Renewable energy system operation scene generation method and system

Abstract

The invention discloses a renewable energy system operation scene generation method. The method comprises the following steps: acquiring historical data information of a target renewable energy system; establishing a diffusion process model for the operation scene of the target renewable energy system; de-noising processing is carried out based on Gaussian noise; modeling a conditional diffusion model; constructing a neural network for a conditional diffusion model; adopting a large model to guide and generate conditional information, and training the established conditional diffusion model to obtain a trained conditional diffusion model; and generating a renewable energy system operation scene by adopting the trained conditional diffusion model. The invention also discloses a system for realizing the renewable energy system operation scene generation method. According to the method, the large language model is used as condition guidance, and the diffusion model is used for directly generating the deterministic operation scene from any Gaussian noise, so that the method not only can realize the generation of the operation scene of the renewable energy system, but also is high in reliability, good in accuracy and high in efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of electrical automation, and in particular relates to a method and system for generating an operating scenario of a renewable energy system. Background Art

[0002] With the development of economy and technology and the improvement of people's living standards, electricity has become an indispensable secondary energy source in people's production and life, bringing endless convenience to people's production and life. Therefore, ensuring the stable and reliable supply of electricity has become one of the most important tasks of the power system.

[0003] At present, more and more renewable energy power generation systems such as wind power and photovoltaic power generation are being integrated into the power system to generate electricity. However, the volatility, intermittency and randomness of the output of renewable energy power generation systems have brought great uncertainty to the short-term operation of the power system. Therefore, it is of great significance for the power system to accurately and reliably quantify the uncertainty of the power system with large-scale renewable energy grid connection and conduct research based on the modeling data.

[0004] At present, common renewable energy system operation scenario generation schemes include sampling-based scenario generation methods and optimization-based scenario generation methods. Sampling-based scenario generation methods are the most common. This type of scheme directly samples the probability distribution to generate a discrete set of scenarios; however, sampling-based methods may suffer from some key problems, that is, the probability distribution of renewable random variables must be known or assumed to approximately follow a certain distribution; this problem greatly affects the reliability of the final generated scenario. Optimization-based methods use scenario reduction technology to extract scenarios from a given large scenario set; however, this type of scheme has low computational efficiency and involves serious high-dimensional solution problems. Summary of the invention

[0005] One of the purposes of the present invention is to provide a method for generating an operation scenario of a renewable energy system with high reliability, good accuracy and high efficiency.

[0006] A second objective of the present invention is to provide a system for realizing the method for generating an operation scenario of a renewable energy system.

[0007] The method for generating a renewable energy system operation scenario provided by the present invention comprises the following steps:

[0008] S1. Obtain historical data information of the target renewable energy system;

[0009] S2. Establishing a diffusion process model for the operation scenario of the target renewable energy system based on the data information obtained in step S1;

[0010] S3. Perform denoising based on Gaussian noise according to the data information obtained in step S1 and the diffusion process model established in step S2;

[0011] S4. Modeling the conditional diffusion model for the data obtained in step S3;

[0012] S5. Construct a neural network for the conditional diffusion model;

[0013] S6. Using the large model to guide, generate conditional information, and train the established conditional diffusion model to obtain a trained conditional diffusion model;

[0014] S7. Using the trained conditional diffusion model obtained in step S6, generate a renewable energy system operation scenario.

[0015] Step S2, according to the data information obtained in step S1, establishes a diffusion process model for the operation scenario of the target renewable energy system, which specifically includes the following steps:

[0016] Assume that the operating scenario of the renewable energy system is x; add Gaussian noise to the operating scenario x through the diffusion process, expressed as

[0017]

[0018] In the formula, q(x 1:T |x0) represents the diffusion process; x 1:T is a sample after 1…T steps of noise diffusion; x0 is the original sample sampled from the real data distribution; T is the total number of diffusion steps; q(x t ∣x t-1 ) represents the tth diffusion step; x t is the sample after t steps of diffusion;

[0019] For the tth diffusion step q(x t ∣x t-1 ), which accepts x t-1 And return x by adding some noise t , expressed as

[0020]

[0021] Where N() is the standard Gaussian noise function; β t is the noise planning table, and β t ∈(0,1); I is the unit matrix;

[0022] After T steps of diffusion, x0 will be transformed into a sample x after T steps of diffusion. T , and obeys an isotropic Gaussian distribution.

