current work is generally based on filtering, sampling and compression, and expands and integrates these three basic methods under specific problems to achieve the purpose of reducing data transmission. However, these works are not suitable for rapidly changing data, and they cannot ensure that the data obtained by cloud computing has sufficient accuracy, which directly affects the performance of the application. Summary of the invention
[0005] In view of the problems existing in the power grid data transmission in the prior art, the present invention proposes an efficient power grid data transmission method based on a dual prediction architecture. The edge gateway collects data from the terminal device and pre-processes the data. The edge gateway sends the pre-processed data to the cloud server. The data transmission process between the edge gateway and the cloud server includes the following steps:
[0006] The same prediction model is deployed on the edge gateway and the cloud server. The edge gateway and the cloud server input the same data into the prediction model to obtain the same output.
[0007] The edge gateway starts the socket server and waits for connection. When the cloud server successfully connects to the edge gateway, the edge gateway collects data from the terminal device and pre-processes it.
[0008] The edge gateway sends one data in each data sending cycle. If the cumulative number of data sent is less than the input sequence length n of the prediction model, the collected data will continue to be sent in the next cycle;
[0009] If the cumulative number of data sent is greater than or equal to the input sequence length n of the prediction model, the edge gateway uses the first n data of the current data as the input of the prediction model to obtain the prediction value of the current data;
[0010] The edge gateway compares the absolute error between the current data and its predicted value. If the absolute error is less than the set threshold, the edge gateway will not send data to the cloud server in the current data sending cycle. Otherwise, the edge gateway will send the real data to the cloud server.
[0011] If the cloud server fails to receive the data sent by the edge gateway in the current cycle, the latest n data are used to predict the data of the current data sending cycle.
[0012] Furthermore, if the edge gateway needs to predict the data collected in the current period, the edge network will use the latest n data sent to the cloud server as input.
[0013] Furthermore, the data transmission period is determined according to the timestamps of the edge gateway sending the data for the first time and the second time, that is, the data transmission period T = the timestamp of the edge gateway sending the data for the second time t 1 -Time stamp t of the first data sent by the edge gateway 0 .
[0014] Furthermore, the prediction model is constructed using the LSTM network. The processing process of the prediction model for an input sequence of length n includes the following steps:
[0015] Process the first data in the input sequence through the cascaded 1st LSTM network and the n+1th LSTM network;
[0016] For the i-th data in the input sequence, n≥i≥2, use the cascaded i-th LSTM network and the n+1-th LSTM network to build a network, concatenate the i-th data with the output of the i-1-th LSTM network as the input of the i-th LSTM network, and concatenate the output of the i-th LSTM network with the output of the n+i-1-th LSTM network as the output of the n+1-th LSTM network;
[0017] The output of the 2nth LSTM network is used as the output of the prediction model.
[0018] Furthermore, the processing of input data by an LSTM network includes the following steps:
[0019] f t =σ(W f ·[h t-1 ,x t ]+b f )
[0020] i t =σ(Wi ·[h t-1 ,x t ]+b i )
[0021]
[0022] o t =σ(W o [h t-1 ,x t ]+b o )
[0023] h t =o t tanh(C t )
[0024] Where σ(·) represents the sigmoid function; W f , W i , W c and W o is the trainable weight matrix in the LSTM network, b f 、b i 、b c and b o is the trainable bias vector in the LSTM network; x t h is the sequence input into the LSTM network at time t; t-1 is the hidden state at time t-1; f t is the output of the forget gate of the LSTM network at time t; i t The output of the LSTM network input threshold layer at time t; is the candidate cell state at time t; tanh(·) is the hyperbolic tangent function; C t is the cell state at time t; o t is the output of the LSTM network output gate at time t; h t is the output of the LSTM network at time t.
