New energy capacity confidence evaluation method based on conditional denoising diffusion model

Abstract

The application discloses a new energy capacity confidence evaluation method based on a conditional denoising diffusion model and belongs to the field of power system reliability analysis and power source planning.The technical scheme that the application adopts is as follows: wind power data and photovoltaic power data are taken as input samples, a noise prediction network is first trained and a conditional denoising diffusion model is constructed based on the noise prediction network, then, the samples are preprocessed and modeled to retain effective information therein, then, according to a label vector of a given scene and pure noise data, a conditional denoising diffusion model that is trained is inferred to generate a scene set of new energy power curves with specific characteristics, and finally, based on power grid system power installed capacity, failure rate parameters, historical load curves and the scene set of new energy power curves, the capacity confidence of the new energy after being connected to the power grid system is evaluated through a generation side COPT convolution method.The application can accurately quantify the power supply guarantee capability of the new energy system, thereby providing guarantee for the safe operation of the power system.

Description

Technical Field

[0001] This invention belongs to the field of power system reliability analysis and power planning, and particularly relates to a method for assessing the confidence level of new energy capacity based on a conditional denoising diffusion model. Background Technology

[0002] In new power systems, new energy sources (such as wind and solar power) are rapidly developing and gradually replacing traditional fossil fuels due to their renewable and environmentally friendly characteristics. However, new energy sources are intermittent and uncertain, and their power generation is affected by various factors such as weather and seasons, leading to severe challenges to the reliability of the power grid after new energy sources are connected to the grid. To ensure the safe and reliable operation of the power system, it is crucial to scientifically and rationally assess the equivalent conventional unit capacity of new energy sources (i.e., capacity confidence level) from a reliability perspective.

[0003] Traditional methods for assessing the confidence level of renewable energy capacity mainly rely on historical data and statistical analysis, such as time series analysis and probabilistic statistical models. While these methods can reflect the fluctuation characteristics of renewable energy to some extent, they often lack accuracy and robustness when dealing with complex and variable renewable energy data. In particular, the performance of traditional methods deteriorates significantly when faced with practical problems such as missing data and noise interference, making it difficult to provide reliable assessment results.

[0004] In recent years, with the rapid development of artificial intelligence technology, especially the remarkable achievements of deep learning in fields such as image processing and natural language processing, researchers have begun to explore its application in the new energy sector. Deep generative models such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) have demonstrated excellent performance in handling complex data distributions. However, these methods typically require large amounts of labeled data and have high computational costs when processing high-dimensional data, making them difficult to directly apply to practical power systems.

[0005] Denoising Diffusion Probabilistic Models (DDPM), as an emerging deep generative model, can generate high-quality data samples through a progressive denoising process. Unlike traditional generative models, DDPM exhibits better generation performance and higher robustness, particularly excelling in handling noisy, high-dimensional data. Therefore, achieving accurate modeling and capacity confidence assessment of new energy power generation scenarios using DDPM, and providing a theoretical basis for reliability analysis and power planning of new power systems, is an urgent problem to be solved. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing renewable energy capacity confidence assessment techniques, which rely on the completeness of historical data, and to provide a renewable energy capacity confidence assessment method based on a conditional denoising diffusion model. This invention uses renewable energy output power time-series scenarios as input. First, it preprocesses and models renewable energy power generation data by constructing and training a conditional denoising diffusion model, preserving the effective information of the data. Then, based on given renewable energy scenario label vectors, it generates a reliable set of renewable energy random scenarios through inference using the trained conditional denoising diffusion model. Finally, based on the power system's power structure, failure rate parameters, and historical load curves, it achieves renewable energy capacity confidence assessment, providing a guarantee for the safe operation of the power system.

[0007] To achieve the above-mentioned objectives, the present invention specifically adopts the following technical solution:

[0008] A method for assessing the confidence level of new energy capacity based on a conditional denoising diffusion model includes the following steps:

[0009] S1: Obtain historical wind power data and historical photovoltaic power data within a preset time period. The historical wind power data includes daily wind power data, and the historical photovoltaic power data includes daily photovoltaic power data. Use both wind power data and photovoltaic power data as samples, calculate the mean and peak-to-valley difference for each sample, convert the mean and peak-to-valley difference for each sample into standardized metrics, and use the metrics of the mean and peak-to-valley difference as the label vector for each sample; construct a training set from each sample and its label vector.

[0010] S2: Train a noise prediction network on the training set, and use the trained noise prediction network to implement the reverse denoising process to construct a conditional denoising diffusion model for generating new energy scenarios; wherein, the noise prediction network is constructed based on an LSTM layer, a first embedding layer, a second embedding layer, an Attention layer and a one-dimensional convolutional layer, the first embedding layer is used to process the label vector, and the second embedding layer is used to process the diffusion time step.