[0023] The step S3 described in which the data information obtained in step S1 and the diffusion process model established in step S2 are subjected to denoising based on Gaussian noise, specifically includes the following steps:

[0024] The denoising process is used to T Remove the noise to recover x0 so that x0 can create new data samples from the Gaussian noise distribution;

[0025] The denoising process is expressed as

[0026] p θ (x t-1 ∣x t )=N(x t-1 ;μ θ (x t ,t),β t I)

[0027] Where p θ (x t-1 ∣x t ) represents the denoising process realized by the neural network; μ θ (x t ,t) represents the neural network processing process with parameter θ;

[0028] According to the denoising steps, the denoising process is expressed as the following formula:

[0029]

[0030] Where p θ (x 0:T ) is the denoising process; p(x T ) is Gaussian noise, and p(x T )=N(0,I);

[0031] Assuming the denoising process is the inverse process of the diffusion process, the objective function of the diffusion model is expressed as

[0032] min(KL(p θ (x t-1 ∣x t ),q(x t-1 ∣x t ,x0)))

[0033] Where KL() is the Kullback-Leibler divergence calculation function; q(x t-1 ∣x t ,x0) is the posterior conditional probability of the diffusion process.

[0034] The step S4 of modeling the conditional diffusion model for the data obtained in step S3 specifically includes the following steps:

[0035] In order to achieve deterministic generation guided by conditions, the diffusion model is extended to a conditional diffusion model to learn conditional information.

[0036] The formula p θ (x t-1 ∣x t )=N(x t-1 ;μ θ (x t ,t),β t I) Expand to

[0037] p θ (x t-1 ∣x t ,x c )=N(x t-1 ;μ θ (x t ,t|x c ),β t I)

[0038] Where p θ (x t-1 ∣x t ,x c ) is to convert x t becomes x t-1 The denoising process of x c is conditional information;

[0039] Will Expand to

[0040]

[0041] Where p θ (x t-1 ∣x t ,x c ) is to convert x t becomes x t-1 The denoising process;

[0042] The undetermined coefficient method is used to To solve:

[0043] set up Where p(x t-1 ∣x t ,x0) is the denoising process, κ t is the first coefficient, λ t is the second coefficient, σ t is the third coefficient;

[0044] Setting σ t is a variable, and where α t is the first intermediate variable, βt is the noise planning table, is the second intermediate variable, is the third intermediate variable;

[0045] So we get the following formula

[0046]

[0047] Where γ is the fourth intermediate variable, and

[0048] If you set And η∈[0,1], η is the weight coefficient, then it can be obtained that when η=0, the variance of the Gaussian distribution is 0, and the generation is deterministic at this time;

[0049] At the same time, the neural network μ θ (x t ,t) is parameterized as

[0050]

[0051] Where ε θ is a diffusion model with parameter θ.

[0052] The step S5 of constructing a neural network for a conditional diffusion model specifically includes the following steps:

[0053] The constructed neural network includes an embedding module, a residual module, and an output module;

[0054] The embedding module is used to project the input features into a high-dimensional space; the embedding module includes a linear layer; the input is set to x t , the condition is x c , then the processing of the embedded module is expressed as

[0055] x emb =FC(x t )+FC(x c )

[0056] Where x emb is the output of the embedding module; FC() is the fully connected layer processing function;

[0057] The residual module is used to connect the input with the module output to alleviate the gradient vanishing problem. The residual module includes four residual submodules, each of which includes a multi-head attention layer and a fully connected layer. The processing process of the i-th residual submodule is expressed as

[0058] x res,i =FC(MHA(x res,i-1 ))+x res,i-1

[0059] Where x res,i is the output of the i-th residual submodule; MHA() is the processing function of the multi-head attention layer; if i is 1, then x res,i-1 is the output of the embedded module;

[0060] The output module is used to connect the outputs of all residual modules and map them to obtain the final output result; the output module includes a fully connected layer; the processing process of the output module is represented by x out =FC(x res,1 ,x res,2 ,x res,3 ,x res,4 ), x out Output module output.