[0025] The present invention also proposes an efficient power grid data transmission system based on a dual prediction architecture, which is characterized by being used to implement an efficient power grid data transmission method based on a dual prediction architecture, transmitting in an architecture consisting of a terminal device, an edge gateway and a cloud server, wherein the edge gateway and the cloud server respectively deploy an identical prediction model, wherein:
[0026] The terminal device uploads the collected data to the edge gateway;
[0027] After the edge network performs prediction processing on the received data, it determines whether the accumulated uploaded data is less than the input sequence length of the prediction model. If so, the predicted data is transmitted through the socket server. Otherwise, the first n data of the most recently sent n current data are used as the input of the prediction model to obtain the predicted value of the current data, and it is determined whether the absolute difference between the predicted value and the true value is less than the set threshold. If so, no data is sent to the cloud server in the current data sending cycle. Otherwise, the true data is sent to the cloud server through the socket server.
[0028] The cloud server is used to receive data sent by the edge gateway. If the data sent by the edge gateway is not received in the current data sending cycle, the data of the most recent n data sending cycles are input into the local prediction model to obtain the predicted value, and the predicted value is used as the data of the current data sending cycle.
[0029] The efficient power grid data transmission solution based on the dual prediction architecture provided by the present invention reduces the amount of rapidly changing time series data transmitted from the edge computing layer to the cloud computing layer while ensuring data accuracy, thereby alleviating the power grid bandwidth pressure. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 This is a diagram of a cloud-edge-end collaborative network architecture in a high-efficiency power grid data transmission method based on a dual prediction architecture of the present invention;
[0031] Figure 2 A schematic diagram of a dual prediction scheme in a high-efficiency power grid data transmission method based on a dual prediction architecture according to the present invention;
[0032] Figure 3 The present invention is a schematic diagram of the overall process of a high-efficiency power grid data transmission method based on a dual prediction architecture of the present invention;
[0033] Figure 4 A chain diagram of the LSTM neural network module used in the present invention;
[0034] Figure 5 This is a schematic diagram of the many-to-one LSTM model used in the present invention. DETAILED DESCRIPTION
[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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.
[0036] The present invention proposes an efficient power grid data transmission method based on a dual prediction architecture. The edge gateway collects data from the terminal device and preprocesses the data. The edge gateway sends the preprocessed data to the cloud server. The data transmission process between the edge gateway and the cloud server includes the following steps:
[0037] The same prediction model is deployed on the edge gateway and the cloud server. The edge gateway and the cloud server input the same data into the prediction model to obtain the same output.
[0038] The edge gateway starts the socket server and waits for connection. When the cloud server successfully connects to the edge gateway, the edge gateway collects data from the terminal device and pre-processes it.
[0039] The edge gateway sends one data in each data sending cycle. If the cumulative number of data sent is less than the input sequence length n of the prediction model, the collected data will continue to be sent in the next cycle;
[0040] If the cumulative number of data sent is greater than or equal to the input sequence length n of the prediction model, the edge gateway uses the first n data of the current data as the input of the prediction model to obtain the prediction value of the current data;
[0041] The edge gateway compares the absolute error between the current data and its predicted value. If the absolute error is less than the set threshold, the edge gateway will not send data to the cloud server in the current data sending cycle. Otherwise, the edge gateway will send the real data to the cloud server.
[0042] If the cloud server fails to receive the data sent by the edge gateway in the current cycle, the latest n data are used to predict the data of the current data sending cycle.
[0043] Smart grids use a cloud-edge-end collaborative network architecture to tightly integrate cloud computing, edge computing, and terminal devices. This cloud-edge-end collaborative architecture is different from the traditional "cloud-end" architecture. The cloud computing model often cannot meet existing needs, while the emergence and development of the edge computing model processes and calculates part of the calculations nearby on the edge side, greatly reducing the communication pressure, computing load, and processing response delay between "cloud-end" to achieve rapid data processing and real-time decision-making, while also reducing network burden and costs.
[0044] The dual prediction architecture is an effective way to reduce data transmission while ensuring data accuracy. It runs the same prediction model on the data sender and receiver. The sender will only transmit data to the receiver when the prediction deviation is greater than a given threshold. In addition, since deep learning models can better capture the temporal correlation and nonlinear relationship between time series data, they are more suitable for predicting rapidly changing industrial data. At the same time, the edge gateway in the smart grid has strong computing power, which can realize the reasoning of deep learning models.
