Artificial intelligence-based new energy power generation prediction model optimization method and system

Abstract

The present invention discloses an artificial intelligence-based new energy power generation prediction model optimization method and system, belonging to the field of artificial intelligence model optimization technology. The method comprises: collecting meteorological data, equipment operating status data, and geographic information data through a multi-source heterogeneous data acquisition module; generating a dynamic coupling feature vector that integrates meteorological dynamics, equipment response characteristics, and terrain stability through a spatiotemporal dynamic feature coupling engine; using an intelligent optimization analysis module to perform two-stage parameter optimization and gradient acceleration training on a preset model based on the feature vector to generate an optimized prediction model; and establishing an adaptive correction closed loop to achieve dynamic fine-tuning of model parameters using power generation prediction data. The present invention adopts a deep spatiotemporal coupling mechanism for multi-source heterogeneous data, combined with a dual optimization strategy of parameter sensitivity analysis and momentum acceleration optimization, to improve the prediction accuracy of new energy power generation and enhance the generalization ability of the model.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence model optimization technology, and in particular to an artificial intelligence-based new energy power generation prediction model optimization method and system. Background Art

[0002] Renewable energy generation forecasting technology builds artificial intelligence models by analyzing meteorological, equipment, and geographic data, providing a core basis for grid dispatch decisions. This field relies on machine learning models to model complex factors such as wind and solar resource fluctuations and equipment response characteristics. The accuracy of these forecasts directly impacts energy consumption efficiency.

[0003] Current mainstream approaches employ a two-stage optimization process, first extracting spatiotemporal features from data using convolutional or recurrent neural networks, followed by training the prediction model parameters using gradient descent algorithms. Some improved approaches incorporate swarm intelligence optimization algorithms to initialize parameters or employ transfer learning to adjust pre-trained models.

[0004] Existing technologies inadequately model the dynamic coupling between multi-source data, making it difficult to capture the interplay between sudden environmental changes and equipment responses. Parameter optimization processes are susceptible to the multimodal nature of complex loss functions, leading to local optima. Deployed models lack online adaptation mechanisms for long-term operating condition drift, requiring frequent retraining to maintain accuracy. Summary of the Invention

[0005] To solve the above problems, the present invention provides an artificial intelligence-based new energy power generation prediction model optimization method and system, which adopts a deep spatiotemporal coupling mechanism of multi-source heterogeneous data and combines the dual optimization strategy of parameter sensitivity analysis and momentum acceleration optimization to improve the prediction accuracy of new energy power generation and enhance the generalization ability of the model.

[0006] The above objectives can be achieved through the following solutions:

[0007] An artificial intelligence-based optimization method and system for a new energy power generation prediction model includes: collecting meteorological data, equipment operating status data, and geographic information data through a multi-source heterogeneous data acquisition module; generating a dynamic coupling feature vector that integrates meteorological dynamics, equipment response characteristics, and terrain stability through a spatiotemporal dynamic feature coupling engine; using an intelligent optimization analysis module to perform two-stage parameter optimization and gradient acceleration training on a preset model based on the feature vector to generate an optimized prediction model; and establishing an adaptive correction closed loop to achieve dynamic fine-tuning of model parameters through power generation prediction data.

[0008] Optionally, the spatiotemporal dynamic feature coupling processing of the multi-source heterogeneous data to generate a dynamic coupling feature vector includes: extracting the spatiotemporal dynamic features of the meteorological data to generate meteorological spatiotemporal features; extracting the spatiotemporal dynamic features of the equipment operation status data to generate equipment status spatiotemporal features; extracting the spatiotemporal dynamic features of the geographic information data to generate geographic spatiotemporal features; calculating the meteorological spatiotemporal features, the equipment status spatiotemporal features and the geographic spatiotemporal features to obtain coupling weights; and fusing the meteorological spatiotemporal features, the equipment status spatiotemporal features and the geographic spatiotemporal features based on the coupling weights to generate a dynamic coupling feature vector.

[0009] Optionally, the calculation of the meteorological spatiotemporal characteristics, the equipment status spatiotemporal characteristics and the geographic spatiotemporal characteristics to obtain the coupling weight includes: identifying the interaction pattern between the meteorological spatiotemporal characteristics to generate an environmental coupling coefficient; monitoring the response relationship of the equipment status spatiotemporal characteristics to environmental factors to generate a response correlation coefficient; performing stability assessment on the topological structure of the geographic spatiotemporal characteristics to generate a terrain stability coefficient; and fusing the environmental coupling coefficient, the response correlation coefficient and the terrain stability coefficient to obtain the coupling weight.

[0010] Optionally, the parameter optimization of the preset new energy power generation prediction model based on the dynamic coupling characteristic vector to generate an optimized new energy power generation prediction model includes: obtaining a model optimization objective function based on the dynamic coupling characteristic vector; performing parameter optimization on the new energy power generation prediction model to generate an initial parameter set; performing gradient acceleration optimization on the initial parameter set to generate a fine-tuning parameter set; based on the model optimization objective function, verifying the generalization ability of the fine-tuning parameter set to generate an optimized new energy power generation prediction model.

