CN116663727A - Photovoltaic power prediction method and system - Google Patents

Photovoltaic power prediction method and system Download PDF

Info

Publication number
CN116663727A
CN116663727A CN202310635698.3A CN202310635698A CN116663727A CN 116663727 A CN116663727 A CN 116663727A CN 202310635698 A CN202310635698 A CN 202310635698A CN 116663727 A CN116663727 A CN 116663727A
Authority
CN
China
Prior art keywords
photovoltaic power
power generation
data
prediction
power prediction
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310635698.3A
Other languages
Chinese (zh)
Inventor
沈开程
贾梦麒
黄杨珏
邰彬
朱远哲
周洪
戴小然
张晓珂
汪进锋
吕鸿
陈鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Power Grid Co Ltd
Electric Power Research Institute of Guangdong Power Grid Co Ltd
Original Assignee
Guangdong Power Grid Co Ltd
Electric Power Research Institute of Guangdong Power Grid Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangdong Power Grid Co Ltd, Electric Power Research Institute of Guangdong Power Grid Co Ltd filed Critical Guangdong Power Grid Co Ltd
Priority to CN202310635698.3A priority Critical patent/CN116663727A/en
Publication of CN116663727A publication Critical patent/CN116663727A/en
Pending legal-status Critical Current

Links

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
    • H02J2300/20—The dispersed energy generation being of renewable origin
    • H02J2300/22—The renewable source being solar energy
    • H02J2300/24—The renewable source being solar energy of photovoltaic origin
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • General Health & Medical Sciences (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Power Engineering (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Biophysics (AREA)
  • Operations Research (AREA)
  • Game Theory and Decision Science (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Development Economics (AREA)
  • Primary Health Care (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Quality & Reliability (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention provides a photovoltaic power prediction method and a photovoltaic power prediction system, wherein the method comprises the following steps: acquiring photovoltaic power generation data; improving the long-term and short-term memory network based on a quantile regression method, and constructing a power prediction model; inputting the photovoltaic power generation data into a power prediction model so that the power prediction model predicts a plurality of predicted output values; and obtaining a photovoltaic power predicted value and a predicted interval based on a plurality of predicted output values, so as to realize the prediction of the photovoltaic power. According to the photovoltaic power prediction method provided by the invention, the traditional long-short-period memory network is improved by using the quantile regression method, a power prediction model for photovoltaic power prediction is obtained, photovoltaic power point prediction and interval probability prediction can be completed, the prediction output can reflect the probability range and fluctuation characteristics of power generation, the uncertainty of power generation can be quantized, the accuracy of a photovoltaic power prediction value and a prediction interval is ensured, and the final prediction accuracy is improved.

