New energy medium and long term output hybrid hidden Markov scene construction method and system

Abstract

The invention discloses a new energy medium and long term output hybrid hidden Markov scene construction method and system, and belongs to the technical field of new energy output scenes. The method comprises the following steps: establishing a new energy output scene prediction multi-dimensional feature data vector set; based on the data vector set, probability clustering is carried out on the new energy power scene; based on the data vector set and the new energy power scene type occurrence probability, establishing an improved long-short-term memory network generation model; and designing a self-adaptive image group algorithm to solve the improved long-short-term memory network generation model to obtain a new energy output scene. According to the invention, the accuracy and reliability of new energy power generation output scene generation are improved, so that stable operation and optimal management of a power system are supported.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy output scenarios, and in particular to a method and system for constructing a hybrid hidden Markov scenario of new energy medium- and long-term output. Background Art

[0002] With the transformation of the global energy structure and the increasing dependence on renewable energy, new energy power generation, as a clean and renewable energy form, continues to increase its share in the global energy market. Effective generation of new energy output scenarios is crucial for the stable operation of the power grid, energy dispatch optimization, and market risk management. Accurately generating the output power of new energy power generation systems can help power grid operators to rationally plan the grid load and reduce losses caused by the volatility and uncertainty of new energy power generation. In addition, accurate generation of new energy output scenarios is also of great practical significance for improving the acceptance capacity of the power grid, reducing operation and maintenance costs, and improving the economic benefits of new energy power generation.

[0003] Although the technology for generating renewable energy output scenarios has developed rapidly in recent years, it still faces a series of challenges. Current generation methods often ignore the impact of environmental climate scenarios on renewable energy power generation, resulting in insufficient generation accuracy under specific climate conditions. Most existing generation models are based on linear or nonlinear modeling based on historical data, and have limited adaptability to complex climate changes. In addition, many models fail to fully utilize advanced artificial intelligence technologies, such as deep learning, to extract deeper features and patterns, thereby achieving high-precision generation in dynamically changing environments. These limitations highlight the shortcomings of existing generation methods in practical applications, which urgently need to be addressed through technological innovation.

[0004] In the existing technology, the impact of environmental climate on the generation of renewable energy power generation output scenarios has been considered, such as Patent 1: A method for generating renewable energy output scenarios based on a dual integrated model, application number: 202210543973.4; Patent 2: A method for generating short-term renewable energy output scenarios based on an LSGAN-DCN network, application number: 202210481070.8, which reflects that environmental climate factors have a relatively important impact on the generation of renewable energy power generation output scenarios. However, the current technology does not consider the probability distribution characteristics of the scenario: in the actual renewable energy generation scenario, there is a certain probability that the environmental climate scenario used for generation will undergo a conversion of the environmental climate scenario. If the converted scenario cannot be corrected in time, the conversion of the environmental climate scenario will have a significant impact on the result of renewable energy generation. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0006] The technical solution of the present invention is: a method for constructing a hybrid hidden Markov scenario for medium- and long-term output of new energy, comprising:

[0007] Step 1: Establish a multi-dimensional feature data vector set for new energy output scenario prediction;

[0008] Step 2: Based on the data vector set, perform probability clustering on the new energy power scenarios;

[0009] Step 3: Based on the data vector set and the probability of occurrence of new energy power scenario types, an improved long short-term memory network generation model is established;

[0010] Step 4: Design an adaptive elephant swarm algorithm to solve the improved long short-term memory network generation model to obtain the new energy output scenario.

[0011] The distributed new energy output scenario basic data set includes relevant characteristic data generated by new energy output scenarios within one year, specifically including irradiance data, temperature data, humidity data, air pressure data, new energy power generation data and load power consumption data.

[0012] The basic data set of the distributed renewable energy output scenario is generated based on the mixed Gaussian hidden Markov method, and includes:

[0013] S1: Initialize the mixed Gaussian model;

[0014] S2: Estimate the parameters of the Gaussian mixture model using the EM algorithm, including:

[0015] S201: According to the given K value, randomly initialize the Gaussian distribution parameters of each cluster and the weight vector ω k ;

[0016] S202: Calculate data point y i For each cluster, the membership degree E[Z ik ], the membership calculation formula is as follows:

[0017]

[0018]

[0019] Where: y i is the i-th data point; μ k is the mean; k is the covariance matrix; K is the number of Gaussian distributions; k is the kth Gaussian distribution;

[0020] S203: Use the calculated membership degree E[Z ik ]Update the model parameters as follows:

[0021]

[0022]

[0023]

[0024] Where: μ' k ,Σ' k ,ω' k are the updated mean, covariance matrix and weight of the k-th multivariate Gaussian distribution model; m is the sample data;

[0025] S3: Iteration: The EM algorithm is iterated using the hidden Markov model based on the Baum-Welch algorithm until the predetermined number of iterations is reached; the iteration formula is:

[0026]

[0027] Among them, λ i+1 ,λ i are the distributed renewable energy output states of the i+1th iteration round and the ith iteration round in the Markov process respectively; W is the distributed renewable energy output observation sequence; T is the distributed renewable energy output hidden observation sequence; P(·|·) represents the conditional probability function.