[0011] Optionally, the step of optimizing the objective function based on the model, verifying the generalization capability of the fine-tuning parameter set, and generating an optimized new energy power generation prediction model includes: optimizing the objective function based on the model, obtaining a generalization capability evaluation matrix; and using the generalization capability evaluation matrix to verify the generalization capability of the fine-tuning parameter set, and generating an optimized new energy power generation prediction model.

[0012] Optionally, based on the predicted power generation data, the parameters of the optimized new energy power generation prediction model are adjusted to generate a fine-tuned new energy power generation prediction model, including: monitoring the continuous deviation between the predicted power generation data and the preset actual power generation data to generate a prediction deviation sequence; using the prediction deviation sequence to generate parameter adjustment suggestions; according to the parameter adjustment suggestions, the parameters of the optimized new energy power generation prediction model are adjusted to generate a fine-tuned new energy power generation prediction model.

[0013] Optionally, adjusting the parameters of the optimized new energy power generation prediction model according to the parameter adjustment suggestions to generate a fine-tuned new energy power generation prediction model includes: obtaining a parameter correction vector matrix according to the parameter adjustment suggestions; obtaining a dynamic parameter response surface according to the key parameters of the optimized new energy power generation prediction model; and using the parameter correction vector matrix to correct the dynamic parameter response surface to generate a fine-tuned new energy power generation prediction model.

[0014] Optionally, the performing power generation prediction on the optimized new energy power generation prediction model to obtain predicted power generation data includes: performing power generation prediction on the optimized new energy power generation prediction model to obtain an original predicted output tensor; and performing dynamic coupling feature post-correction on the original predicted output tensor to obtain predicted power generation data.

[0015] Optionally, the stability assessment of the topological structure of the geographic spatiotemporal features to generate a terrain stability coefficient includes: obtaining a terrain topological correlation matrix based on the topological structure of the geographic spatiotemporal features; performing a spatiotemporal evolution pattern analysis on the terrain topological correlation matrix to obtain a terrain dynamic stability vector; and performing a stability assessment on the terrain dynamic stability vector to generate a terrain stability coefficient.

[0016] Based on the same inventive concept, the present invention also provides an artificial intelligence-based new energy power generation prediction model optimization system, which includes: a multi-source heterogeneous data acquisition module, used to acquire meteorological data, equipment operation status data and geographic information data to obtain multi-source heterogeneous data; a spatiotemporal dynamic coupling engine, used to perform spatiotemporal dynamic feature coupling processing on the multi-source heterogeneous data to generate a dynamic coupling feature vector; a model parameter optimization module, used to optimize the parameters of a preset new energy power generation prediction model based on the dynamic coupling feature vector to generate an optimized new energy power generation prediction model; a power generation prediction execution module, used to perform power generation prediction on the optimized new energy power generation prediction model to obtain predicted power generation data; an adaptive correction module, used to adjust the parameters of the optimized new energy power generation prediction model based on the predicted power generation data to generate a fine-tuned new energy power generation prediction model.

[0017] Compared with the prior art, the present invention has the following advantages:

[0018] The dynamically coupled feature extraction module and the dual-engine optimization module form a closed-loop feedback loop. The environment-device coupling law revealed by the feature vector directly guides the parameter search direction, enabling the optimization process to avoid invalid area exploration, significantly shortening the training cycle and improving the upper limit of model performance, breaking through the efficiency bottleneck of traditional single-stage optimization.

[0019] Through a lightweight online fine-tuning loop, the prediction deviation pattern is analyzed in real time, and complex working condition changes are converted into directional parameter calibration, so that the model can maintain prediction stability in scenarios such as sudden weather fluctuations and equipment performance drift, overcoming the industry problem of performance degradation after static model deployment.

[0020] The wide-area search capability of the macro-exploration engine interacts deeply with the gradient acceleration characteristics of the micro-fine-tuning engine. By dynamically adjusting the search strategy through gradient direction similarity, efficient exploration and fine development of the parameter space are achieved, effectively avoiding local optimal solutions and approaching global optimal performance.

[0021] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 It is a flow chart of an artificial intelligence-based new energy power generation prediction model optimization method according to an embodiment of the present invention.

[0024] Figure 2 2 is a diagram of a parameter optimization process according to an embodiment of the present invention.

[0025] Figure 3 It is a continuous prediction deviation monitoring sequence diagram of an embodiment of the present invention.

[0026] Figure 4 It is a key parameter correction matrix diagram of an embodiment of the present invention.

[0027] Figure 5 It is a model response surface optimization diagram of an embodiment of the present invention.

[0028] Figure 6 This is a photovoltaic power prediction deviation monitoring sequence diagram according to an embodiment of the present invention.

[0029] Figure 7 4 is a parameter correction trigger signal diagram of an embodiment of the present invention.

[0030] Figure 8 4 is a topographic correlation matrix diagram of an embodiment of the present invention.

[0031] Figure 9 is a terrain dynamic stability vector diagram according to an embodiment of the present invention.

[0032] Figure 10 It is a structural diagram of an artificial intelligence-based new energy power generation prediction model optimization system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0034] Reference Figure 1 One embodiment of the present invention proposes an artificial intelligence-based optimization method for a new energy power generation prediction model. It adopts a deep spatiotemporal coupling mechanism of multi-source heterogeneous data and combines a dual optimization strategy of parameter sensitivity analysis and momentum acceleration optimization to improve the prediction accuracy of new energy power generation and enhance the generalization ability of the model.

