Distributed power generation power prediction method, system and device and medium

Abstract

The invention discloses a distributed power generation power prediction method, system and device and a medium, and belongs to the technical field of power generation power prediction.The method comprises the steps that wind speed data and radiation intensity data during power generation of a distributed power supply are collected, and a wind speed influence coefficient and radiation conversion efficiency are calculated based on the wind speed data and the radiation intensity data; the method comprises the following steps: constructing a distributed power supply prediction network comprising an input processing layer, a feature mapping layer, a space-time fusion layer, an optimization feedback layer and a prediction output layer, generating a generated power prediction value, and calculating a confidence interval of a generated power prediction result through the generated power prediction value based on a Bayesian inference method. According to the method, the problem of uncertainty in prediction of the generated power of the distributed power supply is effectively solved by combining spatial-temporal feature extraction and a Bayesian inference method. Compared with a traditional method, the method has the advantages that the volatility and interdependence of radiation conversion and wind speed influence on time and space dimensions can be accurately captured, so that the precision of power generation power prediction is improved.

Description

Technical Field

[0001] This invention relates to the field of power generation prediction technology, specifically to a method, system, device, and medium for predicting the power generation of distributed power sources. Background Technology

[0002] Compared to traditional centralized power grids, distributed generation has the following characteristics: wide distribution, high renewability, and environmental friendliness. However, due to its temporal and spatial volatility, the power output of distributed generation is unstable and difficult to predict, posing a significant challenge to the safe and stable operation of the power grid.

[0003] In traditional power systems, power generation is typically predicted and dispatched by centralized generating units. However, the unpredictability of distributed generation increases the complexity of grid dispatch. This is particularly true in the utilization of wind and solar energy, where weather changes, solar radiation intensity, and wind speed fluctuations directly impact power generation. Therefore, accurate and real-time prediction of distributed generation is crucial for grid load dispatch, backup power activation and deactivation, and grid stability.

[0004] Existing power generation prediction technologies typically rely on statistical models or machine learning methods. Traditional statistical methods, such as time series analysis and regression analysis, can achieve certain predictive results in some scenarios, but they often contain significant errors because they cannot effectively handle the nonlinear characteristics and spatiotemporal dependencies of distributed power generation. On the other hand, although machine learning methods (such as support vector machines and neural networks) can capture nonlinear relationships better, they still face problems such as poor data quality, difficulty in feature selection, and high computational complexity. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by this invention is: how to improve the accuracy of distributed power generation prediction, especially under the influence of the fluctuation and spatiotemporal dependence of environmental factors such as wind speed and radiation intensity, how to accurately predict power generation, and how to ensure the stable satisfaction of grid load demand by adjusting the prediction model in real time.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for predicting the power generation of distributed power sources, comprising, Collect wind speed and radiation intensity data during distributed power generation, and calculate the wind speed influence coefficient and radiation conversion efficiency based on the wind speed and radiation intensity data; A distributed power prediction network is constructed, comprising an input processing layer, a feature mapping layer, a spatiotemporal fusion layer, an optimization feedback layer, and a prediction output layer; The input processing layer receives radiation conversion efficiency and wind speed influence coefficient as input features, performs standardization processing, outlier removal and missing value imputation to generate clean data. The feature mapping layer performs nonlinear mapping on the clean data through radial basis function kernels, transforming the clean data from the original feature space into a high-dimensional feature space and outputting a high-dimensional feature representation; The spatiotemporal fusion layer extracts the dependence of wind speed influence coefficient and radiation conversion efficiency on time and space dimensions in high-dimensional feature representation through convolution operation, and performs dimensionality reduction processing on the output of convolution operation through pooling operation to generate spatiotemporal feature representation. The optimized feedback layer compares the spatiotemporal characteristics with the actual power generation, calculates the error, and optimizes the spatiotemporal characteristic representation. The prediction output layer generates power generation prediction values ​​through linear transformation based on the optimized spatiotemporal feature representation, and calculates the confidence interval of the power generation prediction results based on the power generation prediction values ​​using the Bayesian inference method.

[0008] As a preferred embodiment of the distributed power generation prediction method of the present invention, the calculation of the wind speed influence coefficient and radiation conversion efficiency includes considering the influence of air density and determining the effective transmission of light, expressed as: in, This indicates the corrected effective light intensity. Indicates real-time radiation intensity. Indicates real-time air density. Indicates standard air density, Indicates real-time air pressure. Indicates real-time temperature. Represents the gas constant; Temperature correction formula for photovoltaic panel efficiency: in, This indicates the efficiency of the photovoltaic panel, taking temperature correction into account. This indicates the efficiency of the photovoltaic panel at standard temperature. Indicates the temperature coefficient of the photovoltaic panel. This indicates the difference between the photovoltaic panel temperature and the standard temperature. Indicates the temperature of the photovoltaic panel. Indicates standard temperature; Based on the corrected effective light intensity and the photovoltaic panel efficiency considering temperature correction, the radiative conversion efficiency is determined and expressed as: in, Indicates radiative conversion efficiency. Indicates standard radiation intensity.