[0061] The step S6 of using a large model to guide, generating conditional information, and training the established conditional diffusion model to obtain a trained conditional diffusion model specifically includes the following steps:

[0062] For the diffusion generation task, prompt words are designed: the prompt word format is set as <S, P, D, C>, where S represents the season and the value range is {spring, summer, autumn, winter}, P represents the energy output intensity and the value range is {high, medium, low}, D represents daytime or night, and C represents wind power or solar energy;

[0063] The open source Qwen2-7B model is used as the large model;

[0064] The first round of text dialogue with the big model: The big model uses semantic understanding to confirm whether the input satisfies the format of <S, P, D, C>. If it does, it constructs conditional information to guide the diffusion model training. If not, it enters the second round of dialogue.

[0065] Second round of text dialogue with the big model: Complete the text according to the missing conditions fed back by the big model, and have another text dialogue with the big model;

[0066] The bootstrap process is as follows: construct a vocabulary table to map text vocabulary; the large model formats and extracts the input text, constructs <S, P, D, C> and maps it to the vocabulary table as conditional information x c ; Diffusion of the original sample x0 to obtain x T , and x0, x c and x T The data are inputted together into the conditional diffusion model obtained by modeling in step S4 for training;

[0067] Finally, the trained conditional diffusion model is obtained.

[0068] The step S7 uses the trained conditional diffusion model obtained in step S6 to generate a renewable energy system operation scenario, which specifically includes the following steps:

[0069] Randomly sample a Gaussian noise x from any standard Gaussian distribution T,new , input into the trained conditional diffusion model obtained in step S6, and obtain x generated by the model c , and according to x c Complete the generation of renewable energy system operation scenarios.

[0070] The present invention also provides a system for realizing the method for generating an operation scenario of a renewable energy system, comprising a data acquisition module, a diffusion modeling module, a Gaussian denoising module, a model modeling module, a network construction module, a model guiding module and a scenario generating module; the data acquisition module, the diffusion modeling module, the Gaussian denoising module, the model modeling module, the network construction module, the model guiding module and the scenario generating module are connected in series in sequence; the data acquisition module is used to acquire historical data information of a target renewable energy system, and upload the data information to the diffusion modeling module; the diffusion modeling module is used to establish a diffusion process model for an operation scenario of a target renewable energy system according to the received data information and the acquired data information, and upload the data information to the Gaussian denoising module; the Gaussian denoising module is used to establish a diffusion process model for an operation scenario of a target renewable energy system according to the received data information and the acquired data information The information and the established diffusion process model are de-noised based on Gaussian noise, and the data information is uploaded to the model building module; the model building module is used to model the conditional diffusion model for the obtained data according to the received data information, and upload the data information to the network construction module; the network construction module is used to construct a neural network for the conditional diffusion model according to the received data information, and upload the data information to the model guidance module; the model guidance module is used to generate conditional information based on the received data information and the large model guidance, and train the established conditional diffusion model to obtain the trained conditional diffusion model, and upload the data information to the scenario generation module; the scenario generation module is used to generate the operation scenario of the renewable energy system based on the received data information and the trained conditional diffusion model.

[0071] The renewable energy system operation scenario generation method and system provided by the present invention can use a large language model as a conditional guide and utilize a diffusion model to directly generate deterministic operation scenarios from arbitrary Gaussian noise. Therefore, the present invention can not only realize the generation of renewable energy system operation scenarios, but also has higher reliability, better accuracy and higher efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 The figure is a schematic diagram of the method flow of the present invention.

[0073] Figure 2 A schematic diagram of a generation scenario of an embodiment of the method of the present invention.