[0035] The method of this embodiment specifically includes:

[0036] Acquire meteorological data, equipment operating status data, and geographic information data to obtain multi-source heterogeneous data;

[0037] Performing spatiotemporal dynamic feature coupling processing on the multi-source heterogeneous data to generate a dynamic coupling feature vector;

[0038] Based on the dynamic coupling characteristic vector, the parameters of the preset new energy power generation prediction model are optimized to generate an optimized new energy power generation prediction model;

[0039] Performing power generation prediction on the optimized new energy power generation prediction model to obtain predicted power generation data;

[0040] Based on the predicted power generation data, parameters of the optimized new energy power generation prediction model are adjusted to generate a fine-tuned new energy power generation prediction model.

[0041] Specifically, a meteorological monitoring network is first established to obtain time-series matrix data for wind speed, irradiance, and temperature parameters with a five-minute temporal resolution and a spatial resolution of at least one kilometer. Equipment operating status data is also collected simultaneously, including second-level sampling sequences from generator speed sensors, minute-level recording curves from bearing temperature monitors, and fifteen-minute statistics for inverter efficiency parameters. For geographic information data, a spatial distribution matrix of slope, aspect, and surface roughness is extracted using digital elevation models and remote sensing imagery, with a grid accuracy requirement of 30 meters. These three types of data are input into the data preprocessing module for spatiotemporal alignment. The equipment data is resampled to a five-minute time step using cubic spline interpolation, and bilinear interpolation is applied to align the geographic information data with the meteorological grid resolution. During the fusion storage phase, a multi-source heterogeneous database is created, hierarchically storing meteorological, equipment, and geographic layer data using spatiotemporal location as the primary key. The meteorological layer stores a three-dimensional data cube (time × space × meteorological type), the equipment layer constructs a relational data table (device ID × timestamp × status parameters), and the geographic layer stores a two-dimensional georeferencing matrix (longitude × latitude × terrain attributes). The final generated multi-source heterogeneous data structure definition includes timestamps, geographic coordinates, and nested field combinations of multiple data types, completing data standardization encapsulation.

[0042] Optionally, performing spatiotemporal dynamic feature coupling processing on the multi-source heterogeneous data to generate a dynamic coupling feature vector includes:

[0043] Extracting the spatiotemporal dynamic characteristics of the meteorological data to generate meteorological spatiotemporal characteristics;

[0044] Extracting the spatiotemporal dynamic features of the equipment operation status data to generate equipment status spatiotemporal features;

[0045] Extracting spatiotemporal dynamic features of the geographic information data to generate geographic spatiotemporal features;

[0046] Calculating the meteorological spatiotemporal characteristics, the equipment state spatiotemporal characteristics, and the geographic spatiotemporal characteristics to obtain a coupling weight;

[0047] The meteorological spatiotemporal features, the device state spatiotemporal features, and the geographic spatiotemporal features are fused based on the coupling weights to generate a dynamic coupling feature vector.

[0048] Specifically, we first obtain the wind speed, irradiance, and temperature time series data from the meteorological data, the generator speed and bearing temperature monitoring values ​​from the equipment operation status data, and the elevation and slope matrix from the geographic information data. We extract the spatiotemporal dynamic features of the meteorological data through a convolutional neural network, extract the spatiotemporal dynamic features of the equipment operation status data using a long short-term memory network, and apply a graph neural network to model the spatial correlation of geographic information. The feature coupling process is implemented as follows: Construct a coupling weight calculation module :

[0049] ,

[0050] in Characterize the environmental coupling coefficient between meteorological features, which is calculated through the feature vector self-attention mechanism; Characterize the device state response correlation coefficient and obtain the dependence of device features on meteorological features through the cross-attention mechanism; is the terrain stability coefficient, which is determined by the node degree distribution entropy of the spatial graph convolutional network. is the weight corresponding to the environmental coupling coefficient, is the weight corresponding to the device state response correlation coefficient, is the weight corresponding to the terrain stability coefficient. function After normalization to coupling weights, the dynamic coupling feature vector is generated through the feature fusion operator:

[0051] ,

[0052] is the dynamic coupling eigenvector, are the weight distribution matrices for meteorological, equipment, and geographical features, respectively. represents the Hadamard product, Represents the spatiotemporal characteristics of meteorology. Represents the spatiotemporal characteristics of device status, Representing geographic spatiotemporal features. This vector is input into a pre-built deep neural network prediction model. Parameters are optimized using an adaptive moment estimation optimizer, and the optimal model parameters are determined using an early stopping strategy on the validation set. During the model deployment phase, a sliding time window deviation monitoring mechanism is configured. When the relative error for three consecutive prediction cycles exceeds a preset threshold, a Bayesian optimization algorithm is triggered to dynamically calibrate the model output layer parameters.

