Energy system scheduling method and device based on multi-time scale interval prediction

Abstract

The invention discloses an energy system scheduling method and device based on multi-time scale interval prediction, and belongs to the technical field of distributed energy system scheduling. According to the method, the static photovoltaic power correction model and the dynamic photovoltaic power correction model of each time scale are constructed through the photovoltaic power prediction data of different time scales, so that the static photovoltaic power prediction interval and the dynamic photovoltaic power correction model are generated respectively, and the fused photovoltaic power prediction interval is obtained. The photovoltaic power prediction interval is fused to accurately define the uncertainty range of the photovoltaic power; and then constructing a day-ahead scheduling model, an intra-day hour scheduling model and an intra-day real-time scheduling model of the energy system on different scheduling scales, so that the scheduling scheme of the energy system not only can consider photovoltaic power change characteristics on different time scales, but also can integrate actual operation conditions of the energy system under different scheduling scales. Therefore, the scheduling accuracy of the energy system is improved.

Description

Technical Field

[0001] This invention belongs to the field of distributed energy system scheduling technology, and particularly relates to an energy system scheduling method and apparatus based on multi-timescale interval prediction. Background Technology

[0002] With the continuous development of photovoltaic power generation technology, the proportion of photovoltaic power generation in the energy system has increased rapidly, and it has become the main source of power in the current energy system. However, due to the significant randomness, intermittency and fluctuation of the power generated by photovoltaic power generation, the output of the current energy system is uneven and unstable. Therefore, the photovoltaic power of the energy system is usually predicted, and the energy system is dispatched based on the prediction results to ensure the balance between the energy system and the photovoltaic power, so that the energy system can operate stably.

[0003] Current energy system scheduling schemes based on photovoltaic (PV) power prediction typically rely on neural network models to predict a single PV power value, which is then used for scheduling. However, due to the inherent randomness, intermittency, and volatility of PV power, relying solely on a single PV power value cannot accurately reflect the actual PV power situation within the energy system, leading to scheduling biases. Furthermore, current PV power predictions and energy system scheduling are mostly based on a fixed, single time scale, which also contributes to inaccuracies. Therefore, there is an urgent need for an energy system scheduling method and apparatus based on multi-time-scale interval predictions to address the shortcomings of existing technologies. Summary of the Invention

[0004] The present invention aims to provide an energy system scheduling method and apparatus based on multi-time-scale interval prediction, so as to solve the technical problem of low accuracy of energy system scheduling in the prior art and improve the scheduling accuracy of energy system.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide an energy system scheduling method based on multi-time-scale interval prediction, comprising:

[0006] Obtain photovoltaic power prediction data for the energy system at several different time scales;

[0007] Based on the photovoltaic power prediction data for each time scale, a static photovoltaic power correction model for each time scale is constructed, and the static photovoltaic power prediction interval for each time scale is obtained based on the static photovoltaic power correction model.

[0008] Based on the photovoltaic power prediction data for each time scale, a dynamic photovoltaic power correction model for each time scale is constructed, and the dynamic photovoltaic power prediction interval for each time scale is obtained based on the dynamic photovoltaic power correction model.

[0009] Based on the static photovoltaic power prediction interval and the dynamic photovoltaic power prediction interval for each time scale, the fused photovoltaic power prediction interval for each time scale is determined.

[0010] Based on the fused photovoltaic power prediction interval for each time scale, a day-ahead scheduling model, an intraday hourly scheduling model, and an intraday real-time scheduling model for the energy system are constructed. The day-ahead scheduling model, the intraday hourly scheduling model, and the intraday real-time scheduling model are solved respectively to determine the scheduling scheme of the energy system. The energy system is then scheduled based on the scheduling scheme.

[0011] It is understood that this invention constructs static and dynamic photovoltaic power correction models for each time scale using photovoltaic power prediction data from several different time scales of the energy system. This generates static and dynamic photovoltaic power prediction intervals, respectively, resulting in a fused photovoltaic power prediction interval. The static photovoltaic power correction model optimizes the error of the photovoltaic power prediction data, while the dynamic photovoltaic power correction model adapts to the volatility, randomness, and intermittency of photovoltaic power. This allows the fused photovoltaic power prediction interval to accurately define the uncertainty range of photovoltaic power, avoiding inaccurate prediction results caused by fixed photovoltaic power values. Subsequently, based on the fused photovoltaic power prediction intervals at different time scales, day-ahead scheduling models, intraday hourly scheduling models, and intraday real-time scheduling models of the energy system at different scheduling scales are constructed. This enables the energy system's scheduling scheme to not only consider the photovoltaic power variation characteristics at different time scales but also to comprehensively consider the actual operating conditions of the energy system under different scheduling scales, thereby improving the accuracy of energy system scheduling.

[0012] Accordingly, this invention provides an energy system scheduling device based on multi-timescale interval prediction, including: a photovoltaic power prediction data acquisition module, a static photovoltaic power prediction interval acquisition module, a dynamic photovoltaic power prediction interval acquisition module, a fused photovoltaic power prediction interval acquisition module, and an energy system scheduling module;

[0013] The photovoltaic power prediction data acquisition module is used to acquire photovoltaic power prediction data of the energy system at several different time scales.

[0014] The static photovoltaic power prediction interval acquisition module is used to construct a static photovoltaic power correction model for each time scale based on the photovoltaic power prediction data for each time scale, and obtain the static photovoltaic power prediction interval for each time scale based on the static photovoltaic power correction model.

[0015] The dynamic photovoltaic power prediction interval acquisition module is used to construct a dynamic photovoltaic power correction model for each time scale based on the photovoltaic power prediction data for each time scale, and obtain the dynamic photovoltaic power prediction interval for each time scale based on the dynamic photovoltaic power correction model.

[0016] The integrated photovoltaic power prediction interval acquisition module is used to determine the integrated photovoltaic power prediction interval for each time scale based on the static photovoltaic power prediction interval and the dynamic photovoltaic power prediction interval for each time scale.

[0017] The energy system scheduling module is used to construct a day-ahead scheduling model, an intraday hourly scheduling model, and an intraday real-time scheduling model of the energy system based on the fused photovoltaic power prediction interval for each time scale. The module solves the day-ahead scheduling model, the intraday hourly scheduling model, and the intraday real-time scheduling model to determine the scheduling scheme of the energy system, and schedules the energy system based on the scheduling scheme.

[0018] It is understood that this invention constructs static and dynamic photovoltaic power correction models for each time scale using photovoltaic power prediction data from several different time scales of the energy system. This generates static and dynamic photovoltaic power prediction intervals, respectively, resulting in a fused photovoltaic power prediction interval. The static photovoltaic power correction model optimizes the error of the photovoltaic power prediction data, while the dynamic photovoltaic power correction model adapts to the volatility, randomness, and intermittency of photovoltaic power. This allows the fused photovoltaic power prediction interval to accurately define the uncertainty range of photovoltaic power, avoiding inaccurate prediction results caused by fixed photovoltaic power values. Subsequently, based on the fused photovoltaic power prediction intervals at different time scales, day-ahead scheduling models, intraday hourly scheduling models, and intraday real-time scheduling models of the energy system at different scheduling scales are constructed. This enables the energy system's scheduling scheme to not only consider the photovoltaic power variation characteristics at different time scales but also to comprehensively consider the actual operating conditions of the energy system under different scheduling scales, thereby improving the accuracy of energy system scheduling. Attached Figure Description

[0019] Figure 1 A flowchart illustrating the steps of an energy system scheduling method based on multi-time-scale interval prediction provided in this embodiment of the invention;

[0020] Figure 2 This is a schematic diagram of the structure of an energy system scheduling device based on multi-timescale interval prediction, provided as an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1

[0023] Please refer to Figure 1 , Figure 1 The flowchart of an energy system scheduling method based on multi-time-scale interval prediction provided in this embodiment of the invention includes steps S101 to S105.

[0024] Step S101: Obtain photovoltaic power prediction data for the energy system at several different time scales.

[0025] In one optional embodiment, acquiring photovoltaic power prediction data for the energy system at several different time scales includes: setting time scales including 15-minute time scale, 4-hour time scale, and 24-hour time scale; then setting preset photovoltaic power prediction input feature data including: real-time photovoltaic power of photovoltaic panels, horizontal irradiance, diffuse irradiance, tilted irradiance, and battery temperature, as well as wind speed data, humidity data, and temperature data; then selecting tilted irradiance, battery temperature, horizontal irradiance, diffuse irradiance, wind speed data, humidity data, and temperature data as the photovoltaic power prediction input feature type set for the 15-minute time scale; selecting tilted irradiance, battery temperature, and humidity data as the photovoltaic power prediction input feature type set for the 4-hour time scale; and selecting tilted irradiance, battery temperature, and humidity data as the photovoltaic power prediction input feature type set for the 24-hour time scale.

