Regional power grid wind power photovoltaic power supply guarantee output, low output and guarantee probability prediction method

Abstract

The invention discloses a regional power grid wind power photovoltaic power supply guarantee output, low output and guarantee probability prediction method, and the method comprises the steps: constructing a multi-year historical sample set in which an input variable and a target variable are aligned in time sequence; an off-line regression model is established based on an asymmetric loss function, a multi-year historical sample set is divided into a training set, a verification set and a detection set, and it is ensured that a high-guarantee-rate biased prediction result oriented to the output lower limit is obtained; inputting a data format required by the discrete regression model into an offline regression model, and generating a guarantee output prediction result of a future time period; and evaluating the guarantee output prediction result based on the guarantee rate and the guarantee electric quantity ratio. A wind power / photovoltaic power supply capacity prediction technology system is provided, products of the wind power / photovoltaic power supply capacity prediction technology system are guarantee output prediction, low output process prediction and power supply guarantee probability prediction, and regional power grid dispatching decision application can be effectively supported.

Description

Technical Field

[0001] The present invention relates to the field of new energy power technology, and more specifically, to a method and system for predicting guaranteed output, low output and guaranteed probability of wind power and photovoltaic power supply in a regional power grid. Background Art

[0002] As an important component of clean energy, wind power / photovoltaic plays a vital role in the transformation of the global energy structure. Wind and solar power generation is always affected by meteorological factors such as wind speed, wind direction, and irradiance, and is highly volatile, random, and uncertain. Large-scale wind power and photovoltaic grid connection will significantly affect the stability of the power grid. Wind power and photovoltaic power forecasting can give future wind power and photovoltaic output in advance, and is a key technology for the power system to adjust the dispatching strategy in advance, optimize the energy mix, and effectively reduce system risks.

[0003] In the scenario of power supply guarantee, new energy forecasting faces new demands: First, when incorporating new energy into the power generation plan, considering the uncertainty of the forecast, the dispatching agency needs to know its lower limit of output, that is, how much "guaranteed output" wind power and photovoltaics can provide for power balance; Second, to avoid power outages, when the backup space has been fully tapped and wind power and photovoltaics are needed to provide a specific amount of power support, the dispatching agency needs to grasp the probability that wind power and photovoltaics can provide a specific amount of power, that is, the "power supply guarantee probability"; Third, the long-term low output process of wind power and photovoltaics will put great pressure on the system balance, and the dispatching agency needs to accurately grasp the start and end time, duration and other characteristic parameters of the low output process of wind power and photovoltaics.

[0004] The existing wind power / photovoltaic power forecasting system takes minimizing the power forecast error as its optimization goal and tends to give homogenized forecasting results. Moreover, the single point value results at each time section cannot provide the probabilistic and event-based information required for forecasting objects such as wind power / photovoltaic guaranteed output, wind power / photovoltaic power supply guarantee probability, and wind power / photovoltaic low output process in supply guarantee scenarios.

[0005] Existing wind power / photovoltaic power forecasts are mostly based on single stations as the basic unit, while the dispatching decisions of power grids are usually based on provincial power grids. The total power supply capacity of wind power / photovoltaic power in regional power grids cannot be calculated by summing up single stations due to its event and probability characteristics. Therefore, the power supply capacity forecast based on the total wind power, total photovoltaic power, and total wind power and photovoltaic power of provincial and above regional power grids is the main demand. Summary of the invention

[0006] According to the present invention, a method and system for predicting guaranteed output, low output and guaranteed probability of wind power and photovoltaic power supply in a regional power grid are provided to solve the technical problems of existing wind power / photovoltaic power prediction.

[0007] According to a first aspect of the present invention, a method for predicting wind power and photovoltaic power supply output of a regional power grid is provided, comprising:

[0008] Construct a multi-year historical sample set with time series alignment between input variables and target variables;

[0009] An offline regression model is established based on an asymmetric loss function, and the historical sample sets of many years are divided into training sets, validation sets, and test sets to ensure that biased prediction results with high guarantee rates for the lower limit of output are obtained;

[0010] The data format required by the discrete regression model is input into the offline regression model to generate the guaranteed output prediction results for the future period;

[0011] The guaranteed output forecast result is evaluated based on the guaranteed rate and the guaranteed electricity ratio.

[0012] Optionally, it also includes:

[0013] Using the key meteorological factors that affect wind power / photovoltaic output in numerical weather forecasts or the existing historical prediction results of wind power / photovoltaic power as input variables, and the total wind power, total photovoltaic power, and total wind and solar power actual power of the regional power grid as target variables, a multi-year historical sample set with time series alignment between input variables and target variables is constructed;

[0014] Determine an asymmetric loss function, the positive deviation angle of the asymmetric loss function α The positive deviation angle β is smaller than the negative deviation angle. Under the same deviation amplitude, the error loss value of the positive deviation is larger;

[0015] An offline regression model is established based on an asymmetric loss function. The historical sample sets of many years are divided into training sets, validation sets, and test sets. The optimal α and β combinations are determined by the performance of the discrete regression model on the validation set to ensure that biased prediction results with a high guarantee rate for the lower limit of output are obtained.

[0016] The weather forecast or predicted power for a period of time in the future is organized into the data format required by the offline regression model, and input into the offline regression model to generate the guaranteed output prediction results for the future period.

[0017] Optionally, the guaranteed output prediction result is evaluated based on the guaranteed rate and the guaranteed power ratio, including:

[0018] Based on the guarantee rate S r And the guaranteed power ratio R s Evaluation and guarantee output forecast results:

[0019] Guarantee rate S r It indicates the ratio of actual power to guaranteed output prediction result. The calculation formula is as follows:

[0020]

[0021] Among them, n represents the number of all samples, P Si represents the guaranteed output forecast at time i, P Mi Indicates the measured power at time i, that is, if there is power restriction or abnormality, the available power needs to be used. It means traversing each sample data of i=1,2,...,n and calculating the number of samples that satisfy the conditions in the brackets;

[0022] Guaranteed power ratio R s Calculated by formula (2):

[0023]

[0024] in,

[0025] In the formula, n represents the number of all samples, P Si represents the guaranteed output forecast at time i, P Mi It indicates the measured power at time i, that is, when there is power restriction or abnormality, the available power needs to be used.

