Load long-time power sequence generation method, system and equipment based on sequence combination and medium

Abstract

A sequence combination-based load long-time power sequence generation method, system and device, and a medium, the method comprising: collecting load-related data, and dividing a load long-time power sequence into two segments; constructing a load power prediction model by adopting a deep directed residual map convolutional neural network, and generating a load power sequence based on load prediction, namely a first section of load power sequence; generating a load power sequence by adopting fuzzy C-means clustering and a Markov method, and correcting the load power sequence according to the constructed load power model considering load demand development and the power demand response model to obtain a second section of load power sequence; and splicing the first section of power sequence and the second section of power sequence according to a time sequence to form a load long-time power sequence. The method not only combines the advantage of high short-term load prediction precision, but also fully considers the important influence of rapid increase of electrical load and large-scale application of demand side response on the load power sequence in China at present.

Description

Technical Field

[0001] The present invention belongs to the technical field of production simulation, and in particular relates to a method, system, equipment and medium for generating a long-time power sequence of a load based on sequence combination. Background Art

[0002] The power system is a dynamic system that is in real-time balance, with power generation, transmission, distribution, and consumption completed instantaneously. With the rapid advancement of my country's energy transformation, the power system is undergoing unprecedented changes. Traditional power systems have always followed the "source follows load" operation mode, that is, the power generation side adjusts according to the demand of the load side, which can no longer meet the requirements of the new power system for power and electricity balance. In recent years, the operation mode of my country's power system is undergoing a transformation from "source follows load" to the multi-faceted interaction of "source, grid, load, and storage". Power and electricity balance analysis is a basic task of the power system, and production simulation technology is mainly used for power and electricity balance analysis of the power system. As an important input data for power system production simulation, load power series generation has always been a research focus.

[0003] Currently, some research has been conducted on generating load power series. The paper [A Method and System for Power Grid Load Series Forecasting Based on Generative Adversarial Networks. Application Publication No. CN 114239934 A] encodes meteorological temperature and humidity, economic data, and power load data, which influence grid load, into a long-short-term memory network. The generative adversarial network then learns the coupling between these factors in a purely data-driven manner. Finally, a long-short-term memory network decoder outputs future series predictions. While this approach offers the advantage of high short-term load forecasting accuracy, it significantly decreases as the forecast horizon lengthens, making it unsuitable for generating long-term load series. The paper [A Method for Generating Regional Load Series Scenarios Based on Variational Autoencoders. Application Publication No. CN 117034062 A] combines a Markov chain state transition matrix with a variational autoencoder to generate a daily series of future meteorological day types. Furthermore, a meteorological-load multivariate regression model is established for each region based on the geographical distribution of load. The regional total load series scenario is then derived by superimposing the loads of these regions. At present, my country's power market construction has entered the fast lane, among which demand-side response has developed rapidly. For the long-term load power sequence, load demand development and demand-side response will have an important impact on the load power sequence. However, the above-mentioned results have not taken it into consideration. The generated long-term load power sequence is prone to cause large errors under the current situation of rapid growth of my country's electricity load and large-scale application of demand-side response. Summary of the Invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method, system, device and medium for generating a long-term load power sequence based on sequence combination.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a method for generating a long-term load power sequence based on sequence combination, comprising the following steps:

[0007] Step S1: Collecting load-related data, including load forecast data and demand-side response data;

[0008] Step S2: Based on the load power sequence generation cycle and load forecast accuracy, the long-term load power sequence generation is divided into the first-stage load power sequence generation and the second-stage load power sequence generation;

[0009] Step S3: A load power forecasting model is constructed using a deep directed residual graph convolutional neural network. The load forecasting data collected in step S1 is substituted into the load power forecasting model as input data to generate a load power sequence based on the load forecast, i.e., the first load power sequence.

[0010] Step S4: Fuzzy C-means clustering and Markov method are used to generate a load power sequence, and the load power sequence generated based on the fuzzy C-means clustering and Markov method is corrected according to the load power model and power demand response model that consider the development of load demand, to obtain a second load power sequence;

[0011] Step S5: splicing the first load power sequence and the second load power sequence according to time sequence to form a long-term load power sequence.

[0012] Furthermore, in step S1, the load forecast data includes meteorological data, holiday and seasonal data, macroeconomic data, and key user data. The meteorological data includes historical load power, temperature, humidity, wind speed, and rainfall. The holiday and seasonal data includes seasons, holidays, weekdays, and weekends. The macroeconomic data includes GDP growth rate, population changes, and industrial policy adjustments. The key user data includes major project construction and large user load curves; the demand-side response data includes the initial electricity price, electricity volume, price elasticity coefficient, and negative and positive reserve capacity of the price-type demand response.

[0013] Furthermore, in step S2, the period T for generating the long-term power sequence of the load is determined according to the objectives of the new energy absorption capacity assessment and the research on the power and electricity balance of the new power system, and the prediction duration T1 used for generating the long-term power sequence of the load is determined according to the accuracy of the load prediction at different time scales, and the power sequence of the load with a duration of T is divided into two sections, the first section of the load power sequence has a duration of T1, and the second section of the load power sequence has a duration of T2, and T2=T-T1.

[0014] Furthermore, in step S3, a deep directed residual graph convolutional neural network is used to learn the historical load power, meteorological data, holiday and seasonal data, macroeconomic data, and key user data collected in step S1, and then a load power forecasting model is constructed; the meteorological data, holiday and seasonal data, macroeconomic data, and key user data with a load forecasting duration of T1 are used as input data and substituted into the load power forecasting model to obtain a load power sequence [P1, P2, ..., P n ], where n represents the amount of power contained in the load power sequence with a duration of T1, n=T1 / t, where t is the sampling duration of two adjacent meteorological data in the load forecast;

[0015] The calculation steps of the deep directed residual graph convolutional neural network include:

[0016] First, set the node's autocorrelation weight to 1, and then construct the in-degree adjacency matrix considering the node self-loop and out-degree adjacency matrix To represent the weighted graph The connection relationship and connection weight between the nodes:

[0017]

[0018]

[0019] Where, represents the significance level; Indicates s i and s j There is a connection relationship, the direction is s i ←s j , the weight is ;

[0020] Based on the in-degree adjacency matrix and out-degree adjacency matrix , perform graph convolution operation, and integrate the residual network into the inter-layer propagation process to obtain the residual graph convolutional neural network; considering the established adjacency matrix and It is an asymmetric matrix, and the adjacency matrix is ​​normalized asymmetric. The inter-layer propagation formula of the residual graph convolutional neural network is:

[0021]

[0022] Where, represents the activation function, express The degree matrix of express The degree matrix of 、 Indicates the The feature matrix of the layer input, 、 Indicates the +1 layer input feature matrix, 、 Indicates the The weight matrix of the layer;

[0023] Then, by stacking multiple layers of residual graph convolutional neural networks and integrating the residual network into the output layer, a deep residual graph convolutional neural network is obtained to achieve Deep learning;

[0024] Finally, the outputs of the deep residual graph convolutional neural network based on the in-degree adjacency matrix and the deep residual graph convolutional neural network based on the out-degree adjacency matrix are spliced ​​to obtain a deep directed residual graph convolutional neural network.

