Multi-element load prediction method, system and equipment for integrated energy system

Abstract

The invention provides a multi-element load prediction method, system and device for an integrated energy system, and relates to the field of load prediction of the integrated energy system, and the method comprises the steps: initializing a white whale population representing each connection weight in a multi-element load prediction model based on an improved white whale optimization algorithm, and obtaining a multi-element load prediction model; basic parameters of the improved whale optimization algorithm are set; based on the basic parameters of the improved white whale optimization algorithm, training the multi-element load prediction model according to the initialized white whale population, and generating a trained multi-element load prediction model; and inputting the multivariate load historical data, the weather parameters of the to-be-predicted day, the date type and the time data into the trained multivariate load prediction model, and outputting a multivariate load prediction value at the hourly moment of the to-be-predicted day. According to the method, the multi-element load of the integrated energy system can be accurately predicted, so that multi-energy collaborative optimization scheduling of the integrated energy system is realized.

Description

Technical Field

[0001] The present invention relates to the field of load forecasting for an integrated energy system, and in particular to a method, system and device for forecasting multiple loads for an integrated energy system. Background Art

[0002] The proportion of new energy in the integrated energy system has increased significantly. However, the full absorption of high proportion of new energy is still an urgent problem to be solved. In the integrated energy system, converting some new energy such as wind and light into energy in the form of cold, heat, and electricity is one of the effective ways to promote the absorption of new energy. Multi-load forecasting technology is the premise of multi-energy coordinated optimization and dispatching in the integrated energy system, but there are not many related studies at present, and there are still many problems that need to be solved urgently. Summary of the invention

[0003] The purpose of the present invention is to provide a method, system and device for predicting multiple loads in an integrated energy system, which can accurately predict the multiple loads in the integrated energy system to achieve multi-energy coordinated optimal scheduling of the integrated energy system.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A multi-element load forecasting method for an integrated energy system, comprising:

[0006] Based on the improved Beluga optimization algorithm, the Beluga population that characterizes each connection weight in the multivariate load forecasting model is initialized, and the basic parameters of the improved Beluga optimization algorithm are set; the multivariate load forecasting model is a multivariate load forecasting model based on a bidirectional long-term and short-term neural network; the multivariate load forecasting model based on a bidirectional long-term and short-term neural network is constructed based on the historical data of the integrated energy system and the output tasks that have undergone weight sharing; the historical data includes multivariate load historical data, historical weather parameter data, historical date type data and historical time data; the output task that has undergone weight sharing is obtained by constructing multiple single tasks according to different load types and applying multi-task learning theory to share weights for multiple single tasks; the output task that has undergone weight sharing is a multivariate load column vector that has undergone weight sharing; the basic parameters include the maximum number of iterations and the iteration stop condition;

[0007] Based on the basic parameters of the improved Beluga optimization algorithm, the multivariate load forecasting model is trained according to the initialized Beluga population to generate a trained multivariate load forecasting model;

[0008] The multivariate load historical data, weather parameters of the day to be predicted, date type and time data are input into the trained multivariate load prediction model, and the multivariate load prediction value at the hour of the day to be predicted is output.

[0009] Optionally, based on the improved Beluga optimization algorithm, a Beluga population representing each connection weight in the multivariate load forecasting model is initialized, and basic parameters of the improved Beluga optimization algorithm are set, which also includes:

[0010] The historical data of the integrated energy system is normalized to generate normalized historical data.

[0011] Optionally, the overall optimization function of the multi-task learning theory is:

[0012]

[0013]

[0014] Among them, X t is the output task calculated by multi-task learning at time t; is the mapping function of the output task; is the task weight parameter; ξ(θ sh ,θ l ) is the loss function, θ sh is the shared parameter of the load forecasting task, θ l is a task-related parameter; M is the total number of training data samples in different tasks, m is the mth data in the sample, m∈[1,M]; is the loss function of the mth data and the lth task; f l (y t θ sh ,θ l ) is a multi-task learning function; y t is the normalized value; is the label of the mth data and the lth task.

[0015] Optionally, the multivariate load forecasting model based on bidirectional long-term and short-term neural network is:

[0016]

[0017] Among them, Y t is the multivariate load forecasting result obtained by the multivariate load forecasting model at time t; h t is the hidden state at time t, h t-1 is the hidden state at time t-1; is the hidden state of the forward long short-term neural network at time t, is the hidden state of the backward long-term short-term neural network at time t; D t is the cell state at time t, D t-1 is the cell state at time t-1; σ is the sigmoid activation function; tanh is the hyperbolic tangent activation function; W f ,Wi ,W c ,W 0 are the connection weights of the forget gate, input gate, Tanh gate, and output gate respectively; b f ,b i ,b c ,b o They are the bias vectors of the forget gate, input gate, Tanh gate, and output gate respectively; c 1 ,c 2 are constants, and c 1 +c 2 =1;

[0018] The input-output mapping relationship of the multivariate load forecasting model is Y t =f(X t ), where f(·) is the input-output mapping function of the prediction model, X t is the output task calculated through multi-task learning at time t.