[0009] In a preferred embodiment of the distributed power generation prediction method of the present invention, the calculation of the wind speed influence coefficient and radiation conversion efficiency further includes calculating the wind speed change rate using the maximum, minimum, and average wind speeds, expressed as: in, Indicates the rate of change of wind speed. This indicates the maximum wind speed. This represents the minimum wind speed. This represents the average wind speed. Based on the rate of change of wind speed per unit time, the degree of wind speed change is described, and the rate of change of wind speed is calculated and expressed as: in, For time intervals, This indicates the wind speed at the current moment. Indicates the wind speed at the previous moment; Based on the rate of change of wind speed and the wind speed influence coefficient, the wind speed influence coefficient is calculated and expressed as follows: in, Indicates the wind speed influence coefficient. This indicates the ratio of wind speed fluctuations to the actual wind speed. It represents the ratio of the rate of change of wind speed to the wind speed.

[0010] As a preferred embodiment of the distributed power generation prediction method of the present invention, the output high-dimensional feature representation includes calculating the Euclidean distance for each set of data points and quantifying the differences in radiation conversion efficiency and wind speed influence coefficient. The Euclidean distance is expressed as: in, Indicates a point in time and time point Euclidean distance, Indicates a point in time Radiative conversion efficiency, Indicates a point in time The wind speed influence coefficient, This represents the radiative conversion efficiency of data point i. This represents the radiative conversion efficiency of data point j. This represents the wind speed influence coefficient at data point i. This represents the wind speed influence coefficient at data point j; The calculated Euclidean distance is mapped using a radial basis function kernel to transfer the original features to a high-dimensional feature space, as shown below: in, This represents the result of mapping in a high-dimensional feature space. For kernel width; Constructing the kernel matrix nuclear matrix The dimension is , The number of input data samples, kernel matrix Each element This indicates the temporal dependence and spatial similarity between radiation conversion efficiency and wind speed influence coefficient; Through the constructed kernel matrix The radiation conversion efficiency and wind speed influence coefficient are mapped to a high-dimensional feature space.

[0011] In a preferred embodiment of the distributed power generation prediction method of the present invention, the confidence interval for calculating the power generation prediction result includes obtaining the prediction error by calculating the difference between the predicted power generation and the actual power generation. Define the prior distribution of the error, assume that the error follows a normal distribution, set the mean of the error to be zero, and set the standard deviation of the error to be the standard deviation of the prediction error calculated from historical data; By combining the prior distribution with the actual observed power generation error using Bayes' theorem, the updated posterior distribution is calculated. Based on the updated posterior distribution, the mean and standard deviation of the predicted power generation are calculated, the critical value is calculated by selecting a confidence level, and the confidence interval of the predicted power generation is obtained.

[0012] As a preferred embodiment of the distributed power generation prediction method of the present invention, the calculation of the confidence interval of the power generation prediction result further includes comparing the power generation prediction value with the actual grid demand to determine whether the grid load demand is met; if the power generation prediction value is greater than or equal to the actual grid demand, the current output is maintained. If the predicted power generation is less than the actual grid demand, the number of standby units that need to be activated is calculated based on the gap between the predicted power generation and the actual grid demand and the width of the confidence interval. Based on the calculated number of standby generating units, the power generation capacity is adjusted by controlling the start and stop of the standby generating units.

[0013] As a preferred embodiment of the distributed power generation prediction method of the present invention, the calculation of the number of standby units to be activated includes, based on the confidence interval, calculating the width of the confidence interval, expressed as: in, Indicates the width of the confidence interval; By comparing the predicted power generation with the actual grid demand, the difference between the two is calculated and expressed as: in, This indicates the difference between the predicted power generation and the actual grid demand; Based on the difference between the predicted power generation and the actual grid demand and confidence interval width The number of standby units is calculated and expressed as follows: in, Indicates the number of standby units. This is the maximum generating capacity of each standby unit.

[0014] This invention provides a distributed power generation prediction system.

[0015] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a distributed power generation power prediction system, comprising: a data acquisition unit, used to collect environmental data during the distributed power generation process, including wind speed data and radiation intensity data; The data processing unit is used to calculate the wind speed influence coefficient and radiation conversion efficiency based on wind speed data and radiation intensity data; The distributed power generation prediction network consists of an input processing layer, a feature mapping layer, a spatiotemporal fusion layer, an optimization feedback layer, and a prediction output layer. The dispatch control unit calculates and adjusts the start-up and shutdown status of standby generators in the power grid to ensure that the power grid load demand is met.

[0016] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the distributed power generation prediction method.

[0017] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the distributed power generation prediction method.

[0018] The beneficial effects of this invention are as follows: By combining spatiotemporal feature extraction and Bayesian inference methods, it effectively solves the uncertainty problem in distributed generation power prediction. Compared with traditional methods, this invention can accurately capture the fluctuations and interdependencies of radiation conversion and wind speed effects in the temporal and spatial dimensions, thereby improving the accuracy of power prediction. When the grid load is insufficient, based on the confidence interval and prediction error, it can accurately calculate and activate standby units, thereby ensuring the stability of grid power supply. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 The above is a flowchart of a distributed power generation power prediction method provided in one embodiment of the present invention. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0022] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a method for predicting the power generation of distributed power sources, including: Collect wind speed and radiation intensity data during distributed power generation, and calculate the wind speed influence coefficient and radiation conversion efficiency based on the wind speed and radiation intensity data; Considering the effect of air density, the effective transmission of light is determined as follows: in, This indicates the corrected effective light intensity. Indicates real-time radiation intensity. Indicates real-time air density. Indicates standard air density, This indicates real-time air pressure (unit: Pa). Indicates real-time temperature. This represents the gas constant.

