Power system trigger communication method and device based on prediction model

Abstract

The invention discloses a power system trigger communication method and device based on a prediction model. According to the method, the historical and actual load data are acquired, and the pre-trained power system prediction model is utilized to predict the load. If the similarity between the predicted data and the actual data is lower than a threshold value, transmitting the actual data; otherwise, sending the data of the previous cycle to fill the vacancy of the receiving end. According to the method, the communication efficiency and reliability of the power system are improved, and the problem that in the prior art, system communication cannot be triggered only when the state of the power system changes is solved.

Description

Technical Field

[0001] The present invention relates to the field of power system trigger communication, and in particular to a power system trigger communication method and device based on a prediction model. Background Art

[0002] The role of the power system in maintaining national economic stability, social operation order, and even national security cannot be ignored. However, the contemporary power system faces threats from many aspects, especially the limited communication conditions after unconventional physical attacks, which poses a huge challenge to the safe and stable operation of the power system. Therefore, it is urgent to study the event-triggered communication strategy and interpolation technology of the power system, so that communication can only be carried out when key events or state changes occur, and necessary information can be transmitted, thereby reducing real-time data transmission and reducing the demand for communication bandwidth.

[0003] However, most of the existing research on power system dynamic perception methods under limited communication conditions starts from the perspective of data compression to reduce the number of bits transmitted. The above methods only consider reducing the amount of information transmitted in each round of communication, while the number of communications does not change. This makes it impossible for existing technologies to trigger power system communication only when the state changes. Summary of the invention

[0004] The present invention provides a method and device for triggering communication of an electric power system based on a prediction model, so as to solve the problem in the prior art that system communication cannot be triggered only when the state of the electric power system changes.

[0005] In a first aspect, the present application provides a method for triggering communication of a power system based on a prediction model, comprising:

[0006] Acquire first historical load data, a first actual load sequence, and a second actual load sequence; wherein the first actual load sequence and the second actual load sequence are obtained by dividing the actual load data according to a preset time period, and the first actual load sequence is a sequence of the second actual load sequence in the previous time period;

[0007] Inputting the first historical load data into a pre-trained power system prediction model so that the power system prediction model outputs a load prediction sequence;

[0008] If the similarity distance between the load forecast sequence and the second actual load sequence is less than a preset threshold, triggering power system communication, and sending the second actual load sequence to the receiving end;

[0009] Otherwise, it is determined that the receiving end is a vacant sequence, and the first actual load sequence is sent to the receiving end, so that the receiving end fills the vacant sequence.

[0010] This application first trains the power system prediction model by using historical load data to ensure the model's ability to accurately predict changes in power load. Then, by comparing the similarity between the predicted sequence and the actual sequence, communication is triggered only when the two differ greatly, effectively reducing unnecessary data transmission and reducing the consumption of communication resources. In addition, when there are gaps between the actual sequences, the gaps are filled with the previous complete sequence to ensure the continuity and integrity of the data, thereby making the power system operation more stable and the decision-making more accurate. This application not only optimizes the use of communication resources, but also enhances the power system's ability to respond to emergencies, so as to solve the problem in the prior art that system communication cannot be triggered only when the power system state changes.

[0011] As a preferred embodiment of the first aspect, the pre-trained neural network is specifically:

[0012] Acquiring second historical load data;

[0013] Inputting the second historical load data into the initial power system prediction model for training, and optimizing the initial power system prediction model using a gradient boosting algorithm until the value of the loss function no longer changes, stopping the training, and obtaining a trained power system prediction model;

[0014] The initial power system prediction model is constructed based on the GBDT algorithm. The initial power system prediction model includes multiple decision trees. During the training process, multiple decision trees are continuously traversed and the prediction results of each decision tree are accumulated.

[0015] In this preferred embodiment, the present application achieves high-precision prediction of power load by using a power system prediction model constructed based on a gradient boosting decision tree (GBDT) algorithm and optimizing it in combination with a gradient boosting algorithm. The initial model uses multiple decision trees to learn historical load data. By continuously traversing these decision trees and accumulating their prediction results, the model can capture complex patterns and trends in the load data. Furthermore, the initial model is finely optimized by the gradient boosting algorithm, and the model gradually reduces the value of the loss function during the training process until a stable state is reached, thereby ensuring the generalization ability and prediction accuracy of the model. This method not only improves the accuracy of power system load prediction, but also enhances the model's adaptability to changes in historical load data, so that in the actual operation of the power system, future load changes can be more effectively predicted, thereby optimizing the allocation of power resources and the operation strategy of the power grid, and improving the operation efficiency and reliability of the power grid.

[0016] As a preferred embodiment of the first aspect, if the similarity distance between the load forecast sequence and the second actual load sequence is less than a preset threshold, triggering power system communication is specifically:

[0017] According to the load forecast sequence and the second actual load sequence, the Euclidean distance, the first DTW distance and the second DTW distance are calculated;

[0018] According to the entropy weight method, the weights of the Euclidean distance, the first DTW distance, and the second DTW distance are calculated;

[0019] Calculate the similarity distance according to the weights, the Euclidean distance, the first DTW distance and the second DTW distance;

[0020] The calculation formula of the similarity distance is:

[0021] D total (X D , Y D )=r1D1(X D , Y D )+r2D2(X D , Y D )+r3D3(X D , Y D ),

[0022] In the formula, r1, r2, and r3 are D1(X D , Y D )、D2(X D , Y D )、D3(X D , Y D ) corresponding weight; X D is the second actual load sequence; Y D is the load forecast sequence, D1(X D , Y D ) is the Euclidean distance, D2(X D , Y D ) is the first DTW distance, D3(X D , Y D ) is the second DTW distance;

[0023] If the similarity distance is less than a preset threshold, power system communication is triggered.

[0024] In this preferred embodiment, the present application calculates the similarity distance between the load forecast sequence and the second actual load sequence by comprehensively considering the Euclidean distance, the first DTW distance and the second DTW distance. This comprehensive evaluation method can more comprehensively reflect the similarity between the two sequences. The entropy weight method is used to determine the weights of each distance, which ensures the objectivity and scientificity of the weight distribution, thereby making the similarity evaluation more accurate. When the calculation result of the similarity distance is less than the preset threshold, it indicates that the predicted sequence is very close to the actual sequence. At this time, the communication mechanism is triggered to send the second actual load sequence to the receiving end. The preset threshold is set between zero and one, providing a flexible judgment criterion, making the communication trigger more reasonable. This method not only improves communication efficiency and reduces unnecessary data transmission, but also ensures that communication is only carried out when there is a large deviation between the prediction and the actual through accurate similarity evaluation, thereby optimizing the use of resources and enhancing the power system's responsiveness to changes and overall stability.