[0026] According to the present invention, a method for predicting low output of wind power and photovoltaic power supply in a regional power grid is also provided, comprising:

[0027] Construct a multi-year historical sample set with time series alignment between input variables and target variables;

[0028] An offline binary regression model is established based on the multi-year historical sample set, and the model training and modeling are guided by the cross entropy loss function;

[0029] The weather forecast or predicted power for a period of time in the future is organized into the data format required by the offline binary classification regression model, and input into the offline binary classification regression model to generate the prediction result of the low output process in the future period;

[0030] The low-effort process prediction results are evaluated based on accuracy, critical success index, and fairness skill score.

[0031] Optionally, it also includes:

[0032] The actual power sequences of total wind power, total photovoltaic power, and total wind and solar power are encoded from 0 to 1 and used as target variables. The key meteorological factors affecting wind power / photovoltaic output in numerical weather forecasts or existing historical prediction results of wind power / photovoltaic power are used as input variables to construct a multi-year historical sample set with time series alignment between input variables and target variables.

[0033] Establish an offline binary regression model and use the cross entropy loss function to guide model training and model building;

[0034] The weather forecast or predicted power for a period of time in the future is organized into the data format required by the offline binary regression model, and input into the offline binary regression model to generate the prediction result of the low output process in the future period.

[0035] Optionally, evaluating the low-efficiency process prediction result based on accuracy, key success index, and fairness skill score includes:

[0036] The accuracy rate is the ratio of true low-output events to predicted low-output events. The calculation formula is as follows:

[0037]

[0038] The key success index is the ratio of real low output events to all known low output events, and the calculation formula is as follows:

[0039]

[0040] The fair skill score reflects the proportion of situations where both the forecast and the actual situation occur, and further penalizes false reports and missed reports. The calculation formula is as follows:

[0041]

[0042]

[0043] Among them, N TP N is the number of samples where the low output prediction is true and the low output observation is true; FN is the number of samples where the low output prediction is false and the low output observation is true; N FP is the number of samples where the low output prediction is true and the low output observation is false, N TN is the number of samples where the low output prediction is true and the low output observation is false.

[0044] According to another aspect of the present invention, a method for predicting the probability of wind power and photovoltaic power supply security in a regional power grid is provided, comprising:

[0045] Using the key meteorological factors that affect wind power / photovoltaic output in numerical weather forecasts or the existing historical prediction results of wind power / photovoltaic power as input variables, and the total wind power, total photovoltaic power, and total wind and solar power actual power of the regional power grid as target variables, a multi-year historical sample set with time series alignment between input variables and target variables is constructed;

[0046] A quantile regression prediction model is established based on a multi-year historical sample set. The model output is a set of quantile prediction results with different nominal probabilities for each time section. The corresponding probability is calculated between 0 and the total installed capacity. When the power demand to be calculated falls between discrete quantiles, the corresponding probability is calculated by interpolation.

[0047] Arrange the weather forecast or predicted power for a period of time in the future into the data format required by the quantile regression prediction model, input it into the quantile regression prediction model, and calculate the guarantee probability prediction results at the corresponding time through the model according to the output demand that changes at different times;

[0048] The guarantee probability prediction result is evaluated based on the accuracy of the event estimated probability.

[0049] Optionally, evaluating the guarantee probability prediction result based on the accuracy of the event estimated probability includes:

[0050] The higher the accuracy of the estimated probability of an event, the higher the prediction accuracy of the event probability. The calculation formula is as follows:

[0051]

[0052] Among them, f t is the probability that the predicted power is greater than or equal to the power supply demand at time t, o t It is the result of the actual output meeting the power supply demand at time t. If it meets the demand, it is 1, and if it does not, it is 0. The closer AR is to 1, the better the prediction performance of the power supply guarantee probability is. The closer it is to 0, the less accurate the prediction performance of the power supply guarantee probability is.

[0053] According to another aspect of the present invention, a regional power grid wind power and photovoltaic power supply guarantee output prediction system is provided, which is characterized by comprising:

[0054] The module for building historical sample sets is used to build multi-year historical sample sets with time series alignment of input variables and target variables;

[0055] An offline regression model module is established to establish an offline regression model based on an asymmetric loss function, and to divide the historical sample sets of many years into training sets, validation sets, and test sets to ensure that biased prediction results with a high guarantee rate for the lower limit of output are obtained;

[0056] A module for generating guaranteed output forecast results is used to input the data format required by the discrete regression model into the offline regression model to generate guaranteed output forecast results for future periods;

[0057] The guaranteed output prediction result evaluation module is used to evaluate the guaranteed output prediction result based on the guarantee rate and the guaranteed electricity proportion.

[0058] According to another aspect of the present invention, there is also provided a regional power grid wind power photovoltaic power supply low output prediction system, comprising:

[0059] The module for building historical sample sets is used to build multi-year historical sample sets with time series alignment of input variables and target variables;

[0060] Establishing a binary classification regression model module, which is used to establish an offline binary classification regression model based on the multi-year historical sample set, and guiding model training and modeling through a cross entropy loss function;

[0061] A module for generating low-output process prediction results is used to organize the weather forecast or predicted power for a period of time in the future into the data format required by the offline binary classification regression model, input it into the offline binary classification regression model, and generate low-output process prediction results for the future period;

[0062] The low-efficiency process prediction result evaluation module is used to evaluate the low-efficiency process prediction result based on accuracy, key success index and fairness skill score.

