New energy power prediction method, electronic device and storage medium under extreme climate
Through multi-mode integrated forecast and probability forecast models, meteorological factors and output change characteristics are extracted, and high-precision prediction and early warning of new energy power in extreme climates are achieved, and the problem of difficulty in predicting and early warning of new energy power generation power in extreme climates is solved. Improves grid stability.
Patent Information
- Application Number
- CN202311277606.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-28
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2043-09-28
AI Technical Summary
In extreme climates, the power prediction accuracy and early warning of new energy power generation are difficult, and it is difficult for existing technology to achieve high-precision prediction and early warning.
By obtaining climate observation data in extreme climates and the numerical model forecast results of new energy stations, extracting meteorological elements and output changes characteristics, performing multi-mode integrated forecasts, combining deterministic and probability forecast models, high-precision prediction of new energy power and Early warning.
It improves the prediction accuracy and early warning capabilities of new energy power in extreme climates, and enhances the stability and safety of the power grid.
Smart Images

Figure CN117526274B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of renewable energy power generation, and in particular to a renewable energy power prediction method under extreme climate conditions, an electronic device and a storage medium. Background Art
[0002] As the scale of installed capacity of renewable energy in China continues to increase, the intermittent, random and volatile nature of renewable energy generation has an increasingly significant impact on the stability of the power grid, and the power dispatching system has increasingly stringent requirements for the accuracy of power forecasts. Especially in extreme climates such as typhoons, cold waves, and severe convection, the forecast of renewable energy output requires not only accurate forecasts of meteorological elements such as wind speed, irradiance, cloud cover, and temperature, but also high-precision forecasts and even early warnings of renewable energy output for extreme climates such as typhoons, cold waves, and severe convection.
[0003] Therefore, how to make high-precision predictions or even early warnings of the power of new energy under extreme climate conditions has become a technical problem that needs to be solved urgently. Summary of the invention
[0004] The embodiments of the present application provide a method, electronic device and storage medium for predicting the power of new energy under extreme climate conditions, so as to solve the technical problem existing in the related art of how to perform high-precision prediction or even early warning of the power of new energy under extreme climate conditions.
[0005] According to the first aspect of the present application, a method for predicting new energy power under extreme climate is provided, comprising: obtaining climate observation data and numerical model forecast results of new energy stations under extreme climate; extracting meteorological elements and output change characteristics from the climate observation data and numerical model forecast results of new energy stations under extreme climate; performing multi-mode integrated forecasting on the climate observation data and numerical model forecast results of new energy stations under extreme climate based on the meteorological elements and the output change characteristics to obtain a multi-mode integrated forecast result; using the multi-mode integrated forecast result as input, using a deterministic forecast model to perform a deterministic forecast to obtain a deterministic forecast result; inputting the climate observation data and the deterministic forecast results under extreme climate into a probability forecasting model to obtain a meteorological probability forecast result; and predicting the new energy power based on the meteorological probability forecast result.
[0006] Optionally, the extracting of meteorological elements and output change characteristics from the climate observation data under extreme climate and the numerical model forecast results of new energy stations includes: performing data quality control and uniformity verification on the climate observation data under extreme climate and the numerical model forecast results of new energy stations to obtain valid meteorological data and deviation factors; extracting time characteristics, spatial characteristics and vertical characteristics from the valid meteorological data as the meteorological elements; obtaining output change law data based on statistical analysis of the meteorological elements; and determining the output change characteristics based on the deviation factors and the output change law data.
[0007] Optionally, performing multi-mode integrated forecasting on the climate observation data under the extreme climate and the numerical model forecast results of the new energy stations based on the meteorological elements and the output change characteristics to obtain the multi-mode integrated forecast results includes: obtaining the types of the meteorological elements and the output change characteristics; optimizing the integration weights of each model member in the multi-mode integrated forecast based on the types of the meteorological elements and the output change characteristics; performing multi-mode integrated forecasting on the climate observation data under the extreme climate and the numerical model forecast results of the new energy stations based on the integration weights to obtain the multi-mode integrated forecast results.
[0008] Optionally, the multi-model integrated forecasting of the climate observation data under the extreme climate and the numerical model forecast results of the new energy station based on the integration weights to obtain the multi-model integrated forecast results includes: performing numerical model dynamic downscaling on the numerical model forecast results of the new energy station to obtain meteorological element forecast results; standardizing the climate observation data under the extreme climate to obtain standardized observation data; performing significant factor correlation analysis on the standardized data and the meteorological element forecast results to obtain significant factor analysis results; performing principal component analysis on the significant factor analysis results to obtain principal components; correcting the numerical model forecast results of the new energy station based on the principal components by using a statistical downscaling method to obtain corrected numerical forecast results; performing multi-model integrated forecasting on the corrected numerical forecast results based on the integration weights to obtain multi-model integrated forecast results.
[0009] Optionally, the multi-modal integrated forecast result is used as input, and a deterministic forecast model is used to perform a deterministic forecast to obtain a deterministic forecast result, including: using a self-attention module to extract feature information of forecast results of different modes; fusing the extracted feature information through a residual connection method; and using a prediction unit network to generate a prediction element sequence as the deterministic forecast result.
[0010] Optionally, the inputting the climate observation data under extreme climate and the deterministic forecast results into the probability prediction model to obtain the meteorological probability forecast results includes: constructing an original meteorological sequence based on the climate observation data under extreme climate and the deterministic forecast results; decomposing the original meteorological sequence to obtain multiple meteorological sub-sequences; and performing probability prediction on the meteorological sub-sequences using a probability prediction model constructed using multiple different types of probability prediction methods to obtain a joint meteorological probability forecast result, the probability prediction methods including at least two of a time series residual probability hybrid prediction method, a multi-source Gaussian ensemble kernel fitting prediction method, a fusion prediction method of deep learning and quantile regression, a correlation coefficient data fusion prediction method, and a GRU model prediction method based on particle swarm optimization.
[0011] Optionally, the probability prediction model constructed using multiple different types of probability prediction methods performs probability prediction on the meteorological subsequence to obtain a joint meteorological probability forecast result, including: evaluating the prediction results of each algorithm based on the normalized mean width, average coverage error, cumulative deviation, and reliability and width indicators of the prediction interval coverage to obtain a comprehensive evaluation value; determining the joint weight of each probability prediction method based on the comprehensive evaluation value, wherein the comprehensive evaluation value is positively correlated with the joint weight; and constructing the probability prediction model based on the joint weight.
[0012] According to the second aspect, an embodiment of the present application provides a device for predicting new energy power under extreme climate, including: an acquisition module for acquiring climate observation data and numerical model forecast results of new energy stations under extreme climate; a feature extraction module for extracting meteorological elements and output change characteristics in the climate observation data and numerical model forecast results of new energy stations under extreme climate; an integrated forecasting module for performing multi-mode integrated forecasting on the climate observation data and numerical model forecast results of new energy stations under extreme climate based on the meteorological elements and the output change characteristics to obtain a multi-mode integrated forecasting result; a deterministic forecasting module for taking the multi-mode integrated forecasting result as input, and using a deterministic forecasting model to perform deterministic forecasting to obtain a deterministic forecasting result; a probability forecasting module for inputting the climate observation data and the deterministic forecast results under extreme climate into a probability forecasting model to obtain a meteorological probability forecasting result; and a power prediction module for predicting the new energy power based on the meteorological probability forecasting result.
[0013] According to the third aspect of the present application, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a new energy power prediction method under extreme climate as described in any one of the above items.
[0014] According to a fourth aspect of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute, when running, a new energy power prediction method under extreme climate as described in any one of the above items.
