Photovoltaic power prediction method and system
Through the quantum-evolution hybrid model and spatiotemporal feature fusion technology, the adaptability and uncertainty problems of photovoltaic power prediction are solved, and photovoltaic power prediction with higher accuracy and robustness is achieved.
Patent Information
- Application Number
- CN202510766296.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-10
- Publication Date
- 2025-09-19
Smart Images
Figure CN120675043A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of photovoltaic power generation management, and more particularly to a photovoltaic power prediction method and system. Background Art
[0002] Photovoltaic power generation is a relatively common green energy source. In the management of photovoltaic power generation, photovoltaic power prediction is a key technology for ensuring stable grid operation and optimizing energy scheduling. Traditional prediction methods typically rely on explicit physical modeling, have poor adaptability, and suffer from a single data dimension and insufficient utilization of spatiotemporal features. Mainstream deep learning models such as CNN-LSTM rely solely on historical power series or data from a single meteorological station, failing to effectively integrate high-resolution meteorological grids with sky image data. Existing methods also output deterministic prediction values and fail to model uncertain events such as sudden cloud changes and instantaneous shadows, resulting in inaccurate prediction results and poor scenario adaptability.
[0003] Based on the problems existing in the above-mentioned traditional methods, through the collaborative innovation of quantum probability modeling, multi-source spatiotemporal data fusion and dynamic evolution strategy, the long-standing technical problems such as poor adaptability and lack of uncertainty quantification in photovoltaic power forecasting have been solved. The prediction accuracy, real-time performance and extreme weather robustness are significantly better than existing technologies. Summary of the Invention
[0004] In view of the deficiencies in the prior art, the present invention aims to provide a photovoltaic power prediction method and system to solve the problems existing in the above-mentioned background technology.
[0005] The above technical objectives of the present invention are achieved through the following technical solutions: A photovoltaic power prediction method, comprising: Step A, data collection and preprocessing: collecting operating data of the photovoltaic power station, and preprocessing the data, filling missing values with a sliding window mean, and performing normalization to obtain model input;
[0006] Step B, model prediction: constructing a quantum-evolution hybrid model, inputting the model input into the constructed quantum-evolution hybrid model, and obtaining a model prediction result;
[0007] Step C, predicting power after physical constraints: performing denormalization on the model prediction results, truncating them to the maximum installed capacity of the photovoltaic power station, and outputting the final predicted photovoltaic power value.
[0008] Optionally, the model input includes: historical power data of the photovoltaic power station, an environmental meteorological grid of the photovoltaic power station, and an environmental sky image of the photovoltaic power station.
[0009] Optionally, the step B of constructing a quantum-evolution hybrid model, inputting the model input into the constructed quantum-evolution hybrid model to obtain a model prediction result, is specifically implemented as follows, including:
[0010] B1. Construct a quantum hidden Markov chain model to map the time-series fluctuations of photovoltaic power into quantum state transitions, quantify uncertainty, map the power sequence into a quantum superposition state, and generate probability amplitudes through a Gaussian mixture model.
[0011] B2. Construct a spatiotemporal convolutional attention network, perform multi-scale 3D dilated convolution on the meteorological grid, implement a ring-partitioned attention mechanism on the sky image, achieve spatiotemporal feature fusion, integrate the local features of the meteorological grid and the global cloud motion features of the sky image, and predict the output normalized power value.
[0012] Optionally, in step C, the model prediction result is denormalized and truncated to the maximum installed capacity of the photovoltaic power station to output the final predicted photovoltaic power value. The specific implementation process is as follows, including:
[0013] C1. Based on the prediction results output from steps B1 and B2, Bayesian fusion is used to perform weighted fusion of the probabilistic output of QHMM and the deterministic output of ST-CAN to obtain the fusion result;
[0014] C2. Denormalize the fusion result and impose power station installed capacity constraints, and output the final predicted photovoltaic power value.