[0063] According to another aspect of the present invention, a regional power grid wind power and photovoltaic power supply guarantee probability prediction system is provided, comprising:

[0064] Construct a historical sample set module, which is used to use the key meteorological factors affecting wind power / photovoltaic output in numerical weather forecasts or the existing historical prediction results of wind power / photovoltaic power as input variables, and the total wind power, total photovoltaic power, and total wind and solar power actual power of the regional power grid as target variables to construct a multi-year historical sample set with time series alignment of input variables and target variables;

[0065] Establish a quantile regression prediction model module, which is used to establish a quantile regression prediction model based on a multi-year historical sample set. The model output is a set of quantile prediction results with different nominal probabilities for each time section. The corresponding probability is calculated between 0 and the total installed capacity. When the power demand to be calculated falls between discrete quantiles, the corresponding probability is calculated by interpolation.

[0066] The module for calculating the guarantee probability prediction result is used to organize the weather forecast or predicted power for a period of time in the future into the data format required by the quantile regression prediction model, input it into the quantile regression prediction model, and calculate the guarantee probability prediction results at the corresponding time through the model according to the output demand that changes at different times;

[0067] The guarantee probability prediction result evaluation module is used to evaluate the guarantee probability prediction result based on the accuracy of the event estimated probability.

[0068] Therefore, based on the practical experience of dispatching decisions, a wind power / photovoltaic power supply capacity prediction technology system was proposed, whose products are guaranteed output prediction, low output process prediction and power supply guarantee probability prediction, which can effectively support regional power grid dispatching decision-making applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:

[0070] Figure 1 It is a schematic diagram of a method for predicting output of wind power and photovoltaic power supply guarantee of a regional power grid according to this embodiment;

[0071] Figure 2 It is a schematic diagram of a method for predicting low output of wind power and photovoltaic power supply in a regional power grid according to this embodiment;

[0072] Figure 3 It is a schematic diagram of a method for predicting the probability of wind power and photovoltaic power supply security in a regional power grid according to this embodiment;

[0073] FIG4 is a comparison diagram of asymmetric and symmetric loss functions according to this embodiment;

[0074] Figure 5 This is a schematic diagram of the prediction result of the wind power low output process described in this implementation mode;

[0075] FIG6 is a schematic diagram of the guaranteed output prediction according to this embodiment;

[0076] FIG. 7 is a schematic diagram of the guarantee probability prediction described in this embodiment. DETAILED DESCRIPTION

[0077] Now, exemplary embodiments of the present invention are described with reference to the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to disclose the present invention in detail and completely and to fully convey the scope of the present invention to those skilled in the art. The terms used in the exemplary embodiments shown in the accompanying drawings are not intended to limit the present invention. In the accompanying drawings, the same units / elements are marked with the same reference numerals.

[0078] Unless otherwise specified, the terms (including technical terms) used herein have the commonly understood meanings to those skilled in the art. In addition, it is understood that the terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.

[0079] According to a first aspect of the present invention, a method 100 for predicting the output of wind power and photovoltaic power supply in a regional power grid is provided. Figure 1 As shown, the method 100 includes:

[0080] S101: Construct a multi-year historical sample set with time series alignment of input variables and target variables;

[0081] S102: Establish an offline regression model based on an asymmetric loss function, divide the historical sample set of many years into a training set, a validation set and a test set, and ensure that a biased prediction result with a high guarantee rate facing the lower limit of output is obtained;

[0082] S103: Inputting the data format required by the discrete regression model into the offline regression model to generate a guaranteed output prediction result for the future period;

[0083] S104: Evaluate the guaranteed output prediction result based on the guaranteed rate and the guaranteed power ratio.

[0084] Optionally, it also includes:

[0085] Using the key meteorological factors that affect wind power / photovoltaic output in numerical weather forecasts or the existing historical prediction results of wind power / photovoltaic power as input variables, and the total wind power, total photovoltaic power, and total wind and solar power actual power of the regional power grid as target variables, a multi-year historical sample set with time series alignment between input variables and target variables is constructed;

[0086] Determine an asymmetric loss function, wherein the positive deviation angle α of the asymmetric loss function is smaller than the negative deviation angle β, and under the same deviation amplitude, the error loss value of the positive deviation is larger;

[0087] An offline regression model is established based on an asymmetric loss function. The historical sample sets of many years are divided into training sets, validation sets, and test sets. The optimal α and β combinations are determined by the performance of the discrete regression model on the validation set to ensure that biased prediction results with a high guarantee rate for the lower limit of output are obtained.

[0088] The weather forecast or predicted power for a period of time in the future is organized into the data format required by the offline regression model, and input into the offline regression model to generate the guaranteed output prediction results for the future period.

[0089] Optionally, the guaranteed output prediction result is evaluated based on the guaranteed rate and the guaranteed power ratio, including:

[0090] Based on the guarantee rate S r And the guaranteed power ratio R s Evaluation and guarantee output forecast results:

[0091] Guarantee rate S r It indicates the ratio of actual power to guaranteed output prediction result. The calculation formula is as follows:

[0092]

[0093] Among them, n represents the number of all samples, P Si represents the guaranteed output forecast at time i, P Mi Indicates the measured power at time i, that is, if there is power restriction or abnormality, the available power needs to be used. It means traversing each sample data of i=1,2,...,n and calculating the number of samples that satisfy the conditions in the brackets;

[0094] Guaranteed power ratio R sCalculated by formula (2):

[0095]

[0096] in,

[0097] In the formula, n represents the number of all samples, P Si represents the guaranteed output forecast at time i, P Mi It indicates the measured power at time i, that is, when there is power restriction or abnormality, the available power needs to be used.

[0098] According to the present invention, a method 200 for predicting low output of wind power and photovoltaic power supply in a regional power grid is also provided. Figure 2 As shown, the method 200 includes:

[0099] S201: Construct a multi-year historical sample set with time series alignment of input variables and target variables;

[0100] S202: Establishing an offline binary classification regression model based on the multi-year historical sample set, and guiding model training and modeling through a cross entropy loss function;

[0101] S203: Arrange the weather forecast or predicted power for a period of time in the future into the data format required by the offline binary regression model, input it into the offline binary regression model, and generate the low output process prediction result for the future period;

[0102] S204: Evaluate the low-efficiency process prediction result based on the accuracy, key success index, and fairness skill score.