[0074] Figure 3 Schematic diagram of the functional modules of the system of the present invention. DETAILED DESCRIPTION

[0075] like Figure 1 The method flow chart of the method of the present invention is shown as follows: The method for generating an operation scenario of a renewable energy system disclosed in the present invention comprises the following steps:

[0076] S1. Obtain historical data information of the target renewable energy system;

[0077] The purpose of this step is to mathematically model the diffusion process to achieve the transformation of original samples to noise samples;

[0078] S2. Based on the data information obtained in step S1, a diffusion process model is established for the operation scenario of the target renewable energy system; specifically, the following steps are included:

[0079] Assume that the operating scenario of the renewable energy system is x; add Gaussian noise to the operating scenario x through the diffusion process, expressed as

[0080]

[0081] Where q(x 1:T |x0) represents the diffusion process; x 1:T is a sample after 1…T steps of noise diffusion; x0 is the original sample sampled from the real data distribution; T is the total number of diffusion steps; q(x t ∣x t-1 ) represents the tth diffusion step; x t is the sample after t steps of diffusion;

[0082] In specific implementation, the operation scenario x of the renewable energy system includes the power values ​​of wind power generation and photovoltaic power generation in the actual operation period, and each operation scenario is divided into a time length of 24 hours; Gaussian noise represents a set of random data that conforms to the normal distribution;

[0083] For the tth diffusion step q(x t ∣x t-1 ), which accepts x t-1 And return x by adding some noise t , expressed as

[0084]

[0085] Where N() is the standard Gaussian noise function; β t is the noise planning table, and β t∈(0,1); I is the unit matrix;

[0086] After T steps of diffusion, x0 will be transformed into a sample x after T steps of noise diffusion. T , and obeys an isotropic Gaussian distribution;

[0087] The purpose of this step is to mathematically model the denoising process to guide the training of the model;

[0088] S3. According to the data information obtained in step S1 and the diffusion process model established in step S2, denoising is performed based on Gaussian noise; specifically comprising the following steps:

[0089] The denoising process is used to T Remove the noise to recover x0 so that x0 can create new data samples from the Gaussian noise distribution;

[0090] The denoising process is expressed as

[0091] p θ (x t-1 ∣x t )=N(x t-1 ;μ θ (x t ,t),β t I)

[0092] Where p θ (x t-1 ∣x t ) represents the denoising process realized by the neural network; μ θ (x t ,t) represents the neural network processing process with parameter θ;

[0093] According to the denoising steps, the denoising process is expressed as the following formula:

[0094]

[0095] Where p θ (x 0:T ) is the denoising process; p(x T ) is Gaussian noise, and p(x T )=N(0,I);

[0096] Assuming the denoising process is the inverse process of the diffusion process, the objective function of the diffusion model is expressed as

[0097] min(KL(p θ (x t-1 ∣x t ),q(x t-1 ∣x t ,x0)))

[0098] Where KL() is the Kullback-Leibler divergence calculation function; q(x t-1 ∣x t ,x0) is the posterior conditional probability of the diffusion process;

[0099] The purpose of this step is to mathematically model the conditional diffusion model to guide the model's generation of renewable energy operation scenarios;

[0100] S4. Modeling a conditional diffusion model for the data obtained in step S3; specifically comprising the following steps:

[0101] In order to achieve deterministic generation guided by conditions, the diffusion model is extended to a conditional diffusion model to learn conditional information.

[0102] The formula p θ (x t-1 ∣x t )=N(x t-1 ;μ θ (x t ,t),β t I) Expand to

[0103] p θ (x t-1 ∣x t ,x c )=N(x t-1 ;μ θ (x t ,t|x c ),β t I)

[0104] Where p θ (x t-1 ∣x t ,x c ) is to convert x t becomes x t-1 The denoising process of x c is conditional information;

[0105] Will Expand to

[0106]

[0107] Where p θ (x t-1 ∣x t ,x c ) is to convert x t becomes x t-1 The denoising process;

[0108] The undetermined coefficient method is used to To solve:

[0109] set up Where p(x t-1 ∣x t ,x0) is the denoising process, κ t is the first coefficient, λ t is the second coefficient, σ t is the third coefficient;

[0110] Setting σ t is a variable, and where α t is the first intermediate variable, β t is the noise planning table, is the second intermediate variable, is the third intermediate variable;