[0045] In this embodiment, the smart grid cloud-edge collaborative network architecture is as follows Figure 1 As shown in the figure, the architecture is divided into three layers: cloud computing layer, edge computing layer and terminal devices, among which:
[0046] The cloud computing layer is the core of the entire cloud-edge architecture. The cloud computing layer undertakes the tasks of large-scale data storage and advanced data analysis. In this embodiment, the cloud computing layer is mainly used to receive real-time data or make predictions based on the received real-time data; in the power grid field, this includes processing real-time and historical data from various sensors, monitoring equipment and satellites. Cloud computing is used to perform complex power system modeling, forecasting and planning, as well as to support large-scale power market operations.
[0047] The edge computing layer is located at key nodes of the power grid, such as substations and distribution stations. This layer is used to process data that requires low-latency responses, such as power load monitoring and real-time equipment status updates. Edge computing supports local control and intelligent decision-making of terminal devices to reduce the need to transmit data to the cloud computing layer;
[0048] Terminal devices include various sensors, smart meters, power line communication equipment, etc. Terminal devices are distributed in various parts of the power system. These devices are responsible for real-time monitoring of the grid status, power load and energy production. These devices can also perform local control operations such as load management, equipment switch control, etc. Terminal devices collect data in the power system for transmission to the edge and cloud computing layers for further analysis.
[0049] The present invention focuses on data transmission between edge and cloud. Operations such as model training in cloud computing are not specifically involved in the present invention. In order to simplify the description, the present invention only explains the relevant parts of the data transmission process. During the transmission process, the main work of the cloud computing layer is implemented by the cloud server, and the main work of the edge computing layer is implemented by the edge gateway. Other functions implemented by the cloud computing layer and the edge computing layer are not repeated here.
[0050] In this embodiment, the data transmission process includes:
[0051] The edge gateway will sample, collect and preprocess the raw data from the terminal device, and then store the processed data as a data set in the database. For the preprocessed data in the database, real-time transmission is not adopted, because the transmission time of the data set is much shorter than the collection time of the data set. After the edge gateway collects the complete data set, it will be sent to the cloud at one time, which will be more stable and efficient; the cloud server is responsible for training and distributing the prediction model, and the edge server is responsible for receiving the prediction model and performing model reasoning. Specifically, the larger the data set on the cloud, the higher the prediction accuracy of the training model, and the stronger the computing power of the cloud, the faster the training speed.
[0052] In order to achieve more accurate prediction, the dual prediction architecture adopted by the present invention is as follows Figure 2 As shown, the edge gateway and cloud server first load the same pre-trained prediction model, then the edge gateway starts the socket server and waits for connection. When the cloud server successfully connects to the edge gateway, the edge gateway obtains readings from the sensor and performs sampling and preprocessing.
[0053] The sensor data preprocessed by the edge gateway is divided into timestamp t i and the specific value x i Two parts, for the timestamp, the edge gateway only sends the timestamp t of the first data sent 0 and the timestamp t when the data is sent for the second time 1 , the interval T can be expressed as t 1- t 0 To calculate, the subsequent t i Through t i-1 +T is used for calculation, that is, only the timestamp-related data is sent twice to confirm the data sending cycle, so as to reduce the data transmission of the timestamp part.
[0054] For numerical values, it is divided into the preparation stage and the prediction stage. In the preparation stage, when the sequence number i of the current data to be sent is less than or equal to the sequence length S of the input sequence of the prediction model, it is necessary to continue to transmit the current i-th data x i , because the input length of the prediction model is not enough to make predictions;
[0055] In the prediction phase, when the sequence number i of the current data to be sent is greater than S, the edge gateway and the cloud server can use the same prediction model and the same input data, that is, the S data {x before the sequence number of the current data to be sent i-1 ,x i-2 ,…,x i-S}, and obtain the same predicted value The edge gateway will read the raw data and preprocess it to get the next true value and compare it with the predicted value. If the absolute prediction error If the value is less than the given threshold λ, the edge gateway will not send the true value to the cloud server, and the cloud server will directly use its own predicted value. As the final value at the current moment, since the predicted values of the cloud server and the edge gateway are the same, the deviation of the final value will be less than the given threshold λ. This mechanism avoids transmission and ensures the accuracy of the final data; if If it is greater than λ, the edge gateway will send the true value to the cloud server, and the cloud server will use it as the final value. At this time, the deviation of the final value is 0.
