Search

CN-122026307-A - Photovoltaic potential evaluation and power generation output prediction method integrating GIS and meteorological data

CN122026307ACN 122026307 ACN122026307 ACN 122026307ACN-122026307-A

Abstract

The invention discloses a photovoltaic potential evaluation and power generation output prediction method integrating GIS and meteorological data, and belongs to the field of new energy and geographic information and meteorological data analysis intersection. The method comprises the steps of firstly collecting multi-source data and preprocessing the multi-source data to form a static feature library and a dynamic feature time sequence library, then completing photovoltaic static potential preliminary screening and grading through multi-factor calculation and combination weighting, then obtaining high-resolution photo-thermal data through inclined plane radiation conversion, machine learning downscaling and aerosol and water vapor correction, then constructing an encoder-decoder model containing an attention mechanism, combining physical constraint to realize short-term power generation prediction, and finally completing full life cycle output assessment and leveling degree electric cost calculation based on an environmental stress attenuation model and a financial model. The method solves the problems of poor data fusion, low prediction precision and the like in the prior art, and provides support for photovoltaic site selection, power grid dispatching and investment decision-making.

Inventors

  • LI LINHUAN
  • YU HAOZHENG
  • WANG GANG
  • WANG HONGZHI
  • LI HAORAN

Assignees

  • 国网河南省电力公司安阳县供电公司

Dates

Publication Date
20260512
Application Date
20251216

Claims (7)