[0023] The efficiency of photovoltaic panels is affected by their operating temperature; the higher the temperature, the lower the efficiency. Photovoltaic panel temperature Compared with standard temperature Differences between It is a key factor affecting its efficiency. This is determined by the temperature coefficient. To correct the efficiency of photovoltaic panels.

[0024] Temperature correction formula for photovoltaic panel efficiency: in, This indicates the efficiency of the photovoltaic panel, taking temperature correction into account. This indicates the efficiency of the photovoltaic panel at standard temperature. Indicates the temperature coefficient of the photovoltaic panel. This indicates the difference between the photovoltaic panel temperature and the standard temperature. Indicates the temperature of the photovoltaic panel. Indicates the standard temperature.

[0025] Based on the corrected effective light intensity and the photovoltaic panel efficiency considering temperature correction, the radiative conversion efficiency is determined and expressed as: in, Indicates radiative conversion efficiency. Indicates standard radiation intensity.

[0026] It should be noted that the efficiency of the photovoltaic panel at standard temperature... The power conversion efficiency of the photovoltaic panel is measured under standard test conditions, including a light intensity of [insert value here]. The temperature of the photovoltaic panel is And the measurements were performed under standard spectral conditions. Standard radiation intensity The output power of a photovoltaic (PV) panel, as incident light energy, is obtained by measuring the output power of the PV panel under standard conditions. The efficiency of the PV panel under standard conditions is calculated using the following formula: in, The electrical power output of the photovoltaic panel. The incident radiation power under standard conditions. This refers to the effective area of ​​the photovoltaic panel.

[0027] Temperature coefficient of photovoltaic panels This describes the characteristics of photovoltaic panel efficiency as a function of temperature; the efficiency of the photovoltaic panel decreases as the temperature rises. Temperature coefficient. This represents the percentage decrease in photovoltaic panel efficiency for every 1°C increase in temperature. This value is usually negative, indicating a decrease in efficiency with increasing temperature. The temperature coefficient of a photovoltaic panel can be determined experimentally by testing the panel under different temperature conditions.

[0028] During the experiment, the photovoltaic panel's efficiency was measured for the first time at a standard temperature of 25°C, and the efficiency of the photovoltaic panel at the standard temperature was recorded. Then, by heating the photovoltaic panels and adjusting their temperature to different ambient temperatures, their efficiency was measured at these different temperatures. Based on the data of temperature and efficiency changes, the temperature coefficient was calculated using the following formula: in, The efficiency change caused by temperature variation. This indicates the efficiency of the photovoltaic panel at the current operating temperature. This indicates the efficiency of the photovoltaic panel at 25°C.

[0029] In actual operation, changes in wind speed affect the efficiency of the fan. The efficiency of the fan typically decreases under both low and high wind speed conditions. For example, at low wind speeds, the fan may not be able to reach its starting speed, while at high wind speeds, the fan's power output may be suppressed due to technical limitations (such as rated wind speed).

[0030] The rate of change of wind speed is calculated using the maximum, minimum, and average wind speeds, and is expressed as: in, Indicates the rate of change of wind speed. This indicates the maximum wind speed. This represents the minimum wind speed. This represents the average wind speed.

[0031] Based on the rate of change of wind speed per unit time, the degree of wind speed change is described, and the rate of change of wind speed is calculated and expressed as: in, Indicates the rate of change of wind speed. For time intervals, This indicates the wind speed at the current moment. This indicates the wind speed at the previous moment.

[0032] Based on the rate of change of wind speed and the wind speed influence coefficient, the wind speed influence coefficient is calculated and expressed as follows: in, Indicates the wind speed influence coefficient. This indicates the ratio of wind speed fluctuations to the actual wind speed. It represents the ratio of the rate of change of wind speed to the wind speed.

[0033] It should be noted that traditional wind power generation forecasting methods often use average or instantaneous wind speed values ​​to estimate power generation capacity. However, this method fails to consider the impact of wind speed fluctuations and rates of change on power output, resulting in significant errors when wind speed variations are large. Traditional methods assume a linear relationship between wind speed and power generation, which limits the effective handling of complex wind speed fluctuations, leading to inaccurate power generation predictions during sudden or drastic wind speed changes.

[0034] This invention overcomes the shortcomings of traditional methods by introducing the calculation of wind speed variation rate and variation rate. Specifically, by calculating the fluctuation of wind speed and the wind speed variation rate, this invention dynamically adjusts the impact of wind speed on power generation, thereby more accurately capturing the impact of wind speed fluctuations on power generation capacity. It can promptly correct prediction results when wind speed fluctuations are large, making wind power generation prediction more accurately reflect the actual impact of wind speed on power output.

[0035] Traditional photovoltaic (PV) power generation efficiency calculations are typically based on standard conditions of light intensity and PV panel temperature, neglecting the impact of air temperature and density on PV panel efficiency. Especially under conditions of significant temperature and air pressure fluctuations, traditional methods result in substantial errors in PV power generation efficiency estimates. Traditional methods generally assume that PV panel efficiency decreases with increasing temperature, but fail to adequately account for changes in air pressure and density, making it difficult to accurately capture the actual efficiency changes in extreme weather or high-temperature environments.

[0036] This invention dynamically adjusts the efficiency of photovoltaic (PV) panels based on their actual operating environment by using air density and PV panel temperature as correction factors. Specifically, it incorporates real-time air pressure and temperature data to correct the PV panel's efficiency value, ensuring it accurately reflects the power generation capacity under different meteorological conditions. This method not only considers the impact of temperature on PV panel efficiency but also supplements it with the influence of air density, enabling accurate estimation of PV panel efficiency changes and thus providing the power grid with more precise PV power generation capacity prediction data.