[0056] In the present invention, due to the different processing capabilities caused by the different hardware configurations of the edge and the cloud, in order to ensure data synchronization at both ends, each round of data prediction to the confirmation of the final data must be carried out within the specified period. For the cloud, if no data sent by the edge gateway is received within a period, the predicted data will be used as the final data and will be dormant until the end of the period to start the next round of prediction. If real data is received in the middle, it will be directly replaced with the original predicted data.
[0057] Figure 3 This is a schematic diagram of the overall process running periodically, during which the model input is updated using a sliding window. The green box indicates the transmission value whose prediction error is greater than the threshold. For the first n data, the edge gateway directly sends data to the cloud server to train the deep learning model with a sequence length of n. The n+1th data is then predicted by the first n data. If the predicted value is If the deviation from the true value x(n+1) does not exceed the given threshold, the cloud server will use As the final value of the n+1th data, let the local n+1th data The n+2th data is represented by x 2 to x n+1 These n data are used to predict, and the n+3th data is predicted by x 3 to x n+2 These n data are used to predict, and the predicted value for the n+3th data Assume that it exceeds the threshold, so the edge gateway will send the true value x(n+3) to the cloud server, and the cloud server will use x(n+3) as the final value of the n+3th data, that is, x n+3 =x(n+3). After that, the n+4th data is x 4 to x n+3 These n data are used to predict, and so on.
[0058] Wired transmission is used between the end and the edge to ensure the transmission speed and stability of high-frequency data. Edge gateways are usually deployed at industrial sites, close to the end. Wireless transmission is used between the edge and the cloud because the physical distance between the cloud and the edge is long. Specifically, data is transmitted via WiFi, a wireless protocol between the edge and the cloud.
[0059] For the prediction part of deep learning, the present invention designs a prediction model based on LSTM network to process time series information and make predictions on it, such as Figure 4 The network is a two-layer architecture. Each layer includes the same number of LSTM networks as the input sequence. The LSTM networks of each layer are cascaded, and the LSTM networks of corresponding numbers between the two layers are also cascaded. It can well capture nonlinear relationships and time correlations, and achieve high prediction accuracy for rapidly changing data. Specifically, Figure 5 , the prediction model processes the input sequence of length n in the following steps:
[0060] Process the first data in the input sequence through the cascaded 1st LSTM network and the n+1th LSTM network;
[0061] For the i-th data in the input sequence, n≥i≥2, use the cascaded i-th LSTM network and the n+1-th LSTM network to build a network, concatenate the i-th data with the output of the i-1-th LSTM network as the input of the i-th LSTM network, and concatenate the output of the i-th LSTM network with the output of the n+i-1-th LSTM network as the output of the n+1-th LSTM network;
[0062] The output of the 2nth LSTM network is used as the output of the prediction model.
[0063] LSTM is a special recurrent neural network RNN. Its key lies in the cell state C t LSTM retains and transmits information through cell states. During the entire sequence processing, the cell state can remain relatively unchanged, thereby effectively capturing long-term dependencies. To this end, it introduces input gates, forget gates, and output gates so that it can be executed in longer sequences without changing the overall structure of the RNN network to remove or add information to the cell state. The chain of LSTM neural network modules is as follows: Figure 4 As shown, W f , W i , W c and W o is the input weight matrix, b f 、b i 、b c and b o is the bias vector.