[0025] Furthermore, in step S4, the step of generating a load power sequence using fuzzy C-means clustering and Markov method includes:

[0026] ① Load power data cleaning: The integrated moving average autoregressive model-generalized autoregressive conditional heteroskedasticity model is used to identify abnormal data in the load historical power series collected in step S1 and remove abnormal data. The expression of the integrated moving average autoregressive model is:

[0027]

[0028] Where, represents the input load power sequence; B represents the lag operator, that is, ; Represents the n-order difference calculation process; represents the fixed parameters obtained by fitting; Represents residual; E represents expected value operation; Var represents variance operation, that is, is the residual Variance of the series; 、 The expression is as follows:

[0029]

[0030] Where, P AR , Q MA represents the order of the polynomial; and represents the coefficient obtained by fitting;

[0031] The expression of the generalized autoregressive conditional heteroskedasticity model is:

[0032]

[0033] In the formula, N represents normal distribution; ~ represents that the variable obeys a certain probability distribution; Represents all historical information before time (t-1); represents the conditional variance; are independent and identically distributed variables that obey the standard normal distribution, iid means independent and identically distributed; 、 and is the fixed parameter and coefficient obtained by fitting; P G , Q AC Representation coefficient and coefficients The order of

[0034] ② Load power clustering: The load power sequence after removing abnormal data is clustered using the fuzzy C-means clustering method, and the load daily power sequence is aggregated into a set with multiple typical sequences as the core. The objective function of the fuzzy C-means clustering method is:

[0035]

[0036] Where U is the membership matrix; u i.j is the membership value, u i . j ∈ [ 0 , 1 ] ; is the jth cluster center; is the jth cluster center c j and the i-th load power x i The Euclidean distance of ; m is the weighted index;

[0037] ③ Generate load power sequences using the Markov method: Based on the clustering results obtained from the load power clustering step, calculate the state transition matrix of different typical sequences; based on the state transition matrix, use the Markov method to generate random sequences that convert between multiple typical sequences; reconstruct the random sequences that convert between multiple typical sequences to generate a load power sequence with a duration of T2;

[0038] The calculation formula of the state transfer matrix of different load typical sequences is:

[0039]

[0040] Where, Indicates the load series set category of the previous day; represents the set of typical load sequences of the mth type; i and (i+1) represent time numbers in days; N represents the number of days of historical samples; I represents the state judgment variable, which is defined as:

[0041]

[0042] Using the Markov method, we can generate random sequences that transform multiple typical sequences into each other, as shown in the following formula:

[0043]

[0044] The random sequences converted from multiple typical sequences are reconstructed to generate a load power sequence with a duration of T2. The sequence reconstruction steps are as follows:

[0045] Step 1: Initialize the typical daily load sequence number n=1 extracted by Markov and the random load power sequence ;

[0046] Step 2: Read the Markov state chain ,in accordance with Corresponding load typical sequence set category, randomly extract a load sequence from the corresponding set;

[0047] Step 3: Add the extracted load sequence to the load power sequence P 0 , update the Markov sequence number n=n+1;

[0048] Step 4: Determine the total length of the load power sequence. If the sequence number t is greater than or equal to the amount of data with a duration of T2, truncate the sequence according to the amount of data with a duration of T2 and terminate the loop. Otherwise, proceed to the next step.

[0049] Step 5: Determine the last power value P of the extracted load sequence f (t), and at the same time determine the typical sequence category corresponding to the Markov sequence number n , select the initial power and P from the corresponding load typical sequence set f (t) A temporary set of several relatively close daily load power sequences is randomly selected from them;

[0050] Step 6: Repeat steps 3 to 5 until the termination condition is reached;

[0051] A load power model considering the development of load demand is established. The calculation formula is as follows:

[0052] { P ′ t = P t 0 × C t R / C his / her C t = 1 R = C his / her C t R = ( C ( t 1 + 1 ) M − C t 1 M ) × sort [ rand ( ( t 1 + 1 ) M − t 1 M + 1 ) ] + C t 1 M t 1 M ≤ t < ( t 1 + 1 ) M C M = { C t M | t M ∈ { t 1 , t 2 , ⋯ , t N M } }

[0053] Where C hisIndicates historical load demand; is the load demand at time t in the future considering the load demand development; represents the load demand at the initial moment of the evaluation period; represents the power at time t of the load power sequence generated by fuzzy C-means clustering and Markov method; is the power at time t considering the development of load demand; t1 M and (t1+1) M are two adjacent major time nodes; and represents the load demand corresponding to two adjacent main time nodes; t M Indicates a specific time; Indicates the specified time node; N indicates the number of specified time nodes; represents the expected load demand at the corresponding time node, which is obtained by using a deep directed residual graph convolutional neural network in step S3 to take future meteorological data, holiday and seasonal data, macroeconomic data, and key user data as input data and substitute them into the load power forecasting model; C M Represents the set of load demands at different time nodes in the future;

[0054] And according to the load power model that takes into account the development of load demand, the load power sequence generated by fuzzy C-means clustering and Markov method is corrected for the first time to obtain the load power sequence after the first correction;

[0055] Establish an electricity demand response model, including a price-based demand response model and an incentive-based demand response model; perform a second correction on the load power sequence after the first correction based on the established electricity demand response model to obtain the second load power sequence. The calculation formula is as follows:

[0056]

[0057] Where, represents the load response at time t of price-based demand response, is the reserve capacity called for by the incentive-based demand response at time t;

[0058] The price-based demand response model, that is, the load response quantity at time t of price-based demand response, is The calculation formula is:

[0059]

[0060] D P t DR . rate = [ D u 1 DR . rate D u 2 DR . rate ⋮ D u TD DR . rate ] T × [ ς 1 − 1 ς 1 − 2 … ς 1 − TD ς 2 − 1 ς 2 − 2 … ς 2 − TD ⋮ ⋮ ⋱ ⋮ ς TD − 1 ς TD − 2 … ς TD − TD ]

[0061] Where, represents the change in electricity price of price-type demand response at time t; and They represent the initial load demand and electricity price participating in price-based demand response respectively; and They represent the proportion of load demand and electricity price change at time t respectively; It represents the price elasticity coefficient. The price elasticity coefficient considering the development of load demand is calculated as follows:

[0062] { ς M . DR = { ς t M . DR DR | t M . DR ∈ { t 1 DR , t 2 DR , ⋯ , t N M . DR DR } } ς t R . DR = ( ς ( t 1 + 1 ) M . DR DR − ς t 1 M . DR DR ) × sort [ rand ( ( t 1 + 1 ) M . DR − t 1 M . DR + 1 ) ] + ς t 1 M . DR DR t 1 M . DR ≤ t < ( t 1 + 1 ) M . DR ς 1 R . DR = ς his / her R . DR

[0063] Where, Represents the set of price elasticity coefficients at different time points in the evaluation period; Indicates a specific time. Indicates the time node specified in the evaluation cycle. Indicates the number of specified time nodes; Indicates The price elasticity coefficient when ; and For two adjacent preset time nodes, and Indicates the price elasticity coefficient corresponding to two adjacent preset time nodes, Indicates the price elasticity coefficient at the preset time node t, represents the historical price elasticity coefficient, represents the price elasticity coefficient at the beginning of the evaluation period;

[0064] The incentive demand response model, that is, the reserve capacity called at time t by the incentive demand response The following requirements should be met:

[0065]

[0066] Where, and denote the purchased incentive-based demand response negative reserve and positive reserve capacity respectively; and is the corresponding [0,1] state variable. As electricity market development progresses, industrial users are more willing to participate in incentive-based demand response, and the capacity of incentive-based demand response continues to increase. Its model is a continuous-time stochastic variation model, and its calculation formula refers to the price elasticity coefficient calculation formula that considers load demand development.