[0037] A distributed power prediction network is constructed, comprising an input processing layer, a feature mapping layer, a spatiotemporal fusion layer, an optimization feedback layer, and a prediction output layer.

[0038] The input processing layer receives radiation conversion efficiency and wind speed influence coefficient as input features, performs standardization, outlier removal and missing value imputation to generate clean data; The feature mapping layer performs nonlinear mapping on the clean data through radial basis function kernels, transforming the clean data from the original feature space into a high-dimensional feature space and outputting a high-dimensional feature representation; The spatiotemporal fusion layer extracts the dependence of wind speed influence coefficient and radiation conversion efficiency on the temporal and spatial dimensions in the high-dimensional feature representation through convolution operations, and performs dimensionality reduction processing on the output of the convolution operation through pooling operations to generate spatiotemporal feature representations. The feedback layer is optimized by comparing the spatiotemporal characteristics with the actual power generation, calculating the error, and optimizing the spatiotemporal characteristic representation. The prediction output layer generates power generation prediction values ​​through linear transformation based on the optimized spatiotemporal feature representation. Then, it calculates the confidence interval of the power generation prediction results by combining Bayesian inference methods.

[0039] Specifically, the feature mapping layer includes the following steps: The Euclidean distance is calculated for each pair of data points to quantify the differences in radiation conversion efficiency and wind speed influence coefficient. The Euclidean distance formula is expressed as: in, Indicates a point in time and time point Euclidean distance, Indicates a point in time Radiative conversion efficiency, Indicates a point in time The wind speed influence coefficient, This represents the radiative conversion efficiency of data point i. This represents the radiative conversion efficiency of data point j. This represents the wind speed influence coefficient at data point i. This represents the wind speed influence coefficient at data point j.

[0040] The calculated Euclidean distance is mapped using a radial basis function kernel to transfer the original features to a high-dimensional feature space, as shown below: in, This represents the result of mapping in a high-dimensional feature space. The kernel width.

[0041] Constructing the kernel matrix The dimension of the kernel matrix is ,in The number of input data samples, kernel matrix Each element Reflects the sample and The similarity in the high-dimensional feature space represents the temporal dependence and spatial similarity between radiation conversion efficiency and wind speed influence coefficient.

[0042] Through the constructed kernel matrix The radiation conversion efficiency and wind speed influence coefficient are mapped to a high-dimensional feature space.

[0043] It should be noted that in traditional feature mapping processes, the kernel width is generally a fixed value, which is insufficient to fully capture the fluctuations in wind speed and radiation intensity. In this embodiment of the invention, the kernel width will be adjusted according to the rate of change of wind speed. Adjustments are made based on changes in radiation intensity.

[0044] Radiation intensity fluctuations are calculated by comparing the radiation intensity at the current moment with that at the previous moment: in, Indicates the fluctuation of radiation intensity. This indicates the radiation intensity at the current moment. This indicates the radiation intensity at the previous moment.

[0045] Thresholds are dynamically determined by calculating the average and standard deviation of wind speed change rate and radiation intensity fluctuation in historical data. This allows for subsequent adjustments to the kernel width based on the actual fluctuations in wind speed and radiation intensity over each time period. The wind speed change rate threshold and radiation intensity fluctuation threshold are set by calculating the average and standard deviation, as shown in the following formulas: in, The average value representing the rate of change of wind speed. The fluctuation of radiation intensity is represented by the average value. This represents the adjustment factor, ranging from 1.5 to 2. The standard deviation of the rate of change of wind speed Radiation intensity fluctuation is represented by standard deviation. This represents the threshold for the rate of change of wind speed. This represents the threshold for radiation intensity fluctuation.

[0046] Next, based on the wind speed change rate threshold and the radiation intensity fluctuation threshold, the wind speed change rate and radiation intensity fluctuation are compared at each time step. When both the wind speed change rate and radiation intensity fluctuation are greater than the wind speed change rate threshold and the radiation intensity fluctuation threshold, the kernel width is increased, as shown below: in, The initial kernel width, This represents the adjustment factor for increasing the kernel width, and is taken as 1.5-2; Otherwise, reducing the kernel width is represented as: in, This represents the adjustment factor for reducing the kernel width, which is set to 1.5-2.

[0047] Furthermore, the spatiotemporal fusion layer includes the following steps: The input data is received from the feature mapping layer. The input data contains high-dimensional feature representations of radiative conversion efficiency and wind speed influence coefficient, represented as a three-dimensional matrix. Its size is ,in, This indicates the number of samples in the time dimension, representing data at different points in time. This indicates the number of samples in the spatial dimension, representing feature data from different spatial locations. This indicates the number of features for each sample, including radiation conversion efficiency and wind speed influence coefficient.

[0048] In the spatiotemporal fusion layer, 3D convolution operations are used to extract the interdependence between wind speed influence coefficients and radiation conversion efficiency in both temporal and spatial dimensions. This effectively captures the local features of spatiotemporal data and ensures that these features accurately reflect the dynamic changes between different time points and spatial locations. The following 3D convolution formula is executed: in, This represents the output feature of the convolution operation, located at a position in the spatiotemporal feature map. , Representing a three-dimensional matrix The feature data in the data are specifically as follows: Location feature value Represents the convolution kernel, with a size of It is used to extract spatiotemporal features in the time, space, and feature dimensions. This indicates the position of the convolution kernel in the time dimension. This indicates the position of the convolution kernel in the spatial dimension. This indicates the position of the convolution kernel along the feature dimension. This indicates the boundary range of the convolution kernel in the time dimension. This indicates the boundary range of the convolution kernel in the spatial dimension. This indicates the boundary range of the convolution kernel in the feature dimension.