[0011] S3: Sample pure noise data from a multivariate Gaussian distribution, input the label vector of a given scene and the pure noise data into the conditional denoising diffusion model for inference, and generate a scene set of new energy power curves through stepwise denoising.

[0012] S4: Based on the power generation capacity, failure rate parameters, historical load curves, and renewable energy power curves of the power grid system, the capacity confidence of renewable energy after its access to the power grid system is evaluated by the COPT convolution method on the generation side, thus realizing the capacity confidence assessment of renewable energy.

[0013] Based on the above scheme, each step can be implemented in the following preferred manner.

[0014] Preferably, in step S2, the input to the noise prediction network is a noisy sample, the label vector of the sample, and the diffusion time step corresponding to the sample, and its output is the predicted value of the noise vector contained in the noisy sample.

[0015] Preferably, in step S2, when training the noise prediction network, the following parameters are preset: sample batch size, length of each sample, dimension of sample label vector, training set samples as initial values, and maximum diffusion time step.

[0016] At any diffusion time step n, the noisy training set samples x n It can be represented as:

[0017]

[0018] α i =1-β i

[0019] Where: ε n The Gaussian noise vector added to the training set sample x0, which serves as the initial value, in the nth diffusion step; These are known parameters that do not require training; x n-1 The training set samples contain noise at diffusion time step n-1; α i β represents the intermediate variable in the i-th diffusion step. i It is the diffusion rate at the i-th diffusion step.

[0020] Preferably, in step S2, the loss function of the noise prediction network is the mean square error between the Gaussian noise vector added to all training set samples and the noise vector output by the noise prediction network at all diffusion time steps.

[0021] Preferably, the specific processing flow of the noise prediction network in step S2 is as follows:

[0022] S21. Input the noisy sample into the LSTM layer to obtain the noisy sample expanded to the hidden state;

[0023] S22. Input the label vector into the first embedding layer to obtain the first output embedding; input the diffusion time step into the second embedding layer to obtain the second output embedding; add the first output embedding and the second output embedding to obtain the total output of the embedding layer:

[0024] S23. The noisy samples extended to the hidden state and the total output of the embedding layer are added together as the input of the Attention layer. After processing by the multi-head self-attention mechanism, a tensor containing information on the degree of dependency between data at different locations is obtained.

[0025] S24. Input the tensor containing the dependency information between data at different locations into the one-dimensional convolutional layer, and output the predicted value of the noise vector contained in the noisy sample.

[0026] Preferably, the specific processing flow of the first embedding layer is as follows:

[0027] z c =tanh(Wc+b)

[0028] In the formula: W and b represent the weights and bias matrices to be trained in the linear transformation layer, respectively; c represents the label vector; tanh represents the tanh activation function; z c This indicates the first output embedding.

[0029] Preferably, the second embedding layer is implemented by positional encoding, and the second output embedding z n Specifically, it is expressed as follows:

[0030]

[0031] In the formula: z n (:,:,2i), z n (:,:,2i+1) represent the values ​​of the second output embedded at even indices (:,:,2i) and odd indices (:,:,2i+1), respectively; H is the hidden state dimension consistent with the LSTM layer; n represents the diffusion time step.

[0032] Preferably, the one-dimensional convolutional layer has the following input channels: the hidden state dimension H, the number of output channels: 1, and the kernel size: 1×1.

[0033] As a preferred option, the process for generating the scenario set of new energy power curves in step S3 is as follows:

[0034] First, a vector following a multivariate Gaussian distribution is generated as pure noise data. Then, the label vector of the given scene and the pure noise data are input into the conditional denoising diffusion model for inference. Starting from the pure noise data, the model is recursively pushed to the first denoising step to obtain the result of the first denoising step. Denoising is continuously performed until N denoising steps are completed, and the result of the Nth denoising step is obtained. The result of the Nth denoising step is used as the generated new energy power curve, and the new energy power curve constitutes the scene set of the new energy power curve.

[0035] As a preferred option, the specific process of step S4 is as follows:

[0036] S41. First, obtain the installed power capacity and failure rate parameters of the power grid system from the power generation side of the power grid system, and calculate the outage capacity probability table of all units in the power grid system by convolution method;

[0037] S42. The difference between the historical load curve and the new energy power curve is used as the net load curve. For each time period in the net load curve, the outage capacity probability table is searched to obtain the outage probability and the outage amount. The expected unsupplied power supply of the power grid system under the net load curve is calculated from the outage probability and the outage amount, and the expected unsupplied power supply is used as the reliability quantification index of the power grid system.