[0015] The embodiment of the present application provides a method for predicting new energy power under extreme climate, which obtains climate observation data and numerical model forecast results of new energy stations under extreme climate; extracts meteorological elements and output change characteristics from the climate observation data and numerical model forecast results of new energy stations under extreme climate; performs multi-mode integrated forecasting on the climate observation data and numerical model forecast results of new energy stations under extreme climate based on the meteorological elements and the output change characteristics to obtain a multi-mode integrated forecast result; uses the multi-mode integrated forecast result as input, and uses a deterministic forecast model to perform deterministic forecasting to obtain a deterministic forecast result; inputs the climate observation data and the deterministic forecast results under extreme climate into a probability prediction model to obtain a meteorological probability prediction result. Forecast results; forecast new energy power based on the meteorological probability forecast results; based on the in-depth analysis of the historical forecast errors of multi-mode ensemble deterministic weather forecasts and measured meteorological data, combined with modal decomposition methods, fully adaptive noise ensemble empirical modes and other methods, decompose meteorological elements, use the fitted residual components to use kernel density estimation to establish a probability prediction model, establish a PSO-optimized GRU prediction model for each sequence, and then superimpose each component to achieve multi-step interval prediction; use convolutional neural networks and long short-term memory neural networks to extract meteorological features, perform probability predictions such as quantile regression, and achieve probability / different confidence interval predictions of new energy power stations and grid locations, thereby achieving accurate predictions of new energy power under extreme climates. By analyzing the corresponding meteorological warning characteristic indicators and warning thresholds for extreme climates, extreme weather types / events are classified, and high-precision prediction and warning models for new energy output are established for extreme climates such as typhoons, cold waves and severe convection, and meteorological risk analysis and assessment of new energy output and early warning technology research are carried out to indicate the impact on new energy output. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:
[0017] Figure 1 A schematic flow chart of an optional method for predicting new energy power under extreme climate conditions according to an embodiment of the present application;
[0018] Figure 2 is a schematic diagram of a new energy power prediction device under extreme climate according to an embodiment of the present application;
[0019] Figure 3 It is a structural block diagram of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION
[0020] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the specific embodiments of the present invention are now described with reference to the accompanying drawings. In the drawings, the same reference numerals represent components with the same structure or similar structures but the same functions.
[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.
[0022] In the following description, various aspects of the present invention will be described, but it will be apparent to one of ordinary skill in the art that the present invention may be implemented using only some or all of the structures or processes of the present invention. Specific numbers, configurations, and sequences are described for clarity of explanation, but it is apparent that the present invention may be implemented without these specific details. In other cases, some well-known features will not be described in detail in order not to obscure the present invention.
[0023] According to one aspect of the present application, a method for predicting new energy power under extreme climate is provided. Figure 1 FIG. 1 is a flow chart of an optional method for predicting new energy power under extreme climate conditions according to an embodiment of the present application. Figure 1 As shown, the new energy power prediction method under extreme climate includes:
[0024] S10, obtain climate observation data under extreme climate and numerical model forecast results of new energy stations. As an exemplary embodiment, extreme climate yearbooks such as typhoons, cold waves, and severe convection in the past N years can be obtained; station observations and satellite remote sensing data are used as climate observation data under extreme climate. The numerical model forecast results of new energy stations can be constructed by downloading the corresponding ensemble forecast data of major meteorological centers through the global interactive large ensemble forecast to construct the numerical model forecast results of new energy stations.
[0025] S20, extracting meteorological elements and output change characteristics from the climate observation data under the extreme climate and the numerical model forecast results of the new energy station. In this embodiment, the meteorological elements mainly include time characteristics, spatial characteristics and vertical characteristics. The climate observation data under extreme climate and the numerical model forecast results of the new energy station are time-series aligned, and valid meteorological data are obtained through data quality control and uniformity verification. Based on the four dimensions of daily variation law, monthly variation law, seasonal variation law and annual variation law of the effective meteorological data, the analysis is carried out, and the new energy stations are divided into sections according to the principle of geographical distance such as prefecture-level cities, and the day and night changes, seasonal alternations, seasonal characteristics and annual characteristics in different regions are explored to extract time characteristics.
[0026] Based on the average numerical distribution characteristics of effective meteorological data, several indicators of volatility characteristics and intermittent characteristics are calculated, and regional volatility characteristic distribution maps and intermittent characteristic distribution maps are drawn to determine the spatial characteristics of different regions in the target area.
[0027] Based on the wind shear index of different altitude layers of effective meteorological data, the changing patterns of effective meteorological data at the vertical level under different conditions are studied as vertical characteristic features from three dimensions: different underlying surfaces (mountains, plains, hills, oceans, etc.), different atmospheric stabilities (sunny, cloudy, rainy, snowy, etc.), and different seasons (spring, summer, autumn, winter).
[0028] Based on the time characteristic features, spatial characteristic features and vertical characteristic features, statistical analysis is performed to determine the output change characteristics; among them, the output (power generation) law includes four scales of daily change, monthly change, seasonal change and annual change for statistical analysis. Daily changes show the hourly power generation change law of a single wind turbine and wind farm in a day. Monthly changes are the daily power generation change law within a month, and annual changes are the change law of power generation in each month within a year. Among them, the daily changes include the maximum and minimum output of wind and solar power throughout the day, the average output during the day, the average output at night, and the output statistics of different time periods; monthly changes include the daily average output change curve and the change law of daily power generation, and annual changes show and analyze the change law of output and monthly average power generation in each month of the year; seasonal characteristic output statistics, analysis of strong wind seasons, etc. In addition, with the increase in the scale of wind farm groups, the terrain in the field is complex, and the balance of wind energy distribution becomes more obvious. At the same time, the comprehensive calculation method of time characteristic features, spatial characteristic features and vertical characteristic features is used to determine the output change characteristics.
[0029] S30, based on the meteorological elements and the output change characteristics, a multi-mode integrated forecast is performed on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station to obtain a multi-mode integrated forecast result. Considering different time and space conditions, the weights of each model set member are optimally allocated to improve the accuracy of the forecast of wind and solar meteorological elements under extreme weather or transitional weather. According to the mechanisms such as wind speed climbing and intermittent wind speed points in the sudden change of wind speed, and extreme rainy days caused by severe convective weather, a prediction model is constructed for identification. Specifically, the types of the meteorological elements and the output change characteristics are obtained; based on the types of the meteorological elements and the output change characteristics, the integrated weights of each model member in the multi-mode integrated forecast are optimized; based on the integrated weights, a multi-mode integrated forecast is performed on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station to obtain a multi-mode integrated forecast result.
[0030] As an exemplary embodiment, the types of the meteorological elements and the output change characteristics are obtained. In this embodiment, the fluctuation state of the meteorological elements and the output change characteristics in history is calculated, and the types of the meteorological elements and the output change characteristics are determined based on the fluctuation state. In this embodiment, the types can be divided according to the interval in which the degree of fluctuation is located. Exemplarily, it can include micro-fluctuation type, medium fluctuation type and strong fluctuation type. In this embodiment, different fluctuation types can correspond to different integration weights, and at the same time, different fluctuation types can correspond to different integration methods. Generally, the more violent the fluctuation of the meteorological elements and the output change characteristics, the greater the deviation of the forecast results of each model. Therefore, the types of the meteorological elements and the output change characteristics can be used as the deviation derivatives between the members of each model, thereby reflecting the deviation rate between the members of each model. Therefore, after obtaining the numerical model forecast results of the corresponding area, the integration weights of the corresponding members of each model can be determined based on the types of the meteorological elements and the output change characteristics of the target area.
[0031] As another exemplary embodiment, a first difference between different model forecasts and a second difference between climate observation data under extreme climate and each numerical model forecast result can be calculated; the true deviation between each model member is determined based on the first difference and the second difference, and the integration weight of each model member is determined based on the true deviation.
[0032] Exemplarily, micro-fluctuation types can be integrated and predicted by the multi-model ensemble average method. In this embodiment, using multi-model simulation results and further ensemble averaging is a common method for predicting future regional weather changes. The multi-model ensemble average is a simple arithmetic average of the forecast results of each model involved in the integration. The calculation formula is as follows:
[0033]
[0034] Where Fi is the forecast value of the i-th model; n is the total number of models participating in the ensemble, V EMN It is the integrated forecast result of multi-model ensemble average.
[0035] For medium fluctuation types, a multi-mode sliding training period can be used to eliminate bias. The sliding training period method is used to set the training period to a fixed length and slide backward day by day. Each time, only one day close to the training period is forecasted. The training period window is constantly sliding, and the weight is constantly updated, which can eliminate the forecast bias and make the forecast more accurate. The Bias-Removed Ensemble Mean (BREM) is defined as follows:
[0036]
[0037] Among them, for each grid point, FBREM is the bias-eliminated ensemble mean forecast value; is the average of the observed values during the training period; N is the number of models participating in the set; Fi is the forecast value of the i-th model; is the time average of the i-th model forecast training period. It should be pointed out that this method divides the data into two parts: the training period and the forecaster. The change in the length of the training period will have a greater impact on the final forecast result. Therefore, it is necessary to compare and debug the optimal length of the training period. The training period can be set to 1 to N days for traversal experiments, and the forecast root mean square error and correlation coefficient of meteorological elements such as wind speed and irradiance in the regional forecast period under different training period lengths are calculated.