[0026] Then, the correlation coefficients of the photovoltaic power prediction input feature type sets at 15-minute, 4-hour, and 24-hour time scales were calculated. Specifically, the Pearson correlation coefficient sets and the Spearman correlation coefficient sets for the photovoltaic power prediction input feature type sets at 15-minute, 4-hour, and 24-hour time scales were calculated respectively. The Pearson and Spearman correlation coefficient sets were summed and averaged to obtain the correlation coefficient sets at 15-minute, 4-hour, and 24-hour time scales. For the 15-minute time scale correlation coefficient set, a correlation coefficient curve was plotted, and half the period of the correlation coefficient curve was taken as the length of the 15-minute time scale photovoltaic power prediction input feature sequence. Similarly, for the 4-hour time scale correlation coefficient set, a correlation coefficient curve was plotted, and the period of the correlation coefficient curve was taken as the length of the 4-hour time scale photovoltaic power prediction input feature sequence. Finally, for the 24-hour time scale correlation coefficient set, monotonically decreasing regression was used. The method of regression (MDR) is used to fit the curve, and then a threshold (set to 0.4) is used to truncate the fitted curve. The part that is greater than the threshold is used as the length of the input feature sequence for photovoltaic power prediction on a 24-hour time scale.

[0027] Next, a photovoltaic prediction model is obtained. The photovoltaic prediction model is set as a parallel sub-model of CNN-LSTM and XGBoost. In particular, the photovoltaic prediction model used in this embodiment is only an adaptive example, and the specific model type or model can be changed according to actual needs. After obtaining the photovoltaic prediction model, training sets, validation sets and test sets need to be constructed for 15-minute time scale, 4-hour time scale and 24-hour time scale respectively. Then, the photovoltaic prediction model is trained based on the training set, validation set and test set to obtain photovoltaic prediction models for 15-minute time scale, 4-hour time scale and 24-hour time scale.

[0028] Then, based on the lengths of the photovoltaic power prediction input feature sequences at 15-minute, 4-hour, and 24-hour time scales, the corresponding sets of photovoltaic power prediction input feature types are filtered, i.e., redundant data is removed, so that the data length of the subsequent input photovoltaic prediction models is the same as the length of the photovoltaic power prediction input feature sequences. Next, photovoltaic power prediction is performed based on the photovoltaic prediction models at 15-minute, 4-hour, and 24-hour time scales, yielding photovoltaic power prediction values ​​at these scales. Then, a weighted average is taken of the photovoltaic power prediction values ​​at these scales, and the error between the weighted average and the training, validation, and test sets at the 15-minute time scale is calculated, thus obtaining the photovoltaic power prediction error data at the 15-minute time scale. Similarly, the photovoltaic power prediction error data at the 4-hour and 24-hour time scales can be obtained.

[0029] It should be noted that the parallel sub-models of CNN-LSTM and XGBoost are a hybrid architecture combining the temporal modeling capabilities of deep learning with the structured data advantages of traditional tree models. It includes the CNN-LSTM sub-model (Convolutional Neural Network - Long Short-Term Memory) and the XGBoost sub-model (Extreme Gradient Boosting). The Pearson correlation coefficient is commonly used to measure the degree of linear correlation between two variables. The Spearman's Rank Correlation Coefficient measures the monotonic relationship (whether linear or not) between two variables, calculating the correlation through ranking.

[0030] Step S102: Based on the photovoltaic power prediction data for each time scale, construct a static photovoltaic power correction model for each time scale, and obtain the static photovoltaic power prediction interval for each time scale based on the static photovoltaic power correction model.

[0031] In this embodiment, the step of constructing a static photovoltaic power correction model for each time scale based on photovoltaic power prediction data for each time scale, and obtaining a static photovoltaic power prediction interval for each time scale based on the static photovoltaic power correction model, includes:

[0032] The photovoltaic power prediction data includes: photovoltaic power prediction values ​​and photovoltaic power prediction error data;

[0033] Obtain the number of Gaussian mixture components at each time scale, and construct a Gaussian mixture model for each time scale based on the number of Gaussian mixture components at each time scale;

[0034] The photovoltaic power prediction error data for each time scale is input into the corresponding static photovoltaic power correction model. With maximizing the log-likelihood function as the optimization objective, the Gaussian mixture model for each time scale is iterated using a preset expectation maximization algorithm to determine the static photovoltaic power correction model for each time scale and obtain several Gaussian components output by the static photovoltaic power correction model for each time scale.

[0035] The aggregated Gaussian distribution for each time scale is obtained by weighted averaging of several Gaussian components output by the static photovoltaic power correction model for each time scale.

[0036] Centered on the photovoltaic power prediction value for each time scale, the aggregated Gaussian distribution and the photovoltaic power prediction value for each time scale are superimposed to obtain the static photovoltaic power prediction interval for each time scale.

[0037] In an alternative embodiment, the Gaussian mixture model can be expressed by the following formulas (1) and (2):

[0038]

[0039] In formulas (1) and (2), α represents the photovoltaic power prediction error data, K is the number of mixed Gaussian components, k represents the k-th Gaussian component, and π k Let the mixture weights of the k-th Gaussian component satisfy the following condition: And π k ≥0; The probability density function of the k-th Gaussian component is represented by d; d represents the data dimension of the photovoltaic power prediction error data; μ k Let ∑k represent the mean vector of the k-th Gaussian component; let ∑k represent the covariance matrix; and let T represent the transpose of the matrix.

[0040] Specifically, the number of Gaussian mixture components is set to 3 for the 15-minute timescale, 5 for the 4-hour timescale, and 5 for the 24-hour timescale. The photovoltaic power prediction error data and the number of Gaussian mixture components for the 15-minute, 4-hour, and 24-hour timescales are input into formulas (1) and (2) respectively to obtain the Gaussian mixture model (GMM) for the 15-minute timescale. 15min Gaussian Mixture Model (GMM) with a 4-hour timescale 4hour Gaussian Mixture Model (GMM) with a 24-hour timescale 24hour ;

[0041] Then, with maximizing the log-likelihood function as the optimization objective, the GMM is then optimized using a pre-defined expectation-maximization algorithm (EM algorithm). 15min GMM 4hour and GMM 24hour To perform iterations, specifically, the log-likelihood function is expressed as... Where, α i The i-th data point in the photovoltaic power prediction error data represents the total number of data points in the photovoltaic power prediction error data. The EM algorithm is used to optimize the GMM. 15min GMM 4hour and GMM 24hour parameters The iterative process of the EM algorithm is as follows: First, calculate α. i The posterior probability γ of belonging to the k-th Gaussian component ik Then based on the posterior probability γ ik Update parameters Then iterate through these two operations until the GMM is reached. 15min GMM 4hour and GMM 24hour convergence;

[0042] Furthermore, the convergence criterion uses a log-likelihood change threshold; that is, after each EM iteration, a threshold is set based on the log-likelihood function. Calculate the log-likelihood value of the current EM iteration, then calculate the difference between it and the log-likelihood value of the previous EM iteration. If the difference is less than 10... -3 If the result is positive, then it is considered convergent.

[0043] Posterior probability γ ik The calculation is shown in formula (3):

[0044]

[0045] Update parameters The calculation process is shown in formula (4):

[0046]

[0047] In formulas (3) and (4), α j This represents the j-th data point in the photovoltaic power prediction error data; This represents the mixture weight of the updated k-th Gaussian component. This represents the mean vector of the updated k-th Gaussian component; This represents the updated covariance matrix;

[0048] When GMM 15min GMM 4hour and GMM 24hourAfter convergence, the converged GMM will be... 15min GMM 4hour and GMM 24hour These serve as static photovoltaic power correction models for 15-minute, 4-hour, and 24-hour timescales, respectively, as well as a GMM that directly achieves convergence. 15min GMM 4hour and GMM 24hour Several Gaussian components;

[0049] The Gaussian components are then weighted and averaged to obtain a aggregated Gaussian distribution. Next, using the photovoltaic power prediction value as the center, the aggregated Gaussian distribution and the photovoltaic power prediction value are superimposed using the z-score method to obtain the static photovoltaic power prediction interval. The static photovoltaic power prediction interval is a probability interval centered on the photovoltaic power prediction value; among them, the static photovoltaic power prediction interval on a 15-minute time scale is Y. 15min The static photovoltaic power prediction interval on a 4-hour timescale is Y. 4hour The static photovoltaic power prediction interval on a 24-hour timescale is Y. 24hour .

[0050] It's important to note that a Gaussian Mixture Model (GMM) is a probabilistic model that assumes all data points are generated by a mixture of a finite number of Gaussian (or normal) distributions. It is a widely used generative model. The Expectation-Maximization Algorithm (EM) is an iterative optimization algorithm used to find the maximum likelihood estimate (MLE) or maximum a posteriori estimate (MAP) of parameters in a probabilistic model. The z-score method is a statistical method used to measure the position of a raw data point relative to the mean of its dataset. It represents how many standard deviations the data point differs from the mean.

[0051] This embodiment models the photovoltaic power prediction error data using a Gaussian mixture model, with the goal of maximizing the log-likelihood function. The model parameters are optimized using the expectation-maximization algorithm, which can more accurately describe the statistical characteristics of the photovoltaic power prediction error data. By weighted averaging of multiple Gaussian components to obtain an aggregated Gaussian distribution, and superimposing it with the photovoltaic power prediction value to generate a static prediction interval, the static photovoltaic power prediction interval can effectively capture the uncertainty of photovoltaic power and improve the accuracy of the static photovoltaic power prediction interval.

[0052] Step S103: Based on the photovoltaic power prediction data for each time scale, construct a dynamic photovoltaic power correction model for each time scale, and obtain the dynamic photovoltaic power prediction interval for each time scale based on the dynamic photovoltaic power correction model.