[0019] Optionally, based on the improved Beluga optimization algorithm, a Beluga population representing the weights of each connection in the multivariate load forecasting model is initialized, specifically including:

[0020] An initial beluga whale population is generated by random initialization, with n beluga whale individuals and each beluga whale individual having d-dimensional position information;

[0021] The pinhole imaging theory is used to improve the beluga optimization algorithm, and the corresponding points of the current beluga individual position information are introduced to generate an improved beluga population:

[0022] Compare the fitness of the location information of each beluga whale individual in the initial beluga whale population and the improved beluga whale population, and sort the fitness in ascending or descending order to generate a fitness sequence; the fitness F(W) is calculated as follows: in, is the maximum mean square error of the three output loads of the multivariate load prediction model, namely, the cooling load, the heating load and the electric load; Y is the prediction model output value of the cooling, heating and electric loads at the same time; Y * It is the real value of cold, hot and electric load at the same moment;

[0023] Based on the fitness sequence, beluga individuals whose fitness is greater than the fitness threshold are screened out to form an initialized beluga population; the initialized beluga population is used to characterize each connection weight in the multivariate load forecasting model.

[0024] Optionally, based on the basic parameters of the improved Beluga optimization algorithm, the multivariate load forecasting model is trained according to the initialized Beluga population to generate a trained multivariate load forecasting model, specifically including:

[0025] Initializing the initialized beluga population, and determining whether the current number of iterations is less than the maximum number of iterations;

[0026] If so, determining the location information of the beluga whale individual corresponding to the highest fitness and the location information of the beluga whale individual corresponding to the lowest fitness based on the fitness sequence;

[0027] Calculate the balance index of the position information of the beluga whale individual corresponding to the highest fitness and the position information of the beluga whale individual corresponding to the lowest fitness, as well as the probability of the beluga whale falling;

[0028] Update the position information of individual beluga whales in the initialized beluga whale population according to the balance index and the beluga whale fall probability;

[0029] Determine whether the updated location information meets the preset accuracy condition;

[0030] If so, generate a trained multivariate load forecasting model;

[0031] If not, re-determine whether the current number of iterations is less than the maximum number of iterations, until the updated position information meets the preset accuracy condition or the current number of iterations is not less than the maximum number of iterations, and stop training the multivariate load forecasting model.

[0032] Optionally, updating the position information of individual beluga whales in the initialized beluga whale population according to the balance index and the beluga whale fall probability specifically includes:

[0033] When the balance index is greater than 0.5, using the formula Update the location information of the White Whale search phase; where W i j(z+1) represents the position information of the i-th beluga individual in the j-th dimension at the z+1-th iteration, j = 1, 2, …, d; p j (z) is the dimension randomly selected from the d dimensions in the zth iteration; The i-th beluga individual and the x-th randomly selected beluga individual are j The position information in the dimension; α, β are random factors in the beluga search phase and are any numbers in (0,1); sin(2πβ), cos(2πβ) are any numbers between the average beluga fins; N f is the balance index; even is an even number; odd is an odd number;

[0034] When the balance index is less than or equal to 0.5, use Update the location information of the beluga whale during the predation phase; among them, ε is a constant; is the best individual position of the white whale in the zth iteration; W i j(z) is the position information of the i-th beluga individual in the j-th dimension at the z-th iteration, is the current position of the i-th beluga individual in the z-th iteration; W i z is the current position of a randomly selected beluga individual; φ, are all arbitrary numbers in (0,1); u, v are the values ​​of the control step; δ is the control operator of the Levy flight function; Γ is the gamma function; Zt max is the maximum number of iterations;

[0035] When the balance index is less than or equal to the falling probability, use N f≤ Q f Update the position information of the white whale during the falling stage; where η, κ, and λ are all arbitrary numbers in (0, 1); W step is the step length of the beluga whale individual moving to any position or falling, U b is the upper bound of the initialization variable, L b To initialize the lower bound of the variable, R = 2nN f , R is the step length factor of the beluga whale's individual fall, n is the total number of beluga whales in the beluga whale population; Q f is the probability of falling;

[0036] The position information of the individual beluga whales in the initialized beluga whale population is updated, and the fitness of the individual beluga whales after the position update is calculated.