[0038] For strong fluctuations, the multi-mode Bayesian model averaging method (BMA) can be used. The Bayesian model average combination prediction overcomes the shortcomings of other prediction models that do not consider subjective prior information, do not fully extract the correct prediction information of each prediction method, and do not consider the uncertainty of the model; the key to the Bayesian model average combination prediction is to calculate the posterior probability, and the key to calculating the posterior probability is to calculate the marginal likelihood, which is a high-dimensional and complex integral. Its probability forecast (Probability density function, PDF) for a specific variable is the weighted average of the probability forecasts of a single model after bias correction. Its weight is the posterior probability of the corresponding model, which represents the relative forecasting skill of each model in the model training stage, that is, it can produce a calibrated, highly concentrated forecast PDF, which can predict the percentile of future weather distribution. Take the single-mode ECWMF ensemble forecast as an example: BMA is used to calibrate the probability matching technology of ECWMF51 members to form a single value forecast with a time resolution of 15 minutes and a maximum and minimum forecast of interval wind speed. Assume that the probability density function (PDF) of wind speed in a wind farm follows N 0 (μ,σ 2 ), that is:
[0039]
[0040] g(x|μ,σ)=0,x<0
[0041] Where: φ represents PDF and cumulative distribution function (CDF), μ and σ represent location parameter and scale parameter respectively. Assume that the location parameter μ is a linear function of ECMWF, that is:
[0042] μ=a k +b k f k
[0043] The EM (Expectation Maximization Algorithm) algorithm is used to perform maximum likelihood estimation, solve the log-likelihood function (as shown in the following formula), and calibrate the BMA model parameters.
[0044]
[0045] Let s and t represent the time and space indicators respectively, fkst represents the forecast result of the kth member in the forecast set at space s and time t, and wk is the posterior probability that the kth member is the best during the BMA model training phase, which is non-negative and satisfies Using w k The single value forecast FBMA and probability density function PDFBMA calibrated by probability matching technology are obtained.
[0046]
[0047]
[0048] Determine the optimal training period length, take the width between the 5th percentile and the 95th percentile as the interval of wind speed probability distribution; adopt the regression rolling forecast method, that is, use the previous optimal training period length as the training period for training, apply the model parameters at the training point to the 0-72h forecast, and establish a dynamic BMA forecast model.
[0049] Taking the multi-model BMA model as an example, the European Centre for Medium-Range Weather Forecasts (ECMWF) ensemble forecast, the United States' new generation mesoscale numerical model WRF, and the Italian International Center for Theoretical Physics developed RegCM, a total of three types of model forecast products are integrated (taking meteorological elements such as hub wind speed). It is also assumed that PDF obeys the truncated normal distribution of N0(μ, σ2), and the location parameter μ is a linear function of the three model products. Then there is a linear fitting relationship between μ and the three model products established in the above-mentioned ECMWF single-model ensemble forecast.
[0050] S40. Using the multi-mode integrated forecast result as input, a deterministic forecast model is used to perform deterministic forecasting to obtain a deterministic forecast result. In this embodiment, the deterministic forecast model can use machine learning, such as support vector machine; random forest, gradient boosting regression tree, Bayesian ridge regression algorithm, etc., and can also use deep learning, such as recurrent neural network, long short-term memory neural network, convolutional neural network, Transformer network model, etc.
[0051] In this embodiment, deterministic forecasting can be performed through the Transformer network model. For example, a self-attention module is used to extract feature information of forecast results of different modes; the extracted feature information is fused through a residual connection method; and a prediction unit network is used to generate a prediction element sequence as the deterministic forecast result.
[0052] In this embodiment, the order of the input sequence can be taken into account. Transformer introduces positional encoding vectors at the input of both the encoder and the decoder. The formula for calculating the positional encoding is as follows:
[0053]
[0054]
[0055] Where pos refers to the position of the current state in the sequence, and i is the index of each value in the vector. In fact, sine is used to encode even positions and cosine is used to encode odd positions. The Transformer model is composed of a set of Encoder components and a set of Decoder components, and consists only of Self-Attention and feedforward neural networks.
[0056] In this embodiment, the Transformer model includes a feature extractor, a feature fusion layer and a prediction terminal layer for multiple mode members S modes. The feature extractor follows the Transformer model, the feature fusion layer is an ordinary fully connected network, and the prediction terminal layer is a regression layer. It is expanded in two spatial dimensions, depth and width. The depth expansion helps to enhance the feature extraction ability of the model, thereby obtaining more valuable and useful information from the input data. Width expansion provides a targeted information processing mechanism for multimodal information. First, the structure of the entire model can be flexibly configured because each Transformer is independent. Second, the introduction of a certain number of units can significantly reduce the number of connections in the network. The specific workflow of the model is as follows: The S Transformers at the bottom extract features from S different historical features, and their top layers generate S corresponding high-level features, represented as r(t1), r(t2),..., r(tS). Then, the fusion layer receives these S high-order features, fuses them through nonlinear operations, and generates a more information-rich unified feature ht. Finally, the prediction terminal layer generates the predicted target sequence.
[0057] S50, input the climate observation data under the extreme climate and the deterministic forecast results into the probability prediction model to obtain the meteorological probability forecast results. As an exemplary embodiment, the predictions of new energy wind power and photovoltaic power generation are most highly correlated with wind speed and direction and solar irradiance, respectively. When constructing the prediction model, it is necessary to construct the prediction information space differently. On the basis of the changing characteristics of atmospheric physical indicators, the scheme of joint probability forecasting is carried out by using the ensemble forecast output of physical quantity indicators such as wind speed, wind direction, solar irradiance and temperature, and the probabilistic quantitative wind and light resource forecast of extreme weather is carried out. The extreme weather ensemble forecast case test is carried out using different ensemble methods of multi-source meteorology to improve the forecast and early warning capabilities of extreme weather. The probability prediction method can provide wind speed uncertainty information, which is used to predict the possibility of a certain weather phenomenon in the future, usually represented in the form of quantile and uncertainty interval probability density function. The probability of extreme weather occurring at a given time and place can be predicted.
[0058] For example, on the basis of deterministic forecast results, through probabilistic prediction models including time series residual probability hybrid model, multi-source Gaussian ensemble kernel fitting method, deep learning and quantile regression fusion method, etc., the connotation of renewable energy power prediction is expanded to improve the safety, stability and economy of the power system.
[0059] S60, predicting the new energy power based on the meteorological probability forecast result. In this embodiment, after obtaining the deterministic forecast result and the probabilistic forecast result, a model between the power generation equipment and the meteorological data can be constructed. In this embodiment, the power generation equipment takes a wind turbine as an example:
[0060] Based on data preprocessing by empirical mode decomposition and model optimization by particle swarm optimization (PSO), basic models such as BP neural network, support vector machine (SVM), and long short-term memory network (LSTM) are combined to determine the prediction model for the operating status of wind-level units under different meteorological factors, and then determine the new energy power.
[0061] The added adaptive noise component is subjected to multiple empirical mode decomposition through the complete empirical mode decomposition of adaptive noise, and the unique intrinsic component is obtained by calculating the average value of the results. Specifically, y(t) is obtained by adding the reference standard white noise to the initial data signal x(t), and then EMD decomposition is carried out on y(t), and the results are averaged to obtain the first-order mode function M1(t), and the first-order residual term is calculated. Similarly, the signal after adding noise is subjected to EMD decomposition to obtain the second-order mode function and the second-order residual term. By analogy, the h-th order mode function M is obtained. h (t) and the h-th order residual term (h 2,3,......,H). Repeat the steps of solving the mode function and residual term until the residual term cannot be decomposed, and obtain the final residual term R H(t), its calculation expression is:
[0062]
[0063] The model parameters are optimized based on particle swarm. Specifically, each solution to the problem to be optimized is regarded as a particle in the target search space. In a specific search space, the particle runs at a certain speed, and the speed is dynamically adjusted according to environmental factors and its own flight conditions. The coordinates and flight speed of the particle are recorded. After optimizing each particle, the optimal value that satisfies the objective function is found.
[0064] Assume that in an N-dimensional development target search distribution space, there are M particles forming a biological population. The i-th particle in the population represents an N-dimensional distribution vector, and Pi represents the specific position of the i-th particle. Its calculation expression is:
[0065] P i =(p i1 ,p i1 ,...,p iN ),i=1,2,...,M
[0066] The moving speed of the i-th particle is an N-dimensional distribution vector, and its calculation expression is:
[0067] V i =(v i1 , v i1 , ..., v iN ), i = 1, 2, ..., M
[0068] The optimal reasonable solution of the individual currently searched by the i-th particle, that is, the individual extreme value, is calculated as:
[0069] E best =(p i1 , p i1 , ..., p iN ), i = 1, 2, ..., M
[0070] The global optimal solution searched by the entire particle swarm, that is, the global extreme value, is calculated as follows:
[0071] G best =(p g1 , p g1 , ..., p gN )
[0072] The velocity and position of the particle are updated using individual extrema and global extrema. The calculation expression is:
[0073] V i =W i*V i +C1*R1*(E best -P i )+C2*R2*(E best -P i )
[0074] X i =x0+v i (k+1)
[0075] Where C1 and C2 are accelerations, also called learning ratios, W is the inertia constant, and R1 and R2 are random probability numbers in the control range of [0, 1].