  1. 1. The photovoltaic potential evaluation and power generation output prediction method integrating GIS and meteorological data is characterized by comprising the following steps of: Step 1, multi-source heterogeneous data acquisition and preprocessing, namely acquiring GIS data, multi-dimensional meteorological data, remote sensing images and auxiliary data to obtain acquisition data, and sequentially carrying out spatial reference system unification and resampling, space-time scale matching and data fusion, data cleaning and quality control and management, feature engineering and normalization on the acquisition data to form a standardized static feature library and a standardized dynamic feature time sequence library; Step 2, static photovoltaic potential preliminary screening based on GIS, namely calculating solar radiation receiving potential factors, gradient suitability factors, slope suitability factors, land utilization constraint factors and shielding exclusion factors by utilizing the static feature library data obtained in the step 1, calculating comprehensive suitability indexes and dividing potential grades by adopting a weighted linear combination method of combining weighting by a hierarchical analysis method and an entropy weighting method, and outputting a photovoltaic installation suitability grade distribution map; Step 3, weather data downscaling and photo-thermal resource dynamic correction, namely based on the dynamic characteristic time sequence library in the step 1 and the comprehensive suitability index in the step 2, firstly converting horizontal plane solar radiation into inclined plane solar radiation through a surface photometry model, then utilizing a machine learning model to combine high-resolution geographic features to realize weather data downscaling, and finally introducing aerosol and water vapor transmissivity coefficients to correct the solar radiation to obtain a dynamic corrected high-resolution photo-thermal resource data set; Step 4, generating power prediction integrating the space-time features, namely splicing the static geographic features of the step 2 and the dynamic photo-thermal resource data set of the step 3 into input features integrating the space-time information, constructing an encoder-decoder deep learning model containing an attention mechanism, introducing physical information constraint design loss functions, realizing short-term generating power prediction after training, and outputting a generating power prediction curve and a confidence interval; And 5, evaluating the long-term output by considering the performance attenuation, namely driving the encoder-decoder deep learning model in the step 4 by using the dynamically corrected high-resolution photo-thermal resource data set in the step 3 to determine the first annual energy generation standard, constructing a performance attenuation model based on environmental stress, calculating the annual dynamic attenuation rate by year by combining long-term meteorological data, simulating the full life cycle energy generation, calculating the leveling degree electricity cost by coupling a financial model, and outputting a long-term output evaluation result and an investment decision index.
  2. 2. The method for photovoltaic potential evaluation and power generation output prediction by fusion of GIS and meteorological data according to claim 1, wherein the spatial reference system for collecting data in the step 1 is unified and resampled, space-time scale matching and data fusion, data cleaning and quality control and management, feature engineering and normalization processing, and the method comprises the following specific contents: The spatial reference system unifies and resamples, namely, the collected data is automatically converted into a unified geographic coordinate system and resampled to a preset optimal spatial resolution, a bilinear interpolation method is adopted for continuous grid data to maintain data smoothness, and a nearest neighbor method is adopted for category data to maintain original classification attribute Space-time scale matching and data fusion, namely, a machine learning downscaling method based on terrain assistance is adopted, low-resolution meteorological data is used as a dependent variable, a high-resolution terrain derivative factor is used as an independent variable, a random forest or a geographic weighted regression model is constructed, and a geographic weighted regression model formula is as follows: Wherein, the For the high resolution grid point coordinates, Is a regression coefficient that varies with spatial position, Is residual, by training the model on a low resolution scale and applying it to the high resolution terrain factors of the full area, for predicting high resolution meteorological fields with terrain details The space expressive force of meteorological data is improved from kilometer level to hundred meters level; The data cleaning and quality control method comprises automatically processing the missing and abnormal values in the collected data to obtain cleaning data, specifically based on physical threshold value, surface solar radiation value Satisfy the following requirements Wherein Is irradiated by the outer layer of the atmosphere, Screening for the rule of clear air atmosphere transmissivity, and introducing a data filling model based on a generated countermeasure network for a complex missing mode, wherein the data filling model learns the distribution characteristics of complete data through the countermeasure training of a generator and a discriminator, and generates reasonable filling values which accord with the physical rule and the actual statistical characteristics for the missing area in any form; The method comprises the steps of feature engineering and normalization processing, namely feature derivation and scale unification of cleaning data, specifically, gradient, slope direction and terrain shielding degree are calculated from an original DEM, a theoretical efficiency correction coefficient of photoelectric conversion is calculated preliminarily by combining solar radiation and temperature data, Z-score standardization processing is carried out on all input features to ensure that features with different dimensions play an equal role in a subsequent machine learning model, the processed data are organized into a structured static feature library and a dynamic feature time sequence library, a set of standardized data products with consistent space-time reference and capable of being directly used for model training and reasoning are formed, and data support is provided for subsequent photovoltaic potential evaluation and power generation capacity prediction.