[0053] For example, when this method was applied to power forecasting at the Zhangjiakou wind farm, the system detected a step change of 3.2m / s in the wind speed characteristic vector during the passage of a strong cold front. The equipment status characteristics reflected a sudden drop of 8°C in the gearbox temperature, and the geographical characteristics showed that the turbulence intensity on the leeward slope increased by 40%. The feature coupling module automatically increased the weight of the wind speed feature to 0.62, generating a dynamic coupling feature that highlights the effect of sudden wind speed changes. The model predicted that the power generation would drop by 15MW after 2 hours, and the actual operation data verified that the error of the prediction result was only 1.8MW. When the photovoltaic power station encountered sandstorms, the system increased the weight of the light intensity attenuation feature caused by dust accumulation by 35% through dynamic calibration, which increased the prediction accuracy by 22 percentage points compared with the traditional method.

[0054] Optionally, the calculating the meteorological spatiotemporal characteristics, the device state spatiotemporal characteristics, and the geographic spatiotemporal characteristics to obtain coupling weights includes:

[0055] Identify the interaction pattern between the meteorological spatiotemporal characteristics and generate an environmental coupling coefficient; monitor the response relationship between the equipment status spatiotemporal characteristics and environmental factors and generate a response correlation coefficient;

[0056] Performing stability assessment on the topological structure of the geographic spatiotemporal features to generate a terrain stability coefficient;

[0057] The environmental coupling coefficient, the response correlation coefficient and the terrain stability coefficient are integrated to obtain a coupling weight.

[0058] Specifically, first identify the interaction pattern between meteorological spatiotemporal features, which is achieved by constructing a correlation matrix of elements in the meteorological feature vector. The values ​​of the matrix elements are determined by the Pearson correlation coefficient of the feature components within the time window. When the wind speed component is detected to be significantly negatively correlated with the light component, an environmental coupling coefficient is generated to mark a specific meteorological mutation pattern. Secondly, monitor the response relationship between the spatiotemporal characteristics of the equipment status and environmental factors, calculate the mutual information between the equipment feature vector and the lag time step meteorological feature vector, and generate a response correlation coefficient that characterizes the response delay when the mutual information of the wind turbine speed vector shows an abnormal decrease compared to the historical wind speed vector. Finally, evaluate the stability of the topological structure of geographic spatiotemporal features, construct a spatial autoregressive model based on the geographic feature vector, and perform a variance homogeneity test on the model residual sequence. When the residual variance of the mountain elevation vector exceeds the threshold, a terrain stability coefficient is generated. The above three coefficients are linearly combined by the formula:

[0059] ,

[0060] In the formula is the coupling weight, is the environmental coupling coefficient, is the response correlation coefficient, is the terrain stability coefficient, is the normalized weight factor.

[0061] For example, the coupling weight calculation is performed for coastal wind farms to obtain the wind speed gradient vector and air pressure change vector every five minutes in the current period. It is found that the correlation coefficient between the two during the period of sea and land wind convergence is negative 0.83, generating a high-value environmental coupling coefficient. The mutual information of the gearbox temperature vector and the temperature fluctuation vector is analyzed simultaneously, and a 42% attenuation of the mutual information is detected during the cooling stage, generating an abnormal response correlation coefficient. When constructing a spatial autoregressive model for the altitude difference of the site, it was found that the residual variance in the offshore area exceeded the critical value by 1.6 times, generating a low terrain stability coefficient. The three coefficients are weighted and fused by the normalized weight factor, and finally the fusion weight of the enhanced wind speed mutation effect and the equipment response anomaly is obtained.

[0062] Optionally, the performing parameter optimization on a preset new energy power generation prediction model based on the dynamic coupling eigenvector to generate an optimized new energy power generation prediction model includes:

[0063] Based on the dynamic coupling eigenvector, a model optimization objective function is obtained;

[0064] Optimizing parameters of the new energy power generation prediction model to generate an initial parameter set;

[0065] Performing gradient acceleration optimization on the initial parameter set to generate a fine-tuning parameter set;

[0066] Based on the model optimization objective function, the generalization capability of the fine-tuning parameter set is verified to generate an optimized new energy power generation prediction model.

[0067] Specifically, the parameter optimization process is as follows Figure 2 As shown, firstly, the model optimization objective function is constructed according to the dynamic coupling eigenvector, and the function form is:

[0068]

[0069] in represents the objective function value, and The characteristic vector components representing the weather and equipment status respectively, and is the weight transformation matrix corresponding to the eigenvector, and is the balance factor determined by backpropagation of historical errors. Secondly, the preset model parameters are optimized using an improved particle swarm optimization to generate an initial parameter set consisting of 50 sets of parameter solutions. Each parameter solution contains a weight vector and a set of bias terms. Then, the momentum-accelerated gradient descent method is used to process the initial parameter set. During the iteration process, when the gradient change for 5 consecutive updates is less than the threshold of 1×10 -6The optimization is stopped when , and the fine-tuning parameter set is output. Finally, the generalization ability of the fine-tuning parameter set is tested through Monte Carlo cross-validation. The parameter solution with a verification error rate lower than 15% of the baseline is considered valid, and the final optimization model is generated.