[0076] In this embodiment, the prediction model uses BP network, support vector machine SVM and long short-term memory neural network LSTM. In this embodiment, the prediction process of the algorithm is introduced by taking LSTM network as an example: LSTM network is structurally improved on the basis of RNN, and the improved model repetitive structure is relatively complex. LSTM network consists of memory unit and update gate, forget gate and output gate control gate.
[0077] The forget gate filters the information of the previous moment according to the information input at the current moment, determines the proportion of information to be retained, and decides whether to keep or discard the information. The information of the hidden state of the previous moment and the information input at the current moment are simultaneously passed to the sigmoid function. Through the sigmoid layer and dot product, the output value is controlled between 0 and 1. When the output value is close to 1, it means that the information should be retained; when the output value is close to 0, it means that the information should be discarded.
[0078] The update gate updates the cell state of the memory unit by sharing the information input at the current moment, and determines the information that needs to be updated. It is also called the input gate. The information of the hidden state at the previous moment and the current input information are passed to the sigmoid function, and the output value is controlled between 0 and 1 through the sigmoid layer and point multiplication. When the output value is close to 1, it means that the information is relatively important and should be updated; when the output value is close to 0, it means that the information is not important and should be discarded.
[0079] The output gate passes the hidden state of the previous moment and the input of the current moment to the sigmoid function at the same time, and passes the output of the sigmoid function to the tanh function. The state information of the hidden layer is determined by multiplying and adding the state outputs of the sigmoid layer and the tanh layer, so as to determine and output the value of the output at the next moment.
[0080] In this application, based on the in-depth analysis of the historical prediction errors of multi-mode ensemble deterministic weather forecasts and measured meteorological data, the meteorological elements are decomposed by combining modal decomposition methods, fully adaptive noise ensemble empirical modes and other methods, and a probability prediction model is established using kernel density estimation using the fitted residual components. After establishing a PSO-optimized GRU prediction model for each sequence, each component is superimposed to achieve multi-step interval prediction; convolutional neural networks and long short-term memory neural networks are used to extract meteorological features, and probabilistic predictions such as quantile regression are performed to achieve probability / different confidence interval predictions of new energy power stations and grid locations, thereby achieving accurate predictions of new energy power under extreme climates. By analyzing the meteorological warning characteristic indicators and warning thresholds corresponding to extreme climates, extreme weather types / events are classified, and high-precision prediction and warning models for new energy output are established for extreme climates such as typhoons, cold waves and severe convection, and meteorological risk analysis, assessment and warning technology research on new energy output is carried out to indicate the impact on new energy output.
[0081] As an exemplary embodiment, extracting meteorological elements and output change characteristics from the climate observation data under extreme climate and the forecast results of the numerical model of new energy stations includes:
[0082] Data quality control and homogeneity verification are performed on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station to obtain valid meteorological data and deviation factors. The main contents of the quality control of meteorological observation data when new energy stations face extreme climates such as typhoons, cold waves, and severe convection are to analyze and check the quality of the observation data, and record the quality of the observation data, so that errors can be found and modified in time. In most cases, the acquisition of meteorological observation data has a certain procedure, namely collection, transmission, encoding and decoding, etc., and there may be certain deviations in the meteorological observation data in these links. Therefore, there is always a quality control problem in the process of meteorological observation data and consistency inspection of meteorological observation data.
[0083] Quality control of meteorological observation data under extreme climates such as typhoons, cold waves, and severe convection: static information inspection, unified format of measured data, extreme value inspection, consistency inspection of meteorological elements, and time consistency inspection. Among them, static information inspection includes static information inspection of longitude and latitude information (longitude and latitude format and precise decimal places) of wind farm wind turbines and photovoltaic power station environmental monitoring instruments, power station names, and power station altitudes. Extreme value inspection: After obtaining the measured data in a unified standard format, use the program to determine whether each measured meteorological element exceeds the maximum or minimum value required by the national standard. Of course, according to the actual situation, as the geographical conditions change, the extreme values of the elements may also develop a certain degree of change, so the parameters of the extreme values can also be formulated based on the historical meteorological elements of each station in the early survey. Spatial consistency inspection: The geographical space of the distribution of meteorological elements is correlated. Meteorological stations with relatively close spatial distances have greater similarity in their characteristic values than those with a longer distance. This is one of the theoretical bases of spatial interpolation and the theoretical basis of spatial consistency inspection. Time consistency check: Check the continuity of meteorological element values over a period of time (such as 24 hours). If the element values do not change, it indicates that the observation instrument or transmission equipment may have a fault. Some ground meteorological observation data generally have good consistency with time. By comparing and analyzing the measured values of meteorological observation data before and after the occurrence time, it is possible to understand whether there are abnormalities in the observation data values.
[0084] The data cleaning methods for meteorological observations in new energy stations under extreme climates such as typhoons, cold waves, and severe convection can adopt mathematical methods such as statistical analysis, 3sigma method, wavelet singular point monitoring method, and quartile algorithm. Artificial intelligence algorithms such as support vector machine regression algorithm, K nearest neighbor method, optimal variance method within the group, and outlier detection algorithm can also be used.
[0085] In this embodiment, the original data may be judged for abnormality first, and then the abnormal data may be processed. Exemplarily, the abnormality judgment of the original data may include that there are many missing data of important attributes in the original collected data. If the missing values and wrong values are simply discarded, it may have a serious impact on the application of the data and change the true validity of the original data. For example, misestimating the daily average wind speed of the wind farm, underestimating or overestimating the output of the wind farm, etc., can all mislead the subsequent power generation plan. Data quality control is a prerequisite for correctly evaluating the operation of the wind farm. There are generally four forms of abnormal status of wind farm data: missing numbers, dead numbers, wrong numbers, and verification failure.
[0086] Among them, the judgment rules for missing data are: the time resolution of the original collected data is one point every 5 minutes, with a total of 288 data in a day, and the first point corresponds to 00:00:00 every day. If the time resolution of the actual power, predicted power, wind speed and direction, temperature, humidity and other data of the wind tower and numerical weather forecast exceeds 5 minutes, the data is considered missing. Due to manual operation and other reasons, the power restriction records are missing. The rules for judging power restriction are detailed in the data analysis section.
[0087] Rules for judging dead numbers: To facilitate data identification, all data are retained to 4 significant digits after the decimal point. For wind measurement data, when there is output, if the adjacent absolute error of 6 consecutive data (30 minutes) = 0, it is considered a dead number (wind speed has 6 consecutive data unchanged, such as no wind; power has 6 consecutive data unchanged, such as no output or full power); if there is wind and 6 consecutive power numbers are 0, it is considered a dead number. For light measurement data, when irradiance>0, 6 consecutive unchanged numbers are considered a dead number; when irradiance=0, 6 consecutive non-zero data are considered a dead number.
[0088] Rules for judging wrong numbers: ① The reasonable range of wind speed is [0, 60] m / s, the reasonable range of hourly average wind speed is [0, 40] m / s; the reasonable range of wind direction is [0, 360°); the range of air pressure is [500, 1100] hPa; the reasonable range of humidity is [0, 100%] RH; the reasonable range of ambient temperature is [–60, 60] ℃, and the reasonable range of component temperature is [-60, 150] ℃. ② The reasonable range of photovoltaic irradiance is [0, 2000] W / ㎡.
[0089] Judgment rules for data that fails verification: Data that fails verification often requires the use of data analysis methods.
[0090] Abnormal data processing For the missing, dead, wrong, and check-failed data determined by the above method, data processing is required. Exemplarily, abnormal data processing is performed according to the following principles: the monitoring data of the wind tower should remain consistent. For example, the absolute deviation of the hourly average wind speed at adjacent heights of 70m, 50m and 30m of the wind tower is less than 2m / s; the absolute deviation of the wind speed at adjacent times at the same floor height of the wind tower is less than 20m / s; the ratio of the number of collected data with a deviation of less than 5% from the true data to the number of all collected data; the availability of collected data should be greater than 95%.
[0091] Specifically, for the treatment of missing numbers:
[0092] Processing of wind speed, wind direction, power missing data, and irradiance (photovoltaic) collected by a single machine: missing data within 6 consecutive points are completed with the previous point.
[0093] Processing of missing data for wind speed, wind direction, temperature, humidity and pressure of all layers collected by wind tower data: missing data within 6 consecutive points are completed with the previous point.