[0015] Optionally, step B1 specifically includes:
[0016] S1, quantum state initialization, set N = 15 quantum states, initial μ i Uniformly distributed in [0,100]MW;
[0017] S2, probability amplitude update, update μ through EM algorithm every other prediction cycle i and σ i , the iterative formula is expressed as:
[0018] S3. Calculate the power prediction expected value at the next prediction cycle, expressed as where α i (t) is expressed as the probability of the quantum state, calculated by square of the probability amplitude;
[0019] S4. Output the calculated power prediction expected value as a prediction result.
[0020] Optionally, step B2 specifically includes:
[0021] S1. Meteorological grid processing, input time series grid, calculate dynamic void rate, expressed as G t-6:t ∈R 7×20×20×5 ;
[0022] S2, sky image processing, convert the ambient sky image into HSV space, calculate the cloud coverage by azimuth area, expressed as θ=0°,10°,…,350°;
[0023] S3, based on the calculation results of S1 and S2, the local characteristics of the meteorological grid and the global cloud motion characteristics of the sky image are integrated to predict the power value, which is expressed as P ST-CAN =f CNN-ATT (G t ,I t )(P max -P min )+P min , where f CNN-ATT Denoted as the mapping function of the spatiotemporal convolutional attention network, G t Represents the extracted meteorological raster data, I t Represented as the extracted cloud motion features, P max Expressed as the historical maximum photovoltaic power, P min Expressed as the historical minimum photovoltaic power;
[0024] S4. Output the calculated predicted power value as a prediction result.
[0025] Optionally, the step C1 specifically includes:
[0026] S1. Fusion of prediction results: The probabilistic output of QHMM is combined with the deterministic output of ST-CAN, that is, the prediction results of QHHM and ST-CAN are combined, using Bayesian fusion, expressed as P pred =w1·P QHMM +w2·P ST-CAN , where w1 and w2 represent the weight coefficients that influence the final prediction results;
[0027] S2, Constraint influence weight coefficient: Dynamically adjust the constraints according to the uncertainty of the QHMM model and ST-CAN model, expressed as w2=1-w1,where Expressed as the variance of the QHMM model, It is expressed as the variance of the ST-CAN model;
[0028] S3. Calculate the fusion result and output it.
[0029] Optionally, the step C1 specifically includes:
[0030] S1. Denormalization of fusion results: Denormalize the fusion results and output them;
[0031] S2. Apply actual constraints: Apply the power plant installed capacity constraint to the denormalized fusion result, expressed as P output =min(max(P pred ,0),P max ), where P max It is expressed as the maximum installed capacity of the photovoltaic power station, max(·, 0), to prevent negative power output;
[0032] S3. The fusion result after the constraints is output as the final predicted photovoltaic power value at the moment of the next power generation cycle.
[0033] A photovoltaic power prediction system for executing the above photovoltaic power prediction method comprises:
[0034] The data acquisition module is used to collect the working data of the photovoltaic power station, collect the historical power data in the past invention cycle, and capture and transmit the environmental meteorological grid of the photovoltaic power station and the environmental sky image of the photovoltaic power station;
[0035] A data processing module is used to receive the data transmitted by the data acquisition module, pre-process the transmitted data, and then obtain the model input and transmit it;
[0036] A hybrid model module, configured to calculate and transmit a model prediction result based on the model input;
[0037] The data constraint module is used to constrain the actual invention conditions based on the model prediction results, limit the predicted value to the actual physical range, ensure that it meets the photovoltaic invention characteristics, and then output the final predicted photovoltaic power value.
[0038] Optionally, it also includes a visualization interface module, which is set on the system terminal and is used to visualize the data of each module and transmit it on the interface platform to facilitate management personnel to view the data.
[0039] In summary, the present invention has the following beneficial effects:
[0040] 1. Using quantized modeling, the multimodal characteristics of power fluctuations are quantified by superposition state probability amplitudes. This reduces the uncertainty in the prediction of photovoltaic power from photovoltaic power stations in cloud mutation scenarios, making the predicted photovoltaic frequency more adaptable to actual scenarios and improving extreme weather robustness.