  3. 3. The method for photovoltaic potential assessment and power generation output prediction by fusion of GIS and meteorological data according to claim 1, wherein the solar radiation receiving potential factor, gradient suitability factor, slope suitability factor, land utilization constraint factor and shielding exclusion factor are calculated in the step 2, and the specific contents are as follows: Solar radiation receiving potential factor Calculating theoretical annual total radiation quantity of terrain shielding under ideal atmospheric conditions based on DEM data through a sunlight duration model and terrain shielding analysis, normalizing the theoretical annual total radiation quantity to be a potential index of 0-1, and inputting the potential index as an evaluation basis energy source; Slope suitability factor The gradient determines the installation difficulty and cost, a segment membership function is defined to calculate a gradient suitability score, and a calculation formula can be expressed as follows: Wherein, the For the actual grade of the road, For an optimal grade of the slope, Is the maximum allowable gradient of the vehicle, Parameters for controlling the decay rate; Slope suitability factor Using a fixed photovoltaic array, taking the south direction or the north direction as the optimal direction, and quantifying the slope suitability by adopting a cosine function model: Wherein, the In order to be an actual slope direction, The function makes the score of the positive south direction be 1 and the score of the positive north direction be 0; land utilization constraint factor Is binary constraint factor, and according to land utilization type graph, the areas of water area, forest, wetland, basic farmland and ecological protection area are directly marked as infeasible area, i.e Marking wastelands, worksites, general cultivated lands, roofs, etc. as viable areas, i.e ; Mask exclusion factor Based on high resolution DEM and building contour data, simulating and calculating total annual masking time length generated by each grid unit due to terrain and buildings in main power generation period under solar track data of typical meteorological year, and marking the unit as a serious masking area when the total annual masking time length exceeds a preset threshold value Otherwise, it is an acceptable area 。
  4. 4. The method for photovoltaic potential evaluation and power generation output prediction by fusion of GIS and meteorological data according to claim 1, wherein in the step 2, a weighted linear combination method of combining weighting by a hierarchical analysis method and an entropy weighting method is adopted, comprehensive suitability indexes are calculated, potential grades are divided, and a photovoltaic installation suitability grade distribution map is output, wherein the method comprises the following specific contents: Multi-factor comprehensive integration by adopting weighted linear combination method and comprehensive suitability index The calculation formula of (2) is as follows: Wherein, the The weights of the corresponding factors are respectively, the sum of the weights is 1, the three weights are determined by a subjective and objective combination weighting method combining an analytic hierarchy process and an entropy weighting method, and the expert domain knowledge is considered and the information quantity of the data is respected; finally, the calculated comprehensive suitability index Grading, particularly into high potential areas Zone of intermediate potential Low potential region The output result is a photovoltaic installation suitability grade distribution map with definite spatial distribution.
  5. 5. The method for photovoltaic potential evaluation and power generation output prediction by fusing GIS and meteorological data according to claim 1, wherein the implementation steps of the step3 meteorological data downscaling and photo-thermal resource dynamic correction are as follows: step 3.1, basic data integration, namely extracting core input data, namely low-resolution weather grid data, from the dynamic characteristic time sequence library in step 1, wherein the low-resolution weather grid data comprises time-by-time horizontal plane total solar radiation Near surface air temperature Relative humidity of Wind speed Simultaneously extracting high-resolution geographic characteristic data comprising a digital elevation model DEM and gradient in a suitability area Slope direction Distance to large water body Obtaining a photovoltaic installation suitability grade distribution map from the suitability region data of the step 2 to obtain a high potential region And a medium potential zone Generating a space mask file for the core evaluation range; step 3.2, inclined plane radiation conversion based on a surface photometry model: a. Splitting the radiation component by combining the total solar radiation of the horizontal plane input in the step 1 Split into direct radiation Scattered radiation And ground reflected radiation The resolution formula uses a classical Reindl model: Direct radiation duty cycle: wherein In order to be the solar altitude angle, 、 、 And fitting the local weather station measured data to obtain the regional calibration coefficients, wherein the scattered radiation is as follows: Ground reflected radiation: wherein Taking the land albedo as the land albedo, taking the value according to the land utilization data in the step 1, and barren lands Roof Water body ; B. Calculating each component of inclined plane by direct radiation conversion, namely adopting a core formula of surface photometry to directly radiate horizontal plane Conversion to direct radiation on inclined surfaces : Wherein Is the zenith angle of the sun; for the included angle between the incident light and the inclined surface normal of the photovoltaic panel, the solar position and the topographic parameter to be fused are calculated: Wherein The declination angle of the sun is calculated by the date; taking the latitude of the station from GIS data; Is the inclination angle of the photovoltaic panel; taking the azimuth angle of the photovoltaic panel, and taking the north hemisphere to be 180 degrees; the solar time angle is changed by 15 degrees per hour, and the noon is 0 degree; Scattered radiation conversion by calculating inclined plane scattered radiation using Hay model Consider the sky anisotropic feature: wherein Is the sky scattering anisotropy coefficient; conversion of reflected radiation by calculating the ground reflected radiation received by inclined surfaces : ; C. Combining the total radiation of the inclined plane by superposing the direct radiation, the scattered radiation and the reflected radiation of the inclined plane to obtain the total solar radiation of the inclined plane : Superposing the calculation result and the suitability region space mask in the step 2, and cutting to obtain a 'suitability region inclined plane total radiation grid'; Step 3.3, the meteorological data downscaling based on machine learning: in the suitability area of the step 2, selecting training sample points by adopting a layered random