[0094] Processing of missing data in the entire field: missing data within 6 consecutive points will be filled with the previous point.
[0095] The time length for making up the numbers is no more than half an hour. Missing numbers or dead numbers that last for more than half an hour will not be processed.
[0096] According to the method for determining the missing power-limiting records, if it is determined that there are missing power-limiting records within a certain period of time, the data for that period of time will be filtered out or theoretical missing compensation will be performed in the model application.
[0097] Dealing with wrong numbers:
[0098] If the wind speed data is less than 0m / s, it will be replaced by 0m / s; if the wind speed data is greater than 60m / s, it will be replaced by 60m / s. The error handling method for irradiance, wind direction, temperature, humidity, air pressure, etc. is similar to the handling method for wind speed.
[0099] If the power data of a single machine (the whole field) is less than Tp1, it shall be replaced by Tp1, where Tp1=–P×10%; if the power data of a single machine (the whole field) is greater than Tp2, it shall be replaced by Tp2, where Tp2=P×110%, where P is the rated power of the single machine (the installed capacity of the whole field).
[0100] Status data: If the shutdown or output status is wrong, the status will be changed.
[0101] Processing of data that fails verification:
[0102] Data processing when the absolute deviation of the hourly average wind speed at adjacent storeys of the wind tower exceeds 2m / s: If the wind speed at the 70m storey is correct, the wind speed at that storey shall be taken as the standard. For every 20m decrease or increase, the wind speed shall be subtracted or increased by 2m / s to replace the wind speed at the original storey; If the wind speed at the 70m storey is wrong or a dead number, find the correct wind speed at the adjacent storey at the same time. For every 20m decrease or increase, the wind speed shall be subtracted or increased by 2m / s to replace the wind speed at the original storey; If the wind speeds at all storeys are incorrect, use the correct wind speed of the close-range aircraft head to replace the wind speed at the 70m storey, and process the wind speeds at other storeys in turn.
[0103] If the absolute deviation of wind speed at adjacent times at the same floor height of the wind measurement tower exceeds 20m / s, the wind speed at the previous moment will be increased (or decreased) by 20m / s to replace the wind speed value at the next moment.
[0104] After obtaining valid data, the time characteristic features, spatial characteristic features and vertical characteristic features in the valid meteorological data are extracted as the meteorological elements. In this embodiment, wind resources are taken as an example for explanation:
[0105] First, the temporal variation characteristics of wind speed and direction are explored. In this embodiment, during the influence of extreme weather, the temporal variation and distribution characteristics of the 5-minute average wind speed and wind direction at different heights observed by the wind measuring tower, the difference in the temporal variation of wind speed at each height, the relationship between the bottom and high-level wind speeds between the peak values, and the location and movement trajectory of the typhoon eye area are evaluated based on the changes in wind speed and direction.
[0106] The temporal non-uniformity of the wind field is very large. The difference between the wind speed / direction in a certain period and the previous moment is defined to quantitatively characterize the temporal variation characteristics.
[0107]
[0108] Δ0(t))=0(t)-0(t-1)
[0109] Where V is the horizontal wind vector, θ is the wind direction angle, 0° to the north and 90° to the east, and the interval between the two times is 5 minutes.
[0110] For effective meteorological data, the wind speed model can be used to extract time-varying wind speed, average wind speed and average wind direction; extract turbulence intensity, gust factor and turbulence integral scale; extract the distribution of turbulent pulsation energy at various frequencies, that is, the power spectrum; extract probability density function; wind profile characteristics to characterize the curve of wind speed change with height; extract wind shear index to indicate the change of wind speed in the plane perpendicular to the wind direction; extract short-term changes in wind direction; count the frequency of strong winds, etc.
[0111] The wind speed model can be constructed based on the temporal variation characteristics of wind speed and wind volume. The typhoon average wind speed shows an unsteady characteristic over time. The surrounding environment of the wind farm is complex. The non-steady wind speed model means that the wind speed is decomposed into the time-varying average wind and the steady pulsating wind speed with a mean of 0. The horizontal and vertical wind directions are analyzed according to the non-stationary method. The essence is to split the wind speed into "trend terms" and "steady processes", where the "trend terms" correspond to the average wind speed. The time-varying average wind speed is the trend term established by the wind speed data at low frequency or the sum of several items in the low-frequency band, which can be obtained by signal processing such as EMD and DWT.
[0112] The non-stationary wind speed model expression is:
[0113]
[0114]
[0115] in, represents the time-varying mean wind along the wind direction, u * (t) represents the zero-mean steady random pulsating wind in the downwind direction. represents the time-varying mean wind across the wind direction, v *(t) indicates the mean zero-mean steady random fluctuating wind in the crosswind direction. * indicates the calculation is based on the non-stationary model.
[0116] Specifically, the following method can be used to extract time-varying wind speed based on the wind speed model:
[0117] 1. Perform EMD decomposition on the wind speed sample U(t), which may include: finding all extreme points of U(t); using the cubic spline difference function to draw the extreme envelope; averaging the extreme envelope to obtain m1(t), and then subtracting m1(t) from the original wind speed sample to obtain the first component c1(t); performing IMF conditional judgment on c1(t). If it does not meet the IMF condition, re-screen it. If it meets the IMF condition, c1(t) is the first high-frequency IMF function h1(t):
[0118] c1(t)=U(t)-m1(t)
[0119] h1(t)=c 1k (t)
[0120] Extract h1(t) from U(t) and get the remaining data sequence r1(t)=U(t)-h1(t); take r1(t) as a new template and repeat the S1-S5 process to get the second intrinsic mode function h2(t). And so on. It is known that the remaining component is a monotonic function, that is, the residual generated is r n-1 (t)-h n (t) = r n (t).
[0121] Through EMD decomposition, the non-stationary wind speed process U(t) is decomposed into several IMFs and a residual term r n (t):
[0122]
[0123] In the formula, h j (t) is the intrinsic mode function, r n (t) is the residual component, and n is the natural mode order of the empirical mode wind turbine. When the residual period after screening is greater than the sample time history, the residual is artificially regarded as the time-varying average wind speed.
[0124] 2. Discrete wavelet extraction of time-varying average wind: The discrete wavelet transform and inverse transform of any wind speed sequence U(t) based on DWT theory are represented as:
[0125]
[0126]
[0127] Where U(t) represents the wind speed sequence, ψ j,kIt is a family of functions derived from the basic wavelet ψ(t) by extension and translation, and the symbol * represents conjugation.
[0128] 3. Time-varying average wind speed extraction, in order to accurately evaluate the time-varying average wind, the stability of the pulsating wind is verified to determine:
[0129] For U(t), the expression of fluctuating wind is The stability of the pulsating wind is r = r(t(t)), where is the average wind speed; the time-varying wind speed is extracted from U(t) using EMD and wavelet transform method. Get the fluctuating wind speed under the non-stationary wind speed model And calculate its stability r1=r(u * (t)); verify whether the stability r1 meets the stability requirements.
[0130] For the average wind speed and average wind direction, the horizontal wind speed u(i) and wind direction Wind speed can be divided into two components on the coordinate axis.
[0131]
[0132]
[0133] In the specific calculation and analysis, a certain time interval is taken as the average time interval, such as 5 minutes, then the horizontal average wind speed U and wind direction θ are:
[0134]
[0135]
[0136] In the formula, step() is a step function. When it is a negative value, the function value is 0. When is positive, the function value is . and u x (i) and u y (i) The mean of samples at a certain time interval. Within a certain time interval, the fluctuating wind speed in the downwind and acrosswind directions and It can be expressed as:
[0137]
[0138]
[0139] The average time-distance statistics can better reflect the pulsation and changing patterns of typhoon wind speed, wind direction, turbulence and gust factor.
[0140] The pulsation intensity of the wind can be represented by turbulence intensity and gust factor. By estimating the probability density function of the pulsating wind speed, the Gaussian type of the typhoon data can be determined. In this embodiment, the extraction of turbulence intensity, gust factor and turbulence integral scale can include:
[0141] Turbulence intensity reflects the pulsation characteristics of wind and is a key parameter for determining the fluctuating wind load on the structure. Turbulence intensity is the ratio of the wind speed standard deviation to the horizontal wind speed modulus. The ratio of the wind speed standard deviation at a 5-minute interval to the horizontal average wind speed is used, and the calculation formula is:
[0142]
[0143] Where I is the turbulence intensity, σ is the standard deviation of wind speed at 5-minute intervals, and U is the corresponding 5-minute horizontal average wind speed.
[0144] The gust factor is the gust duration t g The ratio of the maximum average wind speed within a certain time interval to the average wind speed U.