[0070] For example, the model optimization was implemented in a 300MW photovoltaic power station in Gansu Province, and the obtained dynamic coupling eigenvector was decomposed into the light intensity vector and the inverter efficiency vector . Determined by error analysis Take 0.7, Taking 0.3, an objective function dominated by illumination characteristics was constructed. After 200 particle swarm iterations to generate initial parameters, the momentum gradient method converged after 83 iterations, ultimately selecting three sets of parameter solutions with validation errors that outperformed the baseline by 19% to 23%. The resulting optimized model, integrated with these solutions, demonstrated more accurate tracking of rapidly changing cloud shadow effects in subsequent dawn and dusk tests.

[0071] Optionally, the performing generalization capability verification on the fine-tuning parameter set based on the model optimization objective function to generate an optimized new energy power generation prediction model includes:

[0072] Optimizing the objective function based on the model to obtain a generalization ability evaluation matrix;

[0073] The generalization capability evaluation matrix is ​​used to verify the generalization capability of the fine-tuning parameter set, and an optimized new energy power generation prediction model is generated.

[0074] Specifically, when verifying the generalization ability of the fine-tuning parameter set based on the model optimization objective function, it is necessary to first derive the generalization ability evaluation matrix. This matrix is ​​calculated using the following formula:

[0075] ，

[0076] in is a 128×1 dimensional generalization ability evaluation matrix, represents a spatiotemporal feature map matrix of size 128×204 (generated by principal component analysis of the dynamic coupling feature vector), Represents a 204×1 dimensional fine-tuning parameter set vector (formed by concatenating all weight parameters in the fine-tuning parameter set). The matrix performs minimum-maximum normalization and maps its values ​​to the interval [0,1]. Then calculate row by row The variance stability index of matrix elements is used. When the variance value of a row of elements exceeds the preset threshold of 0.25, it is marked as a high-fluctuation parameter group. Finally, the parameter groups with variance stability index below the threshold are retained and mapped back to the original parameter space to generate the optimized new energy power generation prediction model.

[0077] For example, a generalization verification is conducted on the optimization model of a wind farm in Inner Mongolia. The dynamic coupling eigenvector is generated by principal component analysis to generate a 128-dimensional wind speed pulsation correlation. Matrix. The fine-tuning parameter set consists of 204 weight parameters consisting of a vector After matrix multiplication, we get The matrix row corresponding to the sandstorm weather model showed a variance of 0.38. The parameter group corresponding to this row was identified as primarily related to the wind turbine yaw response parameters and was removed as a highly volatile group. The remaining parameters were ultimately used to reconstruct the optimized model, maintaining stable prediction performance during subsequent sandstorm test days.

[0078] Optionally, adjusting parameters of the optimized new energy power generation prediction model based on the predicted power generation data to generate a fine-tuned new energy power generation prediction model includes:

[0079] Monitoring the continuous deviation between the predicted power generation data and the preset actual power generation data to generate a prediction deviation sequence;

[0080] generating parameter adjustment suggestions using the forecast deviation sequence;

[0081] The parameters of the optimized new energy power generation prediction model are adjusted according to the parameter adjustment suggestion to generate a fine-tuned new energy power generation prediction model.

[0082] Specifically, the continuous deviation between the predicted power generation data and the actual power generation data is monitored first, and the continuous prediction deviation monitoring sequence is as follows: Figure 3 As shown. Specifically, the predicted value and actual value of each 15-minute time step are compared through a sliding time window. When the absolute deviation of three consecutive time steps exceeds the preset threshold of 5%, the deviation record is triggered, and a prediction deviation sequence containing a timestamp and deviation amplitude is generated. Subsequently, discrete cosine transform processing is performed on the prediction deviation sequence, and the amplitude of the low-frequency component is extracted as the basis for parameter adjustment recommendations, where the component below 0.1Hz represents the systematic offset, and the component above 0.1Hz reflects the short-term fluctuation characteristics. Parameter adjustment recommendations containing weight attenuation factors and bias correction values ​​are generated based on the spectrum analysis results. Finally, the key parameters of the optimized model are located based on the parameter adjustment recommendations, including the meteorological response layer weight matrix and the output calibration bias term. The parameter correction vector matrix is ​​constructed using the correction values ​​in the parameter adjustment recommendations. , the key parameter correction matrix is ​​as follows Figure 4As shown, the matrix elements are generated by the convolution operation of the deviation spectrum amplitude and the historical correction coefficient. Finally, the surface correction operation is performed:

[0083]

[0084] In the formula is the updated weight, is the original weight, For Hadamard, The learning rate is set to 0.05 to generate a fine-tuned new energy power generation prediction model.

[0085] For example, the winter optimization model of a photovoltaic power station has continuous high prediction deviations. Monitoring found that the actual power generation in the period of 13:00-13:45 at noon increased by 24% due to the melting of snow, while the predicted value only increased by 12%, generating three consecutive negative deviation records exceeding the threshold. The discrete cosine transform identified that the amplitude of the low-frequency component of 0.05Hz reached 0.87, indicating that there is a systematic under-prediction. The parameter adjustment suggestion generated based on this requires increasing the weight of the light intensity sensitive layer by 20%. In the dynamic parameter response surface, it was detected that the weight No. 73 corresponds to the light intensity gradient response, and its sensitivity is =1.8. Construct parameter correction matrix through convolution operation The value at site 73 was 0.2. After correction at a learning rate of 0.05, the weight of this site increased by 1%, and the reconstructed fine-tuning model successfully captured the power generation jump characteristics of subsequent sunny and snowy weather.