sampling method, and ensuring that the samples cover different terrain types and different water body distance intervals; Sample feature assignment, namely extracting low-resolution meteorological features and high-resolution geographic features from the data in the step 1 to form a sample feature matrix for each sample point Simultaneously taking a high-resolution meteorological true value or an interpolation reference value corresponding to the sample point as a label ; Applying a random forest model to all high-resolution grid points in the suitability area of the step 2, and inputting the high-resolution geographic characteristics of each grid point And corresponding low resolution weather data 、 Predicting lattice point by lattice point to obtain high-resolution air temperature field High resolution inclined plane radiation field ; Carrying out post-processing on the result, namely carrying out space smoothing on the high-resolution meteorological field obtained by prediction, eliminating local noise of model prediction, and simultaneously carrying out physical rationality check on the meteorological value of the elevation abnormal region by combining with the DEM data in the step 1, correcting the abnormal value and ensuring that the data accords with local climate rules; step 3.4, dynamically correcting the radiation based on aerosol and water vapor: Irradiating the high-resolution inclined plane obtained in the step 3.3 Multiplying the aerosol transmittance and the water vapor transmittance to obtain the surface effective total solar radiation : Wherein In order to achieve an aerosol transmittance, Is water vapor transmittance, is corrected With high resolution air temperature Relative humidity of Wind speed And (3) integrating to form a dynamically modified high-resolution photo-thermal resource data set, wherein the data format is a TIFF grid, and the time resolution is time-by-time, and the comprehensive suitability index of the step (2) is covered.
  6. 6. The method for photovoltaic potential evaluation and power generation output prediction by fusion of GIS and meteorological data according to claim 1, wherein the step 4 of power generation power prediction by fusion of space-time features is specifically implemented as follows: step 4.1, space-time feature engineering, namely obtaining two input sources, namely, static geographic feature vectors from step 2 I.e. comprehensive suitability index of location Coding of grade, slope, shading coefficient and weather zone type, and dynamic weather feature time sequence from step 3 Efficient solar radiation including inclined surface on an hour-by-hour basis for future prediction periods Ambient temperature Wind speed Relative humidity of Before inputting the model, the static feature vector Copying and associating dynamic characteristics of each time step Splicing to form a complete input characteristic vector fused with space-time information ; Step 4.2, structural design of hybrid deep learning prediction model, namely adopting encoder-decoder deep learning model based on encoder-decoder architecture and integrating attention mechanism, wherein the encoder is composed of any one of multi-layer long-short-term memory network or gating circulation unit and is responsible for inputting characteristic sequences Encoded as a context vector containing past information, and a decoder utilizing the context vector in combination with a future known predictive feature sequence Gradually decoding future generation power sequence Wherein the introduction of the attention mechanism allows the decoder to dynamically and with emphasis review different portions of the encoder output sequence as it predicts the power at each future time instant, thereby capturing key meteorological events in long sequences more effectively; Step 4.3, physical information constraint and loss function design, namely introducing physical constraint into the loss function of the encoder-decoder depth learning model, wherein the power is predicted in the existing loss function And the actual power Increasing the term of physical consistency loss in the mean square error of (2) Then the total loss function The method comprises the following steps: ; Wherein Is a super parameter for balancing the weights of the data driving item and the physical constraint item; step 4.4, model training and rolling prediction, wherein the model is supervised and trained by using historical meteorological data and actual power generation power data of a corresponding power station, and the model can be used for business prediction after training is completed; Finally, the encoder-decoder deep learning model outputs a high-precision power generation prediction curve corresponding to a future time axis, and can give a confidence interval of a predicted value according to uncertainty quantization analysis.
  7. 7. The method for photovoltaic potential evaluation and power generation output prediction by fusing GIS and meteorological data according to claim 1, wherein the specific formula of the characteristic attenuation model in the step 5 is as follows: , wherein, In order to achieve a performance decay rate, Is a manufacturer provided baseline decay rate under standard test conditions; is an additional attenuation caused by temperature stress, which is related to the operating temperature of the panel Cumulative exposure exceeding the reference temperature is positively correlated and can be estimated by the Arrhenius model framework: , Is the activation energy of the catalyst and is used for preparing the catalyst, Is the boltzmann constant; is an additional attenuation caused by damp-heat stress and is connected with the ambient temperature In connection with the co-action of relative humidity RH (t), it is generally in the form of a P-k model: ; is an additional attenuation caused by the stress of ultraviolet radiation and is combined with the accumulated received ultraviolet band radiation energy Is in direct proportion; Driving a performance attenuation model by using the long-term high-resolution weather time sequence data provided in the step 3, and calculating the attenuation rate of dynamic change year by year Wherein Representing the operational chronology. Based on this, full life cycle power generation simulation was performed, in the first The actual output power of the system will be corrected for component attenuation: the accumulated total power generation amount can be obtained by accumulating the whole evaluation period year by year: , The life (years) of the power station is designed, Time step (hours); finally, the long-term power generation amount evaluation result is coupled with a financial model, and a key investment decision index, and most importantly, the leveling degree electric cost is output, wherein the LCOE has a calculation formula as follows: Wherein, the In order to achieve the discount rate, Is the first Annual forecast power production.