[0025] As a preferred embodiment of the first aspect, the Euclidean distance is calculated based on the load forecast sequence and the second actual load sequence, specifically:

[0026] According to the Euclidean distance formula, the Euclidean distance between the sampling points corresponding to the load forecast sequence and the second actual load sequence is calculated;

[0027] The formula for the Euclidean distance is

[0028]

[0029] In the formula, is the load forecast value at time i; is the second actual load value at time i, D1(X D , Y D ) is the Euclidean distance.

[0030] In this preferred embodiment, the present application can accurately quantify the deviation of the corresponding sampling points between the load forecast sequence and the second actual load sequence by adopting the calculation method of the Euclidean distance. The Euclidean distance formula uses the square root of the sum of the squares of the differences between the load forecast value and the actual load value at each moment to represent the distance, thereby evaluating the numerical difference between the two sequences. Specifically, for each time point t, the absolute value of the difference between the load forecast value and the second actual load value is calculated, and then these differences are squared, summed and squared to obtain the Euclidean distance of the entire time series. This calculation method takes into account the differences of all sampling points in the time series and can fully reflect the overall deviation between the predicted sequence and the actual sequence. When the Euclidean distance is small, it indicates that the prediction results of the prediction model are more consistent with the actual load changes; when the Euclidean distance is large, it indicates that there is a large deviation between the predicted results and the actual values. The evaluation mechanism of the present application helps to judge the quality of the prediction and provide a scientific basis for the operation and load management of the power system.

[0031] As a preferred embodiment of the first aspect, the first DTW distance is calculated according to the load prediction sequence and the second actual load sequence, specifically:

[0032] The first DTW distance is calculated according to the objective function, the load forecast sequence, and the second actual load sequence;

[0033] Wherein, the objective function is obtained by determining each moving path of each dynamic point and ensuring that the cumulative distance of each moving path is the shortest;

[0034] The formula of the objective function is

[0035]

[0036] In the formula, is the moving path distance of the e-th dynamic point; D2(X D , Y D ) is the load forecast sequence Y D With the second actual load sequence X D The first DTW distance between them.

[0037] In this preferred embodiment, the first DTW distance mentioned in the present application is realized by the core mechanism of the Dynamic Time Warping (DTW) algorithm, which dynamically adjusts the moving paths of each dynamic point by optimizing an objective function to ensure that the cumulative distance of these paths is the shortest. Specifically, the objective function calculates all possible matching paths between the load forecast sequence and the second actual load sequence, and selects the path that minimizes the sum of the path distances. In this process, the moving path distance of each dynamic point is added up to form the first DTW distance, which reflects the overall similarity of the two sequences in the best matching case. In this way, even if the two sequences are stretched or misaligned in time, the DTW algorithm can effectively evaluate their similarity, thereby providing a reliable measure for the accuracy of power system load forecasting.

[0038] As a preferred embodiment of the first aspect, the second DTW distance is calculated according to the load prediction sequence and the second actual load sequence, specifically:

[0039] According to the difference method, the load forecast sequence and the second actual load sequence, a first load forecast sequence and a third actual load sequence are calculated;

[0040] Wherein, the first load forecast sequence and the third actual load sequence can reflect the fluctuation of the power system;

[0041] A second DTW distance is calculated according to the first DTW distance, the first load prediction sequence and the third actual load sequence.

[0042] In this preferred embodiment, the present application generates a first load forecast sequence and a third actual load sequence by applying the differential method to process the original load forecast sequence and the second actual load sequence. These two new sequences can more sensitively map the load fluctuations of the power system. Then, using these new sequences, combined with the calculation method of the first DTW distance, the second DTW distance is calculated. This process involves evaluating the fluctuation pattern of the time series and dynamically time-warping it with the forecast sequence to find the best time alignment method to obtain the second DTW distance. This calculation method can more accurately reflect the local changes and fluctuations of the power system load, enhance the adaptability and responsiveness of the prediction model to the dynamic changes of the system, and provide more accurate decision support for the stable operation and load management of the power system.

[0043] In a second aspect, the present application provides a power system trigger communication device based on a prediction model. The power system trigger communication device based on a prediction model includes an acquisition module, an input and output module, and a trigger module;

[0044] The acquisition module is used to acquire first historical load data, a first actual load sequence, and a second actual load sequence; wherein the first actual load sequence and the second actual load sequence are obtained by dividing the actual load data according to a preset time period, and the first actual load sequence is a sequence of the second actual load sequence in the previous time period;

[0045] The input-output module is used to input the first historical load data into a pre-trained power system prediction model so that the power system prediction model outputs a load prediction sequence;

[0046] The trigger module is used to trigger power system communication and send the second actual load sequence to a receiving end if the similarity distance between the load forecast sequence and the second actual load sequence is less than a preset threshold;

[0047] Otherwise, it is determined that the receiving end is a vacant sequence, and the first actual load sequence is sent to the receiving end, so that the receiving end fills the vacant sequence.

[0048] This device uses two modules to divide the work and coordinate work to better trigger power system communication when the state changes. This application first trains the power system prediction model by using historical load data to ensure the model's ability to accurately predict changes in power load. Then, by comparing the similarity between the predicted sequence and the actual sequence, communication is triggered only when the two differ greatly, effectively reducing unnecessary data transmission and reducing the consumption of communication resources. In addition, when there are gaps between the actual sequences, the continuity and integrity of the data are ensured by filling the gaps with the previous complete sequence, thereby making the power system operation more stable and the decision-making more accurate. This application not only optimizes the use of communication resources, but also enhances the power system's ability to respond to emergencies, so as to solve the problem in the prior art that system communication cannot be triggered only when the power system state changes.

[0049] As a preferred embodiment of the second aspect, the pre-trained neural network is specifically:

[0050] Acquiring second historical load data;

[0051] Inputting the second historical load data into the initial power system prediction model for training, and optimizing the initial power system prediction model using a gradient boosting algorithm until the value of the loss function no longer changes, stopping the training, and obtaining a trained power system prediction model;

[0052] The initial power system prediction model is constructed based on the GBDT algorithm. The initial power system prediction model includes multiple decision trees. During the training process, multiple decision trees are continuously traversed and the prediction results of each decision tree are accumulated.

[0053] In this preferred embodiment, the present application achieves high-precision prediction of power load by using a power system prediction model constructed based on a gradient boosting decision tree (GBDT) algorithm and optimizing it in combination with a gradient boosting algorithm. The initial model uses multiple decision trees to learn historical load data. By continuously traversing these decision trees and accumulating their prediction results, the model can capture complex patterns and trends in the load data. Furthermore, the initial model is finely optimized by the gradient boosting algorithm, and the model gradually reduces the value of the loss function during the training process until a stable state is reached, thereby ensuring the generalization ability and prediction accuracy of the model. This method not only improves the accuracy of power system load prediction, but also enhances the model's adaptability to changes in historical load data, so that in the actual operation of the power system, future load changes can be more effectively predicted, thereby optimizing the allocation of power resources and the operation strategy of the power grid, and improving the operation efficiency and reliability of the power grid.