[0086] Optionally, adjusting parameters of the optimized new energy power generation prediction model according to the parameter adjustment suggestion to generate a fine-tuned new energy power generation prediction model includes:

[0087] Obtaining a parameter correction vector matrix according to the parameter adjustment suggestion;

[0088] Obtaining a dynamic parameter response surface based on key parameters of the optimized new energy power generation prediction model;

[0089] The parameter correction vector matrix is ​​used to correct the dynamic parameter response surface to generate a fine-tuned new energy power generation prediction model.

[0090] Specifically, the model response surface optimization is as follows Figure 5 As shown in the figure, the weight attenuation factors and bias correction values ​​included in the parameter adjustment suggestions are first constructed into a parameter correction vector matrix. The rows of this matrix correspond to the model weight parameter index numbers, and the columns store the data sources of the correction values ​​of each parameter. At the same time, the key weight parameters of the meteorological response layer in the optimized new energy power generation prediction model are extracted, and the dynamic parameter response surface is calculated based on the parameter sensitivity formula:

[0091] ,

[0092] In the formula For the Number weight For the first Output power generation per time step sensitivity, Indicates The change in weight within the neighborhood, The corresponding power generation change is calculated using the forward difference method at the input value of the real-time meteorological response layer. The parameter correction vector matrix is ​​applied to the dynamic parameter response surface, and through weighted correction operations, a fine-tuned new energy power generation prediction model is ultimately generated.

[0093] For example, during the winter operation of a 150MW wind farm in Inner Mongolia, the optimization model's prediction of blade icing conditions continued to be low. The monitoring system captured three consecutive 15-minute periods in which the actual power generation was 18% to 25% higher than the predicted value, generating a time-stamped prediction deviation sequence. The spectrum analysis extracted a 0.04Hz low-frequency component with an amplitude of 0.92, which represents insufficient system response. Based on this, a parameter adjustment recommendation was generated requiring an increase in the torque response layer weight by 15%. In the dynamic parameter response surface, the weight number 89 corresponding to the wind speed-torque conversion coefficient (basic value =0.74), obtained by forward difference method =23.5 ( = 0.01), calculate the sensitivity =2.35. Construct parameter correction matrix At the 89th position, the value is 0.15. =0.05After the correction, the weight is updated to The fine-tuned model accurately predicted the torque loss recovery curve in subsequent freezing rain weather.

[0094] Optionally, performing power generation prediction on the optimized new energy power generation prediction model to obtain predicted power generation data includes:

[0095] Performing power generation prediction on the optimized new energy power generation prediction model to obtain an original prediction output tensor;

[0096] The original predicted output tensor is corrected after dynamic coupling characteristics to obtain predicted power generation data.

[0097] Specifically, the photovoltaic power forecast deviation monitoring sequence is as follows: Figure 6 As shown in the figure, the pre-processed multi-source heterogeneous data is first input into the optimized new energy generation prediction model. The model generates the original prediction output tensor through forward propagation calculation. ,in It is a three-dimensional tensor containing time dimension and space dimension. The time dimension represents the prediction time step and the space dimension represents the geographical distribution of the power generation unit. Then the dynamic coupling feature post-correction is performed. The parameter correction trigger signal is as follows Figure 7 As shown, it is specifically implemented through the characteristic correction formula:

[0098] ,

[0099] in is the corrected predicted power generation data, is the raw prediction output tensor, is the coupling correction coefficient matrix, is the dynamic coupling eigenvector. The matrix is ​​obtained by analyzing the correlation between historical prediction errors and dynamic coupling eigenvectors, and the least squares fitting method is used to determine the correction weight of each feature. Contains dynamically coupled information on weather, equipment status, and geographic features.

[0100] For example, a cluster of offshore wind farms uses this method to predict power generation. The wind speed forecast for the next 24 hours, wind turbine health status data, and sea state data are input into the optimized prediction model. First, the original prediction output tensor is obtained. The system detects that it is currently in the monsoon transition period, and the dynamic coupling eigenvector It contains significant pressure gradient mutation characteristics and unit response delay characteristics. The matrix is ​​used to correct the prediction results, with an emphasis on enhancing the weight of the predicted values ​​in the gust response area.

[0101] Optionally, performing stability assessment on the topological structure of the geographic spatiotemporal feature to generate a terrain stability coefficient includes:

[0102] Based on the topological structure of the geographic spatiotemporal features, a terrain topological correlation matrix is ​​obtained;

[0103] Performing a spatiotemporal evolution pattern analysis on the terrain topological correlation matrix to obtain a terrain dynamic stability vector;

[0104] A stability evaluation is performed on the terrain dynamic stability vector to generate a terrain stability coefficient.