Description

Photovoltaic potential evaluation and power generation output prediction method integrating GIS and meteorological data Technical Field The invention relates to the field of new energy, in particular to a photovoltaic potential evaluation and power generation capacity prediction method integrating GIS and meteorological data. Background In the global energy transformation background, photovoltaic power generation is an important component of clean energy, and efficient development and utilization of the photovoltaic power generation depend on accurate potential evaluation and output prediction. However, the prior art has the following disadvantages: The evaluation method is single, most researches only rely on GIS data to carry out static solar resource distribution evaluation, and dynamic changes of meteorological factors such as cloud cover, aerosol and the like are ignored, so that the evaluation result has large deviation from the actual result and is too ideal. The data fusion degree is low, GIS data and meteorological data are often analyzed independently, accurate space-time scale matching cannot be achieved, and for example, the low spatial resolution (kilometer level) of the meteorological data is difficult to meet the photovoltaic evaluation requirements of fine areas such as roofs, mountainous areas and the like. The prediction model has insufficient precision, and the traditional physical model or the simple statistical model has weak response capability to complex weather conditions such as instantaneous cloud shielding, rain and snow and the like, has high prediction errors of short-term and ultra-short-term power generation, and cannot support accurate scheduling of a power grid. The full life cycle view angle is lacking, long-term influencing factors such as terrain shielding, component performance attenuation and the like are not considered, only the first annual energy generation is concerned, long-term output evaluation distortion is caused, and project investment decision scientificity is influenced. Aiming at the problems, the photovoltaic evaluation and prediction method with high precision and full flow is constructed through multi-source data depth fusion, space-time feature collaborative modeling and full life cycle attenuation simulation, so that the blank of the prior art is filled. Disclosure of Invention The invention aims to provide a photovoltaic potential evaluation and power generation output prediction method integrating GIS and meteorological data so as to solve the problems in the background technology. In order to achieve the purpose, the invention provides a photovoltaic potential evaluation and power generation output prediction method integrating GIS and meteorological data, which comprises the following steps: Step 1, multi-source heterogeneous data acquisition and preprocessing, namely acquiring GIS data, multi-dimensional meteorological data, remote sensing images and auxiliary data to obtain acquisition data, and sequentially carrying out spatial reference system unification and resampling, space-time scale matching and data fusion, data cleaning and quality control and management, feature engineering and normalization on the acquisition data to form a standardized static feature library and a standardized dynamic feature time sequence library; Step 2, static photovoltaic potential preliminary screening based on GIS, namely calculating solar radiation receiving potential factors, gradient suitability factors, slope suitability factors, land utilization constraint factors and shielding exclusion factors by utilizing the static feature library data obtained in the step 1, calculating comprehensive suitability indexes and dividing potential grades by adopting a weighted linear combination method of combining weighting by a hierarchical analysis method and an entropy weighting method, and outputting a photovoltaic installation suitability grade distribution map; Step 3, weather data downscaling and photo-thermal resource dynamic correction, namely based on the dynamic characteristic time sequence library in the step 1 and the comprehensive suitability index in the step 2, firstly converting horizontal plane solar radiation into inclined plane solar radiation through a surface photometry model, then utilizing a machine learning model to combine high-resolution geographic features to realize weather data downscaling, and finally introducing aerosol and water vapor transmissivity coefficients to correct the solar radiation to obtain a dynamic corrected high-resolution photo-thermal resource data set; Step 4, generating power prediction integrating the space-time features, namely splicing the static geographic features of the step 2 and the dynamic photo-thermal resource data set of the step 3 into input features integrating the space-time information, constructing an encoder-decoder deep learning model containing an attention mechanism, introducing physical information c