[0103] Optionally, it also includes:

[0104] The actual power sequences of total wind power, total photovoltaic power, and total wind and solar power are encoded from 0 to 1 and used as target variables. The key meteorological factors affecting wind power / photovoltaic output in numerical weather forecasts or existing historical prediction results of wind power / photovoltaic power are used as input variables to construct a multi-year historical sample set with time series alignment between input variables and target variables.

[0105] Establish an offline binary regression model and use the cross entropy loss function to guide model training and model building;

[0106] The weather forecast or predicted power for a period of time in the future is organized into the data format required by the offline binary regression model, and input into the offline binary regression model to generate the prediction result of the low output process in the future period.

[0107] Optionally, evaluating the low-efficiency process prediction result based on accuracy, key success index, and fairness skill score includes:

[0108] The accuracy rate is the ratio of true low-output events to predicted low-output events, and the calculation formula is as follows:

[0109]

[0110] The key success index is the ratio of real low output events to all known low output events, and the calculation formula is as follows:

[0111]

[0112] The fair skill score reflects the proportion of situations where both the forecast and the actual situation occur, and further penalizes false reports and missed reports. The calculation formula is as follows:

[0113]

[0114]

[0115] Among them, N TP N is the number of samples where the low output prediction is true and the low output observation is true; FN is the number of samples where the low output prediction is false and the low output observation is true; N FP is the number of samples where the low output prediction is true and the low output observation is false, N TN is the number of samples where the low output prediction is true and the low output observation is false.

[0116] According to another aspect of the present invention, a method 300 for predicting the probability of wind power and photovoltaic power supply security in a regional power grid is provided. Figure 3 As shown, the method 300 includes:

[0117] S301: Using key meteorological factors affecting wind power / photovoltaic output in numerical weather forecasts or existing historical prediction results of wind power / photovoltaic power as input variables, and using the total wind power, total photovoltaic power, and total wind and solar power actual power of the regional power grid as target variables, construct a multi-year historical sample set with time series alignment of input variables and target variables;

[0118] S302: A quantile regression prediction model is established based on a multi-year historical sample set. The model output is a set of quantile prediction results with different nominal probabilities for each time section. The corresponding probability is calculated between 0 and the total installed capacity. When the power demand to be calculated falls between discrete quantiles, the corresponding probability is calculated by interpolation.

[0119] S303: Arrange the weather forecast or predicted power for a period of time in the future into the data format required by the quantile regression prediction model, input it into the quantile regression prediction model, and calculate the guarantee probability prediction results at the corresponding time respectively according to the output demand that changes at different times through the model;

[0120] S304: Evaluate the guarantee probability prediction result based on the accuracy of the event estimated probability.

[0121] Optionally, evaluating the guarantee probability prediction result based on the accuracy of the event estimated probability includes:

[0122] The higher the accuracy of the estimated probability of an event, the higher the prediction accuracy of the event probability. The calculation formula is as follows:

[0123]

[0124] Among them, f t is the probability that the predicted power is greater than or equal to the power supply demand at time t, o t It is the result of the actual output meeting the power supply demand at time t. If it meets the demand, it is 1, and if it does not, it is 0. The closer AR is to 1, the better the prediction performance of the power supply guarantee probability is. The closer it is to 0, the less accurate the prediction performance of the power supply guarantee probability is.

[0125] Specifically, the wind power / photovoltaic power supply capacity prediction technology system consists of three categories: guaranteed output prediction, low output process prediction, and power supply guarantee probability prediction, which will be introduced below:

[0126] Wind power / photovoltaic power supply guaranteed output forecast (abbreviated as: guaranteed output forecast)

[0127] 1) Basic definition of guaranteed output forecast:

[0128] Guaranteed output forecast is to provide a forecast of the lower limit of wind power / photovoltaic output with a high guarantee rate. The total wind power, total photovoltaic power and total wind and solar power of each power station in the regional power grid can be independently modeled and predicted. According to practical experience in dispatching, the guarantee rate (Sr) of the guaranteed output forecast must be greater than or equal to 90%, and provincial power grids usually require 95%.

[0129] 2) Basic methods to ensure output prediction:

[0130] (1) Data preparation stage

[0131] Taking the key meteorological factors affecting wind power / photovoltaic output in numerical weather forecasts, or the existing historical prediction results of wind power / photovoltaic power as input variables, and the total wind power, total photovoltaic power, and total wind and solar power actual power of the regional power grid as target variables, a multi-year historical sample set with time series alignment of input variables and target variables is constructed.

[0132] (2) Offline modeling stage

[0133] Design an asymmetric loss function that has a relatively stronger penalty for positive prediction bias (prediction > actual) to guide regression model training. An example of the design of an asymmetric loss function is shown in Figure 4(a). The positive bias angle of the loss function is αThe positive bias angle β is smaller than the negative bias angle β. Under the same deviation amplitude, the error loss value of the positive bias is larger. Other loss functions that can achieve the same effect can also be used. Figure 4(b) is a symmetric function.

[0134] Common SVM, MLP, XGboost, lightGBM and other methods can be used to build offline regression models. The historical sample set is divided into training set, validation set and test set, and the optimal α and β combination is determined by the performance of the prediction model on the validation set to ensure that a biased prediction result with a high guarantee rate facing the lower limit of output is obtained.

[0135] (3) Online prediction stage

[0136] The weather forecast or predicted power for a period of time in the future is organized into the data format required by the offline model, and input into the offline model to generate the guaranteed output forecast results for the future period.

[0137] 3) Evaluation method for ensuring output prediction results:

[0138] The evaluation method for the guaranteed output forecast results adopts an evaluation method that combines the guarantee rate (Sr) and the guaranteed power ratio (Rs). The guarantee rate is a hard indicator. The forecast result is based on the guarantee rate meeting the demand and increasing the guaranteed power ratio as much as possible. The calculation formulas of the two indicators are shown in (1) and (2) respectively.