[0041] 2. The prediction accuracy of cloudy scenarios is improved by fusing spatiotemporal features. Grid-like atrous convolution is used to dynamically adjust the atrous ratio to make the model adaptive to cloud movement speed, reducing irradiance prediction errors in scenarios with rapid cumulus cover. The sky image circular attention method is used to analyze the HSV features of azimuth partitions (one partition every 10°) to increase the accuracy of cloud motion direction prediction, indirectly improving power prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 It is a schematic flow chart of the method steps of the present invention;
[0043] Figure 2 It is a logical diagram of the system implementation of the present invention. DETAILED DESCRIPTION
[0044] To make the objectives, features, and advantages of the present invention more readily apparent, the following detailed description of the present invention is provided with reference to the accompanying drawings. The accompanying drawings illustrate several embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein.
[0045] In the present invention, unless otherwise expressly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be a communication between the two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. The terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of such features.
[0046] In the present invention, unless otherwise expressly specified and limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features being in contact not directly but through another feature between them. Moreover, a first feature being "above," "above," and "above" a second feature includes the first feature being directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below," "below," and "below" a second feature includes the first feature being directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature. The terms "vertical," "horizontal," "left," "right," "above," "below," and similar expressions are for illustrative purposes only and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore should not be understood as limiting the present invention.
[0047] The present invention is described in detail below with reference to the accompanying drawings and embodiments.
[0048] The present invention provides a photovoltaic power prediction method, such as Figure 1 As shown, the method includes: Step A, data collection and preprocessing: collecting the working data of the photovoltaic power station, preprocessing the data, filling the missing values with the sliding window mean, and performing normalization to obtain the model input;
[0049] Step B, model prediction: constructing a quantum-evolution hybrid model, inputting the model input into the constructed quantum-evolution hybrid model, and obtaining a model prediction result;
[0050] Step C, predicting power after physical constraints: performing denormalization on the model prediction results, truncating them to the maximum installed capacity of the photovoltaic power station, and outputting the final predicted photovoltaic power value.
[0051] Furthermore, the model input includes: historical power data of the photovoltaic power station, an environmental meteorological grid of the photovoltaic power station, and an environmental sky image of the photovoltaic power station.
[0052] Furthermore, the step B of constructing a quantum-evolution hybrid model, inputting the model input into the constructed quantum-evolution hybrid model, and obtaining a model prediction result, is specifically implemented as follows, including:
[0053] B1. Construct a quantum hidden Markov chain model to map the time-series fluctuations of photovoltaic power into quantum state transitions, quantify uncertainty, map the power sequence into a quantum superposition state, and generate probability amplitudes through a Gaussian mixture model.
[0054] B2. Construct a spatiotemporal convolutional attention network, perform multi-scale 3D dilated convolution on the meteorological grid, implement a ring-partitioned attention mechanism on the sky image, achieve spatiotemporal feature fusion, integrate the local features of the meteorological grid and the global cloud motion features of the sky image, and predict the output normalized power value.
[0055] Furthermore, in step C, the model prediction result is denormalized and truncated to the maximum installed capacity of the photovoltaic power station to output the final predicted photovoltaic power value. The specific implementation process is as follows, including:
[0056] C1. Based on the prediction results output from steps B1 and B2, Bayesian fusion is used to perform weighted fusion of the probabilistic output of QHMM and the deterministic output of ST-CAN to obtain the fusion result;
[0057] C2. Denormalize the fusion result and impose power station installed capacity constraints, and output the final predicted photovoltaic power value.
[0058] Furthermore, the step B1 specifically includes:
[0059] S1, quantum state initialization, set N = 15 quantum states, initial μ i Uniformly distributed in [0,100]MW;
[0060] S2, probability amplitude update, update μ through EM algorithm every other prediction cycle i and σ i , the iterative formula is expressed as:
[0061] S3. Calculate the power prediction expected value at the next prediction cycle, expressed as where α i (t) is expressed as the probability of the quantum state, calculated by square of the probability amplitude;
[0062] S4. Output the calculated power prediction expected value as a prediction result.