[0111] So we get the following formula

[0112]

[0113] Where γ is the fourth intermediate variable, and

[0114] If you set And η∈[0,1], η is the weight coefficient, then it can be obtained that when η=0, the variance of the Gaussian distribution is 0, and the generation is deterministic at this time;

[0115] At the same time, the neural network μ θ (x t ,t) is parameterized as

[0116]

[0117] Where ε θ is a diffusion model with parameter θ;

[0118] The purpose of this step is to construct the neural network structure of the conditional diffusion model for model training;

[0119] S5. Construct a neural network for a conditional diffusion model; specifically comprising the following steps:

[0120] The constructed neural network includes an embedding module, a residual module, and an output module;

[0121] The embedding module is used to project the input features into a high-dimensional space; the embedding module includes a linear layer; the input is set to x t , the condition is x c , then the processing of the embedded module is expressed as

[0122] x emb =FC(x t)+FC(x c )

[0123] Where x emb is the output of the embedding module; FC() is the fully connected layer processing function;

[0124] The residual module is used to connect the input with the module output to alleviate the gradient vanishing problem. The residual module includes four residual submodules, each of which includes a multi-head attention layer and a fully connected layer. The processing process of the i-th residual submodule is expressed as

[0125] x res,i =FC(MHA(x res,i-1 ))+x res,i-1

[0126] Where x res,i is the output of the i-th residual submodule; MHA() is the processing function of the multi-head attention layer; if i is 1, then x res,i-1 is the output of the embedded module;

[0127] The output module is used to connect the outputs of all residual modules and map them to obtain the final output result; the output module includes a fully connected layer; the processing process of the output module is represented by x out =FC(x res,1 ,x res,2 ,x res,3 ,x res,4 ), x out is the output of the output module;

[0128] The purpose of this step is to use the large model to identify the conditional parameters of the natural language text and parse it into formatted data as conditional information to guide the training of the diffusion model;

[0129] S6. Using the large model to guide, generate conditional information, and train the established conditional diffusion model to obtain a trained conditional diffusion model; specifically comprising the following steps:

[0130] For the diffusion generation task, prompt words are designed: the prompt word format is set as <S, P, D, C>, where S represents the season and the value range is {spring, summer, autumn, winter}, P represents the energy output intensity and the value range is {high, medium, low}, D represents daytime or night, and C represents wind power or solar energy;

[0131] The open source Qwen2-7B model is used as the large model;

[0132] The first round of text dialogue with the big model: The big model uses semantic understanding to confirm whether the input satisfies the format of <S, P, D, C>. If it does, it constructs conditional information to guide the diffusion model training. If not, it enters the second round of dialogue.

[0133] Second round of text dialogue with the big model: Complete the text according to the missing conditions fed back by the big model, and have another text dialogue with the big model;

[0134] The bootstrap process is as follows: construct a vocabulary table to map text vocabulary; the large model formats and extracts the input text, constructs <S, P, D, C> and maps it to the vocabulary table as conditional information x c ; Diffusion of the original sample x0 to obtain x T , and x0, x c and x T The data are inputted together into the conditional diffusion model obtained by modeling in step S4 for training;

[0135] Finally, the trained conditional diffusion model is obtained;

[0136] The purpose of this step is to achieve the generation of deterministic renewable energy operation scenarios using the trained diffusion model;

[0137] S7. Using the trained conditional diffusion model obtained in step S6, generating a renewable energy system operation scenario; specifically comprising the following steps:

[0138] Randomly sample a Gaussian noise x from any standard Gaussian distribution T,new , the x T,new Corresponding to the noised sample x obtained after T-step diffusion during the training process T It should be noted that the generation process will not use x T Instead, we use sampling to get a new x T,new ;

[0139] Then, the user actively communicates with the big model and inputs the specific scenario to be generated, such as "the output value of the wind turbine on a windy summer night", and the big model extracts the structural condition information x c , and with x T,new The trained diffusion model is sent to S6 to generate the operation scenario of the renewable energy system.