[0139] The guarantee rate (Sr) indicates the ratio of actual power to guaranteed output prediction result. The calculation formula is as follows:

[0140]

[0141] Where:

[0142] n——the number of all samples;

[0143] P Si ——Guaranteed output forecast at time i;

[0144] P Mi ——The measured power at time i (in case of power restriction or abnormality, the available power shall be used);

[0145] ——Traverse the sample data of i=1,2,,n and calculate the number of samples that satisfy the conditions in the brackets.

[0146] The guaranteed power ratio (Rs) is calculated by formula (2).

[0147]

[0148] in,

[0149] Where:

[0150] n——the number of all samples;

[0151] P Si ——Guaranteed output forecast at time i;

[0152] P Mi ——The measured power at time i (in case of power restriction or abnormality, available power should be used).

[0153] Wind power / photovoltaic low output process prediction (abbreviated as: low output process prediction)

[0154] 1) Basic definition of low output process prediction:

[0155] Low output process prediction treats low wind power output, low photovoltaic output, and low wind and solar output as events for prediction. It is recommended to use a classification regression model for prediction. The prediction results can intuitively give the start and end time of the event. Based on the provincial power grid example, the low wind power output process usually defines low output with a threshold of 10% to 20% of installed capacity. Since the output of photovoltaic power is 0 at night, the maximum output of the day is less than 20% to 30% of the installed capacity to define the photovoltaic low output day. The low wind and solar output is usually defined as a threshold of 10% to 20% of the total wind and solar capacity.

[0156] 2) Basic methods for predicting low output processes:

[0157] (1) Data preparation stage

[0158] According to the definition standards in the basic definition, the actual power sequences of total wind power, total photovoltaic power, and total wind and solar power are encoded with "0-1" and used as target variables. The key meteorological factors affecting wind power / photovoltaic output in numerical weather forecasts, or the existing historical prediction results of wind power / photovoltaic power are used as input variables to construct a multi-year historical sample set with time series alignment of input variables and target variables.

[0159] The proportion of low-output samples with a statistical code of 1. Usually, the number of low-output samples is small. In order to improve the prediction performance, the proportion of low-output samples in the sample set can be increased by using sample resampling methods or generative artificial intelligence models to generate virtual samples.

[0160] (2) Offline modeling stage

[0161] Common SVM, MLP, XGboost, lightGBM and other methods can be used to establish an offline binary regression wind power low output process prediction result schematic model, and the cross entropy loss function can be used to guide model training and modeling. Figure 5 As shown, Figure 5 This is a schematic diagram of the prediction results of wind power low output process.

[0162] (3) Online prediction stage

[0163] The weather forecast or predicted power for a period of time in the future is organized into the data format required by the offline model, and input into the offline model to generate the low-output process prediction results coded 0-1 for the future period.

[0164] 3) Evaluation of prediction results

[0165] The result of low-output process prediction is the encoded event result. The number of low-output processes is far less than that of non-low-output processes. In order to highlight the prediction effect of low-output processes, it is recommended to use accuracy, key success index, and fair skill score for evaluation.

[0166] N TP N is the number of samples where the low output prediction is true and the low output observation is true; FN is the number of samples where the low output prediction is false and the low output observation is true; N FP is the number of samples where the low output prediction is true and the low output observation is false, N TN is the number of samples where the low output prediction is true and the low output observation is false.

[0167] Accuracy Ratio: The ratio of real low output events to predicted low output events. The calculation method of this indicator is shown in formula (3).

[0168]

[0169] Critical Success Index: The ratio of real low-output events to all known low-output events. The calculation method of this index is shown in formula (4).

[0170]

[0171] Equitable Threat Score: reflects the proportion of situations where both the forecast and the actual situation occur, and further penalizes false alarms and missed alarms. The calculation method of this indicator is shown in formula (5) and formula (6).

[0172]

[0173]

[0174] 4.3 Wind power / photovoltaic power supply guarantee probability prediction (abbreviated as: guarantee probability prediction)

[0175] 1) Basic definition of guarantee probability prediction:

[0176] The guarantee probability forecast is a method to predict the probability of wind power, photovoltaic power, or wind and solar power combined to ensure power supply under different power demand of the power grid at different moments in the future. During the peak hours in the morning and evening when the balance between supply and demand is tight and deterministic power supply and backup resources have been exhausted, the guarantee probability forecast can estimate the probability of wind power and photovoltaic power filling the power supply gap, making it easier for the dispatching department to assess risks and coordinate resources to ensure power balance.

[0177] 2) Basic methods for guaranteeing probability prediction:

[0178] (1) Data preparation stage

[0179] Taking the key meteorological factors affecting wind power / photovoltaic output in numerical weather forecasts, or the existing historical prediction results of wind power / photovoltaic power as input variables, and the total wind power, total photovoltaic power, and total wind and solar power actual power of the regional power grid as target variables, a multi-year historical sample set with time series alignment of input variables and target variables is constructed.

[0180] (2) Offline modeling stage

[0181] In order to achieve the guaranteed probability prediction, a quantile regression prediction model can be established. The model output is a set of quantile prediction results for different nominal probabilities at each time section, but this result is not the final result of the guaranteed probability prediction. In order to ensure that the corresponding probability can be calculated under any set output demand (usually between 0 and total installed capacity), the selection and division of the nominal probability needs to be as dense as possible. When the power supply demand to be calculated falls between discrete quantiles, the corresponding probability can be conveniently calculated by interpolation, thereby achieving full coverage of the probability results between 0 and 100%.

[0182] (3) Online prediction stage

[0183] The weather forecast or predicted power for a period of time in the future is organized into the data format required by the offline model and input into the offline model. Each time section will get the result of the quantile set. The user can calculate the guarantee probability at the corresponding time by inputting the output demand that changes at different times through the model.