[0053] In this embodiment, the step of constructing a dynamic photovoltaic power correction model for each time scale based on the initial photovoltaic power prediction data for each time scale, and obtaining the dynamic photovoltaic power prediction interval for each time scale based on the dynamic photovoltaic power correction model, includes:

[0054] The photovoltaic power prediction data includes: a set of photovoltaic power prediction input feature types and the length of the photovoltaic power prediction input feature sequence;

[0055] Obtain the number of Gaussian mixture components at each time scale, and construct a mixture density model for each time scale based on the number of Gaussian mixture components at each time scale;

[0056] The preset hybrid density model training set is input into the hybrid density model for each time scale. The negative log-likelihood function is used as the loss function and the conditional log-likelihood function is maximized as the training objective. The preset Adam optimizer is used to train the hybrid density model for each time scale to determine the dynamic photovoltaic power correction model for each time scale.

[0057] Based on the set of photovoltaic power prediction input feature types and the length of photovoltaic power prediction input feature sequence for each time scale, the preset photovoltaic power prediction input feature data is filtered to obtain dynamic photovoltaic power correction input data for each time scale.

[0058] The dynamic photovoltaic power correction input data for each time scale is input into the corresponding dynamic photovoltaic power correction model to obtain the dynamic photovoltaic power prediction interval for each time scale.

[0059] It's important to note that a Mixture Density Network (MDN) is a probabilistic model used to solve one-to-many mappings or highly nonlinear regression problems. Its core idea is to use a mixture of probability distributions (typically a mixture of multiple Gaussian distributions) to model the conditional probability distribution of the output values ​​given the input conditions. The Conditional Log-Likelihood Function (CLF) is a core tool in statistics and machine learning used to evaluate conditional probabilistic models. It is the natural logarithm of the conditional likelihood function.

[0060] In an optional embodiment, a probability model is generally required to generate the prediction interval for the mixture density model. Therefore, this embodiment selects a Gaussian mixture model as the probability density model. Thus, the probability density function of the mixture density model can be expressed as shown in the following formula (5):

[0061]

[0062] In formula (5), p(y|x) represents the probability density function, and K is the number of Gaussian mixture components; π k (x) represents the weight of the k-th Gaussian mixture component, constrained by the Softmax function to be a probability value; μ k (x) is the mean vector of the k-th Gaussian mixture component, ∑ k (x) represents the covariance matrix, which is directly generated by the neural network;

[0063] Then, the negative log-likelihood function is set as the loss function, and the negative log-likelihood function is shown in the following formula (6):

[0064]

[0065] Then, the conditional log-likelihood function is maximized as the training objective, and the Adam optimizer is used to train the mixture density model. The gradient calculation formula of the Adam optimizer is shown in formula (7): Where, θ t-1 The network parameters at the current time (including the mean, variance, and weights in the Gaussian mixture model); It is a negative log-likelihood function; Indicates gradient calculation;

[0066] Furthermore, the training set for the preset hybrid density model is set as the preset photovoltaic power prediction input feature data mentioned in step S101; the preset photovoltaic power prediction input feature data is input into the hybrid density model, and mapped to the hidden layer feature space through the hybrid density model. The mapping formula is h = f θ (x), where θ is the network parameter; x is the input feature data for photovoltaic power prediction, and h represents the output of the hidden layer feature space; then, the hybrid weights, mean, and variance are generated by the hybrid parameters of the hybrid density model, as shown in the following formula (8):

[0067]

[0068] Among them, Z π,k W represents the logarithmic probability of the mixed weights and variance before normalization. π,k W μ,k and W σ,k Let b represent the mixed weights, mean, and variance of the neural network weights, respectively. π,kb μ,k b σ,k The neural network bias term represents the mixed weights, mean, and variance. Then, the mixed weights, mean, and variance are updated using the Adam optimizer with the negative log-likelihood function as the loss function and the maximization of the conditional log-likelihood function as the training objective, to obtain the dynamic photovoltaic power correction model.

[0069] Based on the above process, a dynamic photovoltaic power correction model (MDN) with a 15-minute timescale can be obtained. 15min Gaussian Mixture Model (MDN) with a 4-hour timescale 4hour Gaussian Mixture Model (MDN) with a 24-hour timescale 24hour ;

[0070] Then, based on the photovoltaic power prediction input feature type set and the photovoltaic power prediction input feature sequence length, the photovoltaic power prediction input feature data is filtered. Specifically, the photovoltaic power prediction input feature type set determines which types of data to filter from the photovoltaic power prediction input feature data, and the photovoltaic power prediction input feature sequence length determines the sequence length of each type of data to be filtered from the photovoltaic power prediction input feature data, thereby obtaining the dynamic photovoltaic power correction input data for each time scale.

[0071] Furthermore, for a 15-minute timescale, the selected data on tilted surface irradiance, battery temperature horizontal surface irradiance, diffuse irradiance, wind speed, humidity, and temperature are input into MDN. 15min The dynamic photovoltaic power prediction interval Z on a 15-minute timescale is obtained. 15min For a 4-hour timescale, the selected tilt surface irradiance, battery temperature, and humidity data are input into MDN. 4hour The dynamic photovoltaic power prediction interval Z on a 4-hour timescale is obtained. 4hour For a 24-hour timescale, the selected tilt surface irradiance, battery temperature, and humidity data are input into the MDN. 24hour The dynamic photovoltaic power prediction interval Z on a 24-hour timescale is obtained. 24hour .

[0072] This embodiment improves the accuracy of the dynamic photovoltaic power correction model by constructing a hybrid density model, using the negative log-likelihood function as the loss function and maximizing the conditional log-likelihood as the objective, and training the hybrid density model with the Adam optimizer. Then, based on the photovoltaic power prediction input feature type set and the photovoltaic power prediction input feature sequence length corresponding to different time scales, dynamic photovoltaic power correction input data is generated. This allows the dynamic photovoltaic power correction input data to more accurately represent the variation characteristics of photovoltaic power at different time scales, making the dynamic photovoltaic power prediction interval more adaptable to the volatility, randomness, and intermittency of photovoltaic power at different time scales, thus improving the accuracy of the dynamic photovoltaic power prediction interval.

[0073] Step S104: Based on the static photovoltaic power prediction interval and the dynamic photovoltaic power prediction interval for each time scale, determine the fused photovoltaic power prediction interval for each time scale.

[0074] In this embodiment, determining the fused photovoltaic power prediction interval for each time scale based on the static photovoltaic power prediction interval and the dynamic photovoltaic power prediction interval for each time scale includes:

[0075] Obtain the weighting coefficients for the static photovoltaic power prediction interval and the coefficients for the dynamic photovoltaic power prediction interval;

[0076] Based on the weighting coefficient of the static photovoltaic power prediction interval and the coefficient of the dynamic photovoltaic power prediction interval, the static photovoltaic power prediction interval and the dynamic photovoltaic power prediction interval of each time scale are weighted and fused to obtain the fused photovoltaic power prediction interval of each time scale.

[0077] In one optional embodiment, both the static photovoltaic power prediction interval weighting coefficient and the dynamic photovoltaic power prediction interval coefficient are 0.5. Therefore, the fused photovoltaic power prediction interval X at a 15-minute time scale is... 15min =0.5*Y 15min +0.5*Z 15min ; 4-hour timescale fused photovoltaic power prediction interval X 4hour =0.5*Y 4hour +0.5*Z 4hour ; 4-hour timescale fused photovoltaic power prediction interval X 24hour =0.5*Y 24hour +0.5*Z 24hour .

[0078] This embodiment sets a weighting coefficient for the static photovoltaic power prediction interval and a coefficient for the dynamic photovoltaic power prediction interval, and then performs a weighted fusion of the static and dynamic photovoltaic power prediction intervals. This results in a fused photovoltaic power prediction interval that not only includes short-term changes in photovoltaic power but also retains long-term statistical patterns. This makes the fused photovoltaic power prediction interval more consistent with the actual situation of photovoltaic power in the energy system, improving its accuracy and consequently enhancing the accuracy of subsequent energy system scheduling.

[0079] Step S105: Based on the fused photovoltaic power prediction interval for each time scale, construct the day-ahead scheduling model, intra-day hourly scheduling model, and intra-day real-time scheduling model of the energy system. Solve the day-ahead scheduling model, intra-day hourly scheduling model, and intra-day real-time scheduling model respectively to determine the scheduling scheme of the energy system, and schedule the energy system based on the scheduling scheme of the energy system.

[0080] In this embodiment, based on the fused photovoltaic power prediction interval for each time scale, a day-ahead scheduling model, an intraday hourly scheduling model, and an intraday real-time scheduling model of the energy system are constructed. The day-ahead scheduling model, the intraday hourly scheduling model, and the intraday real-time scheduling model are then solved to determine the scheduling scheme of the energy system, including:

[0081] Obtain the electricity price data and operation data of the energy system;

[0082] Based on the electricity price data of the energy system, and combined with the integrated photovoltaic power prediction interval for each time scale, the integrated electricity price photovoltaic power prediction value for each time scale is determined.

[0083] Based on the operating data of the energy system and the predicted photovoltaic power of the integrated electricity price at each time scale, the day-ahead scheduling constraint set, the intraday hourly scheduling constraint set, and the intraday real-time scheduling constraint set of the energy system are constructed respectively.

[0084] Based on the day-ahead scheduling constraint set, intraday hourly scheduling constraint set, and intraday real-time scheduling constraint set of the energy system, the day-ahead scheduling model, intraday hourly scheduling model, and intraday real-time scheduling model of the energy system are constructed respectively. The day-ahead scheduling model, intraday hourly scheduling model, and intraday real-time scheduling model are solved respectively to determine the scheduling scheme of the energy system.