[0037] A multi-element load forecasting system for an integrated energy system, comprising:

[0038] An initialization module is used to initialize the Beluga population that characterizes each connection weight in the multivariate load forecasting model based on the improved Beluga optimization algorithm, and set the basic parameters of the improved Beluga optimization algorithm; the multivariate load forecasting model is a multivariate load forecasting model based on a bidirectional long-term and short-term neural network; the multivariate load forecasting model based on a bidirectional long-term and short-term neural network is constructed based on the historical data of the integrated energy system and the output tasks that have undergone weight sharing; the historical data includes multivariate load historical data, historical weather parameter data, historical date type data and historical time data; the output task that has undergone weight sharing is obtained by constructing multiple single tasks according to different load types and applying multi-task learning theory to share weights for multiple single tasks; the output task that has undergone weight sharing is a multivariate load column vector that has undergone weight sharing; the basic parameters include the maximum number of iterations and the iteration stop condition;

[0039] A training module, used for training the multivariate load forecasting model based on the basic parameters of the improved Beluga optimization algorithm and the initialized Beluga population to generate a trained multivariate load forecasting model;

[0040] The prediction module is used to input the multivariate load historical data, weather parameters of the day to be predicted, date type and time data into the trained multivariate load prediction model, and output the multivariate load prediction value at the hour of the day to be predicted.

[0041] An electronic device includes a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned integrated energy system multi-load forecasting method.

[0042] Optionally, the memory is a non-transitory computer-readable storage medium, and the non-transitory computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the above-mentioned integrated energy system multi-element load forecasting method is implemented.

[0043] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0044] The embodiment of the present invention is based on an improved Beluga optimization algorithm, initializes a Beluga population that characterizes each connection weight in a multivariate load forecasting model, and sets basic parameters of the improved Beluga optimization algorithm to train the multivariate load forecasting model. The multivariate load forecasting values ​​of cold, hot and electric loads at the hour of the day to be predicted are predicted according to the trained multivariate load forecasting model, providing technical support and data support for multi-energy coordinated scheduling of an integrated energy system, so as to realize multi-energy coordinated optimization scheduling of an integrated energy system and promote the consumption of new energy sources such as wind and solar. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0046] Figure 1 A flow chart of the multi-element load forecasting method for the integrated energy system provided by the present invention;

[0047] Figure 2 The prediction flow chart of the trained multivariate load prediction model provided by the present invention;

[0048] Figure 3 This is a comparison chart of the cold, heat and electricity multi-load prediction effects of the integrated energy system provided by the present invention. DETAILED DESCRIPTION

[0049] 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.

[0050] The purpose of the present invention is to provide a method, system and device for predicting multiple loads in an integrated energy system, which can accurately predict the multiple loads in the integrated energy system to achieve multi-energy coordinated optimal scheduling of the integrated energy system.

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] Embodiment 1

[0053] like Figure 1 As shown, the present invention provides a multi-element load forecasting method for an integrated energy system, comprising:

[0054] Step 101: Based on the improved Beluga optimization algorithm, initialize the Beluga population that characterizes each connection weight in the multivariate load forecasting model, and set the basic parameters of the improved Beluga optimization algorithm; the multivariate load forecasting model is a multivariate load forecasting model based on a bidirectional long-term and short-term neural network; the multivariate load forecasting model based on a bidirectional long-term and short-term neural network is constructed based on the historical data of the integrated energy system and the output tasks that have undergone weight sharing; the historical data includes multivariate load historical data, historical weather parameter data, historical date type data and historical time data; the output task that has undergone weight sharing is obtained by constructing multiple single tasks according to different load types and applying multi-task learning theory to share weights for multiple single tasks; the output task that has undergone weight sharing is a multivariate load column vector that has undergone weight sharing; the basic parameters include the maximum number of iterations and the iteration stop condition.

[0055] In practical applications, multivariate load historical data, corresponding weather parameters, date type and time data are obtained and preprocessed.

[0056] The multi-load historical data includes the historical data of cold, heat and electricity multi-loads in hours; obtain weather parameter data of the same spatiotemporal scale, including weather parameter data in units of the hourly moment of the previous hour; obtain date type data of the same spatiotemporal scale, including date type data in units of the hourly moment of the previous hour; obtain time data representing the load collection time, including data of the hourly moment of 24 hours. The collected historical data of cold, heat and electricity multi-loads and weather parameter data, date type data and time data are used to construct a model data set. The weather parameter data, date type data, time data and part of the historical data of cold, heat and electricity are divided into the model input data of the training set; part of the historical data of cold, heat and electricity multi-loads are divided into the model output comparison data of the training set.

[0057] During data preprocessing, the cold, heat, and electricity multi-loads and the input variables that affect the multi-loads are normalized. Among them, the weather parameter data includes temperature data, humidity data, wind speed data, and atmospheric pressure data; the date type data is divided into holidays and weekdays, and the holidays are mainly statutory holidays and weekends. The rest of the time is processed as weekdays, and the weekends are Saturdays and Sundays. The preprocessed sample value at time t is recorded as x t , is the column vector of temperature data, humidity data, wind speed data, atmospheric pressure data, date type data, time data and historical load data of cold, hot and electricity, then x t It is expressed as:

[0058] x t =[Q t ,A t ,V t ,F t ,G t ,T t ,C t-168 ,...,C t-1 ,H t-168 ,...,H t-1 ,E t-168 ,...,E t-1 ] T

[0059]

[0060] In the formula, Q t ,A t ,V t ,F t ,G t ,T t They are temperature data, humidity data, wind speed data, atmospheric pressure data, date type data and time data at time t respectively; C t-168 ,...,C t-1 ,Ht-168 ,...,H t-1 ,E t-168 ,...,E t-1 The historical load data of cold, hot and electric power for each hour in the seven days before time t; y t , max(x), min(x) are the normalized value, maximum value and minimum value of the indicator respectively; t is the time.