[0105] Specifically, we first construct a terrain topology correlation matrix based on geographic spatiotemporal characteristics. The terrain topology correlation matrix is ​​as follows: Figure 8 The matrix is ​​expressed as an adjacency matrix, where each element represents the connection strength between geographic spatial units, which is obtained by weighting the inverse of the geographic distance:

[0106] ,

[0107] In the formula is the terrain topology correlation matrix element, For geographical units and The three-dimensional Euclidean distance between them is then calculated. The spatiotemporal evolution pattern analysis is then performed on the matrix, and the main director evolution pattern is extracted by eigenvalue decomposition:

[0108] ,

[0109] in is the terrain dynamic stability vector, through the matrix Perform singular value decomposition to obtain the terrain dynamic stability vector as Figure 9 The final terrain stability coefficient is calculated by the stability evaluation formula:

[0110]

[0111] is the terrain stability coefficient, is the vector corresponding to the largest singular value, and the denominator is the sum of the two norms of all eigenvectors, ensuring that the coefficient value is between 0 and 1.

[0112] For example, when constructing the geographic spatiotemporal characteristics of 50 wind turbines in a mountainous wind farm, the geographic units are divided into 1km×1km grids for each wind turbine. The analysis found that the terrain topological correlation matrix of the northeast slope area showed a high-intensity connection feature, and the singular value decomposition was obtained. The vector shows that there is a significant terrain coupling effect in this area. The calculated terrain stability coefficient =0.82, which is much higher than the average value of 0.45 in the field, indicating that the terrain characteristics of the region have a strong dominant effect on the wind field and flow field.

[0113] Based on the same inventive concept, Figure 10 As shown, the present invention also provides an artificial intelligence-based new energy power generation prediction model optimization system, the system comprising:

[0114] Multi-source heterogeneous data acquisition module, used to obtain meteorological data, equipment operation status data and geographic information data, and obtain multi-source heterogeneous data;

[0115] A spatiotemporal dynamic coupling engine, configured to perform spatiotemporal dynamic feature coupling processing on the multi-source heterogeneous data to generate a dynamic coupling feature vector;

[0116] A model parameter optimization module, configured to optimize the parameters of a preset new energy power generation prediction model based on the dynamic coupling eigenvector to generate an optimized new energy power generation prediction model;

[0117] A power generation prediction execution module is used to perform power generation prediction on the optimized new energy power generation prediction model to obtain predicted power generation data;

[0118] The adaptive correction module is used to adjust the parameters of the optimized new energy power generation prediction model based on the predicted power generation data to generate a fine-tuned new energy power generation prediction model.

[0119] It should be noted that the electrical connections between the above-mentioned units do not necessarily mean direct connections of lines. Indirect connections are applicable to the embodiments of the present invention as long as the purpose of the present invention is achieved. The above description is only an exemplary embodiment of the present invention and is not intended to limit the scope of the present invention.

[0120] That is, any equivalent changes and modifications made according to the teachings of the present invention are still within the scope of the present invention. Those skilled in the art will readily conceive of other embodiments of the present invention after considering the disclosure of the specification and practical truths. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary technical means in the art not described herein.

Claims

1. The artificial intelligence-based new energy power generation prediction model optimization method is characterized by: The method comprises: Acquire meteorological data, equipment operating status data, and geographic information data to obtain multi-source heterogeneous data; Performing spatiotemporal dynamic feature coupling processing on the multi-source heterogeneous data to generate a dynamic coupling feature vector; Based on the dynamic coupling characteristic vector, the parameters of the preset new energy power generation prediction model are optimized to generate an optimized new energy power generation prediction model; Performing power generation prediction on the optimized new energy power generation prediction model to obtain predicted power generation data; Based on the predicted power generation data, adjusting parameters of the optimized new energy power generation prediction model to generate a fine-tuned new energy power generation prediction model, including: monitoring the continuous deviation between the predicted power generation data and the preset actual power generation data to generate a prediction deviation sequence; Generate parameter adjustment suggestions using the prediction deviation sequence; adjust parameters of the optimized new energy power generation prediction model according to the parameter adjustment suggestions to generate a fine-tuned new energy power generation prediction model; Among them, adjusting the parameters of the optimized new energy power generation prediction model according to the parameter adjustment suggestion to generate a fine-tuned new energy power generation prediction model includes: constructing the weight attenuation factor and bias correction value contained in the parameter adjustment suggestion into a parameter correction vector matrix, the rows of the matrix correspond to the model weight parameter index number, and the columns store the correction value data source of each parameter; at the same time, extracting the key weight parameters of the meteorological response layer in the optimized new energy power generation prediction model, and calculating the dynamic parameter response surface based on the parameter sensitivity formula: ， In the formula For the Number weight For the first Output power generation per time step sensitivity, Indicates The change in weight within the neighborhood, In order to correspond to the change in power generation, the surface data is calculated at the input value of the real-time meteorological response layer through the forward difference method, and the parameter correction vector matrix is ​​applied to the dynamic parameter response surface. Through the weighted correction operation, a fine-tuned new energy power generation prediction model is finally generated.

2. The artificial intelligence-based new energy power generation prediction model optimization method according to claim 1, characterized in that: The performing spatiotemporal dynamic feature coupling processing on the multi-source heterogeneous data to generate a dynamic coupling feature vector includes: Extracting the spatiotemporal dynamic characteristics of the meteorological data to generate meteorological spatiotemporal characteristics; Extracting the spatiotemporal dynamic features of the equipment operation status data to generate equipment status spatiotemporal features; Extracting spatiotemporal dynamic features of the geographic information data to generate geographic spatiotemporal features; Calculating the meteorological spatiotemporal characteristics, the equipment state spatiotemporal characteristics, and the geographic spatiotemporal characteristics to obtain a coupling weight; The meteorological spatiotemporal features, the device state spatiotemporal features, and the geographic spatiotemporal features are fused based on the coupling weights to generate a dynamic coupling feature vector.