[0085] This embodiment generates a combined electricity price photovoltaic power prediction value by combining electricity price data and the combined photovoltaic power prediction range. This value not only reflects the actual predicted photovoltaic power but also integrates the electricity price situation of the energy system, making it more in line with the actual dispatch needs of the energy system. Subsequently, the energy system dispatch scheme is determined using models with three different dispatch scales: intraday dispatch, intraday hourly dispatch, and intraday real-time dispatch. This ensures that the energy system dispatch scheme not only considers the photovoltaic power variation characteristics at different time scales but also integrates the actual operation of the energy system under different dispatch scales, thereby improving the accuracy of energy system dispatch.

[0086] In this embodiment, determining the predicted photovoltaic power value based on the electricity price data of the energy system, combined with the fused photovoltaic power prediction interval for each time scale, includes:

[0087] The electricity price data of the energy system includes: real-time electricity purchase price, historical minimum electricity purchase price, and historical maximum electricity purchase price for each time scale;

[0088] Based on the real-time electricity purchase price, the lowest historical electricity purchase price, and the highest historical electricity purchase price for each time scale, determine the electricity price coefficient for each time scale;

[0089] Based on the fusion photovoltaic power prediction interval for each time scale, determine the upper limit and lower limit of the fusion photovoltaic power prediction for each time scale.

[0090] Based on the electricity price coefficient, the upper limit of the integrated photovoltaic power prediction, and the lower limit of the integrated photovoltaic power prediction for each time scale, the integrated electricity price photovoltaic power prediction value for each time scale is determined.

[0091] In an alternative embodiment, a real-time electricity purchase price σ is defined for each of the said time scales. buy (t), the lowest historical electricity purchase price σ buy,min (t) and the highest historical electricity purchase price σ buy,max (t);

[0092] Therefore, the electricity price coefficient α at each time scale qj The calculation of (t) is shown in the following formula (9):

[0093]

[0094] Define the upper limit P for the fused photovoltaic power prediction at each time scale. pv,up (t) and the lower limit of the predicted integrated photovoltaic power P pv,low(t); Therefore, the calculation of the photovoltaic power prediction value of the combined electricity price for each time scale is shown in the following formula (10):

[0095] P pv,qj (t)=P pv,low (t)+α qj (t)·(P pv,up (t)-P pv,low (t)) (10);

[0096] In formulas (9) and (10), t represents time t on the time scale;

[0097] Therefore, the fusion photovoltaic power prediction interval X based on a 15-minute time scale 15min The corresponding upper and lower limits of the integrated photovoltaic power prediction are obtained, along with the real-time electricity purchase price, the historical minimum electricity purchase price, and the historical maximum electricity purchase price at a 15-minute time scale. These values ​​are then substituted into formulas (9) and (10) to obtain the integrated electricity price photovoltaic power prediction value P at a 15-minute time scale. pv,qj,15min Similarly, the predicted photovoltaic power output P for the fused electricity price can be obtained on a 4-hour timescale. pv,qj,4hour Combined electricity price photovoltaic power forecast P on a 24-hour timescale pv,qj,24hour .

[0098] This embodiment introduces electricity price coefficients at different time scales, and then combines the upper and lower limits of the integrated photovoltaic power prediction at different time scales to calculate the integrated electricity price photovoltaic power prediction value at different time scales. This fully considers the impact of electricity price fluctuations in the energy system, and also considers the actual photovoltaic power prediction situation of the energy system, making the integrated electricity price photovoltaic power prediction value more consistent with the actual operation of the energy system, thereby improving the accuracy of subsequent energy system dispatch.

[0099] In this embodiment, the construction of the day-ahead scheduling constraint set, the intraday hourly scheduling constraint set, and the intraday real-time scheduling constraint set of the energy system based on the operating data of the energy system and the predicted photovoltaic power at each time scale includes:

[0100] The time scales include: 15-minute time scale, 4-hour time scale and 24-hour time scale;

[0101] The energy system's operational data includes: energy system operational data on a 15-minute timescale, energy system operational data on a 4-hour timescale, and energy system operational data on a 24-hour timescale.

[0102] Based on the energy system operation data at the 24-hour time scale, and combined with the photovoltaic power prediction value of the integrated electricity price at the 24-hour time scale, a set of day-ahead scheduling constraints for the energy system is constructed.

[0103] Based on the energy system operation data at the 4-hour time scale, and combined with the photovoltaic power prediction value of the integrated electricity price at the 4-hour time scale, a set of intraday hourly scheduling constraints for the energy system is constructed.

[0104] Based on the energy system operation data at the 15-minute time scale, and combined with the photovoltaic power prediction value of the integrated electricity price at the 15-minute time scale, a set of intraday real-time scheduling constraints for the energy system is constructed.

[0105] In one optional embodiment, the operating data of the energy system includes: heat pump heating efficiency η Hp The maximum charging power P of the battery Bat,chr,max The maximum discharge power P of the battery Bat,dis,max The rated capacity P of the storage battery Bat,E The charging efficiency η of the storage battery chr The discharge efficiency η of the storage battery dis The lower limit of battery capacity S Bat,min The upper limit of battery capacity S Bat,max The initial capacity S of the battery Bat,star When the battery scheduling is terminated, the capacity S Bat,end The initial capacity Q of the thermal storage tank Hs,star When the scheduling of the thermal storage tank is terminated, the capacity Q Hs,end The self-discharge rate σ of the battery bat The discharge efficiency η of the storage battery dis The self-heat release rate γ of the thermal storage tank h The heat absorption efficiency η of the storage battery Hs,chr The heat dissipation efficiency η of the storage battery Hs,dis The upper limit of the heat absorption power P of the battery Hs,chr,max The lower limit of the heat absorption power P of the battery Hs,chr,min The upper limit of thermal energy stored in the thermal storage tank, Q Hs,max The lower limit of thermal energy stored in the thermal storage tank, Q Hs,min The upper limit of the heat release power P of the thermal storage tank Hs,dis,max The lower limit of heat release power P of the thermal storage tank Hs,dis,min Heat pump heating efficiency η Hp and the upper limit of heat pump heat power output Q Hp,max The scheduling constraint set is defined to include: electrical power balance constraints, thermal power balance constraints, battery operation constraints, energy storage device periodic balance constraints, and other device constraints. Based on the above operational data, the specific scheduling constraint set is as follows:

[0106] The form of the power balance constraint is shown in formula (11):

[0107] P buy (t)-P sell (t)+P pv (t)+P Bat,dis (t)-P Bat,chr (t)-P Hp (t)=P e (t) (11);

[0108] Among them, P buy (t) represents the power purchased by the energy system at time t on the time scale; P sell (t) represents the power sold by the energy system at time t on the time scale; P pv (t) represents the photovoltaic power at time t on the time scale; P Bat,dis (t) represents the battery discharge power at time t on the time scale; P Bat,chr (t) represents the battery charging power at time t on the time scale; P Hp (t) represents the power input from the energy system to the heat pump at time t on the time scale; P e (t) represents the electrical load demand at time t on the time scale;

[0109] The form of the thermal power balance constraint is shown in the following formula (12):

[0110] P Hp (t)·η Hp +P Hs,dis (t)-P Hs,chr (t)=P h (t) (12);

[0111] Among them, P Hp (t) represents the power input from the energy system to the heat pump at time t on the time scale; η Hp Indicates the heating efficiency of a heat pump; P Hs,dis (t) represents the heat dissipation power of the battery at time t on the time scale; P Hs,chr (t) represents the heat absorption power of the battery at time t on the time scale; P h (t) represents the heat load demand at time t on the time scale;

[0112] The operating constraints of the battery are shown in the following formula (13):

[0113]

[0114] Among them, P Bat,chr,max P is the maximum charging power of the battery. Bat,dis,max P represents the maximum discharge power of the battery. Bat,EIndicates the rated capacity of the battery; η chr Indicates the charging efficiency of the battery; η dis Indicates the discharge efficiency of the battery; S Bat,min Indicates the lower limit of the battery capacity; S Bat,max Indicates the upper limit of the battery capacity; S Bat (t) represents the battery capacity at time t on the time scale;

[0115] The cycle balance constraint of the energy storage device is shown in the following formula (14):

[0116]

[0117] Among them, S Bat,star and S Bat,end Q represents the initial capacity and the capacity at the end of the scheduling process, respectively. Hs,star and Q Hs,end These represent the initial capacity and the capacity at the end of the scheduling process, respectively.