[0028] The similarity measurement method is used to perform probability clustering of new energy power scenarios, including:

[0029] S21: Calculate the distance between data points in the data vector set:

[0030]

[0031] In the formula, d(x i ,x j ) is the nth data point pair (x i ,x j ), Γ is the total number of data point pairs in the data vector set; x i Generate relevant feature data for the i-th renewable energy output scenario; x j Generate relevant feature data for the jth renewable energy output scenario;

[0032] S22: Analyze the probability compactness of each cluster during clustering;

[0033]

[0034] Among them, D in is the probability compactness of each class; m j is the center position of the jth class; y j For data belonging to this category; yis the total number of data under this category;

[0035] S23: Analyze the probability distance of each class in the clustering process;

[0036] D out =mind(m i ,m j )

[0037] Among them, D out is the probability distance of each class, m i is the center position of the i-th class;

[0038] S24: Calculate the probability of occurrence of the current new energy power scenario type, specifically:

[0039]

[0040] Among them, Pr is the probability of occurrence of the current new energy power scenario type, and X is the total number of data points in the data vector set;

[0041] S25: determine the probability of occurrence Pr of the current new energy power scenario type by real-time calculation,

[0042] If Pr is greater than the predetermined threshold, the new energy output scenario is generated under the current new energy power scenario;

[0043] If Pr is less than a predetermined threshold, the scenario is corrected to a scenario in which the probability of the new energy power scenario type occurring is greater.

[0044] The obtained probability of occurrence of new energy power scenario types and data vector sets are used as input variables of the long short-term memory network generation model to establish an improved long short-term memory network generation model;

[0045] The improved long short-term memory network generation model is an LSTM new energy output scenario generation model based on the improved whale swarm algorithm.

[0046] The forget gate function f of the LSTM new energy output scenario generation model t for:

[0047] f t =σ g (W f x t +R f h t-1 +b f )

[0048] Among them, σ g is the activation function, W f is the weight equation of the input to the forget gate at the current moment, R fis the cyclic weight matrix between the output and the forget gate at the previous moment, b f is the bias matrix of the forget gate, x t h is the input vector consisting of the probability of occurrence of the new energy power scenario type under the current input and the data vector set; t-1 The information of the previous moment.

[0049] The input gates of the LSTM new energy output scenario generation model include:

[0050] i t =σ g (W i x t +R i h t-1 +b i )

[0051] g t =tanh(W g x t +R g h t-1 +b g )

[0052] Among them, i t is the activation function of the input gate, g t is the input gate state update function, W i is the weight matrix from the input gate to the output gate, R i is the cyclic weight matrix from the output gate to the input gate at the previous moment, b i is the bias matrix of the current input gate, W g is the weight matrix input to the memory cell, R g is the cyclic weight matrix input to the current memory cell at the last moment, b g is the bias matrix of the current memory cell.

[0053] The output gates of the LSTM new energy output scenario generation model include:

[0054] o t =σ g (W o x t +R o h t-1 +b o )

[0055] h t =o t ⊙tanh(c t )

[0056] Among them, t is the activation function of the output gate, h tis the output gate state update function, W o is the weight matrix between the output gate and the input gate, R o is the cyclic weight matrix between the output at the previous moment and the current output gate, b o is the bias matrix of the current output gate, c t is the new cell state.

[0057] The improved whale swarm algorithm is used to optimize the parameters to be adjusted in the LSTM new energy output scenario generation model. The parameters to be adjusted include: W i , R i , b i , W g , R g , b g ; W o , R o , b o .

[0058] The optimization and tuning process of the improved whale swarm algorithm includes:

[0059] S31: Initialize the improved whale group algorithm; including: taking the parameters to be adjusted of the LSTM new energy output scenario generation model as the food position vector x of the improved whale group algorithm m ; Set the number of whales in the improved whale group algorithm n and the maximum number of food searches to I max ;

[0060] S32: Calculate the improved whale group algorithm food location food richness content function F:

[0061]

[0062] Where N p is the total number of generated time points; are the true value and generated value of the i-th point respectively;

[0063] S33: The neighborhood search is performed using the improved whale group algorithm neighborhood search equation that introduces an adaptive coefficient, specifically:

[0064]

[0065] In the formula, is the location of the food in the neighborhood; u(t) is the adaptive function of the improved whale group algorithm; They are the i-th, k-th, and j-th food positions respectively; is the adaptive coefficient; V i g is the optimal solution in the current cycle;

[0066] S34: Determine whether the number of calculations of the improved whale group algorithm has reached the maximum number of explorations Imax ; If the maximum number of explorations is reached, the LSTM parameters currently calculated by the improved whale swarm algorithm are the optimal parameters; if the maximum number of explorations has not yet been reached, continue to use the improved whale swarm algorithm to optimize the LSTM parameters and return to step 32.