[0054] As a preferred embodiment of the second aspect, if the similarity distance between the load forecast sequence and the second actual load sequence is less than a preset threshold, triggering power system communication is specifically:

[0055] According to the load forecast sequence and the second actual load sequence, the Euclidean distance, the first DTW distance and the second DTW distance are calculated;

[0056] According to the entropy weight method, the weights of the Euclidean distance, the first DTW distance, and the second DTW distance are calculated;

[0057] Calculate the similarity distance according to the weights, the Euclidean distance, the first DTW distance and the second DTW distance;

[0058] The calculation formula of the similarity distance is:

[0059] D total (X D , Y D )=r1D1(X D , Y D )+r2D2(X D , Y D )+r3D3(X D , Y D ),

[0060] In the formula, r1, r2, and r3 are D1(X D , Y D )、D2(X D , Y D )、D3(X D , Y D ) corresponding weight; X D is the second actual load sequence; YD is the load forecast sequence, D1(X D , Y D ) is the Euclidean distance, D2(X D , Y D ) is the first DTW distance, D3(X D , Y D ) is the second DTW distance;

[0061] If the similarity distance is less than a preset threshold, power system communication is triggered.

[0062] In this preferred embodiment, the present application calculates the similarity distance between the load forecast sequence and the second actual load sequence by comprehensively considering the Euclidean distance, the first DTW distance and the second DTW distance. This comprehensive evaluation method can more comprehensively reflect the similarity between the two sequences. The entropy weight method is used to determine the weights of each distance, which ensures the objectivity and scientificity of the weight distribution, thereby making the similarity evaluation more accurate. When the calculation result of the similarity distance is less than the preset threshold, it indicates that the predicted sequence is very close to the actual sequence. At this time, the communication mechanism is triggered to send the second actual load sequence to the receiving end. The preset threshold is set between zero and one, providing a flexible judgment criterion, making the communication trigger more reasonable. This method not only improves communication efficiency and reduces unnecessary data transmission, but also ensures that communication is only carried out when there is a large deviation between the prediction and the actual through accurate similarity evaluation, thereby optimizing the use of resources and enhancing the power system's responsiveness to changes and overall stability.

[0063] As a preferred embodiment of the second aspect, the Euclidean distance is calculated based on the load forecast sequence and the second actual load sequence, specifically:

[0064] According to the Euclidean distance formula, the Euclidean distance between the sampling points corresponding to the load forecast sequence and the second actual load sequence is calculated;

[0065] The formula for the Euclidean distance is

[0066]

[0067] In the formula, is the load forecast value at time i; is the second actual load value at time i, D1(X D , Y D ) is the Euclidean distance.

[0068] In this preferred embodiment, the present application can accurately quantify the deviation of the corresponding sampling points between the load forecast sequence and the second actual load sequence by adopting the calculation method of the Euclidean distance. The Euclidean distance formula uses the square root of the sum of the squares of the differences between the load forecast value and the actual load value at each moment to represent the distance, thereby evaluating the numerical difference between the two sequences. Specifically, for each time point t, the absolute value of the difference between the load forecast value and the second actual load value is calculated, and then these differences are squared, summed and squared to obtain the Euclidean distance of the entire time series. This calculation method takes into account the differences of all sampling points in the time series and can fully reflect the overall deviation between the predicted sequence and the actual sequence. When the Euclidean distance is small, it indicates that the prediction results of the prediction model are more consistent with the actual load changes; when the Euclidean distance is large, it indicates that there is a large deviation between the predicted results and the actual values. The evaluation mechanism of the present application helps to judge the quality of the prediction and provide a scientific basis for the operation and load management of the power system.

[0069] As a preferred embodiment of the second aspect, the first DTW distance is calculated according to the load prediction sequence and the second actual load sequence, specifically:

[0070] The first DTW distance is calculated according to the objective function, the load forecast sequence, and the second actual load sequence;

[0071] Wherein, the objective function is obtained by determining each moving path of each dynamic point and ensuring that the cumulative distance of each moving path is the shortest;

[0072] The formula of the objective function is

[0073]

[0074] In the formula, is the moving path distance of the e-th dynamic point; D2(X D , Y D ) is the load forecast sequence Y D With the second actual load sequence X D The first DTW distance between them.

[0075] In this preferred embodiment, the first DTW distance mentioned in the present application is realized by the core mechanism of the Dynamic Time Warping (DTW) algorithm, which dynamically adjusts the moving paths of each dynamic point by optimizing an objective function to ensure that the cumulative distance of these paths is the shortest. Specifically, the objective function calculates all possible matching paths between the load forecast sequence and the second actual load sequence, and selects the path that minimizes the sum of the path distances. In this process, the moving path distance of each dynamic point is added up to form the first DTW distance, which reflects the overall similarity of the two sequences in the best matching case. In this way, even if the two sequences are stretched or misaligned in time, the DTW algorithm can effectively evaluate their similarity, thereby providing a reliable measure for the accuracy of power system load forecasting.

[0076] As a preferred embodiment of the second aspect, the second DTW distance is calculated according to the load prediction sequence and the second actual load sequence, specifically:

[0077] According to the difference method, the load forecast sequence and the second actual load sequence, a first load forecast sequence and a third actual load sequence are calculated;

[0078] Wherein, the first load forecast sequence and the third actual load sequence can reflect the fluctuation of the power system;

[0079] A second DTW distance is calculated according to the first DTW distance, the first load prediction sequence and the third actual load sequence.

[0080] In this preferred embodiment, the present application generates a first load forecast sequence and a third actual load sequence by applying the differential method to process the original load forecast sequence and the second actual load sequence. These two new sequences can more sensitively map the load fluctuations of the power system. Then, using these new sequences, combined with the calculation method of the first DTW distance, the second DTW distance is calculated. This process involves evaluating the fluctuation pattern of the time series and dynamically time-warping it with the forecast sequence to find the best time alignment method to obtain the second DTW distance. This calculation method can more accurately reflect the local changes and fluctuations of the power system load, enhance the adaptability and responsiveness of the prediction model to the dynamic changes of the system, and provide more accurate decision support for the stable operation and load management of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 : A flow chart of an embodiment of a method for triggering communication of a power system based on a prediction model provided in the present application;

[0082] Figure 2: A structural diagram of an embodiment of a power system event-triggered communication process based on a prediction model provided in the present application;

[0083] Figure 3 : A structural diagram of an embodiment of a power system triggering communication device based on a prediction model provided in the present application. DETAILED DESCRIPTION

[0084] 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.

[0085] Embodiment 1

[0086] Please refer to Figure 1 , which is a prediction model-based power system triggering communication method provided in an embodiment of the present invention.

[0087] In this embodiment, the process of the prediction model-based power system triggering communication method in this application is described in detail through steps S01-S03.

[0088] S01: Acquire first historical load data, a first actual load sequence and a second actual load sequence; wherein, the first actual load sequence and the second actual load sequence are obtained by dividing the actual load data according to a preset time period, and the first actual load sequence is the sequence of the second actual load sequence in the previous time period.