[0049] During convolution, the size and weights of the convolution kernel are adaptively adjusted according to the spatiotemporal volatility and feature changes of the input data. For example, when wind speed fluctuates significantly, the window size of the convolution kernel will increase to better capture the changes; in regions where data changes are minimal, the convolution kernel window will decrease to avoid unnecessary computation and improve processing efficiency.

[0050] Pooling is performed on the 3D convolution results using an adaptive pooling method. The specific pooling formula is as follows: in, This represents the spatiotemporal characteristics after pooling. This indicates the size of the pooling window, which is dynamically adjusted based on the rate of change of wind speed and the fluctuation of radiation intensity. The feature values ​​output by the convolution. The pooling operation represents the elements within a sliding window. The pooling window iterates through different locations in the convolution output feature map, pooling the feature values ​​around each location. Perform a weighted summation.

[0051] The size of the pooling window is related to the spatiotemporal volatility of the data. For areas with large wind speed fluctuations, the pooling window will be increased to ensure that sufficient spatiotemporal features are retained; while for areas with smaller fluctuations, the pooling window will be decreased to reduce computation and remove redundant features.

[0052] After 3D convolution and adaptive pooling, the resulting output feature matrix is ​​a spatiotemporal feature representation, containing the dependencies of radiative conversion efficiency and wind speed influence coefficient in both time and space dimensions. The dimensions of the output spatiotemporal features are... ,in, The time dimension after pooling. The spatial dimension after pooling. This represents the feature dimension after pooling.

[0053] Further optimization of the feedback layer includes the following steps: In the optimization feedback layer, the spatiotemporal feature representation output from the spatiotemporal fusion layer is first received. This spatiotemporal feature representation is then compared with the actual power generation data to calculate the prediction error between the two. The actual power generation data comes from historical or real-time measurements and represents the actual power output. The error is calculated by comparing the difference between the predicted power and the actual power at each time point and spatial location, thus obtaining an error metric.

[0054] Based on the calculated error, a loss function is designed to evaluate performance. The loss function combines the prediction error and a regularization term. The regularization term is used to constrain the model's complexity and prevent overfitting. The purpose of the loss function is to minimize the error between the predicted results and the actual power generation by continuously adjusting the model parameters.

[0055] Spatiotemporal feature representations are updated using the backpropagation algorithm based on the gradient information of the loss function. The convolution kernels are then adjusted.

[0056] The predicted power generation value is generated through linear transformation and is expressed as follows: in, This represents the predicted power generation value. This represents the optimized spatiotemporal feature representation. Represents the weight vector. This indicates the bias term.

[0057] The confidence interval for the power generation prediction results is calculated using the Bayesian inference method, including the following steps: Obtain the predicted power generation value and get the actual power generation value from the real-time monitoring system.

[0058] The calculation error, i.e. the difference between the predicted power and the actual generated power, is expressed as: in, Indicates the prediction error. For at any time The predicted power generation capacity, For at any time The actual power generation capacity.

[0059] By analyzing historical data, the prediction error can be reduced. Define a prior distribution such that the prediction error follows a normal distribution, and the mean of the prior distribution is... 0, standard deviation This is estimated using historical data. According to Bayesian theory, this prior distribution is used as a preliminary estimate of the error, expressed as: in, Describe the prior distribution, Indicates a normal distribution. The variance of the error calculated from historical data.

[0060] Based on actual power generation data and prediction results, the posterior distribution of the error is updated according to Bayes' theorem. Bayes' theorem combines the prior distribution and observed data to derive the posterior distribution, thereby correcting the error distribution, as expressed in: in, Let be the posterior distribution of the error, representing the update of the prediction error after observing the actual power generation. It is the likelihood function, representing the prediction error. Under these conditions, the actual power generation was observed. The probability, It is the marginal likelihood.

[0061] Based on the posterior distribution, the confidence interval for the predicted power is calculated, which quantifies the uncertainty of the prediction result. By calculating the mean and standard deviation of the error and using the critical value of the standard normal distribution, the confidence interval for the power generation is obtained.

[0062] The mean and standard deviation of the posterior distribution are calculated from the expectation and variance of the posterior distribution, representing the most likely value and uncertainty of the predicted power, respectively, and are expressed as: in, Let represent the mean of the posterior distribution. This represents the standard deviation of the posterior distribution. Indicates the expected value. Indicates variance.

[0063] Calculate the confidence interval by selecting the confidence level. For The confidence level, using the critical value of the standard normal distribution, is used to calculate the confidence interval for the predicted power, expressed as: in, Indicates the confidence interval. The critical value for the standard normal distribution (for confidence level, ).

[0064] Based on the calculated confidence interval, the confidence interval of the predicted power generation value is output, that is, the confidence range of each prediction result is given.

[0065] The predicted power generation is compared with the actual grid demand to determine whether the grid load demand is met. If the predicted power generation is greater than or equal to the actual grid demand, the current output is maintained. If the predicted power generation is less than the actual grid demand, the number of standby units that need to be activated is calculated based on the gap between the predicted power generation and the actual grid demand and the width of the confidence interval. Based on the calculated number of standby generating units, the power generation capacity is adjusted by controlling the start and stop of the standby generating units.