[0067] Based on the LSTM new energy output scenario generation model, an adaptive elephant group algorithm is designed to solve the problem; specifically:

[0068] Step S41: Initializing the parameters of the adaptive elephant group algorithm; the parameters of the adaptive elephant group algorithm include: the number of elephant groups of the adaptive elephant group algorithm, the optimization dimension of the adaptive elephant group algorithm and the maximum number of searches of the adaptive elephant group algorithm;

[0069] Step S42: Generate the nomadic position of the elephant group of the adaptive elephant group algorithm; specifically: convert the food position of the whale in the improved whale group algorithm generated in step S33 into the nomadic position of the elephant group of the adaptive elephant group algorithm;

[0070] Step S43: Design a nomadic position update function for the elephant herd in the adaptive elephant herd algorithm, specifically:

[0071]

[0072] in, are the nomadic positions of the elephant herd at time t and time t+1 in the adaptive elephant herd algorithm; d g G is the random nomadic distance of the elephant herd in the adaptive elephant herd algorithm; c is the dynamic nomadic constant in the adaptive elephant herd algorithm; is the nomadic position of the optimal elephant herd at t in the adaptive elephant herd algorithm;

[0073] Step S44: Use the updated nomadic position of the elephant herd as the output solution of the LSTM new energy output scenario generation model in the current iteration round.

[0074] The hybrid hidden Markov scenario construction system for new energy medium and long-term output includes:

[0075] Data module, used to establish a multi-dimensional feature data vector set for new energy output scenario prediction;

[0076] A clustering module, used to perform probabilistic clustering of new energy power scenarios based on a data vector set;

[0077] A model module is used to establish an improved long short-term memory network generation model based on a data vector set and the probability of occurrence of new energy power scenario types;

[0078] The solution module is used to design an adaptive elephant swarm algorithm to solve the improved long short-term memory network generation model to obtain new energy output scenarios.

[0079] In operation, the present invention firstly establishes a multi-dimensional feature data vector set for predicting new energy output scenarios taking into account scenario probability correction; secondly, performs probability clustering on new energy power scenarios; then, establishes an improved long short-term memory network generation model; finally, designs an adaptive elephant swarm algorithm to solve the improved long short-term memory network generation model to obtain new energy output scenarios.

[0080] The present invention improves the accuracy of the generated hybrid hidden Markov model output scenario by correcting the scenario probability. At the same time, the use of artificial intelligence technology can further enhance the model's ability to process complex data and improve the generated intelligence level.

[0081] The present invention improves the accuracy and reliability of generating new energy power generation output scenarios to support stable operation and optimized management of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the specific implementation of the present invention or the technical solution in the prior art, the following briefly introduces the drawings required for the specific implementation or the prior art description. In the drawings, each part is not necessarily drawn according to the actual scale.

[0083] Figure 1 is a flow chart of the method of the present invention;

[0084] Figure 2 It is the iterative convergence diagram of the traditional elephant group algorithm;

[0085] Figure 3 It is the iterative convergence diagram of the adaptive elephant swarm algorithm of the present invention;

[0086] Figure 4 A comparison chart is generated for the new energy output scenarios under different algorithms in the embodiments of the present invention. DETAILED DESCRIPTION

[0087] 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 operation and maintenance personnel in this field without creative work are within the scope of protection of the present invention.

[0088] like Figure 1 As shown, the present invention provides a method for constructing a hybrid hidden Markov scenario of medium- and long-term output of new energy, comprising:

[0089] Step 1: Establish a multi-dimensional feature data vector set for new energy output scenario prediction considering scenario probability correction;

[0090] Step 2: Based on the data vector set, perform probability clustering on the new energy power scenarios;

[0091] Step 3: Based on the data vector set and the probability of occurrence of new energy power scenario types, an improved long short-term memory network generation model is established;

[0092] Step 4: Design an adaptive elephant swarm algorithm to solve the improved long short-term memory network generation model to obtain the new energy output scenario.

[0093] The present invention improves the accuracy of the generated hybrid hidden Markov model output scenario by correcting the scenario probability. At the same time, the use of artificial intelligence technology can further enhance the model's ability to process complex data and improve the generated intelligence level.

[0094] The present invention improves the accuracy and reliability of the generation of renewable energy power generation output scenarios to support the stable operation and optimized management of the power system. By introducing a scenario probability correction mechanism, it can better reflect the actual performance of renewable energy power generation under different environmental conditions, thereby reducing the generation error caused by climate change.