[0038] S43. Each time a fully reliable unit with an installed capacity is added to the power grid system to form a new power grid system, the outage capacity probability table is looked up for the load of each time period in the historical load curve to obtain the new outage probability and outage amount. The new expected unsupplied power in the new power grid system under the historical load curve is calculated from the new outage probability and outage amount.

[0039] S44. Installed capacity P m As the horizontal axis, the new expected unsupplied quantity E m As the ordinate, generate a scatter plot (P) m E m A polynomial curve is formed by fitting all scattered points with a polynomial function. The x-coordinate P of the intersection point of the polynomial curve and y = E0 is calculated. CC And calculate the capacity confidence score:

[0040] CC = P CC / P VRE

[0041] In the formula: P VRE It refers to the installed capacity of new energy sources.

[0042] Compared with the prior art, the present invention has the following advantages:

[0043] This invention can capture the complex characteristics of time-series data of renewable energy output power and can generate renewable energy power scenarios based on a given label vector, which can be used for the analysis, evaluation and improvement of renewable energy capacity confidence. Attached Figure Description

[0044] Figure 1 This is a flowchart of the steps of the method of the present invention;

[0045] Figure 2 This is a structural diagram of the noise prediction network in the conditional denoising diffusion model of the present invention;

[0046] Figure 3 This is a flowchart of the new energy capacity confidence assessment method based on convolution method of the present invention;

[0047] Figure 4 This diagram illustrates the variation of the loss function during training of the conditional denoising diffusion model of the present invention in different scenarios; wherein, Figure 4 (a) is a wind power scenario; Figure 4 (b) is a photovoltaic scenario;

[0048] Figure 5 This is a schematic diagram of the directional generation result of a wind power scenario according to an embodiment of the present invention; wherein, Figure 5 (a) is a scenario with low wind power output; Figure 5 (b) is a high-output wind power scenario; Figure 5 (c) is the peak-valley difference scenario for wind power;

[0049] Figure 6 This is a schematic diagram of the directional generation result of a photovoltaic scene according to an embodiment of the present invention; wherein, Figure 6 (a) is a scenario with low photovoltaic output; Figure 6 (b) is a high-output photovoltaic scenario. Detailed Implementation

[0050] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. Technical features in the various embodiments of the present invention can be combined accordingly without mutual conflict.

[0051] In the description of this invention, it should be understood that the terms "first" and "second" are used only for descriptive purposes and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" and "second" may explicitly or implicitly include at least one of those features.

[0052] like Figure 1 As shown, in a preferred embodiment of the present invention, the above-mentioned new energy capacity confidence assessment method based on a conditional denoising diffusion model includes the following steps S1 to S4. The specific implementation process of each step will be described in detail below.

[0053] S1: Obtain historical wind power data and historical photovoltaic power data within a preset time period. The historical wind power data includes daily wind power data, and the historical photovoltaic power data includes daily photovoltaic power data. Use both wind power data and photovoltaic power data as samples, calculate the mean and peak-to-valley difference for each sample, convert the mean and peak-to-valley difference for each sample into standardized metrics, and use the metrics of the mean and peak-to-valley difference as the label vector for each sample. Construct a training set from each sample and its label vector.

[0054] S2: Train a noise prediction network on the training set, and use the trained noise prediction network to implement the reverse denoising process to construct a conditional denoising diffusion model for generating new energy scenarios; wherein, the noise prediction network is constructed based on an LSTM layer, a first embedding layer (first embedding layer), a second embedding layer (second embedding layer), an attention layer and a one-dimensional convolutional layer, the first embedding layer is used to process the label vector, and the second embedding layer is used to process the diffusion time step.

[0055] It should be noted that in step S2, the input to the noise prediction network is the noisy sample x, the label vector c of the sample, and the diffusion time step n corresponding to the sample. The output of the noise prediction network is the predicted value of the noise vector contained in the noisy sample x.

[0056] In this embodiment, the label vector c of each sample has been obtained by step S1; the noisy sample x at each diffusion time step n can be obtained by adding noise to the sample in S1, which can be derived by a simple mathematical formula in the forward diffusion process. The forward diffusion process is common knowledge in the field and will not be described in detail here.

[0057] It should be noted that in step S2, when training the noise prediction network, the following parameters can be preset: sample batch size B, length of each sample S, dimension of sample label vector C, training set sample x0 as initial value, and maximum diffusion time step N.