[0145] The power spectrum is the distribution of turbulent pulsation energy at various frequencies. It is used to describe the proportion of turbulent pulsation kinetic energy of vortices of different scales, and is a representation of the energy distribution of pulsating wind in the frequency domain. Based on the Kolmogorov principle, the pulsation spectrum expression is expressed by VonKarman spectrum, Davenport spectrum, Harris spectrum, Hion spectrum, Kaimal spectrum, Simiu spectrum, Panofsky spectrum, etc.
[0146] For probability density function: six distribution functions are commonly used in wind speed simulation, including generalized extreme value distribution, lognormal distribution, Weibull distribution, and inverse Gaussian distribution:
[0147] Normal distribution Generally speaking, the probability function of artificial fluctuating wind speed in the non-steady wind speed model obeys the normal distribution, and the calculation formula is as follows:
[0148]
[0149] In the formula, u * Indicates the pulsating wind speed, Represents standard deviation.
[0150] Generalized extreme value distribution:
[0151]
[0152] Where v is the wind speed, k, σ and μ are unknown parameters.
[0153] Wind speed profile refers to the curve of wind speed changing with height, which is often used to estimate wind speed at different heights. The reasons for the vertical change of wind speed in the near-surface layer are dynamic factors and thermal factors. The former mainly comes from the friction effect of the underlying surface, that is, the roughness of the underlying surface, and the latter is mainly manifested in the relationship with the atmospheric stability in the near-surface layer. When the atmospheric stratification is neutral, the wind speed in the near-surface layer changes with height and approximately obeys the logarithmic function and the power exponential function.
[0154] The logarithmic function of wind speed changing with height is:
[0155]
[0156] Where u is the wind speed at height z, u1 is the wind speed at height z1, and z0 is the roughness of the underlying surface.
[0157] The power exponential function of wind speed changing with height is:
[0158]
[0159] Where a is the wind speed profile power exponent, which is a parameter that characterizes the roughness of the underlying surface and reflects the strength of the vertical shear of the wind speed.
[0160] The wind shear index indicates the change in wind speed in a plane perpendicular to the wind direction, and its magnitude reflects how fast the wind speed increases with height. Its calculation formula is:
[0161] ①Power law formula:
[0162]
[0163] ② Index formula:
[0164]
[0165] Where α is the wind shear index, z1 and z2 are the known height and the height of the changed wind speed respectively, and v1 and v2 are the wind speeds at the heights of z1 and z2 respectively.
[0166] For short-term changes in wind direction: the 10-minute wind direction turning green is defined as the absolute value of the difference between the wind direction angle in the next 10 minutes and the wind direction angle in the previous 10 minutes.
[0167] For the statistics of strong wind frequency: the cumulative frequency distribution of strong winds in the wind power output gain, the possibility of cut-out, and the period with wind speed greater than the cut-out wind speed are counted separately, and the changing characteristics of the cumulative frequency of strong winds caused by differences in terrain height and mountain shape near wind farms are analyzed.
[0168] Based on the statistical analysis of the meteorological elements, the output change law data is obtained. The power generation law includes four scales for statistical analysis: daily change, monthly change, seasonal change and annual change. Daily changes show the hourly power generation change law of a single wind turbine and wind farm in a day. Monthly changes are the daily power generation change law within a month, and annual changes are the change law of power generation each month within a year. Among them, the daily changes include the maximum and minimum output of wind and solar power throughout the day, the average output during the day, the average output at night, and the output statistics of different time periods; monthly changes include the daily average output change curve and the change law of daily power generation, and annual changes show and analyze the output of each month in a year and the change law of monthly average power generation; seasonal characteristic output statistics, analysis of strong wind seasons, etc. In addition, with the increase in the scale of wind farm groups, the terrain in the field is complex, and the balance of wind energy distribution becomes more obvious, and a comprehensive calculation method of spatiotemporal distribution characteristics is used.
[0169] The output change characteristic is determined based on the deviation factor and the output change law data. In this embodiment, cluster analysis is performed on the data with deviations, key deviation factors in the clusters are extracted, and mechanism analysis is performed on the key deviation factors to obtain the deviation factors. The processing change law data is corrected based on the deviation factors to obtain the output change characteristic.
[0170] The multi-model integrated forecasting is performed on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station based on the integrated weight to obtain the multi-model integrated forecasting results, including:
[0171] The numerical model forecast results of the new energy station are dynamically downscaled to obtain meteorological element forecast results; the climate observation data under the extreme climate are standardized to obtain standardized observation data; significant factor correlation analysis is performed on the standardized data and the meteorological element forecast results to obtain significant factor analysis results; principal component analysis is performed on the significant factor analysis results to obtain principal components; the numerical model forecast results of the new energy station are corrected based on the principal components by using a statistical downscaling method to obtain corrected numerical forecast results; the corrected numerical forecast results are subjected to multi-model integrated forecasting based on the integrated weights to obtain multi-model integrated forecast results.
[0172] Specifically, the dynamic downscaling method is to embed a low-resolution global climate model into a high-resolution regional climate model, use the global model to provide initial and boundary conditions for the regional climate model, and obtain high-resolution prediction information that describes regional climate characteristics.
[0173] In this embodiment, in view of the actual situation of complex terrain in the target area, the WRF terrain data is optimized, and high-resolution terrain data similar to ASTER elevation data is used as the input source for terrain modeling, so as to more realistically depict the geographical distribution characteristics of the Guangxi power grid area and obtain high-precision historical simulation; the driving field (including reanalysis data or GCM data, etc.) provides RCM with initial fields, lateral boundary conditions and surface initial fields (such as soil temperature and humidity and sea surface temperature). During downscaling, RCM is forced through the initial field and lateral boundary conditions to pass the driving field weather background information to the RCM simulation results. When the simulation area is large or the simulation period is long, the weather situation simulated by RCM may deviate from the weather situation of the driving field. Ensure that the RCM simulation results are consistent with the weather situation of the driving field. The consistent methods include segmented simulation, analytical Nudging, spectral Nudging and scale selection bias correction; the Guangxi power grid area is gridded with a horizontal resolution of 1km, and each grid point in the model dynamic downscaling process can more accurately reflect the terrain conditions of the grid point; for the historical observation data of meteorological stations available within the simulation range, quality control is carried out, including outlier removal and missing value interpolation, and Kalman filtering is used to revise reliable historical observation data and mesoscale historical simulation data to better invert the historical climate.
[0174] Select appropriate factors from the historical model output to establish a statistical relationship with the observed value, substitute the corresponding model output into the statistical relationship in the actual forecast, and obtain the revised forecast value. This method can effectively reduce the systematic error of the model. The multivariate linear regression model is often used to establish a statistical relationship, that is, to find a regression model between the observed data and the output results of a certain numerical model, and use this statistical relationship in the actual forecast to transform a single-value model forecast into another single-value forecast.
[0175] Factor selection: During regression analysis, it is necessary to select a group of factors with moderate number and good relationship from the many factors that may affect the dependent variable to establish a regression equation. The stepwise introduction method is often used to select factors. The stepwise introduction method is to select the factor with the largest variance contribution from the p candidate factors given by the model. If the factor passes the significance test, it is added to the regression equation.
[0176] After determining the regression equation of the existing factors, the difference between the observed value and the regression estimate is used as the forecast value, and the steps of factor selection are repeated for the remaining candidate factors, and new factors are introduced until a satisfactory regression equation is obtained. The step-by-step introduction method can effectively reduce the amount of calculation while selecting the best factor solution.
[0177] Taking the solar radiation forecast model as an example, based on the WRF model output results and the total radiation observation data, the MOS forecast process of hourly solar radiation is designed, as shown in the figure. It includes three key steps: low-pass filtering the hourly observation sequence and dividing it by the astronomical radiation, screening and dimension reduction of the model output factors, and establishing the MOS forecast equation. Where: I represents the hourly total radiation observation sequence; Represents the result of low-pass filtering the observation sequence; Represents the result after filtering and dividing by the astronomical radiance.
[0178] How to select the factors of the MOS prediction equation? According to the hourly output of each physical quantity of the WRF model, the output factors significantly correlated with radiation observations are extracted through statistical analysis. The table shows the correlation statistics between radiation observations and model output factors. The principal component analysis is performed on the 13 radiation-related factors output by the model, and each principal component is then compared with The sequence is subjected to correlation analysis, and the principal components with better correlation are selected as factors for establishing the MOS prediction equation.