[0095] The distributed new energy output scenario basic data set includes relevant characteristic data generated by new energy output scenarios within one year, specifically including irradiance data, temperature data, humidity data, air pressure data, new energy power generation data and load power consumption data.

[0096] Furthermore, the basic data set of distributed renewable energy output scenarios is generated based on the hybrid Gaussian hidden Markov method.

[0097] In this way, multiple Gaussian distributions are used to fit the distributed new energy output scenario data that cannot be fitted by a Gaussian distribution, thereby better fitting the distributed new energy output scenario data with uncertainty.

[0098] The specific steps are:

[0099] S1: Initialization. First, the Gaussian mixed model (GMM) needs to be initialized, including determining the number of distributed renewable energy output distributions (K), the mean of each distributed renewable energy output distribution (μ), the covariance matrix of the distributed renewable energy output distribution (∑), and the mixing coefficient (ω), which can be set by random initialization or other methods.

[0100] S2: Expectation-Maximization (EM) algorithm. The core of GMM is to use the EM algorithm to estimate the parameters of the model. The EM algorithm is an iterative optimization algorithm used to maximize the likelihood function to estimate the model parameters. The basic steps are as follows:

[0101] S201: According to the given K value, randomly initialize the Gaussian distribution parameters (mean and variance) of each cluster, as well as the weight vector ω k ;

[0102] S202: Step E. Calculate data point y i For each cluster, the membership degree E[Z ik ], the larger the membership degree, the greater the probability that the sample is generated by the sub-model. The membership degree calculation formula is as follows:

[0103]

[0104]

[0105] Where: E[Z ik ] is the data point y i The membership degree of i is the i-th data point; μ k is the mean; k is the covariance matrix; K is the number of Gaussian distributions, k is the kth Gaussian distribution;

[0106] S203: Step M. Use all the calculated points to identify each sub-model Z k The membership degree of is used to update the model parameters. The formula for calculating the model parameters by the membership degree is as follows:

[0107]

[0108]

[0109]

[0110] Where: μ' k ,Σ' k ,ω' k are the updated mean, covariance matrix and weight of the k-th multivariate Gaussian distribution model; m is the sample data.

[0111] S3: Iteration: Based on the Baum-Welch algorithm, the EM algorithm is iterated using the hidden Markov model (HMM) until the predetermined number of iterations is reached. The iteration formula is:

[0112]

[0113] Among them, λ i+1 ,λ iare the distributed renewable energy output states of the i+1th iteration round and the ith iteration round in the Markov process respectively; W is the distributed renewable energy output observation sequence; T is the distributed renewable energy output hidden observation sequence; P(·|·) represents the conditional probability function.

[0114] Furthermore, the similarity measurement method is used to perform probability clustering on the new energy power scenarios, forming a multi-dimensional feature data vector set for distributed new energy output scenario prediction considering scenario probability correction. The similarity measurement method is used to correct the data under different scenarios during probability clustering, thereby improving data quality.

[0115] Specifically include:

[0116] S21: Calculate the distance between data points in the data vector set:

[0117]

[0118] In the formula, d(x i ,x j ) is the nth data point pair (x i ,x j ), Γ is the total number of data point pairs in the data vector set; x i Generate relevant feature data for the i-th renewable energy output scenario; x j Generate relevant feature data for the jth renewable energy output scenario.

[0119] S22: Analyze the probability compactness of each class in the clustering process, that is, the average distance from each data to its initial cluster center

[0120]

[0121] Among them, D in is the probability compactness of each class; m j is the center position of the jth class; y j For data belonging to this category; y The total number of data under this category.

[0122] S23: Analyze the probability distance of each class in the clustering process, that is, the degree of distance between different classes

[0123] D out =mind(m i ,m j )

[0124] Among them, D out is the probability distance of each class, m i is the center position of the ith class.

[0125] S24: Calculate the probability of occurrence of the current new energy power scenario type.

[0126]

[0127] Among them, Pr is the probability of occurrence of the current new energy power scenario type, and X is the total number of data points in the data vector set.

[0128] S25: Through the above steps, the probability of occurrence of the current new energy power scenario type Pr can be calculated in real time. If Pr is greater than a predetermined threshold (such as 50%), a new energy output scenario is generated under the current new energy power scenario; if Pr is less than the predetermined threshold, the scenario is corrected to a scenario with a higher probability of occurrence of the new energy power scenario type.

[0129] Furthermore, the obtained occurrence probability of the new energy power scenario type and the data vector set are used as input variables of the long short-term memory network generation model to establish an improved long short-term memory network generation model;

[0130] The improved long short-term memory network generation model is an LSTM new energy output scenario generation model based on the improved whale swarm algorithm.