[0058] At any diffusion time step n, the noisy training set samples x n It can be represented as:

[0059]

[0060] α i =1-β i

[0061] In the formula, The Gaussian noise vector added to the training set sample x0, which serves as the initial value, in the nth diffusion step can be derived from a multivariate Gaussian distribution. The sample obtained shows that the mean of the multivariate Gaussian distribution is 0 and the variance is the identity matrix I. These are known parameters that do not require training; x n-1 The training set samples contain noise at diffusion time step n-1; α i β represents the intermediate variable in the i-th diffusion step. i It is the diffusion rate of the i-th diffusion step, which is determined by the noise scheduling scheme.

[0062] It should be noted that in step S2, the loss function of the noise prediction network is the Gaussian noise vector ε added to all training set samples at all diffusion time steps. n The noise vector output by the noise prediction network (i.e., the predicted noise vector value contained in the noisy sample). The mean square error between them, i.e.:

[0063]

[0064] It should be noted that the specific processing flow of the noise prediction network in step S2 is as follows:

[0065] S21. Input the noisy sample x into the LSTM layer to obtain the noisy sample y expanded to the hidden state.

[0066] S22. Input the label vector c into the first embedding layer to obtain the first output embedding z. c The diffusion time step n is input into the second embedding layer to obtain the second output embedding z. n Embed the first output into z c Second output embedding z n Adding them together, we get the total output of the embedding layer:

[0067] z = z c +z n

[0068] S23. The noisy sample y extended to the hidden state and the total output z of the embedding layer are added together (i.e., q = y + z) as the input of the Attention layer. After processing by the multi-head self-attention mechanism, a tensor a containing the degree of dependency between data at different locations is obtained, which has the same dimension as q.

[0069] S24. Input the tensor 'a' containing information about the dependency between data at different locations into the one-dimensional convolutional layer, and output the predicted noise vector value contained in the noisy sample.

[0070] It should be noted that in step S2, as Figure 2 As shown, the noise prediction network consists of an LSTM layer, an Embedding layer, an Attention layer, and a one-dimensional convolutional layer. The specific processing flow of each part of the noise prediction network is described below.

[0071] (1) LSTM layer

[0072] The input to the LSTM layer is a noisy sample. The output is a noisy sample extended to the hidden state. Here, H is the hidden state dimension.

[0073] In this embodiment, we consider the sample x at the (s-1)th position. s-1 The output result y s-1 and long-term memory vector l s-1 Given the given conditions, the output result y at the s-th position s The calculation method is as follows:

[0074] y s =LSTM(x s ,y s-1 ,l s-1 ), s=2,...,S

[0075] The calculation process of LSTM() is as follows:

[0076] i s =σ(W i,s x s +b i,s +U i,s y s-1 +d i,s )

[0077] f s =σ(W f,s x s +b f,s +U f,s y s-1 +d f,s )

[0078] o s =σ(W o,s x s +b o,s +U o,s y s-1 +d o,s )

[0079] m s =tanh(W m,s x s +b m,s +U m,s y s-1 +d m,s )

[0080] l s =l s-1 ⊙f s +m s ⊙i s

[0081] y s =tanh(l s )⊙o s

[0082] In the formula: Wi,s U i,s W f,s U f,s W o,s U o,s W m,s U m,s b are the weight matrices to be trained at the s-th position; i,s d i,s b f,s d f,s b o,s d o,s b m,s d m,s All are bias matrices to be trained at the s-th position; σ() is the sigmoid activation function; tanh() is the tanh activation function; i s f s o s m s l s y s These represent the input gate vector, forget gate vector, output gate vector, input memory, long-term memory, and output at the s-th position, respectively; ⊙ represents element-wise multiplication.

[0083] (2) Embedding layer

[0084] The aforementioned embedding layer specifically includes an embedding layer for the label vector c (i.e., the first embedding layer) and an embedding layer for the diffusion time step n (i.e., the second embedding layer).

[0085] The embedding layer of the label vector c consists of a linear transformation layer and an activation function, and its output z c It can be represented as:

[0086] z c =tanh(Wc+b)

[0087] In the formula, W and b represent the weights and bias matrices to be trained in the linear transformation layer, respectively; c represents the label vector; and tanh represents the tanh activation function.

[0088] The embedding layer at diffusion time step n is implemented using positional encoding, and its output z n It can be represented as:

[0089]

[0090] In the formula, z n (:,:,2i), z n (:,:,2i+1) represent the values ​​of the second output embedded at even indices (:,:,2i) and odd indices (:,:,2i+1), respectively; H is the hidden state dimension consistent with the LSTM layer.

[0091] In this embodiment, tensor z n The first dimension is the sample batch B, and the second dimension is the length S of each sample. In the second dimension, the diffusion time step n is the same, while in the first dimension, the diffusion time step n is different and is randomly generated.