[0067] In a second aspect, the present invention provides a system for generating a long-term load power sequence based on sequence combination, comprising:

[0068] Load data collection module, used to collect load forecast data and demand-side response data;

[0069] A load power sequence division module is used to divide the generation of a long-term load power sequence into a first-segment load power sequence generation and a second-segment load power sequence generation according to the load power sequence generation cycle and load forecast accuracy;

[0070] The first load power sequence generation module is used to build a load power prediction model using a deep directed residual graph convolutional neural network, substitute the load prediction data collected by the load data collection module as input data into the load power prediction model, and generate a load power sequence based on the load prediction, i.e., the first load power sequence;

[0071] The second load power sequence generation module is used to generate a load power sequence using fuzzy C-means clustering and Markov method, and to correct the load power sequence generated based on fuzzy C-means clustering and Markov method according to the constructed load power model and power demand response model considering the development of load demand, to obtain the second load power sequence;

[0072] The load power sequence splicing module is used to splice the first load power sequence and the second load power sequence according to time sequence to form a long-term load power sequence.

[0073] In a third aspect, the present invention provides an electronic device comprising a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the above-mentioned method for generating a long-time load power sequence based on sequence combination.

[0074] In a fourth aspect, the present invention provides a computer-readable storage medium storing at least one instruction, which, when executed by a processor, implements the above-mentioned method for generating a long-time load power sequence based on sequence combination.

[0075] The present invention divides the generation of long-term load power sequences into two stages. In view of the high accuracy of short-term load forecasting, a load power sequence based on load forecasting is generated based on load forecasting and a constructed load power forecasting model, i.e., the first stage of load power sequence. Fuzzy C-means clustering and Markov methods are used to generate load power sequences, and the load power sequence generated based on the fuzzy C-means clustering and Markov methods is corrected based on the constructed load power model and power demand response model that consider the development of load demand, thereby obtaining the second stage of load power sequence. The two stages of load power sequences are then spliced ​​together according to the time sequence to form a long-term load power sequence. The present invention combines the advantages of high short-term load forecasting accuracy for generating load power sequences, and fully considers the important impact of load demand development and demand-side response on load power sequences under the current rapid growth of electricity load in my country and the large-scale application of demand-side response in generating long-term load power sequences. The generated long-term load power sequence is more realistic. At the same time, the present invention adopts a deep directed residual graph convolutional neural network to learn the load forecast data, and then constructs a load measurement model, which not only predicts the short-term load power series, but also predicts the load demand at major time nodes in the medium and long term; the deep directed residual graph convolutional neural network adopted can realize deep learning and feature extraction of weighted graphs, overcoming the problem that convolutional neural networks can only process regularly arranged Euclidean spatial data and cannot process non-Euclidean spatial data, resulting in application limitations. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0077] Figure 1 A flow chart of a method for generating a long-term load power sequence based on sequence combination provided by an exemplary embodiment of the present invention;

[0078] Figure 2 A schematic structural diagram of a load long-term power sequence generation system based on sequence combination provided by an exemplary embodiment of the present invention;

[0079] Figure 3 A structural block diagram of an electronic device provided by an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0080] Below, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0081] It should be noted that the relative arrangement of components and steps, the numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.

[0082] Those skilled in the art will understand that the terms "first" and "second" in the embodiments of the present invention are only used to distinguish different steps, devices or modules, and neither represent any specific technical meaning nor indicate the necessary logical order between them.

[0083] It should also be understood that, in the embodiments of the present invention, “a plurality of” may refer to two or more than two, and “at least one” may refer to one, two or more than two.

[0084] It should also be understood that any component, data or structure mentioned in the embodiments of the present invention can generally be understood as one or more, unless explicitly limited or otherwise indicated in the context.

[0085] In addition, the term "and / or" in the present invention is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, A and B exist at the same time, and B exists alone.

[0086] In addition, the character “ / ” in the present invention generally indicates that the preceding and following related objects are in an “or” relationship.

[0087] It should also be understood that the description of the various embodiments of the present invention focuses on the differences between the various embodiments, and the same or similar aspects thereof can be referenced with each other. For the sake of brevity, they will not be described one by one.

[0088] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0089] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0090] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0091] It should be noted that like reference numerals and letters refer to like items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0092] Embodiments of the present invention can be applied to electronic devices such as terminal devices, computer systems, and servers, and can operate in conjunction with numerous other general-purpose or specialized computing system environments or configurations. Examples of well-known terminal devices, computing systems, environments, and / or configurations suitable for use with terminal devices, computer systems, servers, and other electronic devices include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, minicomputer systems, mainframe computer systems, and distributed cloud computing technology environments including any of the above.

[0093] Electronic devices such as terminal devices, computer systems, and servers can be described in the general context of computer system-executable instructions (such as program modules) executed by the computer system. Generally, program modules can include routines, programs, object programs, components, logic, data structures, and the like that perform specific tasks or implement specific abstract data types. Computer systems / servers can be implemented in distributed cloud computing environments, where tasks are performed by remote processing devices linked via a communications network. In distributed cloud computing environments, program modules can be located on local or remote computer system storage media, including storage devices.

[0094] Exemplary method

[0095] Figure 1 This is a flow chart of a method for generating a long-time load power sequence based on sequence combination provided by an exemplary embodiment of the present invention. This embodiment can be applied to electronic devices, such as Figure 1 As shown, the method for generating a long-term load power sequence based on sequence combination includes the following steps:

[0096] Step S1: Collect load-related data, which includes load forecast data and demand-side response data.

[0097] Specifically, the collected load forecast data include meteorological data, holiday and seasonal data, macroeconomic data, and key user data. Meteorological data include historical load power, temperature, humidity, wind speed, and rainfall. Holiday and seasonal data include seasons, holidays, weekdays, and weekends. Macroeconomic data include GDP growth rate, population changes, and industrial policy adjustments. Key user data include major project construction and large user load curves. The collected demand-side response data include the initial electricity price, electricity volume, price elasticity coefficient, and negative and positive reserve capacity of incentive-based demand response.

[0098] Step S2: Based on the load power sequence generation cycle and load forecast accuracy, the long-term load power sequence generation is divided into the first-stage load power sequence generation and the second-stage load power sequence generation.

[0099] Specifically, according to the objectives of new energy absorption capacity assessment and new power system power balance research, the period T of long-term load power sequence generation is determined, for example, 1 year. According to the accuracy of load forecast at different time scales, the forecast duration T1 used for long-term load power sequence generation is determined. The power sequence of load duration T is divided into two segments. The first segment of load power sequence duration is T1, and the second segment of load power sequence duration is T2, and T2=T-T1.

[0100] Step S3: Use a deep directed residual graph convolutional neural network to build a load power prediction model, substitute the load prediction data collected in step S1 as input data into the load power prediction model, and generate a load power sequence based on the load prediction, that is, the first load power sequence.