[0063] In the specific implementation process, the quantum state is initialized, N=10 quantum states are set, and the initial center value μ i Uniformly distributed in [0,100]MW, expressed as μ1=10MW,μ2=20MW,…,μ 10 =100MW;
[0064] Update the probability amplitude. The current collected power of the photovoltaic power station is P t =56.5MW, calculate for the i=6th state, μ6=60MW,σ6=10MW, Then the power expectation value is calculated, which is expressed as The expected power value is then output as a prediction result.
[0065] Furthermore, the step B2 specifically includes:
[0066] S1. Meteorological grid processing, input time series grid, calculate dynamic void rate, expressed as G t-6:t ∈R 7×20×20×5 ;
[0067] S2, sky image processing, convert the ambient sky image into HSV space, calculate the cloud coverage by azimuth area, expressed as θ=0°,10°,…,350°;
[0068] S3, based on the calculation results of S1 and S2, the local characteristics of the meteorological grid and the global cloud motion characteristics of the sky image are integrated to predict the power value, which is expressed as P ST-CAN =f CNN-ATT (G t ,I t )(P max -P min )+P min , where f CNN-ATT Denoted as the mapping function of the spatiotemporal convolutional attention network, G t Represents the extracted meteorological raster data, I t Represented as the extracted cloud motion features, P max Expressed as the historical maximum photovoltaic power, P min Expressed as the historical minimum photovoltaic power;
[0069] S4. Output the calculated predicted power value as a prediction result.
[0070] In the specific implementation process, the input is 7 frames of meteorological grid, cloud speed v cloud =15m / s, the dynamic void rate is expressed as 3D hole convolution is used to extract features. Then the local features of the meteorological grid are extracted;
[0071] The transmitted ambient sky image is divided into 36 sectors (each 10° is divided into a recognition area), and the attention weight of the recognition area of θ = 90° is calculated, which is expressed as where Q 90 °Generated based on the HSV histogram of the area, V 90° Including 70% cloud coverage; then extracting the global cloud motion features of the sky image;
[0072] By calculating the features obtained above, ST-CAN outputs the normalized value, P ST－CAN =0.68, after denormalization, we get P ST-CAN =0.68×100=68MW, and the predicted power value is obtained at this time.
[0073] Furthermore, the step C1 specifically includes:
[0074] S1. Fusion of prediction results: The probabilistic output of QHMM is combined with the deterministic output of ST-CAN, that is, the prediction results of QHHM and ST-CAN are combined, using Bayesian fusion, expressed as P pred =w1·P QHMM +w2·P ST-CAN , where w1 and w2 represent the weight coefficients that influence the final prediction results;
[0075] S2, Constraint influence weight coefficient: Dynamically adjust the constraints according to the uncertainty of the QHMM model and ST-CAN model, expressed as w2=1-w1,where Expressed as the variance of the QHMM model, It is expressed as the variance of the ST-CAN model;
[0076] S3. Calculate the fusion result and output it.
[0077] In practical applications, the variance of the two models is calculated and expressed as QHMM variance. variance The uncertainty of the corresponding model is determined by the variance of the two models, and the constraint influence weight coefficient is calculated, which is expressed as w2=0.667;
[0078] Then use the influence weight coefficient to calculate the fusion result, which is expressed as P pred =0.333×58.3+0.667×68=64.8MW to obtain the fusion result and output it.
[0079] Optionally, the step C1 specifically includes:
[0080] S1. Denormalization of fusion results: Denormalize the fusion results and output them;
[0081] S2. Apply actual constraints: Apply the power plant installed capacity constraint to the denormalized fusion result, expressed as P output =min(max(P pred ,0),P max ), where P maxIt is expressed as the maximum installed capacity of the photovoltaic power station, max(·, 0), to prevent negative power output;
[0082] S3. The fusion result after the constraints is output as the final predicted photovoltaic power value at the moment of the next power generation cycle.