[0061] In practical applications, multi-task learning includes the following steps.

[0062] Multiple single tasks will be constructed according to different load types, and the multi-task learning theory will be applied to share the weights of multiple single tasks. The above-learned data set will be used as the input information of the multivariate load forecasting model.

[0063] When conducting multi-task learning theory learning, L target tasks are constructed according to different load types, and all data in each task come from a distribution space B×P,B∈R n ,P∈R. Each task contains M data as training data samples, that is, {(B 1l ,P 1l ),(B 2l ,P 2l ),...,(B ml ,P ml )}.

[0064] The overall optimization function of multi-task learning can be expressed as:

[0065]

[0066]

[0067] Where, X t is the output calculated by multi-task learning at time t; is the mapping function of the output task; is the task weight parameter; ξ(θ sh ,θ l ) is the loss function, θ sh is the shared parameter of the load forecasting task, θ l is a task-related parameter; M is the total number of training data samples in different tasks, m is the mth data in the sample, m∈[1,M]; is the loss function of the mth data and the lth task; f l (y t θ sh ,θ l ) is a multi-task learning function; y t is the normalized value; is the label of the mth data and the lth task.

[0068] When constructing a prediction model based on a bidirectional long-term and short-term neural network, the weather parameter data, date type data and time data are taken from the hourly data of a certain hour; the cold, hot and electric load prediction data are respectively the cold load data, hot load data and electric load data measured within one hour before a certain hourly time, and the prediction model output at time t is recorded as Y t , then Y t =[C t ,H t ,E t ] T Where, C t ,H t ,E t They are the cold, hot and electric loads from time t-1 to time t respectively.

[0069] The prediction model input at time t is recorded as X t , is the multivariate load column vector that has been weight-shared through multi-task learning.

[0070] In practical applications, building a multivariate load forecasting model based on a bidirectional long-term and short-term neural network includes the following steps.

[0071] The output tasks that have undergone weight sharing are input into the multi-element load forecasting model, and the cold, heat and electricity multi-element load forecasting data are output.

[0072] Construct a multivariate load forecasting model for an integrated energy system based on a bidirectional long-term and short-term neural network:

[0073]

[0074] h t =σ(W 0 ·[h t-1 ,X t ]+b 0 )·tanh(D t ).

[0075] D t =σ(W f ·[h t-1 ,X t ]+b f )·D t-1 +σ(W i ·[h t-1 ,X t ]+b i )·tanh(W c ·[h t-1 ,X t ]+b c ). Among them, Yt is the output of the bidirectional long-term and short-term neural network operation at time t; h t ,h t-1 are the hidden states at time t and t-1 respectively; are the hidden states of the forward and backward long and short-term neural networks at time t; D t ,D t-1 are the cell states at time t and t-1 respectively; W f ,W i ,W c ,W 0 are the weight matrices of the forget gate, input gate, Tanh gate, and output gate respectively; b f ,b i ,b c ,b o They are the bias vectors of the forget gate, input gate, Tanh gate, and output gate respectively; c 1 ,c 2 are constants, and c 1 +c 2 =1.

[0076] σ is the sigmoid activation function, which is the first activation function used in deep neural networks and is mainly used for logistic regression. The mathematical formula is:

[0077]

[0078] The value range of the sigmoid activation function is (0,1).

[0079] Tanh is the hyperbolic tangent activation function. The hyperbolic tangent "tanh" is derived from the two basic hyperbolic functions of hyperbolic sine and hyperbolic cosine. Its mathematical formula is:

[0080]

[0081] in, For variables The activation function of .

[0082] The input-output mapping relationship of the prediction model is:

[0083] Y t =f(X t ).

[0084] Where f(·) is the input-output mapping function of the prediction model.

[0085] In practical applications, the Beluga population representing the weights of each connection of the established multivariate load forecasting model is initialized, and the basic parameters of the improved Beluga optimization algorithm are set.

[0086] The improved Beluga optimization algorithm is used to train the multivariate load forecasting model. The individual positions of the Beluga in the improved Beluga optimization algorithm reflect the various weight parameters in the multivariate load forecasting model of the integrated energy system. The process of the change of the weight value of the multivariate load forecasting model of the integrated energy system is the process of the change of the individual position value of the Beluga. The training of the forecasting model is completed by finding the optimal position of the Beluga individual under the condition of the highest fitness of the Beluga population. Fitness is used to indicate the improvement of the forecasting model by the algorithm; the better the fitness, the more accurate the forecasting model, and the forecast values ​​of cold, hot and electric loads are closer to the true values.