[0101] Specifically, a deep directed residual graph convolutional neural network is used to learn the historical load power, meteorological data, holiday and seasonal data, macroeconomic data, and key user data collected in step S1, and then a load power forecasting model is constructed; the meteorological data, holiday and seasonal data, macroeconomic data, and key user data with a load forecast duration of T1 are used as input data and substituted into the load power forecasting model to obtain a load power sequence [P1, P2,…, P n ], where n represents the amount of power contained in the load power sequence with a duration of T1, n=T1 / t, where t is the sampling duration of two adjacent meteorological data in the load forecast;

[0102] The calculation steps of the deep directed residual graph convolutional neural network include:

[0103] First, set the node's autocorrelation weight to 1, and then construct the in-degree adjacency matrix considering the node self-loop and out-degree adjacency matrix To represent the weighted graph The connection relationship and connection weight between the nodes:

[0104]

[0105]

[0106] Where, represents the significance level; Indicates s i and s j There is a connection relationship, the direction is s i ←sj , the weight is ;

[0107] Based on the in-degree adjacency matrix and out-degree adjacency matrix , perform graph convolution operation, and integrate the residual network into the inter-layer propagation process to obtain the residual graph convolutional neural network; considering the established adjacency matrix and It is an asymmetric matrix, and the adjacency matrix is ​​normalized asymmetric. The inter-layer propagation formula of the residual graph convolutional neural network is:

[0108]

[0109] Where, represents the activation function, express The degree matrix of express The degree matrix of 、 Indicates the The feature matrix of the layer input, 、 Indicates the +1 layer input feature matrix, 、 Indicates the The weight matrix of the layer;

[0110] Then, by stacking multiple layers of residual graph convolutional neural networks and integrating the residual network into the output layer, a deep residual graph convolutional neural network is obtained to achieve Deep learning;

[0111] Finally, the outputs of the deep residual graph convolutional neural network based on the in-degree adjacency matrix and the deep residual graph convolutional neural network based on the out-degree adjacency matrix are spliced ​​to obtain a deep directed residual graph convolutional neural network.

[0112] Step S4: Fuzzy C-means clustering and Markov method are used to generate a load power sequence, and according to the constructed load power model and power demand response model considering the development of load demand, the load power sequence generated based on fuzzy C-means clustering and Markov method is corrected to obtain the second load power sequence.

[0113] Specifically, the steps of generating load power series using fuzzy C-means clustering and Markov method include:

[0114] ① Load power data cleaning: The integrated moving average autoregressive model-generalized autoregressive conditional heteroskedasticity model is used to identify abnormal data in the load historical power series collected in step S1 and remove abnormal data. The expression of the integrated moving average autoregressive model is:

[0115]

[0116] Where, represents the input load power sequence; B represents the lag operator, that is, ; Represents the n-order difference calculation process; represents the fixed parameters obtained by fitting; Represents residual; E represents expected value operation; Var represents variance operation, that is, is the residual Variance of the series; 、 The expression is as follows:

[0117]

[0118] Where, P AR , Q MA represents the order of the polynomial; and represents the coefficient obtained by fitting;

[0119] The expression of the generalized autoregressive conditional heteroskedasticity model is:

[0120]

[0121] In the formula, N represents normal distribution; ~ represents that the variable obeys a certain probability distribution; Represents all historical information before time (t-1); represents the conditional variance; are independent and identically distributed variables that obey the standard normal distribution, iid means independent and identically distributed; 、 and is the fixed parameter and coefficient obtained by fitting; P G , Q AC Representation coefficient and coefficients The order of

[0122] ② Load power clustering: The load power sequence after removing abnormal data is clustered using the fuzzy C-means clustering method, and the load daily power sequence is aggregated into a set with multiple typical sequences as the core. The objective function of the fuzzy C-means clustering method is:

[0123]

[0124] Where U is the membership matrix; u i.j is the membership value, u i . j ∈ [ 0 , 1 ] ; is the jth cluster center; is the jth cluster center c j and the i-th load power x i The Euclidean distance of ; m is the weighting index.

[0125] ③ Generate load power sequences using the Markov method: Based on the clustering results obtained from the load power clustering step, calculate the state transition matrix of different typical sequences; based on the state transition matrix, use the Markov method to generate random sequences that convert between multiple typical sequences; reconstruct the random sequences that convert between multiple typical sequences to generate a load power sequence with a duration of T2;

[0126] The load sequence has a clear daily periodicity, and the power of the load sequence at the start and end times of two consecutive days is similar. Therefore, when establishing the state transition matrix, the conversion relationship matrix of multiple typical sequence sets can be constructed on a daily basis as follows:

[0127]

[0128] Where, Indicates the load series set category of the previous day; represents the set of typical load sequences of the mth type; i and (i+1) represent time numbers in days; N represents the number of days of historical samples; I represents the state judgment variable, which is defined as:

[0129]

[0130] Using the Markov method, we can generate random sequences that transform multiple typical sequences into each other, as shown in the following formula:

[0131]

[0132] The random sequences converted from multiple typical sequences are reconstructed to generate a load power sequence with a duration of T2. The sequence reconstruction steps are as follows:

[0133] Step 1: Initialize the typical daily load sequence number n=1 extracted by Markov and the random load power sequence ;

[0134] Step 2: Read the Markov state chain ,in accordance with Corresponding load typical sequence set category, randomly extract a load sequence from the corresponding set;

[0135] Step 3: Add the extracted load sequence to the load power sequence P 0 , update the Markov sequence number n=n+1;

[0136] Step 4: Determine the total length of the load power sequence. If the sequence number t is greater than or equal to the amount of data with a duration of T2, truncate the sequence according to the amount of data with a duration of T2 and terminate the loop. Otherwise, proceed to the next step.

[0137] Step 5: Determine the last power value P of the extracted load sequence f (t), and at the same time determine the typical sequence category corresponding to the Markov sequence number n , select the initial power and P from the corresponding load typical sequence set f (t) A temporary set of several relatively close daily load power sequences is randomly selected from them;

[0138] Step 6: Repeat steps 3 to 5 until the termination condition is reached;

[0139] A load power model considering the development of load demand is established. The calculation formula is as follows:

[0140] { P ′ t = P t 0 × C t R / C his / her C t = 1 R = C his / her C t R = ( C ( t 1 + 1 ) M − C t 1 M ) × sort [ rand ( ( t 1 + 1 ) M − t 1 M + 1 ) ] + C t 1 M t 1 M ≤ t < ( t 1 + 1 ) M C M = { C t M | t M ∈ { t 1 , t 2 , ⋯ , t N M } }

[0141] Where C his Indicates historical load demand; is the load demand at time t in the future considering the load demand development; represents the load demand at the initial moment of the evaluation period; represents the power at time t of the load power sequence generated by fuzzy C-means clustering and Markov method; is the power at time t considering the development of load demand; t1 M and (t1+1) M are two adjacent major time nodes; and represents the load demand corresponding to two adjacent main time nodes; t M Indicates a specific time; Indicates the specified time node; N indicates the number of specified time nodes; represents the expected load demand at the corresponding time node, which is obtained by using a deep directed residual graph convolutional neural network in step S3 to take future meteorological data, holiday and seasonal data, macroeconomic data, and key user data as input data and substitute them into the load power forecasting model; C M Represents a set of load demands at different time nodes in the future.