3. The artificial intelligence-based new energy power generation prediction model optimization method according to claim 2, characterized in that: The calculating of the meteorological spatiotemporal characteristics, the equipment state spatiotemporal characteristics, and the geographic spatiotemporal characteristics to obtain coupling weights includes: Identifying interaction patterns between the meteorological spatiotemporal characteristics and generating environmental coupling coefficients; Monitoring the response relationship between the spatiotemporal characteristics of the device state and environmental factors to generate a response correlation coefficient; Performing stability assessment on the topological structure of the geographic spatiotemporal features to generate a terrain stability coefficient; The environmental coupling coefficient, the response correlation coefficient and the terrain stability coefficient are integrated to obtain a coupling weight.

4. The artificial intelligence-based new energy power generation prediction model optimization method according to claim 1, characterized in that: Optimizing parameters of a preset new energy power generation prediction model based on the dynamic coupling eigenvector to generate an optimized new energy power generation prediction model includes: Based on the dynamic coupling eigenvector, a model optimization objective function is obtained; Optimizing parameters of the new energy power generation prediction model to generate an initial parameter set; Performing gradient acceleration optimization on the initial parameter set to generate a fine-tuning parameter set; Based on the model optimization objective function, the generalization capability of the fine-tuning parameter set is verified to generate an optimized new energy power generation prediction model.

5. The artificial intelligence-based new energy power generation prediction model optimization method according to claim 4 is characterized in that: The step of performing generalization capability verification on the fine-tuning parameter set based on the model optimization objective function to generate an optimized new energy power generation prediction model includes: Optimizing the objective function based on the model to obtain a generalization ability evaluation matrix; The generalization capability evaluation matrix is ​​used to verify the generalization capability of the fine-tuning parameter set, and an optimized new energy power generation prediction model is generated.

6. The artificial intelligence-based new energy power generation prediction model optimization method according to claim 1, characterized in that: The power generation prediction of the optimized new energy power generation prediction model is performed to obtain the predicted power generation data, which includes: Performing power generation prediction on the optimized new energy power generation prediction model to obtain an original prediction output tensor; The original predicted output tensor is corrected after dynamic coupling characteristics to obtain predicted power generation data.

7. The artificial intelligence-based new energy power generation prediction model optimization method according to claim 3 is characterized in that: The stability assessment of the topological structure of the geographic spatiotemporal feature to generate a terrain stability coefficient includes: Based on the topological structure of the geographic spatiotemporal features, a terrain topological correlation matrix is ​​obtained; Performing a spatiotemporal evolution pattern analysis on the terrain topological correlation matrix to obtain a terrain dynamic stability vector; A stability evaluation is performed on the terrain dynamic stability vector to generate a terrain stability coefficient.

8. An artificial intelligence-based new energy power generation prediction model optimization system, applied to the artificial intelligence-based new energy power generation prediction model optimization method according to any one of claims 1 to 7, characterized in that: The system comprises: Multi-source heterogeneous data acquisition module, used to obtain meteorological data, equipment operation status data and geographic information data, and obtain multi-source heterogeneous data; A spatiotemporal dynamic coupling module, configured to perform spatiotemporal dynamic feature coupling processing on the multi-source heterogeneous data to generate a dynamic coupling feature vector; A model parameter optimization module, configured to optimize the parameters of a preset new energy power generation prediction model based on the dynamic coupling eigenvector to generate an optimized new energy power generation prediction model; A power generation prediction execution module is used to perform power generation prediction on the optimized new energy power generation prediction model to obtain predicted power generation data; an adaptive correction module for adjusting parameters of the optimized new energy power generation prediction model based on the predicted power generation data to generate a fine-tuned new energy power generation prediction model, including: monitoring the continuous deviation between the predicted power generation data and preset actual power generation data to generate a prediction deviation sequence; using the prediction deviation sequence to generate parameter adjustment suggestions; and adjusting parameters of the optimized new energy power generation prediction model according to the parameter adjustment suggestions to generate a fine-tuned new energy power generation prediction model; Among them, adjusting the parameters of the optimized new energy power generation prediction model according to the parameter adjustment suggestion to generate a fine-tuned new energy power generation prediction model includes: constructing the weight attenuation factor and bias correction value contained in the parameter adjustment suggestion into a parameter correction vector matrix, the rows of the matrix correspond to the model weight parameter index number, and the columns store the correction value data source of each parameter; at the same time, extracting the key weight parameters of the meteorological response layer in the optimized new energy power generation prediction model, and calculating the dynamic parameter response surface based on the parameter sensitivity formula: ， In the formula For the Number weight For the first Output power generation per time step sensitivity, Indicates The change in weight within the neighborhood, In order to correspond to the change in power generation, the surface data is calculated at the input value of the real-time meteorological response layer through the forward difference method, and the parameter correction vector matrix is ​​applied to the dynamic parameter response surface. Through the weighted correction operation, a fine-tuned new energy power generation prediction model is finally generated.

Images