[0087] The pinhole imaging theory is applied to improve the Beluga optimization algorithm and initialize the position of the Beluga population. During the initialization process, the solutions of the Beluga initialized population and the corresponding point Beluga population are compared, and the optimal n solutions are selected as the initial solutions, which improves the prediction accuracy of the model and searches simultaneously within the solution space; the diversity of the population is increased and the ability to escape from the local optimum, search efficiency and convergence performance of the Beluga optimization algorithm are improved.

[0088] Beluga population initialization is divided into three steps:

[0089] Step 1: Randomly initialize and generate a beluga whale population with n beluga whale individuals, each with d-dimensional position information. The position information of the beluga whale individuals is represented by a number between 0 and 1.

[0090] Step 2: Improve the White Whale optimization algorithm through pinhole imaging theory and introduce corresponding points:

[0091] Among them, W i j =(W i 1 ,W i 2 ,...W i d ) is the current position of the beluga whale, then the corresponding point is:

[0092] W i j' =(W i 1' ,W i 2' ,...W i d' ).

[0093]

[0094] Where K is the virtual candle length ratio coefficient; Wi j is the j-dimensional position information of the i-th beluga individual, i∈[1,n], j∈[1,d]; is the position of the current best individual; W i j' is the corresponding point position in pinhole imaging theory; U bi.j ,L bi.j are the dynamic boundaries of the i-th beluga whale individual at the j-th dimension position.

[0095] Step 3: Compare the initial population with the fitness of the individual position information of the beluga whale population improved by the pinhole imaging theory, and select n beluga whale individuals with better fitness to form the final initialization population. The beluga whale population W is expressed as:

[0096]

[0097] In the formula, is the d-dimensional position information of the n-th beluga individual, and the beluga population W represents the connection weight W in the prediction model. f ,W i ,W c ,W 0 .

[0098] The fitness calculation formula is: In the formula, is the maximum mean square error of the model's three outputs: cooling, heating and electricity; Y is the prediction model output value of cooling, heating and electricity load at the same time; Y * It is the true value of cold, hot and electric load at the same moment.

[0099] Step 102: Based on the basic parameters of the improved Beluga optimization algorithm, the multivariate load forecasting model is trained according to the initialized Beluga population to generate a trained multivariate load forecasting model.

[0100] In practical applications, the improved Beluga optimization algorithm is used to train the prediction model of multivariate loads.

[0101] Set the maximum number of iterations and the iteration stop condition, and apply the iterative optimization of the improved Beluga population. If the prediction accuracy requirements of the model are met, the model training is completed. In addition, if the preset accuracy requirements of the model are not met, the process will be repeated until the requirements are met or the number of iterations of the Beluga optimization algorithm exceeds the set maximum number of iterations, and the model training will also be stopped.

[0102] Step 1: Initialize the beluga population W and set the maximum number of iterations Zt max , initialize the upper bound of the variable U b and the lower bound L b .

[0103] Step 2: If it is less than the maximum number of iterations, calculate the fitness value and sort it. Select the individual position of the white whale with the best current fitness and the individual position of the white whale with the worst current fitness, and calculate the balance index N f and the probability of the white whale falling Q f , and then update the position of the beluga population through the formula.

[0104]

[0105]

[0106] Where N o is a constant that changes between (0,1) in each iteration; z is the current iteration number.

[0107] Step 3: When the balance index N f If it is greater than 0.5, the following formula is used to update the position of the white whale in the search phase. The white whale searches for food in the search phase and moves closer to the position of the white whale with high fitness. If the white whale does not find the food it needs, it will go to other places.

[0108]

[0109] Where W i j(z+1) represents the position information of the i-th beluga whale in the j-th dimension at the z+1-th iteration; p j (j=1,2,…,d) is randomly selected from d dimensions; The i-th beluga and the x-th (randomly selected beluga) beluga are respectively j The position information in the dimension; α, β are random factors in the search phase, which are any numbers in (0,1); sin(2πβ), cos(2πβ) are any numbers between the average white whale fins. The position of the white whale in the search phase is updated according to the dimensions selected by odd and even numbers; even is an even number; odd is an odd number.

[0110] Step 4: When the balance index N f Less than or equal to 0.5, the following formula is used to update the position of the white whale predation stage. White whales prey on prey by sharing each other's position information, considering the best candidate solution and other solutions. The Levy flight strategy is introduced in the white whale predation stage to improve convergence.

[0111]

[0112]

[0113] In the formula, ε is a configurable constant; is the best white whale position for the zth iteration; is the current location information of the i-th beluga whale and a randomly selected beluga whale in the z-th iteration; φ, is any number in (0,1); u, v are the values ​​of the control step size, which obey the standard normal distribution mean-variance (0,1); δ is the control operator of the Levy flight function; Γ is the gamma function, which is the extension of the factorial function in the real and complex domains.

[0114] Step 5: When the balance index N f Less than or equal to the probability of falling Q f , use the following formula to update the position of the beluga falling stage. The falling behavior of the beluga is used to simulate the death of the beluga and falling into the sea or moving to other locations in each iteration. In order to ensure that the number of beluga populations remains unchanged, the position of the beluga is updated using the position of the beluga and the step length of the beluga falling.