[0142] And according to the load power model that takes into account the development of load demand, the load power sequence generated by fuzzy C-means clustering and Markov method is corrected for the first time to obtain the load power sequence after the first correction;

[0143] Establish an electricity demand response model, including a price-based demand response model and an incentive-based demand response model; perform a second correction on the load power sequence after the first correction based on the established electricity demand response model to obtain the second load power sequence. The calculation formula is as follows:

[0144]

[0145] Where, represents the load response at time t of price-based demand response, is the reserve capacity called for by the incentive-based demand response at time t;

[0146] The price-based demand response model, that is, the load response quantity at time t of price-based demand response, is The calculation formula is:

[0147]

[0148] D P t DR . rate = [ D u 1 DR . rate D u 2 DR . rate ⋮ D u TD DR . rate ] T × [ ς 1 − 1 ς 1 − 2 … ς 1 − TD ς 2 − 1 ς 2 − 2 … ς 2 − TD ⋮ ⋮ ⋱ ⋮ ς TD − 1 ς TD − 2 … ς TD − TD ]

[0149] Where, represents the change in electricity price of price-type demand response at time t; and They represent the initial load demand and electricity price participating in price-based demand response respectively; and They represent the proportion of load demand and electricity price change at time t respectively; represents the price elasticity coefficient;

[0150] For price-based demand response to participate in power and electricity balance, the main influencing factor is the load's willingness to participate, which is reflected in the price-based demand response model as the change in the elasticity coefficient in the price elasticity matrix. For rigid loads, price will not be able to guide the adjustment of electricity consumption. As the flexibility and controllability of loads in the power market increase, the absolute value of the price elasticity coefficient will increase, and the guiding ability of price will be enhanced. The formula for calculating the price elasticity coefficient considering the development of load demand is as follows:

[0151] { ς M . DR = { ς t M . DR DR | t M . DR ∈ { t 1 DR , t 2 DR , ⋯ , t N M . DR DR } } ς t R . DR = ( ς ( t 1 + 1 ) M . DR DR − ς t 1 M . DR DR ) × sort [ rand ( ( t 1 + 1 ) M . DR − t 1 M . DR + 1 ) ] + ς t 1 M . DR DR t 1 M . DR ≤ t < ( t 1 + 1 ) M . DR ς 1 R . DR = ς his / her R . DR

[0152] Where, Represents the set of price elasticity coefficients at different time points in the evaluation period; Indicates a specific time. Indicates the time node specified in the evaluation cycle. Indicates the number of specified time nodes; Indicates The price elasticity coefficient when ; and For two adjacent preset time nodes, and Indicates the price elasticity coefficient corresponding to two adjacent preset time nodes, Indicates the price elasticity coefficient at the preset time node t, represents the historical price elasticity coefficient, represents the price elasticity coefficient at the beginning of the evaluation period;

[0153] The incentive demand response model, that is, the reserve capacity called at time t by the incentive demand response The following requirements should be met:

[0154]

[0155] Where, and denote the purchased incentive-based demand response negative reserve and positive reserve capacity respectively; and is the corresponding [0,1] state variable. As electricity market development progresses, industrial users are more willing to participate in incentive-based demand response, and the capacity of incentive-based demand response continues to increase. Its model is a continuous-time stochastic variation model, and its calculation formula refers to the price elasticity coefficient calculation formula that considers load demand development.

[0156] Step S5: splicing the first load power sequence and the second load power sequence according to time sequence to form a long-term load power sequence.

[0157] Exemplary system

[0158] Figure 2 This is a structural diagram of a load long-term power sequence generation system based on sequence combination provided by an exemplary embodiment of the present invention. Figure 2 As shown, the apparatus 200 includes:

[0159] Load data collection module 201, used to collect load forecast data and demand-side response data;

[0160] The load power sequence division module 202 is configured to divide the generation of a long-term load power sequence into a first-segment load power sequence generation and a second-segment load power sequence generation according to the load power sequence generation period and the load prediction accuracy;

[0161] The first load power sequence generation module 203 is used to construct a load power prediction model using a deep directed residual graph convolutional neural network, substitute the load prediction data collected by the load data collection module 201 as input data into the load power prediction model, and generate a load power sequence based on the load prediction, i.e., the first load power sequence;

[0162] The second load power sequence generation module 204 is configured to generate a load power sequence using the fuzzy C-means clustering and Markov method, and to modify the load power sequence generated based on the fuzzy C-means clustering and Markov method according to the constructed load power model and power demand response model that consider the development of load demand, to obtain the second load power sequence;

[0163] The load power sequence splicing module 205 is used to splice the first load power sequence and the second load power sequence according to time sequence to form a long-term load power sequence.

[0164] Optionally, the load data collection module 201 collects load forecast data including meteorological data, holiday and seasonal data, macroeconomic data, and key user data, wherein the meteorological data includes historical load power, temperature, humidity, wind speed, and rainfall; the holiday and seasonal data includes seasons, holidays, weekdays, and weekends; the macroeconomic data includes GDP growth rate, population changes, and industrial policy adjustments; the key user data includes major project construction and large user load curves; the collected demand-side response data includes the initial electricity price, electricity volume, price elasticity coefficient, and negative and positive reserve capacity of the price-type demand response.

[0165] Optionally, the load power sequence division module 202 determines the period T for generating the long-term load power sequence according to the objectives of the new energy absorption capacity assessment and the research on the power and electricity balance of the new power system, determines the prediction duration T1 used for generating the long-term load power sequence according to the accuracy of the load prediction at different time scales, and divides the load power sequence of duration T into two segments, the first segment of the load power sequence duration is T1, the second segment of the load power sequence duration is T2, and T2=T-T1.

[0166] Optionally, the first load power sequence generation module 203 uses a deep directed residual graph convolutional neural network to learn the historical load power, meteorological data, holiday and seasonal data, macroeconomic data, and key user data collected by the load data collection module 201, and then construct a load power forecasting model; the meteorological data, holiday and seasonal data, macroeconomic data, and key user data with a load forecasting duration of T1 are substituted into the load power forecasting model to obtain a load power sequence [P1, P2, ..., P n], where n represents the amount of power contained in the load power sequence with a duration of T1, n=T1 / t, where t is the sampling duration of two adjacent meteorological data in the load forecast;

[0167] The calculation steps of the deep directed residual graph convolutional neural network include:

[0168] First, set the node's autocorrelation weight to 1, and then construct the in-degree adjacency matrix considering the node self-loop and out-degree adjacency matrix To represent the weighted graph The connection relationship and connection weight between the nodes:

[0169]

[0170]

[0171] Where, represents the significance level; Indicates s i and s j There is a connection relationship, the direction is s i ←s j , the weight is ;

[0172] Based on the in-degree adjacency matrix and out-degree adjacency matrix , perform graph convolution operation, and integrate the residual network into the inter-layer propagation process to obtain the residual graph convolutional neural network; considering the established adjacency matrix and It is an asymmetric matrix, and the adjacency matrix is ​​normalized asymmetric. The inter-layer propagation formula of the residual graph convolutional neural network is:

[0173]

[0174] Where, represents the activation function, express The degree matrix of express The degree matrix of 、 Indicates the The feature matrix of the layer input, 、 Indicates the +1 layer input feature matrix, 、 Indicates the The weight matrix of the layer;

[0175] Then, by stacking multiple layers of residual graph convolutional neural networks and integrating the residual network into the output layer, a deep residual graph convolutional neural network is obtained to achieve Deep learning;

[0176] Finally, the outputs of the deep residual graph convolutional neural network based on the in-degree adjacency matrix and the deep residual graph convolutional neural network based on the out-degree adjacency matrix are spliced ​​to obtain a deep directed residual graph convolutional neural network.