[0118] Other constraints on the equipment are shown in formulas (15) to (17):

[0119]

[0120] In formulas (15) to (17), S Bat (t) represents the battery capacity at time t on the time scale; S Bat (t-1) represents the battery capacity at time t-1 on the time scale; σ bat P represents the self-discharge rate of the battery. Bat,dis (t) represents the battery discharge power at time t on the time scale; Δt represents the difference between time t and time t-1; η dis P represents the discharge efficiency of the battery. Bat,chr (t) represents the battery charging power at time t on the time scale; η chr Indicates the charging efficiency of the battery; Q Hs (t) represents the thermal energy stored in the thermal storage tank at time t on the time scale; P Hs,dis (t) represents the heat dissipation power of the battery at time t on the time scale; P Hs,chr (t) represents the heat absorption power of the battery at time t on the time scale; γ h Indicates the self-heating rate of the thermal storage tank; η Hs,chr Indicates the heat absorption efficiency of the battery; η Hs,dis Indicates the heat dissipation efficiency of the battery; P Hs,dis (t) represents the heat dissipation power of the battery at time t on the time scale; P Hs,chr (t) represents the heat absorption power of the battery at time t on the time scale; P Hs,chr,max and PHs,chr,min These represent the upper limit and lower limit of the battery's heat absorption power, respectively; Q Hs,max and Q Hs,min These represent the upper limit and lower limit of the thermal energy that the thermal energy storage tank can store, respectively; P Hs,dis,max and P Hs,dis,min These represent the upper and lower limits of the heat release power of the thermal storage tank, respectively; Q Hp Indicates the heat output power of the heat pump; η Hp Surface heat pump heating efficiency; P Hp Q represents the power input from the energy system to the heat pump. Hp,max The upper limit of the heat pump's thermal power output;

[0121] Furthermore, energy system operation data is set at a 15-minute timescale, including energy system power purchase, power sold, battery discharge, battery charging, power input to the heat pump, electrical load demand, battery heat release, battery heat absorption, heat load demand, battery capacity, and thermal energy stored in the thermal storage tank at each moment within the 15-minute timescale; then, the 15-minute timescale photovoltaic power prediction value P based on the integrated electricity price is calculated. pv,qj,15min As the photovoltaic power at each moment on a 15-minute time scale; then these input data are substituted into formulas (11) to (17) to obtain the intraday real-time scheduling constraint set of the energy system;

[0122] Similarly, energy system operation data is set at a 4-hour timescale, including energy system purchased power, energy system sold power, battery discharged power, battery charged power, power input to the heat pump, electrical load demand, battery heat release power, battery heat absorption power, heat load demand, battery capacity, and thermal energy stored in the thermal storage tank at each moment within the 4-hour timescale; then, the 4-hour timescale integrated electricity price photovoltaic power prediction value P is calculated. pv,qj,4hour As the photovoltaic power at each moment on a 4-hour time scale; then these input data are substituted into formulas (11) to (17) to obtain the intraday hourly scheduling constraint set of the energy system;

[0123] The system operates on a 24-hour timescale, including power input, power output, battery discharge, battery charging, power input to the heat pump, electrical load demand, battery heat release, battery heat absorption, heat load demand, battery capacity, and thermal energy stored in the thermal storage tank at each moment within the 24-hour timescale. Then, the 24-hour timescale photovoltaic power prediction value P based on the integrated electricity price is calculated. pv,qj,24hourThe photovoltaic power at each moment on a 24-hour timescale is used as input data. Then, these input data are substituted into formulas (11) to (17) to obtain the day-ahead scheduling constraint set of the energy system.

[0124] This embodiment constructs corresponding scheduling constraint sets using energy system operation data at different time scales and photovoltaic power prediction values ​​based on integrated electricity prices. The hierarchical constraint design enables scheduling models at different scheduling scales to better adapt to the operational requirements of different time dimensions, thereby improving the accuracy of energy system scheduling.

[0125] In this embodiment, based on the day-ahead scheduling constraint set, intraday hourly scheduling constraint set, and intraday real-time scheduling constraint set of the energy system, a day-ahead scheduling model, an intraday hourly scheduling model, and an intraday real-time scheduling model of the energy system are constructed, respectively. The day-ahead scheduling model, the intraday hourly scheduling model, and the intraday real-time scheduling model are then solved to determine the scheduling scheme of the energy system, including:

[0126] Obtain the scheduling cost data of the energy system;

[0127] Based on the scheduling cost data of the energy system, a day-ahead scheduling objective function of the energy system is constructed with the goal of minimizing the total day-ahead scheduling cost.

[0128] Based on the day-ahead scheduling objective function and day-ahead scheduling constraint set of the energy system, a day-ahead scheduling model of the energy system is constructed;

[0129] Solve the day-ahead scheduling model of the energy system to obtain the day-ahead scheduling scheme of the energy system;

[0130] The objective function for intraday hourly scheduling of the energy system is constructed with the minimum deviation cost of intraday hourly scheduling scheme and day-ahead scheduling scheme as the optimization objective.

[0131] Based on the intraday hourly scheduling objective function and intraday hourly scheduling constraint set of the energy system, an intraday hourly scheduling model of the energy system is constructed.

[0132] Solve the intraday hourly scheduling model of the energy system to obtain the intraday hourly scheduling scheme of the energy system;

[0133] The objective function for the intraday real-time scheduling of the energy system is constructed with the minimum deviation cost of the intraday real-time scheduling scheme and the intraday hourly scheduling scheme as the optimization objective.

[0134] Based on the intraday real-time scheduling objective function and intraday real-time scheduling constraint set of the energy system, an intraday real-time scheduling model of the energy system is constructed.

[0135] Solve the intraday real-time scheduling model of the energy system to obtain the intraday real-time scheduling scheme of the energy system;

[0136] Based on the day-ahead scheduling scheme, intraday hourly scheduling scheme, and intraday real-time scheduling scheme of the energy system, the scheduling scheme of the energy system is determined.

[0137] In one optional embodiment, the energy system dispatch cost data includes: the unit maintenance cost λ of photovoltaic power. pv The unit maintenance cost of a heat pump (λ) Hp The unit maintenance cost of batteries (λ) Bat The unit maintenance cost λ of the thermal storage tank Hs The electricity purchase price σ of the energy system at time t on the time scale buy (t) and the electricity price sold by the energy system at time t on the time scale. sell (t);

[0138] With minimizing the total day-ahead scheduling cost as the optimization objective, the day-ahead scheduling objective function of the energy system is constructed as shown in the following formulas (18) to (20):

[0139] CT 24hour =min(C oper +C Gird (18);

[0140]

[0141] In formulas (18) to (20), T represents the number of moments on the time scale; CT 24hour C represents the total cost of day-ahead dispatching; oper Indicates the daily operation and maintenance cost of the energy system; C Gird This represents the interaction cost between the energy system and the power grid;

[0142] Similarly, the 24-hour timescale converged electricity price photovoltaic power forecast value P pv,qj,24hour As the photovoltaic power at each moment on a 24-hour timescale, P is substituted into formula (19). pv (t), and then combined with the day-ahead scheduling constraint set, a day-ahead scheduling model is constructed; then the day-ahead scheduling model is solved to obtain the solution set of energy system operation data at a 15-minute time scale, and the solution set of energy system operation data at a 15-minute time scale is used as the day-ahead scheduling scheme of the energy system.

[0143] Then, taking the minimum deviation cost of the intraday hourly scheduling scheme and the day-ahead scheduling scheme as the optimization objective, the intraday hourly scheduling objective function of the energy system is constructed; since the scheduling of the energy system will generate costs, taking the minimum deviation cost of the intraday hourly scheduling scheme and the day-ahead scheduling scheme as the optimization objective is equivalent to taking the minimum adjustment amount of the intraday hourly scheduling scheme and the day-ahead scheduling scheme as the optimization objective; therefore, the intraday hourly scheduling objective function is shown in the following formulas (21) to (23):

[0144] C T,4hour =min(C p,g +C p,h ) (twenty one);

[0145]

[0146] C T,4hour C represents the cost difference between the intraday hourly scheduling plan and the day-ahead scheduling plan; p,g C represents the penalty cost for adjustments made by the power grid between intraday hourly dispatching schemes and day-ahead dispatching schemes; p,h The penalty cost for adjusting the heating network between the intraday hourly dispatch plan and the day-ahead dispatch plan; ΔP Bat ΔP represents the adjustment amount for the battery between the intraday hourly dispatch plan and the day-ahead dispatch plan. Gird The adjustment amount between the intraday hourly dispatch scheme and the day-ahead dispatch scheme for the interaction between the energy system and the power grid; λ Bat and λ Gird These are the penalty adjustment coefficients for the interaction between the battery, the energy system, and the power grid, respectively; P Bat (t) represents the battery power data for the intraday hourly scheduling plan; P Bat,ah (t) represents the battery power data for the day-ahead dispatch plan; P Gird (t) represents the power interaction data between the energy system and the power grid under the intraday hourly dispatch scheme; P Gird,ah (t) represents the power interaction data between the energy system and the grid under the day-ahead dispatch scheme; ΔP Hs ΔP Hp These represent the adjustments made to the thermal storage tank and heat pump between the intraday hourly scheduling plan and the day-ahead scheduling plan, respectively; λ Hsi and λ Hp P represents the penalty adjustment coefficients for the thermal storage tank and the heat pump, respectively; Hs (t) represents the thermal storage tank power data for the intraday hourly scheduling plan; P Hs,ah (t) represents the power data of the thermal storage tanks in the day-ahead scheduling plan; P Hp (t) represents the heat pump power data of the intraday hourly scheduling plan; P Hp,ah (t) represents the heat pump power data of the day-ahead scheduling plan;

[0147] Based on the intraday hourly scheduling objective function, an intraday hourly scheduling model is constructed by combining the intraday hourly scheduling constraint set. The intraday hourly scheduling model is solved to obtain the solution set of energy system operation data on a 4-hour time scale. The solution set of energy system operation data on a 4-hour time scale is used as the intraday hourly scheduling scheme of the energy system.