[0083] In the specific implementation process, the fusion results are actually constrained, which is expressed as the maximum power of the power station is the power station P max =100MW, P output =min(64.8,100)=64.8MW. When the maximum power is not exceeded, it is determined to be a valid prediction result and output as the final predicted photovoltaic power value at the time of the next power generation cycle.
[0084] A photovoltaic power prediction system for executing the above photovoltaic power prediction method comprises:
[0085] The data acquisition module is used to collect the working data of the photovoltaic power station, collect the historical power data in the past invention cycle, and capture and transmit the environmental meteorological grid of the photovoltaic power station and the environmental sky image of the photovoltaic power station;
[0086] A data processing module is used to receive the data transmitted by the data acquisition module, pre-process the transmitted data, and then obtain the model input and transmit it;
[0087] A hybrid model module, configured to calculate and transmit a model prediction result based on the model input;
[0088] The data constraint module is used to constrain the actual invention conditions based on the model prediction results, limit the predicted value to the actual physical range, ensure that it meets the photovoltaic invention characteristics, and then output the final predicted photovoltaic power value.
[0089] Furthermore, it also includes a visualization interface module, which is set on the system terminal and is used to visualize the data of each module and transmit it on the interface platform to facilitate management personnel to view the data.
[0090] The photovoltaic power prediction method and system of the present invention use quantized modeling to quantize the multimodal characteristics of power fluctuations by superimposing state probability amplitudes. In cloud mutation scenarios, the uncertainty in the prediction of photovoltaic power of photovoltaic power stations is reduced, making the predicted value of photovoltaic frequency more adaptable to actual scenarios and improving extreme weather robustness. The prediction accuracy of cloudy scenarios is improved by fusing spatiotemporal features. Grid dilated convolution is adopted to dynamically adjust the dilation rate to make the model adaptive to the cloud movement speed, reducing the irradiance prediction error in rapid cumulus cloud cover scenarios. The sky image circular attention method is adopted to improve the accuracy of cloud movement direction prediction through HSV feature analysis of azimuth angle partitions (one zone every 10 degrees), thereby indirectly improving power prediction accuracy.
[0091] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.
Claims
1. A photovoltaic power prediction method, characterized in that: include: Step A, Data Collection and Preprocessing: Collect the operating data of the photovoltaic power station and preprocess the data. Use the sliding window mean to fill the missing values and perform normalization to obtain the model input. Step B, model prediction: constructing a quantum-evolution hybrid model, inputting the model input into the constructed quantum-evolution hybrid model, and obtaining a model prediction result; Step C, predicting power after physical constraints: performing denormalization on the model prediction results, truncating them to the maximum installed capacity of the photovoltaic power station, and outputting the final predicted photovoltaic power value.
2. A photovoltaic power prediction method and system according to claim 1, characterized in that: The model input includes: historical power data of the photovoltaic power station, the environmental meteorological grid of the photovoltaic power station and the environmental sky image of the photovoltaic power station.
3. A photovoltaic power prediction method according to claim 2, characterized in that: The step B of constructing a quantum-evolution hybrid model, inputting the model input into the constructed quantum-evolution hybrid model, and obtaining a model prediction result, is specifically implemented as follows, including: B1. Construct a quantum hidden Markov chain model to map the time-series fluctuations of photovoltaic power into quantum state transitions, quantify uncertainty, map the power sequence into a quantum superposition state, and generate probability amplitudes through a Gaussian mixture model. B2. Construct a spatiotemporal convolutional attention network, perform multi-scale 3D dilated convolution on the meteorological grid, implement a ring-partitioned attention mechanism on the sky image, achieve spatiotemporal feature fusion, integrate the local features of the meteorological grid and the global cloud motion features of the sky image, and predict the output normalized power value.