[0140] The method of the present invention is further described below in conjunction with an embodiment:

[0141] 1. Introduce the operation scenarios of renewable energy systems, including the power values ​​of wind power generation and photovoltaic power generation in the actual operation period. Each operation scenario is divided into 24 hours, and the operation scenario dataset of renewable energy systems is constructed, where each sample is represented by x0.

[0142] 2. Establish a diffusion process. For each sample x0, implement x0 to the noise sample x through step S2. TThe transformation of x T It obeys isotropic Gaussian distribution. Here T = 1000, which can reduce the variance introduced by the diffusion process.

[0143] 3. According to step S3, the denoising process is established and the objective function used for model training is set.

[0144] 4. According to step S4, mathematical modeling is performed on the conditional diffusion model used to guide the model to generate renewable energy operation scenarios. c It is expressed in natural language as “season, day and night, intensity, energy type” and other information. t The value is σ t =0.

[0145] 5. According to step S5, a neural network structure of the diffusion model is constructed;

[0146] 6. Construct some natural language description texts and communicate with the big model. The texts represent specific scenarios that need to be generated, such as "the output value of wind turbines in a windy summer night" and "the output value of photovoltaic power generation in a sunny winter day". The big model extracts <S, P, D, C> to construct condition information x c , and the noised renewable energy scenario sample x T The constructed diffusion model neural network is sent together to train the diffusion model with the objective function set in step S3.

[0147] 7. After the diffusion model is trained, the model learns the data distribution of real renewable energy operation scenarios. Users communicate with the big model to propose their own generation requirements, such as "Please generate the output value of the wind turbine for one day in spring with a breeze." The big model identifies the user's needs and extracts the construction condition information x c , and then call the diffusion model to randomly sample a Gaussian noise x from any standard Gaussian distribution T,new , according to the denoising process established in S3, the corresponding scene is generated. The scene diagram is as follows Figure 2 As shown in the figure, Real represents the real scene, Reconstructed represents the generated scene, the vertical axis Power represents the power value, and the horizontal axis Time represents the time.

[0148] pass Figure 2 It can be seen that the generated scenes are very similar to the real scenes, indicating that the diffusion model used is able to generate reliable scene samples.

[0149] like Figure 3The functional module schematic diagram of the system of the present invention is shown as follows: the system disclosed in the present invention for realizing the method for generating the operating scenario of the renewable energy system comprises a data acquisition module, a diffusion modeling module, a Gaussian denoising module, a model modeling module, a network construction module, a model guidance module and a scenario generation module; the data acquisition module, the diffusion modeling module, the Gaussian denoising module, the model modeling module, the network construction module, the model guidance module and the scenario generation module are connected in series in sequence; the data acquisition module is used to acquire historical data information of the target renewable energy system, and upload the data information to the diffusion modeling module; the diffusion modeling module is used to establish a diffusion process model for the operating scenario of the target renewable energy system according to the received data information and the acquired data information, and upload the data information to the Gaussian denoising module; the Gaussian denoising module is used to, according to the received data information, According to the acquired data information and the established diffusion process model, denoising is performed based on Gaussian noise, and the data information is uploaded to the model building module; the model building module is used to model the conditional diffusion model for the obtained data according to the received data information, and upload the data information to the network construction module; the network construction module is used to construct a neural network for the conditional diffusion model according to the received data information, and upload the data information to the model guidance module; the model guidance module is used to generate conditional information based on the received data information using a large model for guidance, and train the established conditional diffusion model to obtain the trained conditional diffusion model, and upload the data information to the scenario generation module; the scenario generation module is used to generate a renewable energy system operation scenario based on the received data information using the trained conditional diffusion model.

Claims

1. A method for generating a renewable energy system operation scenario, comprising the following steps: S1. Obtain historical data information of the target renewable energy system; S2. Establishing a diffusion process model for the operation scenario of the target renewable energy system based on the data information obtained in step S1; S3. Perform denoising based on Gaussian noise according to the data information obtained in step S1 and the diffusion process model established in step S2; S4. Modeling the conditional diffusion model for the data obtained in step S3; S5. Construct a neural network for the conditional diffusion model; S6. Using the large model to guide, generate conditional information, and train the established conditional diffusion model to obtain a trained conditional diffusion model; S7. Using the trained conditional diffusion model obtained in step S6, generate a renewable energy system operation scenario.