[0148] Then, taking the minimum deviation cost of the intraday real-time scheduling scheme and the intraday hourly scheduling scheme as the optimization objective, the intraday real-time scheduling objective function of the energy system is constructed; since the scheduling of the energy system will generate costs, taking the minimum deviation cost of the intraday real-time scheduling scheme and the intraday hourly scheduling scheme as the optimization objective is equivalent to taking the minimum adjustment amount of the intraday real-time scheduling scheme and the intraday hourly scheduling scheme as the optimization objective; therefore, the intraday hourly scheduling objective function is shown in the following formulas (24) to (26):

[0149] C T,15min =min(C p,g,1 +C p,h,1 ) (twenty four);

[0150]

[0151] C T,15min This represents the cost difference between the intraday real-time scheduling scheme and the intraday hourly scheduling scheme; C p,g,1 C represents the penalty cost for adjusting between the intraday real-time dispatch scheme and the intraday hourly dispatch scheme on the power grid side; p,h,1 The penalty cost for adjusting the heating network between the intraday real-time dispatch plan and the intraday hourly dispatch plan; ΔP Bat,1 ΔP represents the adjustment amount between the intraday real-time dispatch plan and the intraday hourly dispatch plan for the battery. Gird,1 The adjustment amount between the intraday real-time dispatch scheme and the intraday hourly dispatch scheme for the interaction between the energy system and the power grid; λ Bat and λ Gird These are the penalty adjustment coefficients for the interaction between the battery, the energy system, and the power grid, respectively; P Bat (t) represents the battery power data for the intraday hourly scheduling plan; P Bat,1 (t) represents the battery power data of the intraday real-time dispatch plan; P Gird (t) represents the power interaction data between the energy system and the power grid under the intraday hourly dispatch scheme; P Gird,1 (t) represents the power interaction data between the energy system and the power grid in the intraday real-time dispatch scheme; ΔP Hs,1 ΔP Hp,1 These represent the adjustment amounts for the thermal storage tank and heat pump between the real-time intraday scheduling plan and the hourly intraday scheduling plan, respectively; λ Hsi and λ Hp P represents the penalty adjustment coefficients for the thermal storage tank and the heat pump, respectively; Hs(t) represents the thermal storage tank power data for the intraday hourly scheduling plan; P Hs,1 (t) represents the power data of the thermal storage tanks in the intraday real-time scheduling scheme; P Hp (t) represents the heat pump power data of the intraday hourly scheduling plan; P Hp,1 (t) represents the heat pump power data of the real-time scheduling scheme within the day;

[0152] Based on the intraday real-time scheduling objective function, an intraday real-time scheduling model is constructed by combining the intraday real-time scheduling constraint set. The intraday real-time scheduling model is solved to obtain the solution set of energy system operation data at a 15-minute time scale. The solution set of energy system operation data at a 15-minute time scale is used as the intraday real-time scheduling scheme of the energy system.

[0153] Based on the day-ahead scheduling scheme, the intraday hourly scheduling scheme, and the intraday real-time scheduling scheme, the scheduling scheme of the energy system is determined.

[0154] This embodiment achieves coordinated optimization across multiple time scales and scheduling scales by using three hierarchical optimization objectives: minimizing the total cost of day-ahead scheduling, minimizing the deviation cost of intraday hourly scheduling schemes and day-ahead scheduling schemes, and minimizing the deviation cost of intraday real-time scheduling schemes and intraday hourly scheduling schemes. Furthermore, the optimization objectives at each level are interconnected, ensuring not only the temporal continuity of the scheduling schemes but also that the scheduling schemes can take into account both long-term and short-term scheduling, thereby improving the accuracy of energy system scheduling.

[0155] This embodiment constructs static and dynamic photovoltaic power correction models for each time scale using photovoltaic power prediction data from several different time scales of the energy system. This generates static and dynamic photovoltaic power prediction intervals, resulting in a fused photovoltaic power prediction interval. The static photovoltaic power correction model optimizes the error in photovoltaic power prediction data, while the dynamic model adapts to the volatility, randomness, and intermittency of photovoltaic power. This allows the fused photovoltaic power prediction interval to accurately define the uncertainty range of photovoltaic power, avoiding inaccurate predictions caused by fixed photovoltaic power values. Based on the fused photovoltaic power prediction intervals at different time scales, day-ahead scheduling models, intraday hourly scheduling models, and intraday real-time scheduling models for the energy system are constructed at different scheduling scales. This ensures that the energy system's scheduling scheme not only considers the photovoltaic power variation characteristics at different time scales but also integrates the actual operating conditions of the energy system under different scheduling scales, thereby improving the accuracy of energy system scheduling.

[0156] Example 2

[0157] Please refer to Figure 2 , Figure 2A schematic diagram of an energy system scheduling device based on multi-timescale interval prediction provided in an embodiment of the present invention includes: a photovoltaic power prediction data acquisition module 201, a static photovoltaic power prediction interval acquisition module 202, a dynamic photovoltaic power prediction interval acquisition module 203, a fused photovoltaic power prediction interval acquisition module 204, and an energy system scheduling module 205.

[0158] The photovoltaic power prediction data acquisition module 201 is used to acquire photovoltaic power prediction data of the energy system at several different time scales.

[0159] The static photovoltaic power prediction interval acquisition module 202 is used to construct a static photovoltaic power correction model for each time scale based on the photovoltaic power prediction data for each time scale, and obtain the static photovoltaic power prediction interval for each time scale based on the static photovoltaic power correction model.

[0160] The dynamic photovoltaic power prediction interval acquisition module 203 is used to construct a dynamic photovoltaic power correction model for each time scale based on the photovoltaic power prediction data for each time scale, and obtain the dynamic photovoltaic power prediction interval for each time scale based on the dynamic photovoltaic power correction model.

[0161] The fusion photovoltaic power prediction interval acquisition module 204 is used to determine the fusion photovoltaic power prediction interval for each time scale based on the static photovoltaic power prediction interval and the dynamic photovoltaic power prediction interval for each time scale.

[0162] The energy system scheduling module 205 is used to construct a day-ahead scheduling model, an intraday hourly scheduling model, and an intraday real-time scheduling model of the energy system based on the fused photovoltaic power prediction interval for each time scale, solve the day-ahead scheduling model, the intraday hourly scheduling model, and the intraday real-time scheduling model respectively, determine the scheduling scheme of the energy system, and schedule the energy system based on the scheduling scheme of the energy system.

[0163] In this embodiment, the static photovoltaic power prediction interval acquisition module 202 includes: a static photovoltaic power prediction interval acquisition unit; in the static photovoltaic power prediction interval acquisition unit, the photovoltaic power prediction data includes: photovoltaic power prediction value and photovoltaic power prediction error data; the static photovoltaic power prediction interval acquisition unit is used to acquire the number of mixed Gaussian components at each time scale, and construct a Gaussian mixture model for each time scale based on the number of mixed Gaussian components at each time scale; the photovoltaic power prediction error data at each time scale is input into the corresponding static photovoltaic power correction model to maximize the log-likelihood function. To optimize the objective, the Gaussian mixture model for each time scale is iterated using a pre-defined expectation-maximization algorithm to determine the static photovoltaic power correction model for each time scale, and several Gaussian components output by the static photovoltaic power correction model for each time scale are obtained. The several Gaussian components output by the static photovoltaic power correction model for each time scale are weighted and averaged to obtain the aggregated Gaussian distribution for each time scale. Centered on the photovoltaic power prediction value for each time scale, the aggregated Gaussian distribution and the photovoltaic power prediction value for each time scale are superimposed to obtain the static photovoltaic power prediction interval for each time scale.

[0164] In this embodiment, the dynamic photovoltaic power prediction interval acquisition module 203 includes: a dynamic photovoltaic power prediction interval acquisition unit; in the dynamic photovoltaic power prediction interval acquisition unit, the photovoltaic power prediction data includes: a photovoltaic power prediction input feature type set and a photovoltaic power prediction input feature sequence length; the dynamic photovoltaic power prediction interval acquisition unit is used to acquire the number of mixed Gaussian components at each time scale, and construct a mixing density model for each time scale based on the number of mixed Gaussian components at each time scale; the preset mixing density model training set is input into the mixing density model at each time scale, respectively, using the negative log-likelihood function. Using the loss function and maximizing the conditional log-likelihood function as training objectives, a mixed density model for each time scale is trained using a preset Adam optimizer to determine the dynamic photovoltaic power correction model for each time scale. Based on the photovoltaic power prediction input feature type set and the photovoltaic power prediction input feature sequence length for each time scale, preset photovoltaic power prediction input feature data is filtered to obtain dynamic photovoltaic power correction input data for each time scale. The dynamic photovoltaic power correction input data for each time scale is then input into the corresponding dynamic photovoltaic power correction model to obtain the dynamic photovoltaic power prediction interval for each time scale.

[0165] In this embodiment, the integrated photovoltaic power prediction interval acquisition module 204 includes: an integrated photovoltaic power prediction interval acquisition unit; the integrated photovoltaic power prediction interval acquisition unit is used to acquire the static photovoltaic power prediction interval weight coefficient and the dynamic photovoltaic power prediction interval coefficient; based on the static photovoltaic power prediction interval weight coefficient and the dynamic photovoltaic power prediction interval coefficient, the static photovoltaic power prediction interval and the dynamic photovoltaic power prediction interval of each time scale are weighted and integrated to obtain the integrated photovoltaic power prediction interval of each time scale.