[0177] Optionally, the second load power sequence generating module 204 generates a load power sequence using fuzzy C-means clustering and Markov method, including the following steps:

[0178] ① Load power data cleaning: The integrated moving average autoregressive model-generalized autoregressive conditional heteroskedasticity model is used to identify abnormal data in the load historical power series collected in step S1 and remove abnormal data. The expression of the integrated moving average autoregressive model is:

[0179]

[0180] Where, represents the input load power sequence; B represents the lag operator, that is, ; Represents the n-order difference calculation process; represents the fixed parameters obtained by fitting; Represents residual; E represents expected value operation; Var represents variance operation, that is, is the residual Variance of the series; 、 The expression is as follows:

[0181]

[0182] Where, P AR , Q MA represents the order of the polynomial; and represents the coefficient obtained by fitting;

[0183] The expression of the generalized autoregressive conditional heteroskedasticity model is:

[0184]

[0185] In the formula, N represents normal distribution; ~ represents that the variable obeys a certain probability distribution; Represents all historical information before time (t-1); represents the conditional variance; are independent and identically distributed variables that obey the standard normal distribution, iid means independent and identically distributed; 、 and is the fixed parameter and coefficient obtained by fitting; P G , Q AC Representation coefficient and coefficients The order of

[0186] ② Load power clustering: The load power sequence after removing abnormal data is clustered using the fuzzy C-means clustering method, and the load daily power sequence is aggregated into a set with multiple typical sequences as the core. The objective function of the fuzzy C-means clustering method is:

[0187]

[0188] Where U is the membership matrix; u i.j is the membership value, u i . j ∈ [ 0 , 1 ] ; is the jth cluster center; is the jth cluster center c j and the i-th load power x i The Euclidean distance of ; m is the weighting index.

[0189] ③ Generate load power sequences using the Markov method: Based on the clustering results obtained from the load power clustering step, calculate the state transition matrix of different typical sequences; based on the state transition matrix, use the Markov method to generate random sequences that convert between multiple typical sequences; reconstruct the random sequences that convert between multiple typical sequences to generate a load power sequence with a duration of T2;

[0190] The calculation formula of the state transfer matrix of different load typical sequences is:

[0191]

[0192] Where, Indicates the load series set category of the previous day; represents the set of typical load sequences of the mth type; i and (i+1) represent time numbers in days; N represents the number of days of historical samples; I represents the state judgment variable, which is defined as:

[0193]

[0194] Using the Markov method, we can generate random sequences that transform multiple typical sequences into each other, as shown in the following formula:

[0195]

[0196] The random sequences converted from multiple typical sequences are reconstructed to generate a load power sequence with a duration of T2. The sequence reconstruction steps are as follows:

[0197] Step 1: Initialize the typical daily load sequence number n=1 extracted by Markov and the random load power sequence ;

[0198] Step 2: Read the Markov state chain ,in accordance with Corresponding load typical sequence set category, randomly extract a load sequence from the corresponding set;

[0199] Step 3: Add the extracted load sequence to the load power sequence P 0 , update the Markov sequence number n=n+1;

[0200] Step 4: Determine the total length of the load power sequence. If the sequence number t is greater than or equal to the amount of data with a duration of T2, truncate the sequence according to the amount of data with a duration of T2 and terminate the loop. Otherwise, proceed to the next step.

[0201] Step 5: Determine the last power value P of the extracted load sequence f (t), and at the same time determine the typical sequence category corresponding to the Markov sequence number n , select the initial power and P from the corresponding load typical sequence set f (t) A temporary set of several relatively close daily load power sequences is randomly selected from them;

[0202] Step 6: Repeat steps 3 to 5 until the termination condition is reached;

[0203] A load power model considering the development of load demand is established. The calculation formula is as follows:

[0204] { P ′ t = P t 0 × C t R / C his / her C t = 1 R = C his / her C t R = ( C ( t 1 + 1 ) M − C t 1 M ) × sort [ rand ( ( t 1 + 1 ) M − t 1 M + 1 ) ] + C t 1 M t 1 M ≤ t < ( t 1 + 1 ) M C M = { C t M | t M ∈ { t 1 , t 2 , ⋯ , t N M } }

[0205] Where C his Indicates historical load demand; is the load demand at time t in the future considering the load demand development; represents the load demand at the initial moment of the evaluation period; represents the power at time t of the load power sequence generated by fuzzy C-means clustering and Markov method; is the power at time t considering the development of load demand; t1 M and (t1+1) M are two adjacent major time nodes; and represents the load demand corresponding to two adjacent main time nodes; t M Indicates a specific time; Indicates the specified time node; N indicates the number of specified time nodes; C represents the expected load demand at the corresponding time node, which is obtained by the first load power sequence generation module 203 using a deep directed residual graph convolutional neural network to take future meteorological data, holiday and seasonal data, macroeconomic data, and key user data as input data and substitute them into the load power prediction model for calculation; M Represents the set of load demands at different time nodes in the future;

[0206] And according to the load power model that takes into account the development of load demand, the load power sequence generated by fuzzy C-means clustering and Markov method is corrected for the first time to obtain the load power sequence after the first correction;

[0207] Establish an electricity demand response model, including a price-based demand response model and an incentive-based demand response model; perform a second correction on the load power sequence after the first correction based on the established electricity demand response model to obtain the second load power sequence. The calculation formula is as follows:

[0208]

[0209] Where, represents the load response at time t of price-based demand response, is the reserve capacity called for by the incentive-based demand response at time t;

[0210] The price-based demand response model, that is, the load response quantity at time t of price-based demand response, is The calculation formula is:

[0211]

[0212] D P t DR . rate = [ D u 1 DR . rate D u 2 DR . rate ⋮ D u TD DR . rate ] T × [ ς 1 − 1 ς 1 − 2 … ς 1 − TD ς 2 − 1 ς 2 − 2 … ς 2 − TD ⋮ ⋮ ⋱ ⋮ ς TD − 1 ς TD − 2 … ς TD − TD ]

[0213] Where, represents the change in electricity price of price-type demand response at time t; and They represent the initial load demand and electricity price participating in price-based demand response respectively; and They represent the proportion of load demand and electricity price change at time t respectively; It represents the price elasticity coefficient. The price elasticity coefficient considering the development of load demand is calculated as follows:

[0214] { ς M . DR = { ς t M . DR DR | t M . DR ∈ { t 1 DR , t 2 DR , ⋯ , t N M . DR DR } } ς t R . DR = ( ς ( t 1 + 1 ) M . DR DR − ς t 1 M . DR DR ) × sort [ rand ( ( t 1 + 1 ) M . DR − t 1 M . DR + 1 ) ] + ς t 1 M . DR DR t 1 M . DR ≤ t < ( t 1 + 1 ) M . DR ς 1 R . DR = ς his / her R . DR

[0215] Where, Represents the set of price elasticity coefficients at different time points in the evaluation period; Indicates a specific time. Indicates the time node specified in the evaluation cycle. Indicates the number of specified time nodes; Indicates The price elasticity coefficient when ; and For two adjacent preset time nodes, and Indicates the price elasticity coefficient corresponding to two adjacent preset time nodes, Indicates the price elasticity coefficient at the preset time node t, represents the historical price elasticity coefficient, represents the price elasticity coefficient at the beginning of the evaluation period;

[0216] The incentive demand response model, that is, the reserve capacity called at time t by the incentive demand response The following requirements should be met:

[0217]

[0218] Where, and denote the purchased incentive-based demand response negative reserve and positive reserve capacity respectively; and is the corresponding [0,1] state variable. As electricity market development progresses, industrial users are more willing to participate in incentive-based demand response, and the capacity of incentive-based demand response continues to increase. Its model is a continuous-time stochastic variation model, and its calculation formula refers to the price elasticity coefficient calculation formula that considers load demand development.