[0089] S02: Inputting the first historical load data into a pre-trained power system prediction model so that the power system prediction model outputs a load prediction sequence.

[0090] As a preferred embodiment of the first embodiment, the pre-trained power system prediction model is obtained by training the initial power system prediction model.

[0091] As a preferred embodiment of the first embodiment, the training process of the initial power system prediction model is specifically as follows:

[0092] (1) The normalized historical power system load data is used as the input of the GBDT prediction model. The collected historical power system load data is normalized and mapped to the range of [0, 1] to prevent the prediction accuracy of the model from being reduced by different orders of magnitude. The specific process is as follows:

[0093]

[0094] Among them, z is the original load data, is the load data after normalization, z max is the maximum original load value, z min is the minimum original load value. Assuming that the short-term load of the power system at the i-th moment of the m-th day needs to be predicted, the normalized historical power system load data of the t-period before this moment is selected as the input of the GBDT prediction model.

[0095] (2) Use the GBDT algorithm to predict the load of the power system. The GBDT algorithm contains hundreds of decision trees. When using the algorithm to predict the load of the power system, it is necessary to set the initial value for the input sample data and traverse each decision tree. It will continuously adjust and update the predicted value, accumulate the prediction results of each decision tree, and obtain the final power system load prediction value. The specific process is as follows:

[0096] F(x)=F0+(β1+β2+…+β M )[T1(X)+T2(X)+T M (X)], where T() is a weak classifier, and its number is M; β is a weight coefficient; F() is a strong classifier composed of all weak classifiers. Through several iterations, the number of M weak classifiers is integrated into a strong classifier. Each iteration can reduce the residual of the previous model, and the new model is created in the gradient direction of the residual reduction.

[0097]

[0098] Describe the loss function to determine the gradient of residual reduction. i ) is the short-term load forecast value of the power system output by the model, y i is the actual value.

[0099] (3) The Boosting gradient boosting algorithm is used to solve the problems of slow training speed and large memory usage caused by the GBDT algorithm needing to traverse the sample data countless times during training. It creates the mth weak classifier in the gradient descent direction of the m-1th weak classifier loss function. The specific process is: Assuming F(x; Q) is the prediction function, Q = {q0, q1, q2...} is the parameter of the prediction function, the expression of the prediction function can be obtained by weighted combination of several weak classifiers:

[0100]

[0101] Among them, α is the parameter of the weak classifier; β is the weight coefficient of the weak classifier. By optimizing the above two parameters, the prediction function can be improved, and finally the framework of the general Boosting gradient boosting algorithm is obtained as follows:

[0102] Fm (x) = F m-1 (x)+ρ m h(x; α m ), ρ is a constant.

[0103] Furthermore, using the power system load forecasting sequence Design a new event-triggered communication mechanism that communicates only when key events or state changes occur. The event-triggered communication mechanism is as follows: if e(t)≤σ, then Where e(t) is the similarity distance between the power system load forecast sequence and the actual sequence; is the current prediction sequence; is the final communication state value; σ is an adjustable trigger parameter, 0<σ<1. The future state of the power system predicted by the prediction model is used to determine whether the trigger condition is met. When the similarity distance between the power system load prediction sequence and the actual sequence is less than the set threshold, the trigger condition is met, the system triggers communication, and the actual sequence of the power system load is transmitted from the sending end to the receiving end. represents the triggering time, where t k <t k+1 In order to maintain generality, assuming that the first trigger time is t0 = t(0) = 0, the next trigger time is

[0104] t k+1 = inf{t>t k |e(t)≤σ}.

[0105] In this preferred embodiment, the present application achieves high-precision prediction of power load by using a power system prediction model constructed based on a gradient boosting decision tree (GBDT) algorithm and optimizing it in combination with a gradient boosting algorithm. The initial model uses multiple decision trees to learn historical load data. By continuously traversing these decision trees and accumulating their prediction results, the model can capture complex patterns and trends in the load data. Furthermore, the initial model is finely optimized by the gradient boosting algorithm, and the model gradually reduces the value of the loss function during the training process until a stable state is reached, thereby ensuring the generalization ability and prediction accuracy of the model. This method not only improves the accuracy of power system load prediction, but also enhances the model's adaptability to changes in historical load data, so that in the actual operation of the power system, future load changes can be more effectively predicted, thereby optimizing the allocation of power resources and the operation strategy of the power grid, and improving the operation efficiency and reliability of the power grid.

[0106] S03: If the similarity distance between the load forecast sequence and the second actual load sequence is less than a preset threshold, trigger the power system communication and send the second actual load sequence to the receiving end;

[0107] Otherwise, it is determined that the receiving end is a vacant sequence, and the first actual load sequence is sent to the receiving end, so that the receiving end fills the vacant sequence.

[0108] As a preferred embodiment of the first embodiment, if the similarity distance between the load prediction sequence and the second actual load sequence is less than a preset threshold, triggering power system communication is specifically as follows:

[0109] According to the load forecast sequence and the second actual load sequence, the Euclidean distance, the first DTW distance and the second DTW distance are calculated;

[0110] According to the entropy weight method, the weights of the Euclidean distance, the first DTW distance, and the second DTW distance are calculated;

[0111] Calculate the similarity distance according to the weights, the Euclidean distance, the first DTW distance and the second DTW distance;

[0112] The calculation formula of the similarity distance is:

[0113] D total (X D , Y D )=r1D1(X D , Y D )+r2D2(X D , Y D )+r3D3(X D , Y D ),

[0114] In the formula, r1, r2, and r3 are D1(X D , Y D )、D2(X D , Y D )、D3(X D , Y D ) corresponding weight; X D is the second actual load sequence; Y D is the load forecast sequence, D1(X D , Y D ) is the Euclidean distance, D2(X D , Y D ) is the first DTW distance, D3(X D , Y D ) is the second DTW distance;

[0115] When the similarity distance between the power system load forecast sequence and the actual sequence is less than the set threshold, it is considered that the power system state has changed, the trigger condition is met, and the system triggers communication.

[0116] In this preferred embodiment, the present application calculates the similarity distance between the load forecast sequence and the second actual load sequence by comprehensively considering the Euclidean distance, the first DTW distance and the second DTW distance. This comprehensive evaluation method can more comprehensively reflect the similarity between the two sequences. The entropy weight method is used to determine the weights of each distance, which ensures the objectivity and scientificity of the weight distribution, thereby making the similarity evaluation more accurate. When the calculation result of the similarity distance is less than the preset threshold, it indicates that the predicted sequence is very close to the actual sequence. At this time, the communication mechanism is triggered to send the second actual load sequence to the receiving end. The preset threshold is set between zero and one, providing a flexible judgment criterion, making the communication trigger more reasonable. This method not only improves communication efficiency and reduces unnecessary data transmission, but also ensures that communication is only carried out when there is a large deviation between the prediction and the actual through accurate similarity evaluation, thereby optimizing the use of resources and enhancing the power system's responsiveness to changes and overall stability.