4. A photovoltaic power prediction method according to claim 3, characterized in that: In step C, the model prediction result is denormalized and truncated to the maximum installed capacity of the photovoltaic power station to output the final predicted photovoltaic power value. The specific implementation process is as follows, including: C1. Based on the prediction results output from steps B1 and B2, Bayesian fusion is used to perform weighted fusion of the probabilistic output of QHMM and the deterministic output of ST-CAN to obtain the fusion result; C2. Denormalize the fusion result and impose power station installed capacity constraints, and output the final predicted photovoltaic power value.
5. A photovoltaic power prediction method according to claim 3, characterized in that: The step B1 specifically includes: S1, quantum state initialization, set N = 15 quantum states, initial μ i Uniformly distributed in [0,100]MW; S2, probability amplitude update, update μ through EM algorithm every other prediction cycle i and σ i , the iterative formula is expressed as: S3. Calculate the power prediction expected value at the next prediction cycle, expressed as where α i (t) is expressed as the probability of the quantum state, calculated by square of the probability amplitude; S4. Output the calculated power prediction expected value as a prediction result.
6. A photovoltaic power prediction method according to claim 3, characterized in that: The step B2 specifically includes: S1. Meteorological grid processing, input time series grid, calculate dynamic void rate, expressed as G t-6:t ∈R 7×20×20×5 ; S2, sky image processing, convert the ambient sky image into HSV space, calculate the cloud coverage by azimuth area, expressed as θ=0°,10°,…,350°; S3, based on the calculation results of S1 and S2, the local characteristics of the meteorological grid and the global cloud motion characteristics of the sky image are integrated to predict the power value, which is expressed as P ST-CAN =f CNN-ATT (G t ,I t ), where f CNN-ATT Denoted as the mapping function of the spatiotemporal convolutional attention network, G t Represents the extracted meteorological raster data, I t Represented as the extracted cloud motion features; S4. Output the calculated predicted power value as a prediction result.
7. A photovoltaic power prediction method according to claim 4, characterized in that: The step C1 specifically includes: S1. Fusion of prediction results: The probabilistic output of QHMM is combined with the deterministic output of ST-CAN, that is, the prediction results of QHHM and ST-CAN are combined, using Bayesian fusion, expressed as P pred =w1·P QHMM +w2·P ST-CAN , where w1 and w2 represent the weight coefficients that influence the final prediction results; S2, Constraint influence weight coefficient: Dynamically adjust the constraints according to the uncertainty of the QHMM model and ST-CAN model, expressed as w2=1-w1,where Expressed as the variance of the QHMM model, It is expressed as the variance of the ST-CAN model; S3. Calculate the fusion result and output it.
8. A photovoltaic power prediction method according to claim 7, characterized in that: The step C1 specifically includes: S1. Denormalization of fusion results: Denormalize the fusion results and output them; S2. Apply actual constraints: Apply the power plant installed capacity constraint to the denormalized fusion result, expressed as P output =min(max(P pred ,0),P max ), where P max It is expressed as the maximum installed capacity of the photovoltaic power station, max(·, 0), to prevent negative power output; S3. The fusion result after the constraints is output as the final predicted photovoltaic power value at the moment of the next power generation cycle.
9. A photovoltaic power prediction system for executing a photovoltaic power prediction method according to any one of claims 1 to 8, characterized in that: include, The data acquisition module is used to collect the working data of the photovoltaic power station, collect the historical power data in the past invention cycle, and capture and transmit the environmental meteorological grid of the photovoltaic power station and the environmental sky image of the photovoltaic power station; A data processing module is used to receive the data transmitted by the data acquisition module, pre-process the transmitted data, and then obtain the model input and transmit it; A hybrid model module, configured to calculate and transmit a model prediction result based on the model input; The data constraint module is used to constrain the actual invention conditions based on the model prediction results, limit the predicted value to the actual physical range, ensure that it meets the photovoltaic invention characteristics, and then output the final predicted photovoltaic power value.
10. A photovoltaic power prediction system according to claim 9, characterized in that: It also includes a visualization interface module, which is set on the system terminal and is used to visualize the data of each module and transmit it on the interface platform to facilitate management personnel to view the data.