[0166] In this embodiment, the energy system scheduling module 205 includes: an energy system scheduling unit; the energy system scheduling unit is used to acquire the electricity price data and operation data of the energy system; based on the electricity price data of the energy system, and combined with the fused photovoltaic power prediction interval for each time scale, determine the fused electricity price photovoltaic power prediction value for each time scale; based on the operation data of the energy system and the fused electricity price photovoltaic power prediction value for each time scale, construct the day-ahead scheduling constraint set, the intraday hourly scheduling constraint set, and the intraday real-time scheduling constraint set of the energy system respectively; based on the day-ahead scheduling constraint set, the intraday hourly scheduling constraint set, and the intraday real-time scheduling constraint set of the energy system, construct the day-ahead scheduling model, the intraday hourly scheduling model, and the intraday real-time scheduling model of the energy system respectively, and solve the day-ahead scheduling model, the intraday hourly scheduling model, and the intraday real-time scheduling model respectively to determine the scheduling scheme of the energy system.

[0167] In this embodiment, the energy system dispatching unit includes: a photovoltaic power prediction value acquisition subunit; in the photovoltaic power prediction value acquisition subunit, the electricity price data of the energy system includes: real-time electricity purchase price, historical minimum electricity purchase price, and historical maximum electricity purchase price for each time scale; the photovoltaic power prediction value acquisition subunit is used to determine the electricity price coefficient for each time scale based on the real-time electricity purchase price, historical minimum electricity purchase price, and historical maximum electricity purchase price for each time scale; determine the upper limit and lower limit of the integrated photovoltaic power prediction for each time scale based on the integrated photovoltaic power prediction interval for each time scale; and determine the integrated electricity price photovoltaic power prediction value for each time scale based on the electricity price coefficient, the upper limit and lower limit of the integrated photovoltaic power prediction for each time scale.

[0168] In this embodiment, the energy system scheduling unit includes a scheduling constraint set construction subunit; in the scheduling constraint set construction subunit, the time scale includes: a 15-minute time scale, a 4-hour time scale, and a 24-hour time scale; the energy system operation data includes: energy system operation data at the 15-minute time scale, energy system operation data at the 4-hour time scale, and energy system operation data at the 24-hour time scale; the scheduling constraint set construction subunit is used to construct the day-ahead scheduling constraint set of the energy system based on the energy system operation data at the 24-hour time scale, combined with the 24-hour time scale photovoltaic power prediction value of the integrated electricity price; to construct the intraday hourly scheduling constraint set of the energy system based on the energy system operation data at the 4-hour time scale, combined with the 4-hour time scale photovoltaic power prediction value of the integrated electricity price; and to construct the intraday real-time scheduling constraint set of the energy system based on the energy system operation data at the 15-minute time scale, combined with the 15-minute time scale photovoltaic power prediction value of the integrated electricity price.

[0169] In this embodiment, the energy system scheduling unit includes: a scheduling scheme acquisition subunit; the scheduling scheme acquisition subunit is used to acquire scheduling cost data of the energy system; based on the scheduling cost data of the energy system, and with minimizing the total day-ahead scheduling cost as the optimization objective, constructing a day-ahead scheduling objective function of the energy system; based on the day-ahead scheduling objective function and the day-ahead scheduling constraint set of the energy system, constructing a day-ahead scheduling model of the energy system; solving the day-ahead scheduling model of the energy system to obtain the day-ahead scheduling scheme of the energy system; constructing an intra-day hourly scheduling objective function of the energy system with the minimum deviation cost between the intra-day hourly scheduling scheme and the day-ahead scheduling scheme as the optimization objective; based on the intra-day hourly scheduling objective function of the energy system... Based on the intraday hourly scheduling constraint set, an intraday hourly scheduling model for the energy system is constructed. The intraday hourly scheduling model is solved to obtain the intraday hourly scheduling scheme for the energy system. An intraday real-time scheduling objective function for the energy system is constructed, with the minimum deviation cost of the intraday real-time scheduling scheme and the intraday hourly scheduling scheme as the optimization objective. Based on the intraday real-time scheduling objective function and the intraday real-time scheduling constraint set, an intraday real-time scheduling model for the energy system is constructed. The intraday real-time scheduling model is solved to obtain the intraday real-time scheduling scheme for the energy system. Based on the day-ahead scheduling scheme, the intraday hourly scheduling scheme, and the intraday real-time scheduling scheme, the scheduling scheme for the energy system is determined.

[0170] This embodiment constructs static and dynamic photovoltaic power correction models for each time scale using photovoltaic power prediction data from several different time scales of the energy system. This generates static and dynamic photovoltaic power prediction intervals, resulting in a fused photovoltaic power prediction interval. The static photovoltaic power correction model optimizes the error in photovoltaic power prediction data, while the dynamic model adapts to the volatility, randomness, and intermittency of photovoltaic power. This allows the fused photovoltaic power prediction interval to accurately define the uncertainty range of photovoltaic power, avoiding inaccurate predictions caused by fixed photovoltaic power values. Based on the fused photovoltaic power prediction intervals at different time scales, day-ahead scheduling models, intraday hourly scheduling models, and intraday real-time scheduling models for the energy system are constructed at different scheduling scales. This ensures that the energy system's scheduling scheme not only considers the photovoltaic power variation characteristics at different time scales but also integrates the actual operating conditions of the energy system under different scheduling scales, thereby improving the accuracy of energy system scheduling.

[0171] In summary, this invention constructs static and dynamic photovoltaic power correction models for each time scale using photovoltaic power prediction data from several different time scales of the energy system. This generates static and dynamic photovoltaic power prediction intervals, resulting in a fused photovoltaic power prediction interval. The static photovoltaic power correction model optimizes the error in photovoltaic power prediction data, while the dynamic photovoltaic power correction model adapts to the volatility, randomness, and intermittency of photovoltaic power. This allows the fused photovoltaic power prediction interval to accurately define the uncertainty range of photovoltaic power, avoiding inaccurate predictions caused by fixed photovoltaic power values. Subsequently, based on the fused photovoltaic power prediction intervals at different time scales, day-ahead scheduling models, intraday hourly scheduling models, and intraday real-time scheduling models of the energy system are constructed at different scheduling scales. This enables the energy system's scheduling scheme to not only consider the photovoltaic power variation characteristics at different time scales but also to comprehensively consider the actual operating conditions of the energy system under different scheduling scales, thereby improving the accuracy of energy system scheduling.

[0172] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An energy system scheduling method based on multi-timescale interval prediction, characterized in that, include: Obtain photovoltaic power prediction data for the energy system at several different time scales; Based on the photovoltaic power prediction data for each time scale, a static photovoltaic power correction model for each time scale is constructed, and the static photovoltaic power prediction interval for each time scale is obtained based on the static photovoltaic power correction model. Based on the photovoltaic power prediction data for each time scale, a dynamic photovoltaic power correction model for each time scale is constructed, and the dynamic photovoltaic power prediction interval for each time scale is obtained based on the dynamic photovoltaic power correction model. Based on the static photovoltaic power prediction interval and the dynamic photovoltaic power prediction interval for each time scale, the fused photovoltaic power prediction interval for each time scale is determined. Based on the fused photovoltaic power prediction interval for each time scale, a day-ahead scheduling model, an intraday hourly scheduling model, and an intraday real-time scheduling model for the energy system are constructed. The day-ahead scheduling model, the intraday hourly scheduling model, and the intraday real-time scheduling model are solved respectively to determine the scheduling scheme of the energy system. The energy system is then scheduled based on the scheduling scheme.

2. The energy system scheduling method based on multi-timescale interval prediction as described in claim 1, characterized in that, The process of constructing a static photovoltaic power correction model for each time scale based on photovoltaic power prediction data for each time scale, and obtaining a static photovoltaic power prediction interval for each time scale based on the static photovoltaic power correction model, includes: The photovoltaic power prediction data includes: photovoltaic power prediction values ​​and photovoltaic power prediction error data; Obtain the number of Gaussian mixture components at each time scale, and construct a Gaussian mixture model for each time scale based on the number of Gaussian mixture components at each time scale; The photovoltaic power prediction error data for each time scale is input into the corresponding static photovoltaic power correction model. With maximizing the log-likelihood function as the optimization objective, the Gaussian mixture model for each time scale is iterated using a preset expectation maximization algorithm to determine the static photovoltaic power correction model for each time scale and obtain several Gaussian components output by the static photovoltaic power correction model for each time scale. The aggregated Gaussian distribution for each time scale is obtained by weighted averaging of several Gaussian components output by the static photovoltaic power correction model for each time scale. Centered on the photovoltaic power prediction value for each time scale, the aggregated Gaussian distribution and the photovoltaic power prediction value for each time scale are superimposed to obtain the static photovoltaic power prediction interval for each time scale.

3. The energy system scheduling method based on multi-timescale interval prediction as described in claim 1, characterized in that, The process involves constructing a dynamic photovoltaic power correction model for each time scale based on the initial photovoltaic power prediction data for each time scale, and obtaining the dynamic photovoltaic power prediction interval for each time scale based on the dynamic photovoltaic power correction model, including: The photovoltaic power prediction data includes: a set of photovoltaic power prediction input feature types and the length of the photovoltaic power prediction input feature sequence; Obtain the number of Gaussian mixture components at each time scale, and construct a mixture density model for each time scale based on the number of Gaussian mixture components at each time scale; The preset hybrid density model training set is input into the hybrid density model for each time scale. The negative log-likelihood function is used as the loss function and the conditional log-likelihood function is maximized as the training objective. The preset Adam optimizer is used to train the hybrid density model for each time scale to determine the dynamic photovoltaic power correction model for each time scale. Based on the set of photovoltaic power prediction input feature types and the length of photovoltaic power prediction input feature sequence for each time scale, the preset photovoltaic power prediction input feature data is filtered to obtain dynamic photovoltaic power correction input data for each time scale. The dynamic photovoltaic power correction input data for each time scale is input into the corresponding dynamic photovoltaic power correction model to obtain the dynamic photovoltaic power prediction interval for each time scale.