[0184] 3) Evaluation of prediction results

[0185] The result of the guarantee probability prediction is in the form of event probability. Therefore, the Brier score-based method is recommended for evaluation. Taking into account the intuitiveness of the indicator results, the original Brier score is improved and an indicator of the accuracy of event estimation probability is proposed. The higher the value, the higher the prediction accuracy of the event probability.

[0186]

[0187] Where: f tis the probability that the predicted power is greater than or equal to the power supply demand at time t, o t It is the result of the actual output meeting the power supply demand at time t (1 if it meets the demand, 0 if it does not). The closer AR is to 1, the better the prediction performance of the power supply guarantee probability is, and the closer it is to 0, the less accurate the prediction performance of the power supply guarantee probability is.

[0188] Therefore, based on the practical experience of dispatching decisions, a wind power / photovoltaic power supply capacity prediction technology system is proposed, whose products are guaranteed output prediction, low output process prediction and power supply guarantee probability prediction, which can effectively support regional power grid dispatching decision-making applications. The patent also provides the basic definitions of guaranteed output prediction, low output process prediction and power supply guarantee probability prediction, the form of results, the basic methods and processes for implementation, and how to evaluate the prediction results. The following is an introduction to the practical effects of the provincial power grid for the three prediction products, and the evaluation calculation methods all use the relevant indicators provided by this patent.

[0189] (1) Example of Provincial Power Grid Calculation to Ensure Output Forecast

[0190] The actual calculation example of Gansu provincial power grid was used for verification and analysis. Under the condition of a high guarantee rate of 95%, the guaranteed power of wind power on the first day can reach 32.78% of the actual power, which is 1.59% higher than the traditional power forecast discount method, and reaches 26.15% on the third day, an increase of 1.19%; the guaranteed power of photovoltaic power on the first day reaches 75.44% of the actual power, an increase of 5.29%, and reaches 64.32% on the third day, an increase of 9.51%; the guaranteed power of total wind and solar power output on the first day reaches 56.21% of the actual power, an increase of 2.23%, and reaches 53.17% on the third day, an increase of 4.15%.

[0191] Therefore, under the premise of a high guarantee rate, for different forecast objects of provincial wind power, photovoltaic power and the total of wind power and photovoltaic power, the guaranteed output forecast method can absorb an average of 1-9 percentage points more new energy electricity and reduce fossil energy consumption.

[0192] (2) Example of provincial power grid prediction for low output process:

[0193] The actual calculation example of Gansu provincial power grid was used for verification and analysis. The CSI of the wind power low output process (threshold is 10% installed capacity) in the first three days was 67.44%, which was 20.67% higher than the method of directly calibrating the low output process based on traditional power prediction. The CSI of the photovoltaic low output process in the first three days was 85.38%, which was 10.08% higher. The CSI of the wind and solar combined low output process (threshold is 10% installed capacity) in the first three days was 76.29%, which was 29.68% higher. More accurate prediction of the occurrence of low output process can better coordinate resources to deal with the risk of insufficient power supply when low output process occurs, and reduce the waste of standby resources caused by misjudgment.

[0194] (3) Example of provincial power grid guarantee probability prediction:

[0195] The actual calculation example of Gansu provincial power grid was used for verification and analysis. The accuracy of wind power 1-3 days power supply guarantee probability prediction under different power supply output requirements was 70.42%-83.78%, and that of photovoltaic power was 77.64%.

[0196] —87.77%, and the total of wind power and photovoltaic power is 76.16%-83.37%.

[0197] Refer to Figure 6, which shows the forecast results of the guaranteed output of wind power, photovoltaic power, and wind and solar power combined in the regional power grid. Refer to Figure 7, which shows the total wind power of the provincial power grid and the total new energy of the provincial power grid.

[0198] Based on the practical experience of dispatching decision-making, a wind power / photovoltaic power supply capacity prediction technology system is proposed. Its products are guaranteed output prediction, low output process prediction and power supply guarantee probability prediction, which can effectively support the application of regional power grid dispatching decision-making. At the same time, the basic definitions, result forms, basic methods and processes for implementation of guaranteed output prediction, low output process prediction and power supply guarantee probability prediction are given, as well as methods for evaluating the prediction results. The following introduces the practical effects of the provincial power grid for the three prediction products. The evaluation calculation methods all use the relevant indicators provided in this implementation method.

[0199] (1) Example of Provincial Power Grid Calculation to Ensure Output Forecast

[0200] The actual calculation example of Gansu provincial power grid was used for verification and analysis. Under the condition of a high guarantee rate of 95%, the guaranteed power of wind power on the first day can reach 32.78% of the actual power, which is 1.59% higher than the traditional power forecast discount method, and reaches 26.15% on the third day, an increase of 1.19%; the guaranteed power of photovoltaic power on the first day reaches 75.44% of the actual power, an increase of 5.29%, and reaches 64.32% on the third day, an increase of 9.51%; the guaranteed power of total wind and solar power output on the first day reaches 56.21% of the actual power, an increase of 2.23%, and reaches 53.17% on the third day, an increase of 4.15%.

[0201] Therefore, under the premise of a high guarantee rate, for different forecast objects of provincial wind power, photovoltaic power and the total of wind power and photovoltaic power, the guaranteed output forecast method can absorb an average of 1-9 percentage points more new energy electricity and reduce fossil energy consumption.

[0202] (2) Example of provincial power grid prediction for low output process:

[0203] The actual calculation example of Gansu provincial power grid was used for verification and analysis. The CSI of the wind power low output process (threshold is 10% installed capacity) in the first three days was 67.44%, which was 20.67% higher than the method of directly calibrating the low output process based on traditional power prediction. The CSI of the photovoltaic low output process in the first three days was 85.38%, which was 10.08% higher. The CSI of the wind and solar combined low output process (threshold is 10% installed capacity) in the first three days was 76.29%, which was 29.68% higher. More accurate prediction of the occurrence of low output process can better coordinate resources to deal with the risk of insufficient power supply when low output process occurs, and reduce the waste of standby resources caused by misjudgment.