[0131] The model takes into account both the probability of occurrence of new energy power scenario types and the data vector set. After the improvement, it can more effectively consider the impact of the probability of occurrence of new energy power scenario types on the model and enhance the effectiveness of new energy power in uncertain scenarios.

[0132] LSTM controls the flow of information by introducing a gating mechanism. Each LSTM unit is equipped with three important gates: input gate, forget gate, and output gate. These gates work using the Sigmoid activation function, and their outputs are between 0 and 1, thereby regulating the amount of information transmitted.

[0133] The forget gate of the LSTM unit receives the current input and the previous hidden state as input, and uses the Sigmoid function to calculate a value between 0 and 1, which determines the degree of retention of the information at the previous moment. 0 represents complete forgetfulness, and 1 represents complete retention.

[0134] The forget gate function f of the LSTM new energy output scenario generation model t for:

[0135] f t =σ g (W f x t +R f h t-1 +b f )

[0136] Among them, σ gis the activation function, W f is the weight equation of the input to the forget gate at the current moment, R f is the cyclic weight matrix between the output and the forget gate at the previous moment, b f is the bias matrix of the forget gate; x t h is the input vector consisting of the probability of occurrence of the new energy power scenario type under the current input and the data vector set; t-1 The information of the previous moment.

[0137] The input gate plays a role in LSTM to determine which new information should be included in the cell state. This process consists of two steps: first, a sigmoid layer is used to determine which information needs to be retained; then a tanh function is used to generate new candidate values, which are then added to the cell state.

[0138] The input gates of the LSTM new energy output scenario generation model include:

[0139] i t =σ g (W i x t +R i h t-1 +b i )

[0140] g t =tanh(W g x t +R g h t-1 +b g )

[0141] Among them, i t is the activation function of the input gate, g t is the input gate state update function, W i is the weight matrix from the input gate to the output gate, R i is the cyclic weight matrix from the output gate to the input gate at the previous moment, b i is the bias matrix of the current input gate, W g is the weight matrix input to the memory cell, R g is the cyclic weight matrix input to the current memory cell at the last moment, b g is the bias matrix of the current memory cell.

[0142] The old cell state c t-1 Combined with the output of the forget gate, and then added with the output of the input gate and the candidate value, the new cell state c is obtained. t The cell state update is specifically shown in the following formula:

[0143] ct =f t ⊙c t-1 +i t ⊙g t

[0144] where ⊙ represents the Hadamard product.

[0145] The current input x(t) and the previous output h(t-1) are processed by the output gate. The cell state is processed by a Tanh layer. The processed cell state is then multiplied by the output gate result to obtain the final selected output information. The cell output formula is as follows:

[0146] o t =σ g (W o x t +R o h t-1 +b o )

[0147] h t =o t ⊙tanh(c t )

[0148] Among them, t is the activation function of the output gate, h t is the output gate state update function, W o is the weight matrix between the output gate and the input gate, R o is the cyclic weight matrix between the output at the previous moment and the current output gate, b o is the bias matrix of the current output gate.

[0149] Furthermore, the following parameters to be adjusted for the LSTM new energy output scenario generation model are considered, including: W i , R i , b i , W g , R g , b g ; W o , R o , b o .

[0150] Furthermore, the improved whale swarm algorithm is used to optimize and adjust the above parameters to be adjusted. After adjustment, the performance of the LSTM new energy output scenario generation model can be optimized.

[0151] Furthermore, the optimization and tuning process of the improved whale group algorithm is as follows:

[0152] S31: Initialize the improved whale group algorithm. This includes: using the parameters to be adjusted of the LSTM new energy output scenario generation model as the food position vector x of the improved whale group algorithm m ; Set the number of whales in the improved whale group algorithm n and the maximum number of food searches to I max .

[0153] S32: Calculate the food position and food richness content function F of the improved whale group algorithm to quantify the quality of the current long short-term memory network parameters to be optimized.

[0154] The food position and food abundance function F of the improved whale group algorithm can be expressed as:

[0155]

[0156] Where N p is the total number of generated time points; are the true value and generated value of the i-th point respectively;

[0157] S33: Use the improved whale group algorithm neighborhood search equation with the introduction of adaptive coefficients to perform neighborhood search. Specifically:

[0158]

[0159] In the formula, is the location of the food in the neighborhood; u(t) is the adaptive function of the improved whale group algorithm; They are the i-th, k-th, and j-th food positions respectively; is the adaptive coefficient; V i g is the optimal solution in the current cycle.

[0160] Furthermore, in order to balance the global exploration ability and local mining ability of the algorithm at different stages of the iterative process of the improved whale swarm algorithm, the adaptive coefficient is designed as follows: in the early stage of the algorithm iteration process, the adaptive coefficient is set to a higher value (such as 0.9). At this time, the algorithm is in a free exploration state and can better explore the global optimal solution; in the later stage of the algorithm iteration process, the adaptive coefficient is set to a lower value (such as 0.1). At this time, the algorithm is in a state of following the current local optimal solution, thereby improving the algorithm's ability to mine local solutions.