[0066] Specifically, based on the confidence interval, the width of the confidence interval is calculated and expressed as: in, This indicates the width of the confidence interval.

[0067] By comparing the predicted power generation with the actual grid demand, the difference between the two is calculated and expressed as: in, This indicates the difference between the predicted power generation and the actual grid demand.

[0068] Based on the difference between the predicted power generation and the actual grid demand and confidence interval width The number of standby units is calculated and expressed as follows: in, Indicates the number of standby units. It is the maximum generating power of each standby unit (i.e., the maximum output power of a single standby unit).

[0069] Based on the calculated number of standby units The start / stop status of all standby generating units: if the number of standby generating units is positive, additional units are activated to meet the grid demand; if the number of standby generating units is zero or negative, no additional units need to be activated.

[0070] Traditional feature mapping methods use a fixed kernel width for radial basis function mapping. This method cannot adapt to dynamic changes in wind speed and radiation intensity, resulting in poor adaptability of the mapping results in the face of environmental fluctuations. This invention addresses this problem by dynamically adjusting the kernel width. The kernel width is adjusted based on real-time data of wind speed fluctuations and radiation intensity changes, accurately adapting to environmental changes. In this way, the mapped high-dimensional feature space can effectively capture the uncertainties of wind speed and radiation intensity in both time and space, providing more accurate power generation predictions under extreme weather or sudden events, improving the accuracy of distributed power generation prediction, and avoiding the errors caused by the fixed kernel width of traditional methods.

[0071] Traditional convolutional neural networks typically use two-dimensional convolutional layers to process spatiotemporal data, but neglect the complex spatiotemporal dependencies between wind speed and radiation intensity. This invention employs 3D convolutional operations to extract the dependencies between wind speed and radiation intensity in both temporal and spatial dimensions, capturing more complex features at different time and spatial points. Combined with an adaptive pooling strategy, this invention can automatically adjust the size of the convolutional kernels according to data volatility, improving the model's ability to process multi-scale data.

[0072] Traditional forecasting methods often provide only a single predicted value, failing to provide information on the uncertainty of the forecast. In grid dispatching, a single predicted value carries significant risk, potentially leading to over- or under-generation. This invention introduces a Bayesian inference method, calculating a confidence interval for the forecast result based on the prior distribution of the forecast error and observed data on actual power generation. This allows grid dispatching to assess the reliability of the forecast based on the width of the confidence interval, enabling more precise decision-making. A smaller confidence interval indicates a more accurate forecast, allowing the current output to be maintained; a larger confidence interval indicates higher uncertainty, necessitating the activation of standby units for adjustment.

[0073] Example 2 is an embodiment of the present invention, which provides a method for predicting the power generation of distributed power sources. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0074] To verify the beneficial effects of this invention, a comparative experiment was conducted using actual operating data from a distributed power system in a certain region. The dataset includes 8 photovoltaic power stations and 6 wind power stations, spanning the entire year of 2023, with a sampling interval of 15 minutes, totaling 35,040 sample points. These samples were divided into training, validation, and test sets in a ratio of 7:1.5:1.5.

[0075] The following comparative methods were selected for the experiment: Comparison Method 1 used a traditional LSTM neural network, with raw wind speed and radiation intensity data as input; Comparison Method 2 used a support vector machine combined with an RBF kernel function, with a fixed kernel width of 0.5; Comparison Method 3 used a convolutional neural network, employing 2D convolution to process spatial features; the method of this invention is a complete dynamic spatiotemporal fusion prediction network. Mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) were used as evaluation metrics.

[0076] The overall prediction accuracy comparison results of the test set show that the MAE of method 1 is 342.7kW, RMSE is 478.3kW, and MAPE is 18.4%; the MAE of method 2 is 389.6kW, RMSE is 521.9kW, and MAPE is 21.7%; the MAE of method 3 is 318.5kW, RMSE is 445.2kW, and MAPE is 16.8%; and the MAE of the method of this invention is 158.2kW, RMSE is 221.6kW, and MAPE is 8.3%.

[0077] Further, extreme operating conditions were selected from the test set, with wind speed fluctuations exceeding 30% or radiation intensity changes exceeding 400 W / m² within one hour, for specialized testing, totaling 687 sample points. Under extreme conditions, the MAE of comparative method 1 was 687.3 kW, the MAE of comparative method 3 was 621.8 kW, and the MAE of the method of this invention was 213.7 kW. Simultaneously, the errors of each method under normal operating conditions were statistically analyzed: comparative method 1 had an error of 268.3 kW, with an error growth rate of 156%; comparative method 3 had an error of 245.7 kW, with an error growth rate of 153%; and the method of this invention had an error of 112.8 kW, with an error growth rate of 89%.