[0204] (3) Example of provincial power grid guarantee probability prediction:

[0205] The actual example of Gansu provincial power grid was used for verification and analysis. The accuracy of the 1-3 day power supply guarantee probability prediction for wind power under different power supply output demands was 70.42%-83.78%, for photovoltaic power it was 77.64%-87.77%, and for the total of wind power and photovoltaic power it was 76.16%-83.37%.

[0206] According to another aspect of the present invention, a regional power grid wind power and photovoltaic power supply guarantee output prediction system is provided, which is characterized by comprising:

[0207] The module for building historical sample sets is used to build multi-year historical sample sets with time series alignment of input variables and target variables;

[0208] An offline regression model module is established to establish an offline regression model based on an asymmetric loss function, and to divide the historical sample sets of many years into training sets, validation sets, and test sets to ensure that biased prediction results with a high guarantee rate for the lower limit of output are obtained;

[0209] A module for generating guaranteed output forecast results is used to input the data format required by the discrete regression model into the offline regression model to generate guaranteed output forecast results for future periods;

[0210] The guaranteed output prediction result evaluation module is used to evaluate the guaranteed output prediction result based on the guarantee rate and the guaranteed electricity proportion.

[0211] According to another aspect of the present invention, there is also provided a regional power grid wind power photovoltaic power supply low output prediction system, comprising:

[0212] The module for building historical sample sets is used to build multi-year historical sample sets with time series alignment of input variables and target variables;

[0213] Establishing a binary classification regression model module, which is used to establish an offline binary classification regression model based on the multi-year historical sample set, and guiding model training and modeling through a cross entropy loss function;

[0214] A module for generating low-output process prediction results is used to organize the weather forecast or predicted power for a period of time in the future into the data format required by the offline binary classification regression model, input it into the offline binary classification regression model, and generate low-output process prediction results for the future period;

[0215] The low-efficiency process prediction result evaluation module is used to evaluate the low-efficiency process prediction result based on accuracy, key success index and fairness skill score.

[0216] According to another aspect of the present invention, a regional power grid wind power and photovoltaic power supply guarantee probability prediction system is provided, comprising:

[0217] Construct a historical sample set module, which is used to use the key meteorological factors affecting wind power / photovoltaic output in numerical weather forecasts or the existing historical prediction results of wind power / photovoltaic power as input variables, and the total wind power, total photovoltaic power, and total wind and solar power actual power of the regional power grid as target variables to construct a multi-year historical sample set with time series alignment of input variables and target variables;

[0218] Establish a quantile regression prediction model module, which is used to establish a quantile regression prediction model based on a multi-year historical sample set. The model output is a set of quantile prediction results with different nominal probabilities for each time section. The corresponding probability is calculated between 0 and the total installed capacity. When the power demand to be calculated falls between discrete quantiles, the corresponding probability is calculated by interpolation.

[0219] The module for calculating the guarantee probability prediction result is used to organize the weather forecast or predicted power for a period of time in the future into the data format required by the quantile regression prediction model, input it into the quantile regression prediction model, and calculate the guarantee probability prediction results at the corresponding time through the model according to the output demand that changes at different times;

[0220] The guarantee probability prediction result evaluation module is used to evaluate the guarantee probability prediction result based on the accuracy of the event estimated probability.

[0221] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0222] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0223] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0224] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0225] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0226] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for predicting the output of wind power and photovoltaic power supply in a regional power grid, characterized in that: include: Construct a multi-year historical sample set with time series alignment between input variables and target variables; An offline regression model is established based on an asymmetric loss function, and the historical sample sets of many years are divided into training sets, validation sets, and test sets to ensure that biased prediction results with high guarantee rates for the lower limit of output are obtained; The data format required by the discrete regression model is input into the offline regression model to generate the guaranteed output prediction results for the future period; The guaranteed output forecast result is evaluated based on the guaranteed rate and the guaranteed electricity ratio.

2. The method according to claim 1, characterized in that Also includes: Using the key meteorological factors that affect wind power / photovoltaic output in numerical weather forecasts or the existing historical prediction results of wind power / photovoltaic power as input variables, and the total wind power, total photovoltaic power, and total wind and solar power actual power of the regional power grid as target variables, a multi-year historical sample set with time series alignment between input variables and target variables is constructed; Determine an asymmetric loss function, wherein the positive deviation angle α of the asymmetric loss function is smaller than the negative deviation angle β, and under the same deviation amplitude, the error loss value of the positive deviation is larger; An offline regression model is established based on an asymmetric loss function. The historical sample sets of many years are divided into training sets, validation sets, and test sets. The optimal α and β combinations are determined by the performance of the discrete regression model on the validation set to ensure that biased prediction results with a high guarantee rate for the lower limit of output are obtained. The weather forecast or predicted power for a period of time in the future is organized into the data format required by the offline regression model, and input into the offline regression model to generate the guaranteed output prediction results for the future period.

3. The method according to claim 1, characterized in that The guaranteed output forecast results are evaluated based on the guaranteed rate and guaranteed power ratio, including: Based on the guarantee rate S r And the guaranteed power ratio R s Evaluation and guarantee output forecast results: Guarantee rate S r It indicates the ratio of actual power to guaranteed output prediction result. The calculation formula is as follows: Among them, n represents the number of all samples, P Si represents the guaranteed output forecast at time i, P Mi Indicates the measured power at time i, that is, if there is power restriction or abnormality, the available power needs to be used. It means traversing each sample data of i=1,2,...,n and calculating the number of samples that satisfy the conditions in the brackets; Guaranteed power ratio R s Calculated by formula (2): in, In the formula, n represents the number of all samples, P Si represents the guaranteed output forecast at time i, P Mi It indicates the measured power at time i, that is, when there is power restriction or abnormality, the available power needs to be used.