Description

Photovoltaic power prediction method and system
Technical Field
The invention relates to the technical field of photovoltaics, in particular to a photovoltaic power prediction method and system.
Background
Along with the rapid development of new energy technology, green energy power generation such as photovoltaic, wind power and the like becomes a new research hot spot due to the clean and renewable characteristics of the green energy power generation, and has good development prospect. As a key module for new energy power generation, the installed capacity of photovoltaic power generation has also been rapidly increased in recent years. However, with large-scale photovoltaic grid connection, the randomness and fluctuation of power generation of the photovoltaic grid connection can cause huge impact on the operation of the power grid, and the accurate photovoltaic power prediction can optimize the operation scheduling plan of the power grid, so that the safety and stability of the operation of the power grid are improved.
Existing photovoltaic power prediction methods can be divided into physical methods, statistical methods, machine learning methods, and hybrid methods. The physical method is to establish a mathematical model according to the photovoltaic power generation principle, and directly calculate meteorological data obtained by means of numerical weather forecast and the like to obtain photovoltaic power generation power; the statistical method is to build the association mapping relation between input and output data through curve fitting, parameter estimation, association relation analysis and other means to realize future photovoltaic power generation output prediction; machine learning generally refers to learning a mapping relationship between input and output by a neural network through a correlation algorithm to perform prediction; the hybrid method is a method of combining two or more predictive models. The machine learning method is simple and convenient and has high accuracy, and is a mainstream prediction method of photovoltaic prediction at present, and common models comprise regression trees, support vector machines, artificial neural networks and the like.
However, most existing photovoltaic power prediction methods adopt a single model, and the prediction model has a simple structure, such as a physical model, a machine learning model and the like. Each model has its own limitations, so that the existing prediction model still has various problems, such as lag in prediction effect, difficulty in tracking the output of a severely fluctuating scene, inaccurate prediction effect, and the like.
Although the machine learning method, i.e. the photovoltaic power prediction based on the neural network, achieves a certain effect from the viewpoint of the prediction model, the prediction model is easy to have the problems of over fitting, gradient disappearance, gradient explosion and the like. From the prediction result, most of the existing photovoltaic power prediction methods are point predictions. The point prediction is also called deterministic prediction, and outputs as a single-point power value, so that it is difficult for the point prediction to accurately output an actual value of the generated power, and uncertainty of power generation cannot be quantified, so that accuracy of final prediction is low.
Disclosure of Invention
The invention aims to provide a photovoltaic power prediction method and a photovoltaic power prediction system, which are used for solving the technical problems, and a long-term and short-term memory network is improved by a quantile regression method to obtain a power prediction model for photovoltaic power prediction, so that the accuracy of a photovoltaic power prediction value and a prediction interval is ensured, and the final prediction accuracy is improved.
In order to solve the technical problems, the invention provides a photovoltaic power prediction method, which comprises the following steps:
acquiring photovoltaic power generation data;
improving the long-term and short-term memory network based on a quantile regression method, and constructing a power prediction model;
inputting the photovoltaic power generation data into a power prediction model so that the power prediction model predicts a plurality of predicted output values;
and obtaining a photovoltaic power predicted value and a predicted interval based on a plurality of predicted output values, so as to realize the prediction of the photovoltaic power.
According to the scheme, the traditional long-short-period memory network is improved by using the quantile regression method, a power prediction model for photovoltaic power prediction is obtained, photovoltaic power point prediction and interval probability prediction can be completed, the prediction output can reflect the probability range and fluctuation characteristics of power generation, the uncertainty of power generation can be quantized, the accuracy of a photovoltaic power prediction value and a prediction interval is guaranteed, and the final prediction accuracy is improved.
Further, in the modification of the long-term and short-term memory network based on the quantile regression method, a power prediction model is constructed, wherein the power prediction model corresponds to the quantile tau, and a loss function of the power prediction model corresponding to the quantile tau is specifically expressed as:
wherein: l (·) represents a loss function; w (τ) = (W) f (τ),W i (τ),W c (τ),W o (τ)) represents a weight matrix for a long and short term memory network at quantiles τ, where W f (. Cndot.) a weight matrix representing the forgetting gate, W i (. Cndot.) represents the weight matrix of the input gates, W c (. Cndot.) represents the weight matrix of the memory element, W o (. Cndot.) represents the weight matrix of the output gates; b (τ) = (b) f (τ),b i (τ),b c (τ),b o (τ)) representing the long and short term memory network at quantiles τBias matrix, where b f (. Cndot.) the bias matrix representing the forgetting gate, b i (. Cndot.) represents the bias matrix of the input gate, b c (. Cndot.) represents the bias matrix of the memory element, b o (-) represents the bias matrix of the output gate; ρ τ Is a checking function; y is Y t A response variable representing the time t; x is X t The input variable at time t is indicated.
Further, in the modification of the long-term and short-term memory network based on the quantile regression method, a calculation formula of a parameter vector beta (tau) corresponding to the quantile tau in the quantile regression method is as follows:
wherein, the optimal value of the parameter vector beta (tau) corresponding to the quantile tau is determined by minimizing the function; x' is the transposed matrix of the input variables; y is a response variable.
Further, the obtaining photovoltaic power generation data includes: acquiring photovoltaic power generation data, dividing the photovoltaic power generation data according to time in proportion, and acquiring corresponding combined similar moment data under all time for being used as input of a power prediction model; the photovoltaic power generation data comprise photovoltaic power generation output power, historical meteorological data and numerical weather forecast NWP data corresponding to time and all times.
In the scheme, since the photovoltaic power output is closely related to meteorological factors, the photovoltaic power output on similar days has great similarity. Therefore, the historical data (i.e., similar day data) with high correlation with the day data to be predicted can be selected for model training to improve the accuracy of photovoltaic power generation power prediction.