[0115]

[0116]

[0117] Where η, κ, λ are any numbers between (0, 1); W step Indicates the step size of the white whale moving to other places or falling, which is related to the boundary of the design variable, the number of iterations and the step size factor; R = 2nN f , R is the step factor of beluga whale fall, which is related to the probability of beluga whale fall and population size, and n is the total number of beluga whales in the beluga whale population.

[0118] Step 6: Calculate the fitness value and update the current white whale position.

[0119] Step 7: Check whether the preset accuracy condition is met. If so, exit and output the result. Otherwise, repeat steps 2 to 6 until the preset accuracy is met or the maximum number of iterations is reached.

[0120] Step 103: Input the multivariate load historical data, weather parameters of the day to be predicted, date type and time data into the trained multivariate load prediction model, and output the multivariate load prediction value at the hour of the day to be predicted.

[0121] In practical applications, such as Figure 2 As shown, the trained prediction model is used to predict the multi-element load of cold, heat and electricity, which specifically includes: multi-task learning is performed on the weather parameters, date type, time data and historical cold, heat and electricity multi-element load data of the predicted day, and then input into the trained prediction model to obtain the predicted values ​​of cold, heat and electricity load at the hour of the predicted day. Due to the proportion of the attached figure, Figure 2 The “?” in the judgment box indicates “whether the maximum number of iterations or the preset accuracy has been reached”.

[0122] The multi-load forecasting results of the cold, heat and electricity multi-load forecasting of the integrated energy system output by the trained multi-load forecasting model are as follows: Figure 3 As shown, Figure 3 It is a curve chart of the predicted and actual values ​​of cold, hot and electric loads at the hour.

[0123] The prediction results of cold, hot and electric loads at the hour and the actual cold, hot and electric load values ​​are shown in Table 1:

[0124] Table 1: Some cold, hot and electric load prediction results

[0125]

[0126] The mean square prediction accuracy of the prediction results of cold, hot and electric loads at the whole time is calculated respectively, and the calculation formula is:

[0127]

[0128] In the formula, θ υ It indicates the mean square prediction accuracy of cold, heat and electricity multi-load; Indicates the actual value of the cold, hot and electric loads at a certain moment; Y υ,τ Represents the predicted value of cold, hot and electric load at a certain moment; υ∈[1,3], τ∈[1,24].

[0129] After calculation, the mean square prediction accuracy of the cold load prediction results at the hour is 96.56%; the mean square prediction accuracy of the heat load prediction results at the hour is 96.25%; the mean square prediction accuracy of the electric load prediction results at the hour is 95.28%. The results show that the multivariate load prediction method for the integrated energy system based on the improved Beluga optimization algorithm proposed in the present invention has the advantages of high prediction accuracy, high prediction reliability, high efficiency, and jumping out of local optimality, which meets the actual application needs.

[0130] Embodiment 2

[0131] The difference between this embodiment and the first embodiment is that:

[0132] In practical applications, the cooling and heating loads are converted into the same unit as the electrical power, which is kW.

[0133] In practical applications, the virtual candle length ratio coefficient K can be set to 1.5×104.

[0134] In practical applications, the maximum number of iterations Zt can be set max is 100.

[0135] In practical applications, ε can be 1.5.

[0136] In practical applications, set the model preset accuracy The judgment is as follows:

[0137] when Once the model’s preset accuracy requirements are met and model training is completed, the model can be used for prediction.

[0138] when If the model has not yet reached the preset accuracy requirement, the iterative process of improving the White Whale optimization algorithm will be repeated or the number of iterations of the improved White Whale optimization algorithm exceeds the set maximum number of iterations, the model training will also be stopped.

[0139] In this embodiment, the model convergence accuracy is set For 10 -6 .

[0140] Embodiment 3

[0141] In order to execute the method corresponding to the above-mentioned embodiment 1 to achieve the corresponding functions and technical effects, a multi-load forecasting system for an integrated energy system is provided below.

[0142] A multi-element load forecasting system for an integrated energy system, comprising:

[0143] An initialization module is used to initialize the Beluga population that characterizes each connection weight in the multivariate load forecasting model based on the improved Beluga optimization algorithm, and set the basic parameters of the improved Beluga optimization algorithm; the multivariate load forecasting model is a multivariate load forecasting model based on a bidirectional long-term and short-term neural network; the multivariate load forecasting model based on a bidirectional long-term and short-term neural network is constructed based on the historical data of the integrated energy system and the output tasks that have undergone weight sharing; the historical data includes multivariate load historical data, historical weather parameter data, historical date type data and historical time data; the output task that has undergone weight sharing is obtained by constructing multiple single tasks according to different load types and applying multi-task learning theory to share weights for multiple single tasks; the output task that has undergone weight sharing is a multivariate load column vector that has undergone weight sharing; the basic parameters include the maximum number of iterations and the iteration stop condition.