[0092] (3) Attention layer

[0093] The Attention layer is built based on a multi-head self-attention mechanism, and its computation process is as follows:

[0094] d k =H / head

[0095]

[0096] a = concat(h1,...,h) head W O

[0097] In the formula: head represents the number of segments in the multi-head self-attention mechanism; H is the hidden state dimension consistent with the LSTM layer; d k h represents the hidden state dimension after partitioning. j The output of the j-th attention head is represented by ; softmax represents the softmax activation function; the superscript T represents the transpose of the matrix; h1,...,h head W represents the output of the 1st, ..., head attention head; Q W K W V and W O Both are weight matrices to be trained, W Q W K and W V Its function is to divide q along the H dimension, W O Its function is to set h1,...,h head Concatenation is performed along the H dimension; concat indicates the concatenation of matrices or tensors.

[0098] (4) One-dimensional convolutional layer (Conv 1D)

[0099] The input to the one-dimensional convolutional layer is a tensor 'a', output by the Attention layer, containing information about the dependencies between data at different locations. The output is the predicted noise vector of the noisy sample with the same dimension as tensor 'a'. And This is the final output of the noise prediction network.

[0100] In this embodiment, the number of input channels of the one-dimensional convolutional layer is the hidden state dimension H, the number of output channels is 1, and the convolutional kernel size is 1×1.

[0101] In summary, the noise prediction network can be represented as: That is, the output of the noise prediction network It is a function of the noisy sample x, the label vector c, and the diffusion time step n.

[0102] Additionally, it should be noted that after the noise prediction network is built, once the initial learning rate, batch size, maximum number of diffusion steps, and maximum number of training generations are set, the noise prediction network can be trained on the training set, and the .ckpt file with the minimum loss function can be saved.

[0103] S3: Sample pure noise data from a multivariate Gaussian distribution, input the label vector of a given scene and the pure noise data into the conditional denoising diffusion model for inference, and generate a scene set of new energy power curves through stepwise denoising.

[0104] It should be noted that in step S3, the process of generating the scenario set of new energy power curves using the trained conditional denoising diffusion model is as follows:

[0105] First, generate vectors that follow a multivariate Gaussian distribution. As pure noise data, the label vector of the given scene and the pure noise data are then input into the conditional denoising diffusion model for inference, from the pure noise data x′ N Starting from the first denoising step, we recursively proceed to the first denoising step to obtain the result x′ of the first denoising step. N-1 The noise is continuously reduced until N noise reduction steps are completed, and the result of the Nth noise reduction step x0′ is obtained. The result of the Nth noise reduction step is used as the generated new energy power curve, and the new energy power curve constitutes the scene set of the new energy power curve.

[0106] Wherein, x′ is the result of the Nn-th denoising step. n The result x′ obtained by recursively applying the algorithm to the (N-n+1)th denoising step n-1 The specific process is as follows:

[0107]

[0108] In the formula: This represents a random vector sampled from a multivariate Gaussian distribution.

[0109] S4: Based on the scenario set of power generation capacity, failure rate parameters, historical load curves and renewable energy power curves of the power grid system, the capacity confidence of renewable energy after being connected to the power grid system is evaluated by the generation-side COPT (Capacity Outage Probability Table) convolution method, thereby realizing the capacity confidence assessment of renewable energy.

[0110] It should be noted that, as Figure 3 As shown, the specific process of step S4 is as follows:

[0111] S41. First, obtain the installed power capacity and failure rate parameters of the power grid system from the power generation side, and calculate the outage capacity probability table of all units in the power grid system using the convolution method.

[0112] In this embodiment, the range and probability distribution of total available power generation capacity can be queried from the outage capacity probability table.

[0113] S42. The difference between the historical load curve and the new energy power curve is taken as the net load curve. For each time period in the net load curve, the outage capacity probability table is looked up to obtain the outage probability and the outage amount. The expected energy not supplied (EENS) E0 of the power grid system under the net load curve is calculated from the outage probability and the outage amount. The expected energy not supplied (EENS) E0 is used as the reliability quantification index of the power grid system.

[0114] S43. Each time an installed capacity P is added to the power grid system m The fully reliable generating units form a new power grid system. For each time period in the historical load curve, the outage capacity probability table is consulted to obtain the new outage probability and outage amount. Based on the new outage probability and outage amount, the new expected unsupplied power volume E of the new power grid system under the historical load curve is calculated. m Where m = 1, ..., M.