2. The renewable energy system operation scenario generation method according to claim 1, characterized in that Step S2, according to the data information obtained in step S1, establishes a diffusion process model for the operation scenario of the target renewable energy system, which specifically includes the following steps: Assume that the operating scenario of the renewable energy system is x; add Gaussian noise to the operating scenario x through the diffusion process, expressed as Where q(x 1:T |x0) represents the diffusion process; x 1:T is a sample after 1…T steps of noise diffusion; x0 is the original sample sampled from the real data distribution; T is the total number of diffusion steps; q(x t ∣x t-1 ) represents the tth diffusion step; x t is the sample after t steps of diffusion; For the tth diffusion step q(x t ∣x t-1 ), which accepts x t-1 And return x by adding some noise t , expressed as Where N() is the standard Gaussian noise function; β t is the noise planning table, and β t ∈(0,1); I is the unit matrix; After T steps of diffusion, x0 will be transformed into a sample x after T steps of diffusion. T , and obeys an isotropic Gaussian distribution.

3. The renewable energy system operation scenario generation method according to claim 2, characterized in that The step S3 described in which the data information obtained in step S1 and the diffusion process model established in step S2 are subjected to denoising based on Gaussian noise, specifically includes the following steps: The denoising process is used to T Remove the noise to recover x0 so that x0 can create new data samples from the Gaussian noise distribution; The denoising process is expressed as p θ (x t-1 ∣x t )=N(x t-1 ;μ θ (x t ,t),β t I) Where p θ (x t-1 ∣x t ) represents the denoising process realized by the neural network; μ θ (x t ,t) represents the neural network processing process with parameter θ; According to the denoising steps, the denoising process is expressed as the following formula: Where p θ (x 0:T ) is the denoising process; p(x T ) is Gaussian noise, and p(x T )=N(0,I); Assuming the denoising process is the inverse process of the diffusion process, the objective function of the diffusion model is expressed as min(KL(p θ (xt-1|xt),q(xt-1|xt,x0))) Where KL() is the Kullback-Leibler divergence calculation function; q(x t-1 ∣x t ,x0) is the posterior conditional probability of the diffusion process.

4. The renewable energy system operation scenario generation method according to claim 3, characterized in that The step S4 of modeling the conditional diffusion model for the data obtained in step S3 specifically includes the following steps: In order to achieve deterministic generation guided by conditions, the diffusion model is extended to a conditional diffusion model to learn conditional information. The formula p θ (x t-1 ∣x t )=N(x t-1 ;μ θ (x t ,t),β t I) Expand to p θ (x t-1 ∣x t ,x c )=N(x t-1 ;μ θ (x t ,t|x c ),β t I) Where p θ (x t-1 ∣x t ,x c ) is to convert x t becomes x t-1 The denoising process of x c is conditional information; Will Expand to Where p θ (x t-1 ∣x t ,x c ) is to convert x t becomes x t-1 The denoising process; The undetermined coefficient method is used to To solve: set up Where p(x t-1 ∣x t ,x0) is the denoising process, κ t is the first coefficient, λ t is the second coefficient, σ t is the third coefficient; Setting σ t is a variable, and where α t is the first intermediate variable, β t is the noise planning table, is the second intermediate variable, is the third intermediate variable; So we get the following formula Where γ is the fourth intermediate variable, and If set And η∈[0,1], η is the weight coefficient, then it can be obtained that when η=0, the variance of the Gaussian distribution is 0, and the generation is deterministic at this time; At the same time, the neural network μ θ (x t ,t) is parameterized as Where ε θ is a diffusion model with parameter θ.