[0144] The training module is used to train the multivariate load forecasting model based on the basic parameters of the improved Beluga optimization algorithm and the initialized Beluga population to generate a trained multivariate load forecasting model.

[0145] The prediction module is used to input the multivariate load historical data, weather parameters of the day to be predicted, date type and time data into the trained multivariate load prediction model, and output the multivariate load prediction value at the hour of the day to be predicted.

[0146] Embodiment 4

[0147] An electronic device includes a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned integrated energy system multi-load forecasting method.

[0148] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned method for predicting multiple loads in an integrated energy system.

[0149] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0150] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A multi-element load forecasting method for an integrated energy system, characterized in that: include: Based on the improved Beluga optimization algorithm, the Beluga population that characterizes each connection weight in the multivariate load forecasting model is initialized, and the basic parameters of the improved Beluga optimization algorithm are set; the multivariate load forecasting model is a multivariate load forecasting model based on a bidirectional long-term and short-term neural network; the multivariate load forecasting model based on a bidirectional long-term and short-term neural network is constructed based on the historical data of the integrated energy system and the output tasks that have undergone weight sharing; the historical data includes multivariate load historical data, historical weather parameter data, historical date type data and historical time data; the output task that has undergone weight sharing is obtained by constructing multiple single tasks according to different load types and applying multi-task learning theory to share weights for multiple single tasks; the output task that has undergone weight sharing is a multivariate load column vector that has undergone weight sharing; the basic parameters include the maximum number of iterations and the iteration stop condition; Based on the basic parameters of the improved Beluga optimization algorithm, the multivariate load forecasting model is trained according to the initialized Beluga population to generate a trained multivariate load forecasting model; The multivariate load historical data, weather parameters of the day to be predicted, date type and time data are input into the trained multivariate load prediction model, and the multivariate load prediction value at the hour of the day to be predicted is output.

2. The multi-element load forecasting method for an integrated energy system according to claim 1, characterized in that: Based on the improved Beluga optimization algorithm, the Beluga population representing each connection weight in the multivariate load forecasting model is initialized, and the basic parameters of the improved Beluga optimization algorithm are set, which also includes: The historical data of the integrated energy system is normalized to generate normalized historical data.

3. The multi-element load forecasting method for an integrated energy system according to claim 1, characterized in that: The overall optimization function of the multi-task learning theory is: Among them, X t is the output task calculated by multi-task learning at time t; is the mapping function of the output task; is the task weight parameter; ξ(θ sh ,θ l ) is the loss function, θ sh is the shared parameter of the load forecasting task, θ l is a task-related parameter; M is the total number of training data samples in different tasks, m is the mth data in the sample, m∈[1,M]; is the loss function of the mth data and the lth task; f l (y t θ sh ,θ l ) is a multi-task learning function; y t is the normalized value; is the label of the mth data and the lth task.

4. The method for multi-element load forecasting of an integrated energy system according to claim 1, characterized in that: The multivariate load forecasting model based on bidirectional long-term and short-term neural network is: h t =σ(W0·[h t-1 ,X t ]+b0)·tanh(D t ); Among them, Y t is the multivariate load forecasting result obtained by the multivariate load forecasting model at time t; h t is the hidden state at time t, h t-1 is the hidden state at time t-1; is the hidden state of the forward long short-term neural network at time t, is the hidden state of the backward long-term short-term neural network at time t; D t is the cell state at time t, D t-1 is the cell state at time t-1; σ is the sigmoid activation function; tanh is the hyperbolic tangent activation function; W f ,W i ,W c , W0 are the connection weights of forget gate, input gate, Tanh gate and output gate respectively; b f ,b i ,b c ,b o are the bias vectors of forget gate, input gate, Tanh gate and output gate respectively; c1 and c2 are both constants, and c1+c2=1; The input-output mapping relationship of the multivariate load forecasting model is Y t =f(X t ), where f(·) is the input-output mapping function of the prediction model, X t is the output task calculated through multi-task learning at time t.

5. The method for multi-element load forecasting of an integrated energy system according to claim 1, characterized in that: Based on the improved Beluga optimization algorithm, the Beluga population representing the weights of each connection in the multivariate load forecasting model is initialized, including: An initial beluga whale population is generated by random initialization, with n beluga whale individuals and each beluga whale individual having d-dimensional position information; The pinhole imaging theory is used to improve the beluga optimization algorithm, and the corresponding points of the current beluga individual position information are introduced to generate an improved beluga population. Compare the fitness of the location information of each beluga whale individual in the initial beluga whale population and the improved beluga whale population, and sort the fitness in ascending or descending order to generate a fitness sequence; the fitness F(W) is calculated as follows: in, is the maximum mean square error of the three output loads of the multivariate load prediction model, namely, the cooling load, the heating load and the electric load; Y is the prediction model output value of the cooling, heating and electric loads at the same time; Y * It is the real value of cold, hot and electric load at the same moment; Based on the fitness sequence, beluga individuals whose fitness is greater than the fitness threshold are screened out to form an initialized beluga population; the initialized beluga population is used to characterize each connection weight in the multivariate load forecasting model.