4. A method for predicting low output of wind power and photovoltaic power supply in a regional power grid, characterized in that: include: Construct a multi-year historical sample set with time series alignment between input variables and target variables; An offline binary regression model is established based on the multi-year historical sample set, and the model training and modeling are guided by the cross entropy loss function; The weather forecast or predicted power for a period of time in the future is organized into the data format required by the offline binary classification regression model, and input into the offline binary classification regression model to generate the prediction result of the low output process in the future period; The low-effort process prediction results are evaluated based on accuracy, critical success index, and fairness skill score.

5. The method according to claim 4, characterized in that Also includes: The actual power sequences of total wind power, total photovoltaic power, and total wind and solar power are encoded from 0 to 1 and used as target variables. The key meteorological factors affecting wind power / photovoltaic output in numerical weather forecasts or existing historical prediction results of wind power / photovoltaic power are used as input variables to construct a multi-year historical sample set with time series alignment between input variables and target variables. Establish an offline binary regression model and use the cross entropy loss function to guide model training and model building; The weather forecast or predicted power for a period of time in the future is organized into the data format required by the offline binary regression model, and input into the offline binary regression model to generate the prediction result of the low output process in the future period.

6. The method according to claim 4, characterized in that The low-efficiency process prediction results are evaluated based on accuracy, key success index, and fairness skill score, including: The accuracy rate is the ratio of true low-output events to predicted low-output events, and the calculation formula is as follows: The key success index is the ratio of real low output events to all known low output events, and the calculation formula is as follows: The fair skill score reflects the proportion of situations where both the forecast and the actual situation occur, and further penalizes false reports and missed reports. The calculation formula is as follows: Among them, N TP N is the number of samples where the low output prediction is true and the low output observation is true; FN is the number of samples where the low output prediction is false and the low output observation is true; N FP is the number of samples where the low output prediction is true and the low output observation is false, N TN is the number of samples where the low output prediction is true and the low output observation is false.

7. A method for predicting the probability of wind power and photovoltaic power supply security in a regional power grid, characterized in that: include: Using the key meteorological factors that affect wind power / photovoltaic output in numerical weather forecasts or the existing historical prediction results of wind power / photovoltaic power as input variables, and the total wind power, total photovoltaic power, and total wind and solar power actual power of the regional power grid as target variables, a multi-year historical sample set with time series alignment between input variables and target variables is constructed; A quantile regression prediction model is established based on a multi-year historical sample set. The model output is a set of quantile prediction results with different nominal probabilities for each time section. The corresponding probability is calculated between 0 and the total installed capacity. When the power demand to be calculated falls between discrete quantiles, the corresponding probability is calculated by interpolation. Arrange the weather forecast or predicted power for a period of time in the future into the data format required by the quantile regression prediction model, input it into the quantile regression prediction model, and calculate the guarantee probability prediction results at the corresponding time through the model according to the output demand that changes at different times; The guarantee probability prediction result is evaluated based on the accuracy of the event estimated probability.

8. The method according to claim 7, characterized in that The guarantee probability prediction result is evaluated based on the accuracy of the event estimated probability, including: The higher the accuracy of the estimated probability of an event, the higher the prediction accuracy of the event probability. The calculation formula is as follows: Among them, f t is the probability that the predicted power is greater than or equal to the power supply demand at time t, o t It is the result of the actual output meeting the power supply demand at time t. If it meets the demand, it is 1, and if it does not, it is 0. The closer AR is to 1, the better the prediction performance of the power supply guarantee probability is. The closer it is to 0, the less accurate the prediction performance of the power supply guarantee probability is.

9. A regional power grid wind power and photovoltaic power supply guarantee output prediction system, characterized in that: include: The module for building historical sample sets is used to build multi-year historical sample sets with time series alignment of input variables and target variables; An offline regression model module is established to establish an offline regression model based on an asymmetric loss function, and to divide the historical sample sets of many years into training sets, validation sets, and test sets to ensure that biased prediction results with a high guarantee rate for the lower limit of output are obtained; A module for generating guaranteed output forecast results is used to input the data format required by the discrete regression model into the offline regression model to generate guaranteed output forecast results for future periods; The guaranteed output prediction result evaluation module is used to evaluate the guaranteed output prediction result based on the guarantee rate and the guaranteed electricity proportion.

10. A regional power grid wind power and photovoltaic power supply low output prediction system, characterized in that: include: The module for building historical sample sets is used to build multi-year historical sample sets with time series alignment of input variables and target variables; Establishing a binary classification regression model module, which is used to establish an offline binary classification regression model based on the multi-year historical sample set, and guiding model training and modeling through a cross entropy loss function; A module for generating low-output process prediction results is used to organize the weather forecast or predicted power for a period of time in the future into the data format required by the offline binary classification regression model, input it into the offline binary classification regression model, and generate low-output process prediction results for the future period; The low-efficiency process prediction result evaluation module is used to evaluate the low-efficiency process prediction result based on accuracy, key success index and fairness skill score.

11. A regional power grid wind power and photovoltaic power supply guarantee probability prediction system, characterized in that: include: Construct a historical sample set module, which is used to use the key meteorological factors affecting wind power / photovoltaic output in numerical weather forecasts or the existing historical prediction results of wind power / photovoltaic power as input variables, and the total wind power, total photovoltaic power, and total wind and solar power actual power of the regional power grid as target variables to construct a multi-year historical sample set with time series alignment of input variables and target variables; Establish a quantile regression prediction model module, which is used to establish a quantile regression prediction model based on a multi-year historical sample set. The model output is a set of quantile prediction results with different nominal probabilities for each time section. The corresponding probability is calculated between 0 and the total installed capacity. When the power demand to be calculated falls between discrete quantiles, the corresponding probability is calculated by interpolation. The module for calculating the guarantee probability prediction result is used to organize the weather forecast or predicted power for a period of time in the future into the data format required by the quantile regression prediction model, input it into the quantile regression prediction model, and calculate the guarantee probability prediction results at the corresponding time through the model according to the output demand that changes at different times; The guarantee probability prediction result evaluation module is used to evaluate the guarantee probability prediction result based on the accuracy of the event estimated probability.

Images