5. The method for generating a renewable energy system operation scenario according to claim 4, characterized in that The step S5 of constructing a neural network for a conditional diffusion model specifically includes the following steps: The constructed neural network includes an embedding module, a residual module, and an output module; The embedding module is used to project the input features into a high-dimensional space; the embedding module includes a linear layer; the input is set to x t , the condition is x c , then the processing of the embedded module is expressed as x emb =FC(x t )+FC(x c ) Where x emb is the output of the embedding module; FC() is the fully connected layer processing function; The residual module is used to connect the input with the module output to alleviate the gradient vanishing problem. The residual module includes four residual submodules, each of which includes a multi-head attention layer and a fully connected layer. The processing process of the i-th residual submodule is expressed as x res,i =FC(MHA(x res,i-1 ))+x res,i-1 Where x res,i is the output of the i-th residual submodule; MHA() is the processing function of the multi-head attention layer; if i is 1, then x res,i-1 is the output of the embedded module; The output module is used to connect the outputs of all residual modules and map them to obtain the final output result; the output module includes a fully connected layer; the processing process of the output module is represented by x out =FC(x res,1 ,x res,2 ,x res,3 ,x res,4 ), x out Output module output.

6. The renewable energy system operation scenario generation method according to claim 5, characterized in that The step S6 of using a large model to guide, generating conditional information, and training the established conditional diffusion model to obtain a trained conditional diffusion model specifically includes the following steps: For the diffusion generation task, prompt words are designed: the prompt word format is set as <S, P, D, C>, where S represents the season and the value range is {spring, summer, autumn, winter}, P represents the energy output intensity and the value range is {high, medium, low}, D represents daytime or night, and C represents wind power or solar energy; The Qwen2-7B model was used as the large model; The first round of text dialogue with the big model: The big model uses semantic understanding to confirm whether the input satisfies the format of <S, P, D, C>. If it does, it constructs conditional information to guide the diffusion model training. If not, it enters the second round of dialogue. Second round of text dialogue with the big model: Complete the text according to the missing conditions fed back by the big model, and have another text dialogue with the big model; The bootstrap process is as follows: construct a vocabulary table to map text vocabulary; the large model formats and extracts the input text, constructs <S, P, D, C> and maps it to the vocabulary table as conditional information x c ; Diffusion of the original sample x0 to obtain x T , and x0, x c and x T The data are inputted together into the conditional diffusion model obtained by modeling in step S4 for training; Finally, the trained conditional diffusion model is obtained.

7. The method for generating a renewable energy system operation scenario according to claim 6, characterized in that The step S7 uses the trained conditional diffusion model obtained in step S6 to generate a renewable energy system operation scenario, which specifically includes the following steps: Randomly sample a Gaussian noise x from any standard Gaussian distribution T,new , input into the trained conditional diffusion model obtained in step S6, and obtain x generated by the model c , and according to x c Complete the generation of renewable energy system operation scenarios.

8. A system for implementing the method for generating a renewable energy system operation scenario according to any one of claims 1 to 7, characterized in that It includes a data acquisition module, a diffusion modeling module, a Gaussian denoising module, a model modeling module, a network construction module, a model guidance module and a scenario generation module; the data acquisition module, the diffusion modeling module, the Gaussian denoising module, the model modeling module, the network construction module, the model guidance module and the scenario generation module are connected in series in sequence; the data acquisition module is used to obtain historical data information of the target renewable energy system, and upload the data information to the diffusion modeling module; The diffusion modeling module is used to establish a diffusion process model for the operation scenario of the target renewable energy system according to the received data information and the acquired data information, and upload the data information to the Gaussian denoising module; the Gaussian denoising module is used to perform denoising based on Gaussian noise according to the received data information and the established diffusion process model, and upload the data information to the model building module; The model building module is used to model the conditional diffusion model for the obtained data according to the received data information, and upload the data information to the network building module; the network building module is used to build a neural network for the conditional diffusion model according to the received data information, and upload the data information to the model guidance module; The model guidance module is used to generate conditional information based on the received data information and adopt the large model guidance, and train the established conditional diffusion model to obtain the trained conditional diffusion model, and upload the data information to the scene generation module; The scenario generation module is used to generate the renewable energy system operation scenario based on the received data information and the trained conditional diffusion model.

Images