[0161] S34: Determine whether the number of calculations of the improved whale group algorithm has reached the maximum number of explorations I max If the maximum number of explorations is reached, the LSTM parameters currently calculated by the improved whale swarm algorithm are the optimal parameters. If the maximum number of explorations is not reached, the improved whale swarm algorithm is continued to be used to optimize the LSTM parameters, and the process returns to step 2102.

[0162] Further,

[0163] Based on the LSTM new energy output scenario generation model, an adaptive elephant swarm algorithm is designed to improve the local search capability of the optimal solution of the generation model parameters. The specific steps are:

[0164] Step S41: Initializing the parameters of the adaptive elephant group algorithm. The parameters of the adaptive elephant group algorithm include: the number of elephant groups of the adaptive elephant group algorithm, the optimization dimension of the adaptive elephant group algorithm, and the maximum number of searches of the adaptive elephant group algorithm.

[0165] Step S42: Generate the nomadic position of the elephant group of the adaptive elephant group algorithm. Specifically, the food position of the whale in the improved whale group algorithm generated in step S33 in step 3 is converted into the nomadic position of the elephant group of the adaptive elephant group algorithm.

[0166] Step S43: Design a nomadic position update function for the elephant herd in the adaptive elephant herd algorithm to find the optimal solution for the parameters of the LSTM new energy output scenario generation model.

[0167]

[0168] in, are the nomadic positions of the elephant herd at time t and time t+1 in the adaptive elephant herd algorithm; d g G is the random nomadic distance of the elephant herd in the adaptive elephant herd algorithm; c is the dynamic nomadic constant in the adaptive elephant herd algorithm. is the nomadic position of the optimal elephant herd at t in the adaptive elephant herd algorithm.

[0169] Step S44: The updated nomadic position of the elephant herd is used as the output solution of the LSTM new energy output scenario generation model in the current iteration round, that is, the new energy output scenario is obtained.

[0170] In the present invention, a new energy output scenario within one year in a certain place is selected to generate a data set for training, and the data includes irradiance data, temperature data, humidity data, air pressure data, new energy power generation data and load power consumption data. The number of whales in the improved whale group algorithm is set to 200, and the maximum number of iterations is set to 100. At the same time, the number of elephant groups is 100, and the maximum number of iterations is also 100.

[0171] In the present invention, two improved LSTM new energy output scenario generation models based on two different elephant group algorithm optimizations are compared. Method 1: an improved LSTM new energy output scenario generation model based on traditional elephant group algorithm optimization; Method 2: an improved LSTM new energy output scenario generation model based on the adaptive elephant group algorithm optimization proposed in the present invention. Figure 2 and Figure 3The convergence of the improved LSTM new energy output scenario generation model based on the two methods is shown respectively. It can be seen that the adaptive elephant herd algorithm proposed by the present invention can converge to a better result faster and can continuously optimize local parameters. This significant improvement in performance stems from the fact that the present invention can adaptively optimize and adjust the nomadic position of the elephant herd, thereby exploring the optimal solution.

[0172] Figure 4 The generation error of the new energy output scenario generation model under different algorithms is demonstrated. Including: the new energy power generation output scenario generation algorithm (AEHO) considering the scenario probability correction proposed in the present invention, the new energy power generation output scenario generation algorithm (EHO) based on the traditional elephant swarm algorithm, and the new energy power generation output scenario generation algorithm (ABC) based on the bee colony algorithm. As can be seen from the figure, when the scene undergoes a sudden change, the new energy power generation output scenario generation algorithm considering the scenario probability correction proposed in the present invention can still maintain a low generation error. This is because the present invention incorporates the scenario probability into the data set for overall consideration, and can adjust the algorithm model parameters in a timely manner when the scene undergoes a sudden change, thereby ensuring that the proposed method can accurately generate output scenarios in more new energy output scenario generation scenarios.

[0173] The new energy medium- and long-term output hybrid hidden Markov scenario construction system of the present invention includes:

[0174] Data module, used to establish a multi-dimensional feature data vector set for new energy output scenario prediction;

[0175] A clustering module, used to perform probabilistic clustering of new energy power scenarios based on a data vector set;

[0176] A model module is used to establish an improved long short-term memory network generation model based on a data vector set and the probability of occurrence of new energy power scenario types;

[0177] The solution module is used to design an adaptive elephant swarm algorithm to solve the improved long short-term memory network generation model to obtain new energy output scenarios.

[0178] The present invention corrects the generation result by comprehensively considering the probability distribution of the environmental climate scene to address the problem of insufficient accuracy of existing generation technology under specific climate conditions.