[0117] As a preferred embodiment of the first embodiment, the Euclidean distance is calculated based on the load forecast sequence and the second actual load sequence, specifically:

[0118] According to the Euclidean distance formula, the Euclidean distance between the sampling points corresponding to the load forecast sequence and the second actual load sequence is calculated;

[0119] The formula for the Euclidean distance is

[0120]

[0121] In the formula, is the load forecast value at time i; is the second actual load value at time i, D1(X D , Y D ) is the Euclidean distance.

[0122] In this preferred embodiment, the present application can accurately quantify the deviation of the corresponding sampling points between the load forecast sequence and the second actual load sequence by adopting the calculation method of the Euclidean distance. The Euclidean distance formula uses the square root of the sum of the squares of the differences between the load forecast value and the actual load value at each moment to represent the distance, thereby evaluating the numerical difference between the two sequences. Specifically, for each time point t, the absolute value of the difference between the load forecast value and the second actual load value is calculated, and then these differences are squared, summed and squared to obtain the Euclidean distance of the entire time series. This calculation method takes into account the differences of all sampling points in the time series and can fully reflect the overall deviation between the predicted sequence and the actual sequence. When the Euclidean distance is small, it indicates that the prediction results of the prediction model are more consistent with the actual load changes; when the Euclidean distance is large, it indicates that there is a large deviation between the predicted results and the actual values. The evaluation mechanism of the present application helps to judge the quality of the prediction and provide a scientific basis for the operation and load management of the power system.

[0123] As a preferred embodiment of the first embodiment, the first DTW distance is calculated according to the load prediction sequence and the second actual load sequence, specifically:

[0124] Step 1: Construct a matrix D of size n×m, whose element d ij for:

[0125]

[0126] Where, d ED () is a distance calculation function, and the present invention adopts Euclidean distance.

[0127] Step 2: Construct set P D To store the point coordinates of all moving paths of the moving point, recorded as For the element d ij The moving point of the position, the position of the element it reaches next time can only be d i+1，j d i，j+1 d i+1，j+1 The position of an element in the elements. Repeat the movement of the moving point until the moving point reaches element d n，m until.

[0128] Step 3: During the movement of the moving point, the following objective function needs to be satisfied to make its total moving path the shortest:

[0129]

[0130] In the formula, is the path distance of the e-th point; D2(X D , Y D ) is the power system load forecast sequence Y D With the actual sequence XD The similarity distance between them is the first DTW distance.

[0131] In this preferred embodiment, the first DTW distance mentioned in the present application is realized by the core mechanism of the Dynamic Time Warping (DTW) algorithm, which dynamically adjusts the moving paths of each dynamic point by optimizing an objective function to ensure that the cumulative distance of these paths is the shortest. Specifically, the objective function calculates all possible matching paths between the load forecast sequence and the second actual load sequence, and selects the path that minimizes the sum of the path distances. In this process, the moving path distance of each dynamic point is added up to form the first DTW distance, which reflects the overall similarity of the two sequences in the best matching case. In this way, even if the two sequences are stretched or misaligned in time, the DTW algorithm can effectively evaluate their similarity, thereby providing a reliable measure for the accuracy of power system load forecasting.

[0132] As a preferred embodiment of the first embodiment, the second DTW distance is calculated according to the load prediction sequence and the second actual load sequence, specifically:

[0133] According to the difference method, the load forecast sequence and the second actual load sequence, a first load forecast sequence and a third actual load sequence are calculated;

[0134] Wherein, the first load forecast sequence and the third actual load sequence can reflect the fluctuation of the power system;

[0135] A second DTW distance is calculated according to the first DTW distance, the first load prediction sequence and the third actual load sequence;

[0136] More specifically, the second DTW distance is calculated by: D , Y D Through forward difference, we can obtain two sequences X′ that can map the fluctuations D , Y′ D , and then describe the local fluctuation characteristics by calculating the DTW distance, and record it as D3(X D , Y D ); D3(X D , Y D ) is the second DTW distance.

[0137] In this preferred embodiment, the present application generates a first load forecast sequence and a third actual load sequence by applying the differential method to process the original load forecast sequence and the second actual load sequence. These two new sequences can more sensitively map the load fluctuations of the power system. Then, using these new sequences, combined with the calculation method of the first DTW distance, the second DTW distance is calculated. This process involves evaluating the fluctuation pattern of the time series and dynamically time-warping it with the forecast sequence to find the best time alignment method to obtain the second DTW distance. This calculation method can more accurately reflect the local changes and fluctuations of the power system load, enhance the adaptability and responsiveness of the prediction model to the dynamic changes of the system, and provide more accurate decision support for the stable operation and load management of the power system.

[0138] As a preferred embodiment of the first embodiment, the sending of the first actual load sequence to the receiving end so that the receiving end fills the missing sequence is specifically as follows:

[0139] The missing sequence is replaced by the complete sequence of the previous period of the vacant sequence (i.e., the first actual load sequence), so that multiple sequence segments at the receiving end are combined into a reliable and complete reconstructed sequence. The specific process is as follows:

[0140] x(t k )=x(t k-1 ),

[0141] Among them, x(t k ) is the current missing sequence, x(t k-1 ) is the complete sequence of the period before the missing sequence.

[0142] The overall process of the entire technical solution of this application is as follows Figure 2 As shown, the present application first trains the power system prediction model by using historical load data to ensure the model's ability to accurately predict changes in power load. Then, by comparing the similarity between the predicted sequence and the actual sequence, communication is triggered only when the two differ greatly, effectively reducing unnecessary data transmission and reducing the consumption of communication resources. In addition, when there are gaps between actual sequences, the gaps are filled with the previous complete sequence to ensure the continuity and integrity of the data, thereby making the power system operation more stable and the decision-making more accurate. The present application not only optimizes the use of communication resources, but also enhances the power system's ability to respond to emergencies, so as to solve the problem in the prior art that system communication cannot be triggered only when the power system state changes.

[0143] Embodiment 2

[0144] Please refer to Figure 3 , which is a power system triggering communication device based on a prediction model provided in an embodiment of the present application.

[0145] In this embodiment, the prediction model-based power system trigger communication device includes an acquisition module 10 , an input-output module 20 and a trigger module 30 .

[0146] The acquisition module 10 is used to acquire first historical load data, a first actual load sequence and a second actual load sequence; wherein, the first actual load sequence and the second actual load sequence are obtained by dividing the actual load data according to a preset time period, and the first actual load sequence is the sequence of the second actual load sequence in the previous time period.

[0147] The input-output module 20 is used to input the first historical load data into a pre-trained power system prediction model so that the power system prediction model outputs a load prediction sequence.

[0148] As a preferred embodiment of the first embodiment, the pre-trained power system prediction model is obtained by training the initial power system prediction model.