4. An energy system scheduling method based on multi-timescale interval prediction as described in any one of claims 1 to 3, characterized in that, The determination of the fused photovoltaic power prediction interval for each time scale, based on the static photovoltaic power prediction interval and the dynamic photovoltaic power prediction interval for each time scale, includes: Obtain the weighting coefficients for the static photovoltaic power prediction interval and the coefficients for the dynamic photovoltaic power prediction interval; Based on the weighting coefficient of the static photovoltaic power prediction interval and the coefficient of the dynamic photovoltaic power prediction interval, the static photovoltaic power prediction interval and the dynamic photovoltaic power prediction interval of each time scale are weighted and fused to obtain the fused photovoltaic power prediction interval of each time scale.

5. The energy system scheduling method based on multi-timescale interval prediction as described in claim 1, characterized in that, Based on the fused photovoltaic power prediction interval for each time scale, a day-ahead scheduling model, an intraday hourly scheduling model, and an intraday real-time scheduling model of the energy system are constructed. The day-ahead scheduling model, the intraday hourly scheduling model, and the intraday real-time scheduling model are solved respectively to determine the scheduling scheme of the energy system, including: Obtain the electricity price data and operation data of the energy system; Based on the electricity price data of the energy system, and combined with the integrated photovoltaic power prediction interval for each time scale, the integrated electricity price photovoltaic power prediction value for each time scale is determined. Based on the operating data of the energy system and the predicted photovoltaic power of the integrated electricity price at each time scale, the day-ahead scheduling constraint set, the intraday hourly scheduling constraint set, and the intraday real-time scheduling constraint set of the energy system are constructed respectively. Based on the day-ahead scheduling constraint set, intraday hourly scheduling constraint set, and intraday real-time scheduling constraint set of the energy system, the day-ahead scheduling model, intraday hourly scheduling model, and intraday real-time scheduling model of the energy system are constructed respectively. The day-ahead scheduling model, intraday hourly scheduling model, and intraday real-time scheduling model are solved respectively to determine the scheduling scheme of the energy system.

6. The energy system scheduling method based on multi-timescale interval prediction as described in claim 5, characterized in that, The method of determining the fused electricity price photovoltaic power prediction value for each time scale, based on the electricity price data of the energy system and combined with the fused photovoltaic power prediction interval for each time scale, includes: The electricity price data of the energy system includes: real-time electricity purchase price, historical minimum electricity purchase price, and historical maximum electricity purchase price for each time scale; Based on the real-time electricity purchase price, the lowest historical electricity purchase price, and the highest historical electricity purchase price for each time scale, determine the electricity price coefficient for each time scale; Based on the fusion photovoltaic power prediction interval for each time scale, determine the upper limit and lower limit of the fusion photovoltaic power prediction for each time scale. Based on the electricity price coefficient, the upper limit of the integrated photovoltaic power prediction, and the lower limit of the integrated photovoltaic power prediction for each time scale, the integrated electricity price photovoltaic power prediction value for each time scale is determined.

7. An energy system scheduling method based on multi-timescale interval prediction as described in claim 5 or 6, characterized in that, Based on the operating data of the energy system and the predicted photovoltaic power at each time scale, the day-ahead scheduling constraint set, the intraday hourly scheduling constraint set, and the intraday real-time scheduling constraint set of the energy system are constructed, respectively, including: The time scales include: 15-minute time scale, 4-hour time scale and 24-hour time scale; The energy system's operational data includes: energy system operational data on a 15-minute timescale, energy system operational data on a 4-hour timescale, and energy system operational data on a 24-hour timescale. Based on the energy system operation data at the 24-hour time scale, and combined with the photovoltaic power prediction value of the integrated electricity price at the 24-hour time scale, a set of day-ahead scheduling constraints for the energy system is constructed. Based on the energy system operation data at the 4-hour time scale, and combined with the photovoltaic power prediction value of the integrated electricity price at the 4-hour time scale, a set of intraday hourly scheduling constraints for the energy system is constructed. Based on the energy system operation data at the 15-minute time scale, and combined with the photovoltaic power prediction value of the integrated electricity price at the 15-minute time scale, a set of intraday real-time scheduling constraints for the energy system is constructed.

8. The energy system scheduling method based on multi-timescale interval prediction as described in claim 7, characterized in that, Based on the day-ahead scheduling constraint set, intraday hourly scheduling constraint set, and intraday real-time scheduling constraint set of the energy system, a day-ahead scheduling model, an intraday hourly scheduling model, and an intraday real-time scheduling model of the energy system are constructed, respectively. The day-ahead scheduling model, the intraday hourly scheduling model, and the intraday real-time scheduling model are solved to determine the scheduling scheme of the energy system, including: Obtain the scheduling cost data of the energy system; Based on the scheduling cost data of the energy system, a day-ahead scheduling objective function of the energy system is constructed with the goal of minimizing the total day-ahead scheduling cost. Based on the day-ahead scheduling objective function and day-ahead scheduling constraint set of the energy system, a day-ahead scheduling model of the energy system is constructed; Solve the day-ahead scheduling model of the energy system to obtain the day-ahead scheduling scheme of the energy system; The objective function for intraday hourly scheduling of the energy system is constructed with the minimum deviation cost of intraday hourly scheduling scheme and day-ahead scheduling scheme as the optimization objective. Based on the intraday hourly scheduling objective function and intraday hourly scheduling constraint set of the energy system, an intraday hourly scheduling model of the energy system is constructed. Solve the intraday hourly scheduling model of the energy system to obtain the intraday hourly scheduling scheme of the energy system; The objective function for the intraday real-time scheduling of the energy system is constructed with the minimum deviation cost of the intraday real-time scheduling scheme and the intraday hourly scheduling scheme as the optimization objective. Based on the intraday real-time scheduling objective function and intraday real-time scheduling constraint set of the energy system, an intraday real-time scheduling model of the energy system is constructed. Solve the intraday real-time scheduling model of the energy system to obtain the intraday real-time scheduling scheme of the energy system; Based on the day-ahead scheduling scheme, intraday hourly scheduling scheme, and intraday real-time scheduling scheme of the energy system, the scheduling scheme of the energy system is determined.

9. An energy system dispatching device based on multi-timescale interval prediction, characterized in that, include: The system includes a photovoltaic power prediction data acquisition module, a static photovoltaic power prediction range acquisition module, a dynamic photovoltaic power prediction range acquisition module, a fused photovoltaic power prediction range acquisition module, and an energy system scheduling module. The photovoltaic power prediction data acquisition module is used to acquire photovoltaic power prediction data of the energy system at several different time scales. The static photovoltaic power prediction interval acquisition module is used to construct a static photovoltaic power correction model for each time scale based on the photovoltaic power prediction data for each time scale, and obtain the static photovoltaic power prediction interval for each time scale based on the static photovoltaic power correction model. The dynamic photovoltaic power prediction interval acquisition module is used to construct a dynamic photovoltaic power correction model for each time scale based on the photovoltaic power prediction data for each time scale, and obtain the dynamic photovoltaic power prediction interval for each time scale based on the dynamic photovoltaic power correction model. The integrated photovoltaic power prediction interval acquisition module is used to determine the integrated photovoltaic power prediction interval for each time scale based on the static photovoltaic power prediction interval and the dynamic photovoltaic power prediction interval for each time scale. The energy system scheduling module is used to construct a day-ahead scheduling model, an intraday hourly scheduling model, and an intraday real-time scheduling model of the energy system based on the fused photovoltaic power prediction interval for each time scale. The module solves the day-ahead scheduling model, the intraday hourly scheduling model, and the intraday real-time scheduling model to determine the scheduling scheme of the energy system, and schedules the energy system based on the scheduling scheme.

10. An energy system dispatching device based on multi-timescale interval prediction as described in claim 9, characterized in that, The static photovoltaic power prediction range acquisition module includes: a static photovoltaic power prediction range acquisition unit; In the static photovoltaic power prediction range acquisition unit, the photovoltaic power prediction data includes: photovoltaic power prediction value and photovoltaic power prediction error data; The static photovoltaic power prediction interval acquisition unit is used to acquire the number of mixed Gaussian components at each time scale, and to construct a Gaussian mixture model for each time scale based on the number of mixed Gaussian components at each time scale. The photovoltaic power prediction error data for each time scale is input into the corresponding static photovoltaic power correction model. With maximizing the log-likelihood function as the optimization objective, the Gaussian mixture model for each time scale is iterated using a preset expectation maximization algorithm to determine the static photovoltaic power correction model for each time scale and obtain several Gaussian components output by the static photovoltaic power correction model for each time scale. The aggregated Gaussian distribution for each time scale is obtained by weighted averaging of several Gaussian components output by the static photovoltaic power correction model for each time scale. Centered on the photovoltaic power prediction value for each time scale, the aggregated Gaussian distribution and the photovoltaic power prediction value for each time scale are superimposed to obtain the static photovoltaic power prediction interval for each time scale.

Images