[0064] The forget gate determines what information will be discarded from the cell state. The forget gate reads the previous output h t-1 and the current input x t , do a sigmoid nonlinear mapping, and then output a vector f t , the value of each dimension of this vector is between 0 and 1, 1 means completely retained, 0 means completely discarded, which is equivalent to remembering the important ones and forgetting the unimportant ones, and finally with the cell state C t-1 Multiply. The output of the forget gate is:
[0065] f t =σ(W f ·[h t-1 ,x t ]+b f )
[0066] The input gate determines what new information is stored in the cell state. There are two parts here: the sigmoid layer is called the "input gate layer", which determines what value will be updated; the tanh layer creates a new candidate value vector will be added to the cell state. The output of the input gate is:
[0067] i t =σ(W i ·[h t-1 ,x t ]+b i )
[0068]
[0069] With the above two gates, the cell state (memory) can be updated. t-1 Updated to C t . Put the old state with f t Multiply, discard the information we are sure we need to discard, and then add This is the new candidate value, which changes depending on how much we decide to update each state. The cell state update formula is as follows:
[0070]
[0071] Finally, we need to determine what value will be output, and this output will be based on the cell state. First, we run a sigmoid layer to determine which part of the cell state will be output. Next, we process the cell state through a tanh layer and multiply it with the output of the sigmoid layer, and finally output the part we determined to output. The output formula of the output gate is as follows:
[0072] o t =σ(W o [h t-1 ,x t ]+b o )
[0073] h t =o t tanh(C t )
[0074] The present invention utilizes Figure 5The many-to-one LSTM model predicts the data. It consists of an input layer, some hidden layers and an output layer. Each LSTM unit in the hidden layer corresponds to an LSTM neural network module. The hidden layer consists of two layers of several LSTM units. The length is equal to the sequence length S of the model. The output of the previous module is used as the input of the next module. By inputting data of the LSTM sequence length into the model, the output of the last LSTM unit is the predicted value.
[0075] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. An efficient power grid data transmission method based on a dual prediction architecture, characterized in that: The edge gateway collects data from the terminal device and pre-processes the data. The edge gateway sends the pre-processed data to the cloud server. The data transmission process between the edge gateway and the cloud server includes the following steps: The same prediction model is deployed on the edge gateway and the cloud server. The edge gateway and the cloud server input the same data into the prediction model to obtain the same output. The edge gateway starts the socket server and waits for connection. When the cloud server successfully connects to the edge gateway, the edge gateway collects data from the terminal device and pre-processes it. The edge gateway sends one data in each data sending cycle. If the cumulative number of data sent is less than the input sequence length n of the prediction model, the collected data will continue to be sent in the next cycle; If the cumulative number of data sent is greater than or equal to the input sequence length n of the prediction model, the edge gateway uses the first n data of the current data as the input of the prediction model to obtain the prediction value of the current data; The edge gateway compares the absolute error between the current data and its predicted value. If the absolute error is less than the set threshold, the edge gateway will not send data to the cloud server in the current data sending cycle. Otherwise, the edge gateway will send the real data to the cloud server. If the cloud server fails to receive the data sent by the edge gateway in the current cycle, the latest n data are used to predict the data of the current data sending cycle.
2. The method for efficient power grid data transmission based on a dual prediction architecture according to claim 1, characterized in that: If the edge gateway needs to predict the data collected in the current period, the edge network will use the latest n data sent to the cloud server as input.
3. The efficient power grid data transmission method based on a dual prediction architecture according to claim 1, characterized in that: The data transmission cycle is determined according to the timestamps of the first and second data transmissions by the edge gateway, that is, the data transmission cycle T=timestamp t1 of the second data transmission by the edge gateway-timestamp t0 of the first data transmission by the edge gateway.
4. The efficient power grid data transmission method based on a dual prediction architecture according to claim 1, characterized in that: The prediction model is constructed using the LSTM network. The prediction model processes the input sequence of length n in the following steps: Process the first data in the input sequence through the cascaded 1st LSTM network and the n+1th LSTM network; For the i-th data in the input sequence, n≥i≥2, use the cascaded i-th LSTM network and the n+1-th LSTM network to build a network, concatenate the i-th data with the output of the i-1-th LSTM network as the input of the i-th LSTM network, and concatenate the output of the i-th LSTM network with the output of the n+i-1-th LSTM network as the output of the n+1-th LSTM network; The output of the 2nth LSTM network is used as the output of the prediction model.