[0179] In this embodiment, the probability distribution of the required predicted weather variables is first given, and then the parameters of the predicted distribution are connected to the predicted members using a function, and the model parameters are solved by an optimization algorithm to achieve prediction correction. For example, the original EMOS method uses a Gaussian distribution to model temperature and sea level pressure, and the mean of the Gaussian distribution is the mapping function of the predicted members, and its variance is the mapping function of the set variance. This method is conceptually simpler and clearer than the currently very popular Bayesian model averaging method, and the required amount of data samples is much less than that of the Bayesian model averaging method, so it will be more flexible and convenient in use.
[0180] Inputting the climate observation data under extreme climate and the deterministic forecast results into the probability prediction model to obtain the meteorological probability forecast results includes:
[0181] Based on the climate observation data under the extreme climate and the deterministic forecast results, an original meteorological sequence is constructed; the original meteorological sequence is decomposed to obtain multiple meteorological sub-sequences; the meteorological sub-sequences are probabilistically predicted by a probability prediction model constructed using multiple different types of probability prediction methods to obtain a joint meteorological probability forecast result, wherein the probability prediction method includes at least two of a time series residual probability hybrid prediction method, a multi-source Gaussian ensemble kernel fitting prediction method, a fusion prediction method of deep learning and quantile regression, a correlation coefficient data fusion prediction method, and a GRU model prediction method based on particle swarm optimization.
[0182] In this embodiment, the prediction results of each algorithm are evaluated based on the normalized mean width, average coverage error, cumulative deviation, and reliability and width indicators of the prediction interval coverage to obtain a comprehensive evaluation value; the joint weight of each probability prediction method is determined based on the comprehensive evaluation value, wherein the comprehensive evaluation value is positively correlated with the joint weight; and the probability prediction model is constructed based on the joint weight.
[0183] As an exemplary embodiment, the probability prediction results can be output based on historical data using a time series residual probability hybrid prediction method, a multi-source Gaussian set kernel fitting prediction method, a fusion prediction method of deep learning and quantile regression, a correlation coefficient data fusion prediction method, and a GRU model prediction method based on particle swarm optimization. Each probability prediction result is evaluated based on the normalized mean width, average coverage error, cumulative deviation, and reliability and width indicators of the prediction interval coverage to obtain an evaluation value, and the joint weight of each prediction method is determined based on the evaluation value. Afterwards, after obtaining the new climate observation data and the deterministic forecast results, the probability prediction results output by the time series residual probability hybrid prediction method, the multi-source Gaussian ensemble kernel fitting prediction method, the fusion prediction method of deep learning and quantile regression, the correlation coefficient data fusion prediction method and the GRU model prediction method based on particle swarm optimization are fused based on the joint weights, or the time series residual probability hybrid prediction method, the multi-source Gaussian ensemble kernel fitting prediction method, the fusion prediction method of deep learning and quantile regression, the correlation coefficient data fusion prediction method and the GRU model prediction method based on particle swarm optimization are fused into a probability prediction model based on the joint weights, and the new climate observation data and the deterministic forecast results are input into the probability prediction model for probability prediction.
[0184] In this embodiment, the prediction method based on the time series residual probability hybrid model can use the improved variational mode decomposition method to decompose the wind speed into components with different frequency characteristics; then use kernel density estimation to establish a probability prediction model for the fitting residual components.
[0185] The multi-source Gaussian ensemble kernel fitting method can first construct the joint probability density of multi-mode integrated prediction and determine the joint probability density of the m-dimensional random column vector X.
[0186] The fusion method of deep learning and quantile regression uses convolutional neural network and long short-term memory neural network to extract the features of wind speed, and then uses the extracted features for quantile regression.
[0187] The correlation coefficient data fusion method uses the fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) to decompose the original wind speed, sample entropy to perform data fusion, and the traditional confidence interval method to predict the wind speed range.
[0188] The multi-step interval prediction of wind speed based on VMD-FIG and parameter optimized GRU, that is, the multi-step interval prediction method of wind speed based on VMD-FIG combined with parameter optimized GRU, firstly uses VMD to decompose the wind speed sequence into K relatively stable subsequences, and reconstructs them into three parts of trend, oscillation and noise according to SE; FIG is used to extract the minimum, average and maximum value of each window for the noise part; then a PSO optimized GRU prediction model is established for each sequence separately, and then the prediction results of each component are superimposed to realize the multi-step interval prediction of wind speed.
[0189] As an exemplary embodiment, the normalized average width is represented by PINAW, the average coverage error is represented by ACE, the cumulative deviation is represented by ADI, and the reliability and width index of the prediction interval coverage are represented by the comprehensive index WS, where:
[0190] PINAW evaluates the width of the prediction interval and the calculation formula is:
[0191]
[0192] In the formula, and They are the upper and lower limits of the τ confidence level interval; Max p and Min p are the maximum and minimum values of the predicted values respectively; k m is the scaling factor, which is generally set to 1.5; m is the dimension, and τ is the delay time. While meeting the reliability requirement, the smaller the PINAW is, the more accurate the prediction information can be provided, which is beneficial for decision making.
[0193] The reliability of the ACE indicator reaction interval is calculated as follows:
[0194]
[0195] Where x0(i) is the true value. The smaller the absolute value of ACE, the more accurate the coverage and the more reliable the probability prediction results. ACE can only quantitatively analyze the interval prediction results and cannot accurately and directly reflect the degree of deviation between the actual wind speed and the prediction interval.
[0196] ADI describes the interval envelope characteristics, and the calculation formula is:
[0197]
[0198] The smaller the ADI value, the smaller the deviation of the predicted interval and the better the prediction result.
[0199] WS takes into account both the reliability and width of the forecast interval coverage, and the calculation formula is:
[0200]
[0201]
[0202] The smaller the absolute value of WS is, the more likely the prediction model is to guarantee high coverage within a lower range.
[0203] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM (Read-Only Memory) / RAM (Random Access Memory), a disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.
[0204] The present application also provides a device for predicting new energy power under extreme climate conditions. Figure 2 A device for predicting new energy power under extreme climate conditions according to an embodiment of the present application includes:
[0205] An acquisition module 10 is used to obtain climate observation data under extreme climate and forecast results of numerical models of new energy stations;
[0206] A feature extraction module 20 is used to extract meteorological elements and output change characteristics from the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station;
[0207] An integrated forecasting module 30 is used to perform multi-mode integrated forecasting on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station based on the meteorological elements and the output change characteristics to obtain a multi-mode integrated forecasting result;
[0208] The deterministic forecast module 40 is used to take the multi-model integrated forecast result as input, use the deterministic forecast model to perform deterministic forecasting, and obtain a deterministic forecast result;
[0209] The probability forecast module 50 is used to input the climate observation data under extreme climate and the deterministic forecast result into the probability prediction model to obtain the meteorological probability forecast result;
[0210] The power prediction module 60 is used to predict the new energy power based on the meteorological probability forecast result.
[0211] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the contents disclosed in the above embodiments. It should be noted that the above modules as part of the device can be implemented by software or hardware, wherein the hardware environment includes a network environment.
[0212] According to a fourth aspect of the present application, there is provided an electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, and the memory is used to store a computer program; and the processor is used to execute a method in any one of the embodiments described above by running the computer program stored in the memory.
[0213] Figure 3 is a structural block diagram of an optional electronic device according to an embodiment of the present application, such as Figure 3 As shown, it includes a processor 302, a communication interface 304, a memory 306 and a communication bus 308, wherein the processor 302, the communication interface 304 and the memory 306 communicate with each other through the communication bus 308, wherein,
[0214] Memory 306, used for storing computer programs;
[0215] The processor 302 is used to implement the following steps when executing the computer program stored in the memory 306:
[0216] Obtain climate observation data under extreme climate conditions and numerical model forecast results of new energy stations;
[0217] Extract meteorological elements and output change characteristics from climate observation data and new energy station numerical model forecast results under the extreme climate;
[0218] Based on the meteorological elements and the output change characteristics, a multi-model integrated forecast is performed on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station to obtain a multi-model integrated forecast result;
[0219] Taking the multi-model integrated forecast result as input, using a deterministic forecast model to perform deterministic forecasting to obtain a deterministic forecast result;
[0220] Inputting the climate observation data under the extreme climate and the deterministic forecast results into a probability prediction model to obtain a meteorological probability forecast result;
[0221] The new energy power is predicted based on the meteorological probability forecast results.
[0222] Optionally, in this embodiment, the communication bus may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.
[0223] The communication interface is used for communication between the above electronic device and other devices.
[0224] The memory may include RAM, or may include non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.
[0225] The above-mentioned processor can be a general-purpose processor, which can include but not be limited to: CPU (Central Processing Unit), NP (Network Processor), etc.; it can also be DSP (Digital Signal Processing), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
[0226] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and this embodiment will not be described in detail here.