[0219] Exemplary electronic equipment

[0220] Figure 3 FIG1 is a block diagram of an electronic device provided by an exemplary embodiment of the present invention. Figure 3 As shown, the electronic device 300 includes one or more processors 301 and a memory 302 .

[0221] The processor 301 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0222] The memory 302 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include random access memory (RAM), cache memory, etc. The non-volatile memory may include read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 301 may execute the program instructions to implement the methods of the software programs of the various embodiments of the present invention described above and / or other desired functions. In one example, the electronic device may further include: an input device 303 and an output device 304, and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0223] In addition, the input device 303 may also include a keyboard, a mouse, etc.

[0224] The output device 304 can output various information to the outside, and can include a display, a speaker, a printer, a communication network and a remote output device connected thereto.

[0225] Of course, to simplify, Figure 3 Only some of the components related to the present invention in the electronic device are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, the electronic device may further include any other appropriate components according to specific application conditions.

[0226] Example computer program products and computer readable storage media

[0227] In addition to the above-mentioned methods and devices, an embodiment of the present invention may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to perform the steps of the method according to various embodiments of the present invention described in the above "Exemplary Method" section of this specification.

[0228] The computer program product may be written in any combination of one or more programming languages ​​to implement the operations of embodiments of the present invention, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as C or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's computing device, as a stand-alone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0229] In addition, an embodiment of the present invention may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present invention described in the above "Exemplary Method" section of this specification.

[0230] The computer-readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, include but is not limited to a system, system or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0231] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in the present invention are merely illustrative and non-limiting, and should not be construed as necessarily possessed by each embodiment of the present invention. Furthermore, the specific details disclosed above are provided for illustrative purposes and to facilitate understanding, and are not intended to be limiting. These details do not necessarily limit the present invention to being implemented using these specific details.

[0232] Each embodiment in this specification is described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the device embodiments, since they are essentially identical to the method embodiments, their descriptions are relatively simple. For relevant parts, reference can be made to the descriptions of the method embodiments.

[0233] The block diagrams of the devices, systems, equipment, and systems involved in the present invention are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, systems, equipment, and systems can be connected, arranged, or configured in any manner. Words such as "including," "comprising," "having," and the like are open-ended words, meaning "including but not limited to," and can be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or" and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and can be used interchangeably therewith.

[0234] The method and system of the present invention may be implemented in many ways. For example, the method and system of the present invention may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above sequence of steps for the method is for illustration only, and the steps of the method of the present invention are not limited to the sequence specifically described above, unless otherwise specified. In addition, in some embodiments, the present invention may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present invention. Thus, the present invention also covers recording media that store programs for executing the method according to the present invention.

[0235] It should also be noted that, in the system, device and method of the present invention, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. The above description of the disclosed aspects is provided to enable any technician in this field to make or use the present invention. Various modifications to these aspects will be very obvious to those skilled in the art, and the general principles defined here can be applied to other aspects without departing from the scope of the present invention. Therefore, the present invention is not intended to be limited to the aspects shown here, but according to the widest scope consistent with the principles disclosed here and novel features.

[0236] The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present invention to the forms disclosed herein. While various exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize that certain variations, modifications, alterations, additions, and sub-combinations thereof are intended to fall within the scope of the claims of the present invention.

Claims

1. A method for generating a long-term load power sequence based on sequence combination, characterized in that: include: Step S1: Collecting load-related data, including load forecast data and demand-side response data; Step S2: Based on the load power sequence generation cycle and load forecast accuracy, the long-term load power sequence generation is divided into the first-stage load power sequence generation and the second-stage load power sequence generation; Step S3: A load power forecasting model is constructed using a deep directed residual graph convolutional neural network. The load forecasting data collected in step S1 is substituted into the load power forecasting model as input data to generate a load power sequence based on the load forecast, i.e., the first load power sequence. Step S4: Fuzzy C-means clustering and Markov method are used to generate a load power sequence, and the load power sequence generated based on the fuzzy C-means clustering and Markov method is corrected according to the load power model and power demand response model that consider the development of load demand, to obtain a second load power sequence; Step S5: splicing the first load power sequence and the second load power sequence according to time sequence to form a long-term load power sequence.

2. The method according to claim 1, characterized in that In step S1, the load forecast data includes meteorological data, holiday and seasonal data, macroeconomic data, and key user data. The meteorological data includes historical load power, temperature, humidity, wind speed, and rainfall. The holiday and seasonal data includes seasons, holidays, weekdays, and weekends. The macroeconomic data includes GDP growth rate, population changes, and industrial policy adjustments. The key user data includes major project construction and large user load curves. The demand-side response data includes the initial electricity price, electricity volume, price elasticity coefficient, and negative and positive reserve capacity of the price-type demand response.

3. The method according to claim 1, characterized in that In step S2, the period T for generating the long-term load power sequence is determined according to the objectives of the new energy absorption capacity assessment and the research on the power and electricity balance of the new power system. The prediction duration T1 used for generating the long-term load power sequence is determined according to the accuracy of the load forecast at different time scales. The power sequence of the load with a duration of T is divided into two sections, the first section of the load power sequence has a duration of T1, and the second section of the load power sequence has a duration of T2, and T2=T-T1.

4. The method according to claim 1, wherein In step S3, a deep directed residual graph convolutional neural network is used to learn the historical load power, meteorological data, holiday and seasonal data, macroeconomic data, and key user data collected in step S1, and then a load power forecasting model is constructed; the meteorological data, holiday and seasonal data, macroeconomic data, and key user data with a load forecasting duration of T1 are used as input data and substituted into the load power forecasting model to obtain a load power sequence [P1, P2, ..., P n ], where n represents the amount of power contained in the load power sequence with a duration of T1, n=T1 / t, where t is the sampling duration of two adjacent meteorological data in the load forecast; The calculation steps of the deep directed residual graph convolutional neural network include: First, set the node's autocorrelation weight to 1, and then construct the in-degree adjacency matrix considering the node self-loop and out-degree adjacency matrix To represent the weighted graph The connection relationship and connection weight between the nodes: ； ； Where, represents the significance level; Indicates s i and s j There is a connection relationship, the direction is s i ←s j , the weight is ; Based on the in-degree adjacency matrix and out-degree adjacency matrix , perform graph convolution operation, and integrate the residual network into the inter-layer propagation process to obtain the residual graph convolutional neural network; considering the established adjacency matrix and It is an asymmetric matrix, and the adjacency matrix is ​​normalized asymmetric. The inter-layer propagation formula of the residual graph convolutional neural network is: ； Where, represents the activation function, express The degree matrix of express The degree matrix of 、 Indicates the The feature matrix of the layer input, 、 Indicates the +1 layer input feature matrix, 、 Indicates the The weight matrix of the layer; Then, by stacking multiple layers of residual graph convolutional neural networks and integrating the residual network into the output layer, a deep residual graph convolutional neural network is obtained to achieve Deep learning; Finally, the outputs of the deep residual graph convolutional neural network based on the in-degree adjacency matrix and the deep residual graph convolutional neural network based on the out-degree adjacency matrix are spliced ​​to obtain a deep directed residual graph convolutional neural network.