5. The method for efficient power grid data transmission based on dual prediction architecture according to claim 4, characterized in that: The process of processing input data by an LSTM network includes the following steps: f t =σ(W f ·[h t-1 ,x t ]+b f ) i t =σ(W i ·[h t-1 ,x t ]+b i ) the t =σ(W o [h t-1 ,x t ]+b o ) h t =o t ·tanh(C t ) Where σ(·) represents the sigmoid function; W f , W i , W c and W o is the trainable weight matrix in the LSTM network, b f , b i , b c and b o is the trainable bias vector in the LSTM network; x t h is the sequence input into the LSTM network at time t; t-1 is the hidden state at time t-1; f t is the output of the forget gate of the LSTM network at time t; i t The output of the LSTM network input threshold layer at time t; is the candidate cell state at time t; tanh(·) is the hyperbolic tangent function; C t is the cell state at time t; o t is the output of the LSTM network output gate at time t; h t is the output of the LSTM network at time t.
6. The high-efficiency power grid data transmission system based on dual prediction architecture according to claim 1, characterized in that: The method for efficiently transmitting power grid data based on a dual prediction architecture according to claim 1 is used to transmit data in an architecture composed of a terminal device, an edge gateway and a cloud server, wherein the edge gateway and the cloud server respectively deploy an identical prediction model, wherein: The terminal device uploads the collected data to the edge gateway; After the edge network performs prediction processing on the received data, it determines whether the accumulated uploaded data is less than the input sequence length of the prediction model. If so, the predicted data is transmitted through the socket server. Otherwise, the first n data of the most recently sent n current data are used as the input of the prediction model to obtain the predicted value of the current data, and it is determined whether the absolute difference between the predicted value and the true value is less than the set threshold. If so, no data is sent to the cloud server in the current data sending cycle. Otherwise, the true data is sent to the cloud server through the socket server. The cloud server is used to receive data sent by the edge gateway. If the data sent by the edge gateway is not received in the current data sending cycle, the data of the most recent n data sending cycles are input into the local prediction model to obtain the predicted value, and the predicted value is used as the data of the current data sending cycle.
7. The high-efficiency power grid data transmission system based on dual prediction architecture according to claim 6, characterized in that: The data transmission cycle is determined according to the timestamps of the first and second data transmissions by the edge gateway, that is, the data transmission cycle T=timestamp t1 of the second data transmission by the edge gateway-timestamp t0 of the first data transmission by the edge gateway.
8. The high-efficiency power grid data transmission system based on dual prediction architecture according to claim 6, characterized in that: The prediction model is constructed using the LSTM network. The prediction model processes the input in the following steps: Process the first data in the input sequence through the cascaded 1st LSTM network and the n+1th LSTM network; For the i-th data in the input sequence, n≥i≥2, use the cascaded i-th LSTM network and the n+1-th LSTM network to build a network, concatenate the i-th data with the output of the i-1-th LSTM network as the input of the i-th LSTM network, and concatenate the output of the i-th LSTM network with the output of the n+i-1-th LSTM network as the output of the n+1-th LSTM network; The output of the 2nth LSTM network is used as the output of the prediction model.
9. The high-efficiency power grid data transmission system based on dual prediction architecture according to claim 6, characterized in that: The process of processing input data by an LSTM network includes the following steps: f t =σ(W f ·[h t-1 ,x t ]+b f ) i t =σ(W i ·[h t-1 ,x t ]+b i ) the t =σ(W o [h t-1 ,x t ]+b o ) h t =o t ·tanh(C t ) Where σ(·) represents the sigmoid function; W f , W i , W c and W o is the trainable weight matrix in the LSTM network, b f 、b i 、b c and b o is the trainable bias vector in the LSTM network; x t h is the sequence input into the LSTM network at time t; t-1 is the hidden state at time t-1; f t is the output of the forget gate of the LSTM network at time t; i t The output of the LSTM network input threshold layer at time t; is the candidate cell state at time t; tanh(·) is the hyperbolic tangent function; C t is the cell state at time t; o t is the output of the LSTM network output gate at time t; h t is the output of the LSTM network at time t.