[0115] S44. Installed capacity P m As the horizontal axis, the new expected unsupplied quantity E m As the ordinate, generate a scatter plot (P) m E m A polynomial curve is formed by fitting all scattered points with a polynomial function. The x-coordinate P of the intersection point of the polynomial curve and y = E0 is calculated. CC And calculate the capacity confidence score:

[0116] CC = P CC / P VRE

[0117] In the formula: P VRE It refers to the installed capacity of new energy sources.

[0118] To better demonstrate the specific implementation and technical effects of the present invention, the new energy capacity confidence assessment method based on the conditional denoising diffusion model shown in steps S1 to S4 of the above preferred implementation is applied to a specific example.

[0119] Example

[0120] The implementation process of the new energy capacity confidence assessment method based on the conditional denoising diffusion model used in this embodiment is as described above. This embodiment uses the Python programming language, the PyTorch machine learning framework, and the CUDA GPU computing architecture to write the above steps S1 to S4 of the invention, which will not be repeated here. The following mainly demonstrates its specific technical effects using case data.

[0121] Operating environment:

[0122] 12th Gen Intel® Core™ i7-12700H 2.30GHz CPU, 16GB RAM

[0123] Microsoft Windows 11 x64

[0124] NVIDIA GeForce RTX 3060Laptop GPU

[0125] Python 3.11.5

[0126] PyTorch 2.2.1

[0127] CUDA 12.1

[0128] Implementation results:

[0129] This embodiment is based on the IEEE RTS-GMLC standard test system, which includes three regional power grids. The power supply structure and capacity parameters of the system are shown in Table 1. The peak load of the system is 8191.8MW, and the historical time series data of wind power and photovoltaic power has a resolution of 1 hour and a duration of 1 year. Some parameters of the conditional denoising diffusion model training are set as follows: initial learning rate of 1×10-4, batch size of 4, hidden layer dimension of 48, number of partitions of multi-head attention network of 4, maximum diffusion steps of 50, initial diffusion rate of 1×10-4, final diffusion rate of 0.5, Adam optimization algorithm is used, a total of 200 epochs are iterated, and the weights of the conditional denoising diffusion model of the epoch with the minimum loss function are used for subsequent inference.

[0130] Table 1. System power supply structure and installed capacity parameters

[0131]

[0132] Figure 4 This demonstrates how the loss function changes during training of the conditional denoising diffusion model of this invention. The loss function type chosen here is mean square error (MSE).

[0133] Figure 5 The results are generated for wind power scenarios: (a) represents a low-output wind power scenario, (b) represents a high-output wind power scenario, and (c) represents a peak-valley difference wind power scenario, each containing 50 generated scenario curves. The scenario label vector is formatted as [low-output magnitude, high-output magnitude, peak-valley difference magnitude], with each magnitude ranging from [0,1]. The closer the value is to 1, the closer the scenario features are to the corresponding description. When generating wind power scenarios, the labels for low-output wind power scenarios are specified as [0.1,0.9,0.5], high-output wind power scenarios as [0.9,0.1,0.5], and peak-valley difference wind power scenarios as [0.5,0.5,0.9].

[0134] Figure 6 The results are the targeted generation results for photovoltaic scenarios. (a) is a low-output photovoltaic scenario, and (b) is a high-output photovoltaic scenario, each containing 50 generated scenario curves. The format of the scenario label vector is the same as that for wind power. The label for the low-output photovoltaic scenario is specified as [0.1, 0.9, 0.5], and the label for the high-output photovoltaic scenario is specified as [0.9, 0.1, 0.5].

[0135] Table 2 shows the capacity confidence results under various new energy scenarios. According to... Figure 5 and Figure 6 The five generated scenarios—low wind power output scenario, high wind power output scenario, wind power peak-valley difference scenario, low photovoltaic output scenario, and high photovoltaic output scenario—are used to evaluate the mean, lower limit, and upper limit of the confidence level of new energy capacity, respectively.

[0136] Table 2. Capacity confidence results under various new energy scenarios

[0137]

[0138]

[0139] First, based on labeled historical wind power and photovoltaic scene samples, conditional denoising diffusion models for each were trained. Their loss functions decreased rapidly in the first few epochs, then slowed down as they approached their optimum, resulting in a gradual and fluctuating decrease in the loss function. Analysis showed that the lowest loss function for the wind power scene's conditional denoising diffusion model occurred in the 164th epoch, while for the photovoltaic scene it occurred in the 157th epoch. The weight files for these two models were saved and can be used for subsequent wind power and photovoltaic scene generation.