[0179] By integrating scene probability correction and artificial intelligence algorithms, the present invention can not only overcome the challenges faced by existing generation technologies, but also provide stronger guarantees for the reliability of renewable energy power generation and the stability of the power grid.

[0180] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, ordinary operation and maintenance personnel in the field should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for constructing a hybrid hidden Markov scenario for medium- and long-term output of new energy, characterized in that: include: Step 1: Establish a multi-dimensional feature data vector set for new energy output scenario prediction; Step 2: Based on the data vector set, perform probability clustering on the new energy power scenarios; Step 3: Based on the data vector set and the probability of occurrence of new energy power scenario types, an improved long short-term memory network generation model is established; Step 4: Design an adaptive elephant swarm algorithm to solve the improved long short-term memory network generation model to obtain the new energy output scenario.

2. The method for constructing a hybrid hidden Markov scenario of new energy medium- and long-term output according to claim 1 is characterized in that: The distributed new energy output scenario basic data set includes relevant characteristic data generated by new energy output scenarios within one year, specifically including irradiance data, temperature data, humidity data, air pressure data, new energy power generation data and load power consumption data.

3. The method for constructing a hybrid hidden Markov scenario of new energy medium- and long-term output according to claim 1 is characterized in that: The basic data set of distributed renewable energy output scenarios is generated based on a mixed Gaussian hidden Markov method, and includes: S1: Initialize the mixed Gaussian model; S2: Use the EM algorithm to estimate the parameters of the Gaussian mixture model, including: S201: According to the given Gaussian distribution number K value, randomly initialize the Gaussian distribution parameters of each cluster and the weight vector ω k ; S202: Calculate data point y i For each cluster, the membership degree E[Z ik ], the membership calculation formula is as follows: Where: y i is the i-th data point; μ k is the mean; k is the covariance matrix; k is the kth Gaussian distribution; S203: Use the calculated membership degree E[Z ik ]Update the model parameters as follows: Where: μ' k ,Σ' k ,ω' k are the updated mean, covariance matrix and weight of the k-th multivariate Gaussian distribution model; m is the sample data; S3: Iteration: Based on the Baum-Welch algorithm, the EM algorithm is iterated using the hidden Markov model until the predetermined number of iterations is reached; the iteration formula is: Among them, λ i+1 ,λ i are the distributed renewable energy output states of the i+1th iteration round and the ith iteration round in the Markov process respectively; W is the distributed renewable energy output observation sequence; T is the distributed renewable energy output hidden observation sequence; P(·|·) represents the conditional probability function.

4. The method for constructing a hybrid hidden Markov scenario of new energy medium- and long-term output according to claim 1 is characterized in that: The similarity measurement method is used to perform probability clustering of new energy power scenarios, including: S21: Calculate the distance between data points in the data vector set: In the formula, d(x i ,x j ) is the nth data point pair (x i ,x j ), Γ is the total number of data point pairs in the data vector set; x i Generate relevant feature data for the i-th renewable energy output scenario; x j Generate relevant feature data for the jth renewable energy output scenario; S22: Analyze the probability compactness of each cluster during clustering; Among them, D in is the probability compactness of each class; m j is the center position of the jth class; y j For data belonging to this category; y is the total number of data under this category; S23: Analyze the probability distance of each class in the clustering process; D out =mind(m i ,m j ) Among them, D out is the probability distance of each class, m i is the center position of the i-th class; S24: Calculate the probability of occurrence of the current new energy power scenario type, specifically: Among them, Pr is the probability of occurrence of the current new energy power scenario type, and X is the total number of data points in the data vector set; S25: determine the probability of occurrence Pr of the current new energy power scenario type by real-time calculation, If Pr is greater than the predetermined threshold, the new energy output scenario is generated under the current new energy power scenario; If Pr is less than a predetermined threshold, the scenario is corrected to a scenario in which the probability of the new energy power scenario type occurring is greater.

5. The method for constructing a hybrid hidden Markov scenario of new energy medium- and long-term output according to claim 1 is characterized in that: The obtained probability of occurrence of new energy power scenario types and data vector sets are used as input variables of the long short-term memory network generation model to establish an improved long short-term memory network generation model; The improved long short-term memory network generation model is an LSTM new energy output scenario generation model based on the improved whale swarm algorithm.

6. The method for constructing a hybrid hidden Markov scenario of new energy medium- and long-term output according to claim 5 is characterized in that: The forget gate function f of the LSTM new energy output scenario generation model t for: f t =σ g (W f x t +R f h t-1 +b f ) Among them, σ g is the activation function, W f is the weight equation of the input to the forget gate at the current moment, R f is the cyclic weight matrix between the output and the forget gate at the previous moment, b f is the bias matrix of the forget gate, x t h is the input vector consisting of the probability of occurrence of the new energy power scenario type under the current input and the data vector set; t-1 The information of the previous moment.