[0078] First, there is a strong coupling and synergistic effect between dynamic kernel width adjustment and 3D spatiotemporal fusion. Those skilled in the art generally believe that RBF mapping with a fixed kernel width is already effective in extracting nonlinear features, and further optimization of the feature mapping has limited improvement on prediction accuracy. However, experimental results show that compared to the comparative method 2 using a fixed kernel width, the MAE of this invention decreased from 389.6kW to 158.2kW, an improvement of 59.4%, far exceeding the expected 20%-30%. The underlying reason is that when the kernel width is adjusted in real time according to the wind speed change rate and radiation fluctuations, the mapping structure of the high-dimensional feature space undergoes a fundamental change: during periods of drastic environmental fluctuations, the expanded kernel width makes the similarity measurement between feature points smoother, avoiding fragmentation of the feature space; at this time, 3D convolution extracts spatiotemporal dependencies in a better feature space, and the convolution kernel can capture the temporal patterns that were originally masked by noise. The two reinforce each other, producing a nonlinear synergistic improvement effect, a mechanism not revealed in existing technologies.

[0079] Second, the robustness under extreme conditions surpasses conventional understanding in this field. Traditional methods typically experience an error growth rate exceeding 150% under extreme conditions because extreme samples deviate from the training set distribution, drastically reducing the model's generalization ability. However, the error growth rate of this invention is only 89%, and the absolute error of 213.7kW under extreme conditions is even lower than the error level of traditional methods under normal conditions. This unexpected effect stems from the "adaptive defense" characteristic of the dynamic kernel width adjustment mechanism: when the wind speed change rate is detected to exceed the threshold ∆V_th, the system automatically expands the kernel width from the initial value of 0.5 to 1.8-2.1 times, significantly expanding the receptive field of the RBF kernel function. This allows extreme samples to establish a mapping relationship with "sub-extreme samples" in the training set, maintaining the continuity of the feature space. Simultaneously, the adaptive pooling window of 3D convolution automatically expands under extreme conditions, enhancing the time tolerance to abrupt signals. The synergistic effect of this dual adaptive mechanism enables the model to maintain stable predictions even when the data distribution undergoes drastic shifts, which is impossible to achieve with traditional methods using fixed parameters.

[0080] Third, the confidence interval calculated by Bayesian inference achieves high-quality uncertainty quantification. The 95% confidence interval calculated by the method of this invention achieves an actual coverage of 94.8%, deviating from the theoretical value by only 0.2 percentage points; at the same time, the normalized average width of the confidence interval is 0.087, significantly narrower than the typical value of 0.15-0.25 for existing Bayesian prediction methods. The key to this effect lies in the fact that this invention optimizes the error correction of the feedback layer, ensuring that the spatiotemporal feature representation entering Bayesian inference is sufficiently optimized, resulting in a smaller variance in the prior distribution; simultaneously, the posterior distribution update process utilizes the temporal correlation of historical prediction errors, avoiding drastic jumps in the confidence interval. In the experiment simulating 30 days of power grid dispatch, the method of this invention reduced the number of standby unit false starts from 41 times compared to the comparative method to 5 times, and the power outage duration from 10.3 hours to 0.9 hours, achieving a dispatch accuracy of 97.3%. This proves that the confidence interval width W CI Practical value in the standby unit calculation formula: When the prediction uncertainty is large, W CI Increase, the system starts the backup unit in advance; when the prediction is reliable, W CI By reducing redundancy and avoiding redundant startups, the system achieves an optimal balance between grid stability and economy. Traditional forecasting methods cannot provide reliable confidence intervals, resulting in a lack of quantitative basis for the start-up and shutdown decisions of standby units, which can only rely on empirical safety margins.

[0081] Example 3 is an embodiment of the present invention, which provides a distributed power generation prediction system, including: The data acquisition unit is used to collect environmental data during the distributed power generation process, including wind speed data and radiation intensity data. The data processing unit is used to calculate the wind speed influence coefficient and radiation conversion efficiency based on wind speed data and radiation intensity data; The distributed power generation prediction network consists of an input processing layer, a feature mapping layer, a spatiotemporal fusion layer, an optimization feedback layer, and a prediction output layer. The dispatch control unit calculates and adjusts the start-up and shutdown status of standby generators in the power grid to ensure that the power grid load demand is met.

[0082] This embodiment also provides an electronic device applicable to a distributed power generation prediction method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the distributed power generation prediction method proposed in the above embodiment.

[0083] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a distributed power generation prediction method as proposed in the above embodiments.

[0084] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for predicting the power generation of a distributed power source proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0085] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0086] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for distributed power generation power prediction, characterized in that: include, Collect wind speed and radiation intensity data during distributed power generation, and calculate the wind speed influence coefficient and radiation conversion efficiency based on the wind speed and radiation intensity data; A distributed power prediction network is constructed, comprising an input processing layer, a feature mapping layer, a spatiotemporal fusion layer, an optimization feedback layer, and a prediction output layer; The input processing layer receives radiation conversion efficiency and wind speed influence coefficient as input features, performs standardization processing, outlier removal and missing value imputation to generate clean data. The feature mapping layer performs nonlinear mapping on the clean data through radial basis function kernels, transforming the clean data from the original feature space into a high-dimensional feature space and outputting a high-dimensional feature representation; The spatiotemporal fusion layer extracts the dependence of wind speed influence coefficient and radiation conversion efficiency on time and space dimensions in high-dimensional feature representation through convolution operation, and performs dimensionality reduction processing on the output of convolution operation through pooling operation to generate spatiotemporal feature representation. The optimized feedback layer compares the spatiotemporal characteristics with the actual power generation, calculates the error, and optimizes the spatiotemporal characteristic representation. The prediction output layer generates power generation prediction values ​​through linear transformation based on the optimized spatiotemporal feature representation, and calculates the confidence interval of the power generation prediction results based on the power generation prediction values ​​using the Bayesian inference method.