[0140] Then, to verify the effectiveness of the conditional denoising diffusion model, three labels—low output, high output, and peak-valley difference—were selected for wind power, and two labels—low output and high output—were selected for photovoltaics (because the daily minimum power of photovoltaics is zero, and peak-valley difference is essentially equivalent to high output). This was used to generate targeted new energy scenarios. Low and high output were determined based on the average daily power of wind / photovoltaics, while peak-valley difference was determined based on the difference between the maximum and minimum daily power of wind / photovoltaics. The metrics for these three dimensions are normalized based on historical scenarios. For example, the label for a low-output wind power scenario is [0.1, 0.9, 0.5], meaning that a metric of 0.1 for the low-output dimension represents a level that is relatively closer to the lowest average wind power level in historical scenarios, and the same applies to other dimensions. Figure 5 As can be seen, all three types of wind power scenarios generally retain the characteristic of higher power output at night compared to daytime, which is consistent with the characteristic of higher wind speed at night and lower wind speed during the day. The overall wind power output in the low-output scenario is relatively low (all below 0.4 pu), while the overall wind power output in the high-output scenario is relatively high (generally above 0.4 pu), and the peak-to-valley difference is not large in either case; however, the peak-to-valley difference scenario exhibits a large fluctuation range (the highest value can reach 1.0 pu, and the lowest value can reach 0.0 pu). These characteristics demonstrate the correctness of the wind power condition denoising and diffusion model in scenario-oriented generation.

[0141] Similarly, the low-output and high-output scenarios of photovoltaics can be derived from... Figure 6 Intuitively, the highest daily photovoltaic power in low-output scenarios is around 0.6 pu, while the highest daily photovoltaic power in high-output scenarios is around 0.8 pu, demonstrating the correctness of the photovoltaic conditional denoising diffusion model in scenario-oriented generation.

[0142] Finally, based on the various scenarios obtained above, the capacity confidence level of new energy sources (i.e., the equivalent capacity of conventional units in terms of reliability) was evaluated, and the results are shown in Table 2. In the low-output wind power scenario, the average capacity confidence level is only 2.55%, with a lower limit as low as 0%, meaning it contributes nothing to reliability. This is because wind power has no output power during peak daytime loads, and although it outputs power at night, there is no risk of load shedding. The average capacity confidence level in the high-output wind power scenario can reach 39.24%. The average capacity confidence level in the peak-valley difference wind power scenario is 22.97%, falling between the two, but the difference between the upper and lower limits is 61.96%, higher than the corresponding values ​​for the low-output and high-output wind power scenarios, reflecting its large peak-valley difference characteristic. For photovoltaics, because its regularity is stronger than that of wind power, the lower limit of the capacity confidence level in both high-output and low-output scenarios is around 12%, while the mean and upper limit of the capacity confidence level in the high-output scenario are much higher than those in the low-output scenario.

[0143] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A method for assessing the confidence level of new energy capacity based on a conditional denoising diffusion model, characterized in that, Includes the following steps: S1: Obtain historical wind power data and historical photovoltaic power data within a preset time period. The historical wind power data includes daily wind power data, and the historical photovoltaic power data includes daily photovoltaic power data. Use both wind power data and photovoltaic power data as samples, calculate the mean and peak-to-valley difference for each sample, convert the mean and peak-to-valley difference for each sample into standardized metrics, and use the metrics of the mean and peak-to-valley difference as the label vector for each sample; construct a training set from each sample and its label vector. S2: Train a noise prediction network on the training set, and use the trained noise prediction network to implement the reverse denoising process to construct a conditional denoising diffusion model for generating new energy scenarios; wherein, the noise prediction network is constructed based on an LSTM layer, a first embedding layer, a second embedding layer, an Attention layer and a one-dimensional convolutional layer, the first embedding layer is used to process the label vector, and the second embedding layer is used to process the diffusion time step. S3: Sample pure noise data from a multivariate Gaussian distribution, input the label vector of a given scene and the pure noise data into the conditional denoising diffusion model for inference, and generate a scene set of new energy power curves through stepwise denoising. S4: Based on the scenario set of power generation capacity, failure rate parameters, historical load curves and new energy power curves of the power grid system, the capacity confidence of new energy after being connected to the power grid system is evaluated by the COPT convolution method on the generation side, so as to realize the capacity confidence assessment of new energy. The specific process of step S4 is as follows: S41. First, obtain the installed power capacity and failure rate parameters of the power grid system from the power generation side of the power grid system, and calculate the outage capacity probability table of all units in the power grid system by convolution method; S42. The difference between the historical load curve and the new energy power curve is used as the net load curve. For each time period in the net load curve, the outage capacity probability table is searched to obtain the outage probability and the outage amount. The expected unsupplied power supply of the power grid system under the net load curve is calculated from the outage probability and the outage amount, and the expected unsupplied power supply is used as the reliability quantification index of the power grid system. S43. Each time a fully reliable unit with an installed capacity is added to the power grid system to form a new power grid system, the outage capacity probability table is looked up for the load of each time period in the historical load curve to obtain the new outage probability and outage amount. The new expected unsupplied power in the new power grid system under the historical load curve is calculated from the new outage probability and outage amount. S44. Installed capacity As the horizontal axis, the new expected unsupplied quantity As the ordinate, generate a scatter plot. A polynomial curve is formed by fitting all scattered points with a polynomial function, and the polynomial curve is then calculated to... x-coordinate of the intersection point And calculate the capacity confidence score: ； In the formula: It refers to the installed capacity of new energy sources.