6. The method for multi-element load forecasting of an integrated energy system according to claim 5, characterized in that: Based on the basic parameters of the improved Beluga optimization algorithm, the multivariate load forecasting model is trained according to the initialized Beluga population to generate a trained multivariate load forecasting model, which specifically includes: Initializing the initialized beluga population, and determining whether the current number of iterations is less than the maximum number of iterations; If so, determining the location information of the beluga whale individual corresponding to the highest fitness and the location information of the beluga whale individual corresponding to the lowest fitness based on the fitness sequence; Calculate the balance index of the position information of the beluga whale individual corresponding to the highest fitness and the position information of the beluga whale individual corresponding to the lowest fitness, as well as the probability of the beluga whale falling; Update the position information of individual beluga whales in the initialized beluga whale population according to the balance index and the beluga whale fall probability; Determine whether the updated location information meets the preset accuracy condition; If so, generate a trained multivariate load forecasting model; If not, re-determine whether the current number of iterations is less than the maximum number of iterations, until the updated position information meets the preset accuracy condition or the current number of iterations is not less than the maximum number of iterations, and stop training the multivariate load forecasting model.

7. The method for multi-element load forecasting of an integrated energy system according to claim 6, characterized in that: Updating the position information of individual beluga whales in the initialized beluga whale population according to the balance index and the beluga whale fall probability specifically includes: When the balance index is greater than 0.5, using the formula Update the location information of the White Whale search phase; where W i j(z+1) represents the position information of the i-th beluga individual in the j-th dimension at the z+1-th iteration, j = 1, 2, …, d; p j (z) is the dimension randomly selected from the d dimensions in the zth iteration; The i-th beluga individual and the x-th randomly selected beluga individual are j The position information in the dimension; α, β are random factors in the beluga search phase and are any numbers in (0,1); sin(2πβ), cos(2πβ) are any numbers between the average beluga fins; N f is the balance index; even is an even number; odd is an odd number; When the balance index is less than or equal to 0.5, use Update the location information of the beluga whale during the predation phase; among them, ε is a constant; is the best individual position of the white whale in the zth iteration; W i j(z) is the position information of the i-th beluga individual in the j-th dimension at the z-th iteration, is the current position of the i-th beluga individual in the z-th iteration; W i z is the current position of a randomly selected beluga individual; φ, are all arbitrary numbers in (0,1); u, v are the values ​​of the control step; δ is the control operator of the Levy flight function; Γ is the gamma function; Zt max is the maximum number of iterations; When the balance index is less than or equal to the falling probability, use Update the position information of the white whale during the falling stage; where η, κ, and λ are all arbitrary numbers in (0, 1); W step is the step length of the beluga whale individual moving to any position or falling, U b is the upper bound of the initialization variable, L b To initialize the lower bound of the variable, R = 2nN f , R is the step length factor of the beluga whale's individual fall, n is the total number of beluga whales in the beluga whale population; Q f is the probability of falling; The position information of the individual beluga whales in the initialized beluga whale population is updated, and the fitness of the individual beluga whales after the position update is calculated.

8. A multi-element load forecasting system for an integrated energy system, characterized in that: include: An initialization module is used to initialize the Beluga population that characterizes each connection weight in the multivariate load forecasting model based on the improved Beluga optimization algorithm, and to set the basic parameters of the improved Beluga optimization algorithm; the multivariate load forecasting model is a multivariate load forecasting model based on a bidirectional long-term and short-term neural network; the multivariate load forecasting model based on a bidirectional long-term and short-term neural network is constructed based on the historical data of the integrated energy system and the output tasks that have undergone weight sharing; the historical data includes multivariate load historical data, historical weather parameter data, historical date type data, and historical time data; the output task that has undergone weight sharing is obtained by constructing multiple single tasks according to different load types and applying multi-task learning theory to share weights for multiple single tasks; the output task that has undergone weight sharing is a multivariate load column vector that has undergone weight sharing; The basic parameters include the maximum number of iterations and the iteration stop condition; A training module, used for training the multivariate load forecasting model based on the basic parameters of the improved Beluga optimization algorithm and the initialized Beluga population to generate a trained multivariate load forecasting model; The prediction module is used to input the multivariate load historical data, weather parameters of the day to be predicted, date type and time data into the trained multivariate load prediction model, and output the multivariate load prediction value at the hour of the day to be predicted.

9. An electronic device, characterized in that: It includes a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the multi-load forecasting method for an integrated energy system as described in any one of claims 1-7.

10. The electronic device according to claim 9, characterized in that: The memory is a non-transitory computer-readable storage medium, and the non-transitory computer-readable storage medium stores a computer program. When the computer program is executed by the processor, the multi-element load forecasting method for the integrated energy system as described in any one of claims 1-7 is implemented.

Images