2. The method of claim 1, wherein: The calculation of the wind speed influence coefficient and radiation conversion efficiency includes considering the influence of air density to determine the effective transmission of light, expressed as follows: wherein, represents the modified effective light intensity, represents the real-time radiation intensity, represents the real-time air density, represents the standard air density, represents the real-time air pressure, represents the real-time air temperature, represents the gas constant; Temperature correction formula for photovoltaic panel efficiency: wherein, represents the efficiency of the photovoltaic panel considering the temperature correction, represents the efficiency of the photovoltaic panel at standard temperature, represents the temperature coefficient of the photovoltaic panel, represents the difference between the temperature of the photovoltaic panel and the standard temperature, represents the temperature of the photovoltaic panel, represents the standard temperature; Based on the corrected effective light intensity and the photovoltaic panel efficiency considering temperature correction, the radiative conversion efficiency is determined and expressed as: wherein represents the radiation conversion efficiency, represents the standard radiation intensity.

3. The method of claim 2, wherein: The calculation of the wind speed influence coefficient and radiation conversion efficiency also includes calculating the rate of change of wind speed using the maximum, minimum, and average wind speeds, expressed as: wherein, represents the wind speed change rate, represents the wind speed maximum value, represents the wind speed minimum value, represents the wind speed average value; Based on the rate of change of wind speed per unit time, the degree of wind speed change is described, and the rate of change of wind speed is calculated and expressed as: wherein is a time interval, denotes the wind speed at the current time instant, denotes the wind speed at the previous time instant; Based on the rate of change of wind speed and the wind speed influence coefficient, the wind speed influence coefficient is calculated and expressed as follows: wherein, represents a wind speed influence coefficient, represents a proportion of wind speed fluctuation to wind speed, represents a proportion of wind speed change rate to wind speed.

4. The distributed power generation prediction method as described in claim 3, characterized in that: The output high-dimensional feature representation includes calculating the Euclidean distance for each set of data points and quantifying the differences in radiation conversion efficiency and wind speed influence coefficient. The Euclidean distance is expressed as: in, Indicates a point in time and time point Euclidean distance, Indicates a point in time Radiative conversion efficiency, Indicates a point in time The wind speed influence coefficient, This represents the radiative conversion efficiency of data point i. This represents the radiative conversion efficiency of data point j. This represents the wind speed influence coefficient at data point i. This represents the wind speed influence coefficient at data point j; The calculated Euclidean distance is mapped using a radial basis function kernel to transfer the original features to a high-dimensional feature space, as shown below: in, This represents the result of mapping in a high-dimensional feature space. For kernel width; Constructing the kernel matrix nuclear matrix The dimension is , The number of input data samples, kernel matrix Each element This indicates the temporal dependence and spatial similarity between radiation conversion efficiency and wind speed influence coefficient; Through the constructed kernel matrix The radiation conversion efficiency and wind speed influence coefficient are mapped to a high-dimensional feature space.

5. The distributed power generation prediction method as described in claim 4, characterized in that: The confidence interval for the calculated power generation prediction result includes the prediction error obtained by calculating the difference between the predicted power generation and the actual power generation. Define the prior distribution of the error, assume that the error follows a normal distribution, set the mean of the error to be zero, and set the standard deviation of the error to be the standard deviation of the prediction error calculated from historical data; By combining the prior distribution with the actual observed power generation error using Bayes' theorem, the updated posterior distribution is calculated. Based on the updated posterior distribution, the mean and standard deviation of the predicted power generation are calculated, the critical value is calculated by selecting a confidence level, and the confidence interval of the predicted power generation is obtained.

6. The distributed power generation prediction method as described in claim 5, characterized in that: The confidence interval for calculating the power generation prediction result also includes comparing the power generation prediction value with the actual grid demand to determine whether the grid load demand is met. If the power generation prediction value is greater than or equal to the actual grid demand, the current output is maintained. If the predicted power generation is less than the actual grid demand, the number of standby units that need to be activated is calculated based on the gap between the predicted power generation and the actual grid demand and the width of the confidence interval. Based on the calculated number of standby generating units, the power generation capacity is adjusted by controlling the start and stop of the standby generating units.

7. The distributed power generation prediction method as described in claim 6, characterized in that: The calculation of the number of standby units to be activated includes, based on the confidence interval, calculating the width of the confidence interval, expressed as: in, Indicates the width of the confidence interval; By comparing the predicted power generation with the actual grid demand, the difference between the two is calculated and expressed as: in, This indicates the difference between the predicted power generation and the actual grid demand; Based on the difference between the predicted power generation and the actual grid demand and confidence interval width The number of standby units is calculated and expressed as follows: in, Indicates the number of standby units. This is the maximum generating capacity of each standby unit.

8. A distributed power generation prediction system, employing the distributed power generation prediction method as described in any one of claims 1 to 7, characterized in that, include: The data acquisition unit is used to collect environmental data during the distributed power generation process, including wind speed data and radiation intensity data. The data processing unit is used to calculate the wind speed influence coefficient and radiation conversion efficiency based on wind speed data and radiation intensity data; The distributed power generation prediction network consists of an input processing layer, a feature mapping layer, a spatiotemporal fusion layer, an optimization feedback layer, and a prediction output layer. The dispatch control unit calculates and adjusts the start-up and shutdown status of standby generators in the power grid to ensure that the power grid load demand is met.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the distributed power generation prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the distributed power generation prediction method according to any one of claims 1 to 7.

Images