2. The method for assessing the confidence level of new energy capacity based on a conditional denoising diffusion model as described in claim 1, characterized in that, In step S2, the input to the noise prediction network is the noisy sample, the label vector of the sample, and the diffusion time step corresponding to the sample. Its output is the predicted value of the noise vector contained in the noisy sample.

3. The method for assessing the confidence level of new energy capacity based on a conditional denoising diffusion model as described in claim 1, characterized in that, In step S2, during the training of the noise prediction network, the following parameters are preset: sample batch size, length of each sample, dimension of the sample label vector, training set samples used as initial values, and maximum diffusion time step. At any diffusion time step Below are noisy training set samples. Represented as: ； ； ； In the formula: In the first In each diffusion step, the training set samples used as initial values ​​are... Added Gaussian noise vector; These are known parameters that do not require training; To be in diffusion time step The training set samples containing noise; Indicates the first Intermediate variables in each diffusion step; It is the first The diffusion rate of each diffusion step.

4. The method for assessing the confidence level of new energy capacity based on a conditional denoising diffusion model as described in claim 3, characterized in that, In step S2, the loss function of the noise prediction network is the mean square error between the Gaussian noise vector added to all training set samples and the noise vector output by the noise prediction network at all diffusion time steps.

5. The method for assessing the confidence level of new energy capacity based on a conditional denoising diffusion model as described in claim 2, characterized in that, In step S2, the specific processing flow of the noise prediction network is as follows: S21. Input the noisy sample into the LSTM layer to obtain the noisy sample expanded to the hidden state; S22. Input the label vector into the first embedding layer to obtain the first output embedding; input the diffusion time step into the second embedding layer to obtain the second output embedding; add the first output embedding and the second output embedding to obtain the total output of the embedding layer: S23. The noisy samples extended to the hidden state and the total output of the embedding layer are added together as the input of the Attention layer. After processing by the multi-head self-attention mechanism, a tensor containing information on the degree of dependency between data at different locations is obtained. S24. Input the tensor containing the dependency information between data at different locations into the one-dimensional convolutional layer, and output the predicted value of the noise vector contained in the noisy sample.

6. The method for assessing the confidence level of new energy capacity based on a conditional denoising diffusion model as described in claim 5, characterized in that, The specific processing flow of the first embedding layer is as follows: ； In the formula: and These represent the weights and bias matrices to be trained in the linear transformation layer, respectively. Represents a label vector; This represents the tanh activation function; This indicates the first output embedding.

7. The method for assessing the confidence level of new energy capacity based on a conditional denoising diffusion model as described in claim 5, characterized in that, The second embedding layer is implemented by positional encoding, and the second output embedding... Specifically, it is expressed as follows: ； In the formula: , These respectively indicate that the second output is embedded in an even index. Odd index The possible values ​​of ; It is the same hidden state dimension as the LSTM layer; Indicates the diffusion time step.

8. The method for assessing the confidence level of new energy capacity based on a conditional denoising diffusion model as described in claim 5, characterized in that, The number of input channels of the one-dimensional convolutional layer is equal to the number of hidden state dimensions. The number of output channels is 1, and the kernel size is 1×1.

9. The method for assessing the confidence level of new energy capacity based on a conditional denoising diffusion model as described in claim 1, characterized in that, In step S3, the process for generating the scenario set of new energy power curves is as follows: First, a vector following a multivariate Gaussian distribution is generated as pure noise data. Then, the label vector of the given scene and the pure noise data are input into the conditional denoising diffusion model for inference. Starting from the pure noise data, the model recursively proceeds to the first denoising step, obtaining the result of the first denoising step. This denoising process is repeated until the noise is completely removed. After the denoising step, we obtain the first... The result of the first denoising step will be the first step. The result of each denoising step is used as the generated new energy power curve, and the new energy power curve constitutes the scene set of the new energy power curve.

Images