5. The method according to claim 1, wherein In step S4, the step of generating a load power sequence using fuzzy C-means clustering and Markov method includes: ① Load power data cleaning: The integrated moving average autoregressive model-generalized autoregressive conditional heteroskedasticity model is used to identify abnormal data in the load historical power series collected in step S1 and remove abnormal data. The expression of the integrated moving average autoregressive model is: ； Where, represents the input load power sequence; B represents the lag operator, that is, ; Represents the n-order difference calculation process; represents the fixed parameters obtained by fitting; Represents residual; E represents expected value operation; Var represents variance operation, that is, is the residual Variance of the series; 、 The expression is as follows: ； Where, P AR , Q MA represents the order of the polynomial; and represents the coefficient obtained by fitting; The expression of the generalized autoregressive conditional heteroskedasticity model is: ； In the formula, N represents normal distribution; ~ represents that the variable obeys a certain probability distribution; Represents all historical information before time (t-1); represents the conditional variance; are independent and identically distributed variables that obey the standard normal distribution, iid means independent and identically distributed; 、 and is the fixed parameter and coefficient obtained by fitting; P G , Q AC Representation coefficient and coefficients The order of ② Load power clustering: The load power sequence after removing abnormal data is clustered using the fuzzy C-means clustering method, and the load daily power sequence is aggregated into a set with multiple typical sequences as the core. The objective function of the fuzzy C-means clustering method is: ； Where U is the membership matrix; u i.j is the membership value, ; is the jth cluster center; is the jth cluster center c j and the i-th load power x i The Euclidean distance of ; m is the weighted index; ③ Generate load power sequences using the Markov method: Based on the clustering results obtained from the load power clustering step, calculate the state transition matrix of different typical sequences; based on the state transition matrix, use the Markov method to generate random sequences that convert between multiple typical sequences; reconstruct the random sequences that convert between multiple typical sequences to generate a load power sequence with a duration of T2; The calculation formula of the state transfer matrix of different load typical sequences is: ； Where, Indicates the load series set category of the previous day; represents the set of typical load sequences of the mth type; i and (i+1) represent time numbers in days; N represents the number of days of historical samples; I represents the state judgment variable, which is defined as: ； Using the Markov method, we can generate random sequences that transform multiple typical sequences into each other, as shown in the following formula: ； The random sequences converted from multiple typical sequences are reconstructed to generate a load power sequence with a duration of T2. The sequence reconstruction steps are as follows: Step 1: Initialize the typical daily load sequence number n=1 extracted by Markov and the random load power sequence ; Step 2: Read the Markov state chain ,in accordance with Corresponding load typical sequence set category, randomly extract a load sequence from the corresponding set; Step 3: Add the extracted load sequence to the load power sequence P 0 , update the Markov sequence number n=n+1; Step 4: Determine the total length of the load power sequence. If the sequence number t is greater than or equal to the amount of data with a duration of T2, truncate the sequence according to the amount of data with a duration of T2 and terminate the loop. Otherwise, proceed to the next step. Step 5: Determine the last power value P of the extracted load sequence f (t), and at the same time determine the typical sequence category corresponding to the Markov sequence number n , select the initial power and P from the corresponding load typical sequence set f (t) A temporary set of several relatively close daily load power sequences is randomly selected from them; Step 6: Repeat steps 3 to 5 until the termination condition is reached; A load power model considering the development of load demand is established. The calculation formula is as follows: ； Where C his Indicates historical load demand; is the load demand at time t in the future considering the load demand development; represents the load demand at the initial moment of the evaluation period; represents the power at time t of the load power sequence generated by fuzzy C-means clustering and Markov method; is the power at time t considering the development of load demand; t1 M and (t1+1) M are two adjacent major time nodes; and represents the load demand corresponding to two adjacent main time nodes; t M Indicates a specific time; Indicates the specified time node; N indicates the number of specified time nodes; represents the expected load demand at the corresponding time node, which is obtained by using a deep directed residual graph convolutional neural network in step S3 to take future meteorological data, holiday and seasonal data, macroeconomic data, and key user data as input data and substitute them into the load power forecasting model; C M Represents the set of load demands at different time nodes in the future; And according to the load power model that takes into account the development of load demand, the load power sequence generated by fuzzy C-means clustering and Markov method is corrected for the first time to obtain the load power sequence after the first correction; Establish an electricity demand response model, including a price-based demand response model and an incentive-based demand response model; perform a second correction on the load power sequence after the first correction based on the established electricity demand response model to obtain the second load power sequence. The calculation formula is as follows: ； Where, represents the load response at time t of price-based demand response, is the reserve capacity called for by the incentive-based demand response at time t; The price-based demand response model, that is, the load response quantity at time t of price-based demand response, is The calculation formula is: ； ； Where, represents the change in electricity price of price-type demand response at time t; and They represent the initial load demand and electricity price participating in price-based demand response respectively; and They represent the proportion of load demand and electricity price change at time t respectively; It represents the price elasticity coefficient. The price elasticity coefficient considering the development of load demand is calculated as follows: ； Where, Represents the set of price elasticity coefficients at different time points in the evaluation period; Indicates a specific time. Indicates the time node specified in the evaluation cycle. Indicates the number of specified time nodes; Indicates The price elasticity coefficient when ; and For two adjacent preset time nodes, and Indicates the price elasticity coefficient corresponding to two adjacent preset time nodes, Indicates the price elasticity coefficient at the preset time node t, represents the historical price elasticity coefficient, represents the price elasticity coefficient at the beginning of the evaluation period; The incentive demand response model, that is, the reserve capacity called at time t by the incentive demand response The following requirements should be met: ； Where, and denote the purchased incentive-based demand response negative reserve and positive reserve capacity respectively; and is the corresponding [0,1] state variable. As electricity market development progresses, industrial users are more willing to participate in incentive-based demand response, and the capacity of incentive-based demand response continues to increase. Its model is a continuous-time stochastic variation model, and its calculation formula refers to the price elasticity coefficient calculation formula that considers load demand development.

6. A load long-term power sequence generation system based on sequence combination, characterized in that: include: Load data collection module, used to collect load forecast data and demand-side response data; A load power sequence division module is used to divide the generation of a long-term load power sequence into a first-segment load power sequence generation and a second-segment load power sequence generation according to the load power sequence generation cycle and load forecast accuracy; The first load power sequence generation module is used to build a load power prediction model using a deep directed residual graph convolutional neural network, substitute the load prediction data collected by the load data collection module as input data into the load power prediction model, and generate a load power sequence based on the load prediction, i.e., the first load power sequence; The second load power sequence generation module is used to generate a load power sequence using fuzzy C-means clustering and Markov method, and to correct the load power sequence generated based on fuzzy C-means clustering and Markov method according to the constructed load power model and power demand response model considering the development of load demand, to obtain the second load power sequence; The load power sequence splicing module is used to splice the first load power sequence and the second load power sequence according to time sequence to form a long-term load power sequence.

7. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the method for generating a long-time load power sequence based on sequence combination according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the method for generating a long-time load power sequence based on sequence combination according to any one of claims 1 to 5 is implemented.

Images