[0149] As a preferred embodiment of the first embodiment, the training process of the initial power system prediction model is specifically as follows:

[0150] (1) The normalized historical power system load data is used as the input of the GBDT prediction model. The collected historical power system load data is normalized and mapped to the range of [0, 1] to prevent the prediction accuracy of the model from being reduced by different orders of magnitude. The specific process is as follows:

[0151]

[0152] Among them, z is the original load data, is the load data after normalization, z max is the maximum original load value, z min is the minimum original load value. Assuming that the short-term load of the power system at the i-th moment of the m-th day needs to be predicted, the normalized historical power system load data of the t-period before this moment is selected as the input of the GBDT prediction model.

[0153] (2) Use the GBDT algorithm to predict the load of the power system. The GBDT algorithm contains hundreds of decision trees. When using the algorithm to predict the load of the power system, it is necessary to set the initial value for the input sample data and traverse each decision tree. It will continuously adjust and update the predicted value, accumulate the prediction results of each decision tree, and obtain the final power system load prediction value. The specific process is as follows:

[0154] F(x)=F0+(β1+β2+…+β M )[T1(X)+T2(X)+T M(X)],

[0155] Where T() is a weak classifier, and its number is M; β is a weight coefficient; F() is a strong classifier composed of all weak classifiers. Through several iterations, the number of M weak classifiers is integrated into a strong classifier. Each iteration can reduce the residual of the previous model, and create a new model in the gradient direction of the residual reduction.

[0156]

[0157] Describe the loss function to determine the gradient of residual reduction. i ) is the short-term load forecast value of the power system output by the model, y i is the actual value.

[0158] (3) The Boosting gradient boosting algorithm is used to solve the problems of slow training speed and large memory usage caused by the GBDT algorithm needing to traverse the sample data countless times during training. It creates the mth weak classifier in the gradient descent direction of the m-1th weak classifier loss function. The specific process is: Assuming F(x; Q) is the prediction function, Q = {q0, q1, q2...} is the parameter of the prediction function, the expression of the prediction function can be obtained by weighted combination of several weak classifiers:

[0159]

[0160] Among them, α is the parameter of the weak classifier; β is the weight coefficient of the weak classifier. By optimizing the above two parameters, the prediction function can be improved, and finally the framework of the general Boosting gradient boosting algorithm is obtained as follows:

[0161] F m (x) = F m-1 (x)+ρ m h(x; α m ), ρ is a constant.

[0162] Furthermore, using the power system load forecasting sequence Design a new event-triggered communication mechanism that communicates only when key events or state changes occur. The event-triggered communication mechanism is as follows: if e(t)≤σ, then Where e(t) is the similarity distance between the power system load forecast sequence and the actual sequence; is the current prediction sequence; is the final communication state value; σ is an adjustable trigger parameter, 0<σ<1. The future state of the power system predicted by the prediction model is used to determine whether the trigger condition is met. When the similarity distance between the power system load prediction sequence and the actual sequence is less than the set threshold, the trigger condition is met, the system triggers communication, and the actual sequence of the power system load is transmitted from the sending end to the receiving end. represents the triggering time, where t k <t k+1 In order to maintain generality, assuming that the first trigger time is t0 = t(0) = 0, the next trigger time is

[0163] t k+1 = inf{t>t k |e(t)≤σ}.

[0164] In this preferred embodiment, the present application achieves high-precision prediction of power load by using a power system prediction model constructed based on a gradient boosting decision tree (GBDT) algorithm and optimizing it in combination with a gradient boosting algorithm. The initial model uses multiple decision trees to learn historical load data. By continuously traversing these decision trees and accumulating their prediction results, the model can capture complex patterns and trends in the load data. Furthermore, the initial model is finely optimized by the gradient boosting algorithm, and the model gradually reduces the value of the loss function during the training process until a stable state is reached, thereby ensuring the generalization ability and prediction accuracy of the model. This method not only improves the accuracy of power system load prediction, but also enhances the model's adaptability to changes in historical load data, so that in the actual operation of the power system, future load changes can be more effectively predicted, thereby optimizing the allocation of power resources and the operation strategy of the power grid, and improving the operation efficiency and reliability of the power grid.

[0165] Trigger module 30: If the similarity distance between the load forecast sequence and the second actual load sequence is less than a preset threshold, trigger the power system communication and send the second actual load sequence to the receiving end;

[0166] Otherwise, it is determined that the receiving end is a vacant sequence, and the first actual load sequence is sent to the receiving end, so that the receiving end fills the vacant sequence.

[0167] As a preferred embodiment of the first embodiment, if the similarity distance between the load prediction sequence and the second actual load sequence is less than a preset threshold, triggering power system communication is specifically as follows:

[0168] According to the load forecast sequence and the second actual load sequence, the Euclidean distance, the first DTW distance and the second DTW distance are calculated;

[0169] According to the entropy weight method, the weights of the Euclidean distance, the first DTW distance, and the second DTW distance are calculated;

[0170] Calculate the similarity distance according to the weights, the Euclidean distance, the first DTW distance and the second DTW distance;

[0171] The calculation formula of the similarity distance is:

[0172] D total (X D , Y D )=r1D1(X D , Y D )+r2D2(X D , Y D )+r3D3(X D , Y D ),

[0173] In the formula, r1, r2, and r3 are D1(X D , Y D )、D2(X D , Y D )、D3(X D , Y D ) corresponding weight; X D is the second actual load sequence; Y D is the load forecast sequence, D1(X D , Y D ) is the Euclidean distance, D2(X D , Y D ) is the first DTW distance, D3(X D , Y D ) is the second DTW distance;

[0174] When the similarity distance between the power system load forecast sequence and the actual sequence is less than the set threshold, it is considered that the power system state has changed, the trigger condition is met, and the system triggers communication.

[0175] In this preferred embodiment, the present application calculates the similarity distance between the load forecast sequence and the second actual load sequence by comprehensively considering the Euclidean distance, the first DTW distance and the second DTW distance. This comprehensive evaluation method can more comprehensively reflect the similarity between the two sequences. The entropy weight method is used to determine the weights of each distance, which ensures the objectivity and scientificity of the weight distribution, thereby making the similarity evaluation more accurate. When the calculation result of the similarity distance is less than the preset threshold, it indicates that the predicted sequence is very close to the actual sequence. At this time, the communication mechanism is triggered to send the second actual load sequence to the receiving end. The preset threshold is set between zero and one, providing a flexible judgment criterion, making the communication trigger more reasonable. This method not only improves communication efficiency and reduces unnecessary data transmission, but also ensures that communication is only carried out when there is a large deviation between the prediction and the actual through accurate similarity evaluation, thereby optimizing the use of resources and enhancing the power system's responsiveness to changes and overall stability.