7. The method for constructing a hybrid hidden Markov scenario for medium- and long-term output of new energy according to claim 6 is characterized in that: The input gates of the LSTM new energy output scenario generation model include: i t =σ g (W i x t +R i h t-1 +b i ) g t =tanh(W g x t +R g h t-1 +b g ) Among them, i t is the activation function of the input gate, g t is the input gate state update function, W i is the weight matrix from the input gate to the output gate, R i is the cyclic weight matrix from the output gate to the input gate at the previous moment, b i is the bias matrix of the current input gate, W g is the weight matrix input to the memory cell, R g is the cyclic weight matrix input to the current memory cell at the last moment, b g is the bias matrix of the current memory cell.

8. The method for constructing a hybrid hidden Markov scenario of new energy medium- and long-term output according to claim 7 is characterized in that: The output gates of the LSTM new energy output scenario generation model include: o t =σ g (W o x t +R o h t-1 +b o ) h t =o t ⊙tanh(c t ) Among them, t is the activation function of the output gate, h t is the output gate state update function, W o is the weight matrix between the output gate and the input gate, R o is the cyclic weight matrix between the output at the previous moment and the current output gate, b o is the bias matrix of the current output gate, c t is the new cell state.

9. The method for constructing a hybrid hidden Markov scenario of new energy medium- and long-term output according to claim 8 is characterized in that: The improved whale swarm algorithm is used to optimize the parameters to be adjusted in the LSTM new energy output scenario generation model. The parameters to be adjusted include: W i , R i , b i , W g , R g , b g ; W o , R o , b o .

10. The method for constructing a hybrid hidden Markov scenario of new energy medium- and long-term output according to claim 9, characterized in that: The optimization and tuning process of the improved whale swarm algorithm includes: S31: Initialize the improved whale group algorithm; including: taking the parameters to be adjusted of the LSTM new energy output scenario generation model as the food position vector x of the improved whale group algorithm m ; Set the number of whales in the improved whale group algorithm to n and the maximum number of food searches to I max ; S32: Calculate the improved whale group algorithm food location food richness content function F: Where N p is the total number of generated time points; are the true value and generated value of the i-th point respectively; S33: The neighborhood search is performed using the improved whale group algorithm neighborhood search equation that introduces an adaptive coefficient, specifically: In the formula, is the location of the food in the neighborhood; u(t) is the adaptive function of the improved whale group algorithm; They are the i-th, k-th, and j-th food positions respectively; is the adaptive coefficient; V i g is the optimal solution in the current cycle; S34: Determine whether the number of calculations of the improved whale group algorithm has reached the maximum number of explorations I max ; If the maximum number of explorations is reached, the LSTM parameters currently calculated by the improved whale swarm algorithm are the optimal parameters; if the maximum number of explorations has not yet been reached, continue to use the improved whale swarm algorithm to optimize the LSTM parameters and return to step 32.

11. The method for constructing a hybrid hidden Markov scenario of new energy medium- and long-term output according to claim 10, characterized in that: Based on the LSTM renewable energy output scenario generation model, an adaptive elephant group algorithm is designed to solve the problem; Specifically include: Step S41: Initializing the parameters of the adaptive elephant group algorithm; the parameters of the adaptive elephant group algorithm include: the number of elephant groups of the adaptive elephant group algorithm, the optimization dimension of the adaptive elephant group algorithm and the maximum number of searches of the adaptive elephant group algorithm; Step S42: Generate the nomadic position of the elephant group of the adaptive elephant group algorithm; specifically: convert the food position of the whale in the improved whale group algorithm generated in step S33 into the nomadic position of the elephant group of the adaptive elephant group algorithm; Step S43: Design a nomadic position update function for the elephant herd in the adaptive elephant herd algorithm, specifically: in, are the nomadic positions of the elephant herd at time t and time t+1 in the adaptive elephant herd algorithm; d g G is the random nomadic distance of the elephant herd in the adaptive elephant herd algorithm; c is the dynamic nomadic constant in the adaptive elephant herd algorithm; is the nomadic position of the optimal elephant herd in the adaptive elephant herd algorithm at t; Step S44: Use the updated nomadic position of the elephant herd as the output solution of the LSTM new energy output scenario generation model in the current iteration round.

12. A hybrid hidden Markov scenario construction system for new energy medium and long-term output, characterized by: include: Data module, used to establish a multi-dimensional feature data vector set for new energy output scenario prediction; A clustering module, used to perform probabilistic clustering of new energy power scenarios based on a data vector set; A model module is used to establish an improved long short-term memory network generation model based on a data vector set and the probability of occurrence of new energy power scenario types; The solution module is used to design an adaptive elephant swarm algorithm to solve the improved long short-term memory network generation model to obtain new energy output scenarios.

Images