[0227] It can be understood by those skilled in the art that Figure 3 The structure shown is for illustration only. The device for implementing the method described in any one of the above embodiments may be a terminal device, which may be a smart phone (such as an Android phone, an IOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (Mobile Internet Devices, MID), a PAD, and other terminal devices. Figure 3 It does not limit the structure of the above electronic device. For example, the terminal device may also include Figure 3 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Figure 3Different configurations shown.
[0228] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, which can include: a flash drive, ROM, RAM, a magnetic disk or an optical disk, etc.
[0229] As an exemplary embodiment, the present application also provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the method steps described in any one of the embodiments when running.
[0230] Optionally, in this embodiment, the above-mentioned storage medium can be used to execute the program code of the method steps of the embodiment of the present application.
[0231] Optionally, in this embodiment, the storage medium may be located on at least one network device among a plurality of network devices in the network shown in the above embodiment.
[0232] Optionally, in this embodiment, the storage medium is configured to store program codes for executing the following steps:
[0233] Obtain climate observation data under extreme climate conditions and numerical model forecast results of new energy stations;
[0234] Extract meteorological elements and output change characteristics from climate observation data and new energy station numerical model forecast results under the extreme climate;
[0235] Based on the meteorological elements and the output change characteristics, a multi-model integrated forecast is performed on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station to obtain a multi-model integrated forecast result;
[0236] Taking the multi-model integrated forecast result as input, using a deterministic forecast model to perform deterministic forecasting to obtain a deterministic forecast result;
[0237] Inputting the climate observation data under the extreme climate and the deterministic forecast results into a probability prediction model to obtain a meteorological probability forecast result;
[0238] The new energy power is predicted based on the meteorological probability forecast results.
[0239] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, which will not be described in detail in this embodiment.
[0240] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media that can store program codes, such as a USB flash drive, a ROM, a RAM, a mobile hard disk, a magnetic disk, or an optical disk.
[0241] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.
[0242] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.
[0243] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.
[0244] In the several embodiments provided in the present application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.
[0245] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution provided in this embodiment.
[0246] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0247] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.
[0248] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.
Claims
1. A method for predicting new energy power under extreme climate, characterized in that: include: Obtain climate observation data under extreme climate conditions and numerical model forecast results of new energy stations; Extract meteorological elements and output change characteristics from climate observation data and new energy station numerical model forecast results under extreme climate, including: Conduct data quality control and homogeneity verification on the climate observation data and the forecast results of the numerical model of the new energy station under the extreme climate to obtain effective meteorological data and deviation factors; Extracting time characteristic features, space characteristic features and vertical characteristic features from the effective meteorological data as the meteorological elements; Based on the statistical analysis of the meteorological elements, output variation law data is obtained; Determining the output change characteristic based on the deviation factor and the output change law data; Based on the meteorological elements and the output change characteristics, a multi-mode integrated forecast is performed on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station to obtain a multi-mode integrated forecast result, including: Acquire the meteorological elements and the types of the output change characteristics; Optimizing the integration weight of each model member in the multi-model integrated forecast based on the meteorological elements and the types of the output change characteristics; Based on the integration weights, multi-model integrated forecasting is performed on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station to obtain a multi-model integrated forecast result; Based on the integration weight, a multi-model integrated forecast is performed on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station, and the multi-model integrated forecast results obtained include: Performing numerical model dynamic downscaling on the numerical model forecast results of the new energy station to obtain meteorological element forecast results; Standardizing the climate observation data under the extreme climate to obtain standardized observation data; Performing significant factor correlation analysis on the standardized observation data and the meteorological element forecast results to obtain significant factor analysis results; Performing principal component analysis on the significant factor analysis results to obtain principal components; Based on the principal component, a statistical downscaling method is used to correct the numerical model forecast result of the new energy station to obtain a corrected numerical forecast result; The corrected numerical forecast results are used for multi-model integrated forecasting based on the integrated weights to obtain multi-model integrated forecast results; Taking the multi-model integrated forecast result as input, using a deterministic forecast model to perform deterministic forecasting to obtain a deterministic forecast result; Inputting the climate observation data under the extreme climate and the deterministic forecast results into a probability prediction model to obtain a meteorological probability forecast result; The new energy power is predicted based on the meteorological probability forecast results.
2. The method for predicting new energy power under extreme climate conditions as claimed in claim 1, characterized in that: The method of using the multi-model integrated forecast result as input and using a deterministic forecast model to perform deterministic forecasting to obtain a deterministic forecast result includes: Use the self-attention module to extract the feature information of different mode prediction results; The extracted feature information is fused through the residual connection method; A prediction element sequence is generated using a prediction unit network as the deterministic forecast result.
3. The method for predicting new energy power under extreme climate conditions as claimed in claim 1, characterized in that: The step of inputting the climate observation data under extreme climate and the deterministic forecast result into a probability prediction model to obtain a meteorological probability forecast result includes: Constructing an original meteorological sequence based on the climate observation data under the extreme climate and the deterministic forecast results; Decomposing the original meteorological sequence to obtain multiple meteorological subsequences; A probability prediction model constructed using multiple different types of probability prediction methods is used to perform probability prediction on the meteorological subsequence to obtain a joint meteorological probability forecast result. The probability prediction methods include at least two of a time series residual probability hybrid prediction method, a multi-source Gaussian ensemble kernel fitting prediction method, a deep learning and quantile regression fusion prediction method, a correlation coefficient data fusion prediction method, and a GRU model prediction method based on particle swarm optimization.
4. The method for predicting new energy power under extreme climate conditions as claimed in claim 3, characterized in that: The probability prediction model constructed by using multiple different types of probability prediction methods performs probability prediction on the meteorological subsequence to obtain a joint meteorological probability forecast result including: The prediction results of each algorithm are evaluated based on the normalized average width, average coverage error, cumulative deviation, and the reliability and width of the prediction interval coverage to obtain a comprehensive evaluation value; Determining a joint weight of each probability prediction method based on the comprehensive evaluation value, wherein the comprehensive evaluation value is positively correlated with the joint weight; The probability prediction model is constructed based on the joint weight.
5. A new energy power prediction device under extreme climate, characterized in that: include: The acquisition module is used to obtain climate observation data under extreme climate conditions and the forecast results of numerical models of new energy stations; A feature extraction module, used to extract meteorological elements and output change characteristics from the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station; The feature extraction module is also used to perform data quality control and homogeneity verification on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station to obtain valid meteorological data and deviation factors; extract the time characteristic features, spatial characteristic features and vertical characteristic features in the valid meteorological data as the meteorological elements; obtain the output change law data based on the statistical analysis of the meteorological elements; determine the output change characteristics based on the deviation factors and the output change law data; An integrated forecasting module, used for performing multi-mode integrated forecasting on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station based on the meteorological elements and the output change characteristics, to obtain a multi-mode integrated forecasting result; The integrated forecast module is also used to obtain the types of the meteorological elements and the output change characteristics; Optimizing the integration weight of each model member in the multi-model integrated forecast based on the meteorological elements and the types of the output change characteristics; Based on the integration weights, multi-model integrated forecasting is performed on the climate observation data under the extreme climate and the forecast results of the numerical model of the new energy station to obtain a multi-model integrated forecast result; Wherein, based on the integration weight, multi-model integrated forecasting is performed on the climate observation data under the extreme climate and the numerical model forecasting results of the new energy station, and the multi-model integrated forecasting results obtained include: Performing numerical model dynamic downscaling on the numerical model forecast results of the new energy station to obtain meteorological element forecast results; Standardizing the climate observation data under the extreme climate to obtain standardized observation data; Performing significant factor correlation analysis on the standardized observation data and the meteorological element forecast results to obtain significant factor analysis results; Performing principal component analysis on the significant factor analysis results to obtain principal components; Based on the principal component, a statistical downscaling method is used to correct the numerical model forecast result of the new energy station to obtain a corrected numerical forecast result; The corrected numerical forecast results are used for multi-model integrated forecasting based on the integrated weights to obtain multi-model integrated forecast results; A deterministic forecasting module is used to take the multi-model integrated forecasting result as input, use a deterministic forecasting model to perform deterministic forecasting, and obtain a deterministic forecasting result; A probability forecast module, used for inputting the climate observation data under the extreme climate and the deterministic forecast results into a probability prediction model to obtain a meteorological probability forecast result; The power prediction module is used to predict the new energy power based on the meteorological probability forecast result.
6. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the new energy power prediction method under extreme climate as described in any one of claims 1 to 4.
7. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method for predicting new energy power under extreme climate conditions as claimed in any one of claims 1 to 4 when running.