[0176] As a preferred embodiment of the first embodiment, the Euclidean distance is calculated based on the load forecast sequence and the second actual load sequence, specifically:

[0177] According to the Euclidean distance formula, the Euclidean distance between the sampling points corresponding to the load forecast sequence and the second actual load sequence is calculated;

[0178] The formula for the Euclidean distance is

[0179]

[0180] In the formula, is the load forecast value at time i; is the second actual load value at time i, D1(X D , Y D ) is the Euclidean distance.

[0181] In this preferred embodiment, the present application can accurately quantify the deviation of the corresponding sampling points between the load forecast sequence and the second actual load sequence by adopting the calculation method of the Euclidean distance. The Euclidean distance formula uses the square root of the sum of the squares of the differences between the load forecast value and the actual load value at each moment to represent the distance, thereby evaluating the numerical difference between the two sequences. Specifically, for each time point t, the absolute value of the difference between the load forecast value and the second actual load value is calculated, and then these differences are squared, summed and squared to obtain the Euclidean distance of the entire time series. This calculation method takes into account the differences of all sampling points in the time series and can fully reflect the overall deviation between the predicted sequence and the actual sequence. When the Euclidean distance is small, it indicates that the prediction results of the prediction model are more consistent with the actual load changes; when the Euclidean distance is large, it indicates that there is a large deviation between the predicted results and the actual values. The evaluation mechanism of the present application helps to judge the quality of the prediction and provide a scientific basis for the operation and load management of the power system.

[0182] As a preferred embodiment of the first embodiment, the first DTW distance is calculated according to the load prediction sequence and the second actual load sequence, specifically:

[0183] Step 1: Construct a matrix D of size n×m, whose element d ij for:

[0184]

[0185] Where, d ED () is a distance calculation function, and the present invention adopts Euclidean distance.

[0186] Step 2: Construct set P D To store the point coordinates of all moving paths of the moving point, recorded as For the element d ij The moving point of the position, the position of the element it reaches next time can only be d i+1，j d i，j+1 d i+1，j+1 The position of an element in the elements. Repeat the movement of the moving point until the moving point reaches element d n，m until.

[0187] Step 3: During the movement of the moving point, the following objective function needs to be satisfied to make its total moving path the shortest:

[0188]

[0189] In the formula, is the path distance of the e-th point; D2(X D , Y D ) is the power system load forecast sequence Y D With the actual sequence XD The similarity distance between them is the first DTW distance.

[0190] In this preferred embodiment, the first DTW distance mentioned in the present application is realized by the core mechanism of the Dynamic Time Warping (DTW) algorithm, which dynamically adjusts the moving paths of each dynamic point by optimizing an objective function to ensure that the cumulative distance of these paths is the shortest. Specifically, the objective function calculates all possible matching paths between the load forecast sequence and the second actual load sequence, and selects the path that minimizes the sum of the path distances. In this process, the moving path distance of each dynamic point is added up to form the first DTW distance, which reflects the overall similarity of the two sequences in the best matching case. In this way, even if the two sequences are stretched or misaligned in time, the DTW algorithm can effectively evaluate their similarity, thereby providing a reliable measure for the accuracy of power system load forecasting.

[0191] As a preferred embodiment of the first embodiment, the second DTW distance is calculated according to the load prediction sequence and the second actual load sequence, specifically:

[0192] According to the difference method, the load forecast sequence and the second actual load sequence, a first load forecast sequence and a third actual load sequence are calculated;

[0193] Wherein, the first load forecast sequence and the third actual load sequence can reflect the fluctuation of the power system;

[0194] A second DTW distance is calculated according to the first DTW distance, the first load prediction sequence and the third actual load sequence;

[0195] More specifically, the second DTW distance is calculated by: D , Y D Through forward difference, we can obtain two sequences X′ that can map the fluctuations D , Y′ D , and then describe the local fluctuation characteristics by calculating the DTW distance, and record it as D3(X D , Y D ); D3(X D , Y D ) is the second DTW distance.

[0196] In this preferred embodiment, the present application generates a first load forecast sequence and a third actual load sequence by applying the differential method to process the original load forecast sequence and the second actual load sequence. These two new sequences can more sensitively map the load fluctuations of the power system. Then, using these new sequences, combined with the calculation method of the first DTW distance, the second DTW distance is calculated. This process involves evaluating the fluctuation pattern of the time series and dynamically time-warping it with the forecast sequence to find the best time alignment method to obtain the second DTW distance. This calculation method can more accurately reflect the local changes and fluctuations of the power system load, enhance the adaptability and responsiveness of the prediction model to the dynamic changes of the system, and provide more accurate decision support for the stable operation and load management of the power system.

[0197] As a preferred embodiment of the first embodiment, the sending of the first actual load sequence to the receiving end so that the receiving end fills the missing sequence is specifically as follows:

[0198] The missing sequence is replaced by the complete sequence of the previous period of the vacant sequence (i.e., the first actual load sequence), so that multiple sequence segments at the receiving end are combined into a reliable and complete reconstructed sequence. The specific process is as follows:

[0199] x(t k )=x(t k-1 ),

[0200] Among them, x(t k ) is the current missing sequence, x(t k-1 ) is the complete sequence of the previous period of the vacant sequence.

[0201] The overall process of the entire technical solution of this application is as follows Figure 2 As shown, the device uses two modules to divide the work and coordinate the work to better trigger power system communication when the state changes. The present application first trains the power system prediction model by using historical load data to ensure the model's ability to accurately predict changes in power load. Then, by comparing the similarity between the predicted sequence and the actual sequence, communication is triggered only when the two differ greatly, effectively reducing unnecessary data transmission and reducing the consumption of communication resources. In addition, when there are gaps between the actual sequences, the continuity and integrity of the data are ensured by filling the gaps with the previous complete sequence, thereby making the power system operation more stable and the decision-making more accurate. The present application not only optimizes the use of communication resources, but also enhances the power system's ability to respond to emergencies, so as to solve the problem in the prior art that system communication cannot be triggered only when the power system state changes.

[0202] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A power system trigger communication method based on a prediction model, characterized in that: include: Acquire first historical load data, a first actual load sequence, and a second actual load sequence; wherein the first actual load sequence and the second actual load sequence are obtained by dividing the actual load data according to a preset time period, and the first actual load sequence is a sequence of the second actual load sequence in the previous time period; Inputting the first historical load data into a pre-trained power system prediction model so that the power system prediction model outputs a load prediction sequence; If the similarity distance between the load forecast sequence and the second actual load sequence is less than a preset threshold, triggering power system communication, and sending the second actual load sequence to the receiving end; Otherwise, it is determined that the receiving end is a vacant sequence, and the first actual load sequence is sent to the receiving end, so that the receiving end fills the vacant sequence.

2. The power system trigger communication method based on prediction model according to claim 1, characterized in that: The pre-trained neural network is specifically: Acquiring second historical load data; Inputting the second historical load data into the initial power system prediction model for training, and optimizing the initial power system prediction model using a gradient boosting algorithm until the value of the loss function no longer changes, stopping the training, and obtaining a trained power system prediction model; The initial power system prediction model is constructed based on the GBDT algorithm. The initial power system prediction model includes multiple decision trees. During the training process, multiple decision trees are continuously traversed and the prediction results of each decision tree are accumulated.

3. The power system trigger communication method based on prediction model according to claim 1, characterized in that: If the similarity distance between the load forecast sequence and the second actual load sequence is less than a preset threshold, triggering power system communication is specifically: According to the load forecast sequence and the second actual load sequence, the Euclidean distance, the first DTW distance and the second DTW distance are calculated; According to the entropy weight method, the weights of the Euclidean distance, the first DTW distance, and the second DTW distance are calculated; Calculate the similarity distance according to the weights, the Euclidean distance, the first DTW distance and the second DTW distance; The calculation formula of the similarity distance is: D total (X D ,Y D )=r1D1(X D ,Y D )+r2D2(X D ,Y D )+r3D3(X D ,Y D ), In the formula, r1, r2, and r3 are D1(X D , Y D )、D2(X D , Y D )、D3(X D , Y D ) corresponding weight; X D is the second actual load sequence; Y D is the load forecast sequence, D1(X D , Y D ) is the Euclidean distance, D2(X D , Y D ) is the first DTW distance, D3(X D , Y D ) is the second DTW distance; If the similarity distance is less than a preset threshold, power system communication is triggered.

4. The power system trigger communication method based on prediction model according to claim 3, characterized in that: The Euclidean distance is calculated based on the load forecast sequence and the second actual load sequence, specifically: According to the Euclidean distance formula, the Euclidean distance between the sampling points corresponding to the load forecast sequence and the second actual load sequence is calculated; The formula for the Euclidean distance is In the formula, is the load forecast value at time i; is the second actual load value at time i, D1(X D , Y D ) is the Euclidean distance.

5. The power system trigger communication method based on prediction model according to claim 3, characterized in that: The first DTW distance is calculated based on the load forecast sequence and the second actual load sequence, specifically: The first DTW distance is calculated according to the objective function, the load forecast sequence, and the second actual load sequence; Wherein, the objective function is obtained by determining each moving path of each dynamic point and ensuring that the cumulative distance of each moving path is the shortest; The formula of the objective function is In the formula, is the moving path distance of the e-th dynamic point; D2(X D , Y D ) is the load forecast sequence Y D With the second actual load sequence X D The first DTW distance between them.

6. The power system trigger communication method based on prediction model according to claim 3, characterized in that: The second DTW distance is calculated based on the load prediction sequence and the second actual load sequence, specifically: According to the difference method, the load forecast sequence and the second actual load sequence, a first load forecast sequence and a third actual load sequence are calculated; Wherein, the first load forecast sequence and the third actual load sequence can reflect the fluctuation of the power system; A second DTW distance is calculated according to the first DTW distance, the first load prediction sequence and the third actual load sequence.

7. A power system trigger communication device based on a prediction model, characterized in that: It includes an acquisition module, an input and output module, and a trigger module; The acquisition module is used to acquire first historical load data, a first actual load sequence, and a second actual load sequence; wherein the first actual load sequence and the second actual load sequence are obtained by dividing the actual load data according to a preset time period, and the first actual load sequence is a sequence of the second actual load sequence in the previous time period; The input-output module is used to input the first historical load data into a pre-trained power system prediction model so that the power system prediction model outputs a load prediction sequence; The trigger module is used to trigger power system communication and send the second actual load sequence to a receiving end if the similarity distance between the load forecast sequence and the second actual load sequence is less than a preset threshold; Otherwise, it is determined that the receiving end is a vacant sequence, and the first actual load sequence is sent to the receiving end, so that the receiving end fills the vacant sequence.

8. The power system trigger communication device based on prediction model according to claim 7, characterized in that: The pre-trained neural network is specifically: Acquiring second historical load data; Inputting the second historical load data into the initial power system prediction model for training, and optimizing the initial power system prediction model using a gradient boosting algorithm until the value of the loss function no longer changes, stopping the training, and obtaining a trained power system prediction model; The initial power system prediction model is constructed based on the GBDT algorithm. The initial power system prediction model includes multiple decision trees. During the training process, multiple decision trees are continuously traversed and the prediction results of each decision tree are accumulated.

9. The power system trigger communication device based on prediction model according to claim 7, characterized in that: If the similarity distance between the load forecast sequence and the second actual load sequence is less than a preset threshold, triggering power system communication is specifically: According to the load forecast sequence and the second actual load sequence, the Euclidean distance, the first DTW distance and the second DTW distance are calculated; According to the entropy weight method, the weights of the Euclidean distance, the first DTW distance, and the second DTW distance are calculated; Calculate the similarity distance according to the weights, the Euclidean distance, the first DTW distance and the second DTW distance; The calculation formula of the similarity distance is: D total (XD , AND D )=r1D1(X D ,AND D )+r2D2(X D ,AND D )+r3D3(X D ,AND D ), In the formula, r1, r2, and r3 are D1(X D , Y D )、D2(X D , Y D )、D3(X D , Y D ) corresponding weight; X D is the second actual load sequence; Y D is the load forecast sequence, D1(X D ,Y D ) is the Euclidean distance, D2(X D , Y D ) is the first DTW distance, D3(X D , Y D ) is the second DTW distance; If the similarity distance is less than a preset threshold, power system communication is triggered.

10. The power system trigger communication device based on prediction model according to claim 9, characterized in that: The Euclidean distance is calculated based on the load forecast sequence and the second actual load sequence, specifically: According to the Euclidean distance formula, the Euclidean distance between the sampling points corresponding to the load forecast sequence and the second actual load sequence is calculated; The formula for the Euclidean distance is In the formula, is the load forecast value at time i; is the second actual load value at time i, D1(X D , Y D ) is the Euclidean distance.

11. The power system trigger communication device based on prediction model according to claim 9, characterized in that: The first DTW distance is calculated based on the load forecast sequence and the second actual load sequence, specifically: The first DTW distance is calculated according to the objective function, the load forecast sequence, and the second actual load sequence; Wherein, the objective function is obtained by determining each moving path of each dynamic point and ensuring that the cumulative distance of each moving path is the shortest; The formula of the objective function is In the formula, is the moving path distance of the e-th dynamic point; D2(X D , Y D ) is the load forecast sequence Y D With the second actual load sequence X D The first DTW distance between them.

12. The power system trigger communication device based on prediction model according to claim 9, characterized in that: The second DTW distance is calculated based on the load prediction sequence and the second actual load sequence, specifically: According to the difference method, the load forecast sequence and the second actual load sequence, a first load forecast sequence and a third actual load sequence are calculated; Wherein, the first load forecast sequence and the third actual load sequence can reflect the fluctuation of the power system; A second DTW distance is calculated according to the first DTW distance, the first load prediction sequence and the third actual load sequence.

Images