Further, the obtaining the corresponding combined similar time data at all times includes: and obtaining the similar moment j of the moment i and the combined similar moment data corresponding to the similar moment j by calculating the minimum Euclidean distance between the photovoltaic power generation data corresponding to the moment i and different moments.
According to the scheme, the traditional similarity day selection algorithm is improved, the fluctuation characteristics of the photovoltaic data in reality can be reflected more effectively by acquiring the combined similarity time data, the similarity of each period is further guaranteed, the severe fluctuation scene is tracked, and therefore the prediction accuracy of the model is improved.
The scheme utilizes combination of similar day and fractional number regression to improve the traditional long-short-period memory network, and further obtains a power prediction model for photovoltaic power prediction. The similarity of each period is guaranteed by combining the similarity days, namely the similarity of the data at similar moments is combined, so that the prediction accuracy is improved; and the quantile regression realizes the point output and probability output of photovoltaic power prediction, ensures the accuracy of a photovoltaic power predicted value and a predicted interval, and improves the final prediction accuracy.
The invention provides a photovoltaic power prediction system, comprising:
the data acquisition module is used for acquiring photovoltaic power generation data;
the model construction module is used for improving the long-term and short-term memory network based on a quantile regression method and constructing a power prediction model;
the model execution module is used for inputting the photovoltaic power generation data into the power prediction model so that the power prediction model predicts a plurality of predicted output values;
and the photovoltaic power prediction module is used for acquiring a photovoltaic power predicted value and a predicted interval based on the plurality of predicted output values.
The system is simple in architecture and easy to build, application and popularization, a traditional long-short-period memory network is improved by using a quantile regression method, a power prediction model for photovoltaic power prediction is obtained, and photovoltaic power point prediction and interval probability prediction are completed. The prediction output can reflect the probability range and fluctuation characteristics of power generation, and can also quantify the uncertainty of power generation, ensure the accuracy of the photovoltaic power prediction value and the prediction interval, and improve the final prediction accuracy.
Further, the model construction module is used for improving the long-term and short-term memory network based on a quantile regression method, and constructing a power prediction model, specifically: the power prediction model corresponds to the quantile tau, and in the time t epsilon [1, T ], the loss function of the power prediction model corresponding to the quantile tau is specifically expressed as:
wherein: l (·) represents a loss function; w (τ) = (W) f (τ),W i (τ),W c (τ),W o (τ)) represents a weight matrix for a long and short term memory network at quantiles τ, where W f (. Cndot.) a weight matrix representing the forgetting gate, W i (. Cndot.) represents the weight matrix of the input gates, W c (. Cndot.) represents the weight matrix of the memory element, W o (. Cndot.) represents the weight matrix of the output gates; b (τ) = (b) f (τ),b i (τ),b c (τ),b o (τ)) represents the bias matrix of the long and short term memory network at quantiles τ, where b f (. Cndot.) the bias matrix representing the forgetting gate, b i (. Cndot.) represents the bias matrix of the input gate, b c (. Cndot.) represents the bias matrix of the memory element, b o (-) represents the bias matrix of the output gate; ρ τ Is a checking function; y is Y t A response variable representing the time t; x is X t The input variable at time t is indicated.
Further, in the model building module, a calculation formula of a parameter vector β (τ) corresponding to the quantile τ in the quantile regression method is as follows:
wherein, the optimal value of the parameter vector beta (tau) corresponding to the quantile tau is determined by minimizing the function; x' is the transposed matrix of the input variables; y is a response variable.
Further, the method is used for acquiring photovoltaic power generation data, and specifically comprises the following steps: acquiring photovoltaic power generation data, dividing the photovoltaic power generation data according to time in proportion, and acquiring corresponding combined similar moment data under all time for being used as input of a power prediction model; the photovoltaic power generation data comprise photovoltaic power generation output power, historical meteorological data and numerical weather forecast NWP data corresponding to time and all times.
Further, in the data obtaining module, the obtaining the corresponding combined similar time data at all times includes: and obtaining the similar moment j of the moment i and the combined similar moment data corresponding to the similar moment j by calculating the minimum Euclidean distance between the photovoltaic power generation data corresponding to the moment i and different moments.
Drawings
Fig. 1 is a schematic flow chart of a photovoltaic power prediction method according to an embodiment of the present invention;
fig. 2 is a schematic application flow chart of a photovoltaic power prediction method according to an embodiment of the present invention;
FIG. 3 is a block diagram of a long and short term memory LSTM according to an embodiment of the present invention;
FIG. 4 is a schematic diagram showing the prediction effect of an improved method according to an embodiment of the present invention;
FIG. 5 is a schematic diagram showing the prediction effect of the prior art according to an embodiment of the present invention;
fig. 6 is a schematic diagram of a photovoltaic power prediction system according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, the present embodiment provides a photovoltaic power prediction method, which includes the following steps:
s1: acquiring photovoltaic power generation data;
s2: improving the long-term and short-term memory network based on a quantile regression method, and constructing a power prediction model;
s3: inputting the photovoltaic power generation data into a power prediction model so that the power prediction model predicts a plurality of predicted output values;
s4: and obtaining a photovoltaic power predicted value and a predicted interval based on a plurality of predicted output values, so as to realize the prediction of the photovoltaic power.
According to the embodiment, a traditional long-short-period memory network is improved by using a quantile regression method, a power prediction model for photovoltaic power prediction is obtained, photovoltaic power point prediction and interval probability prediction can be completed, the prediction output can reflect the probability range and fluctuation characteristics of power generation, the uncertainty of power generation can be quantized, the accuracy of a photovoltaic power prediction value and a prediction interval is ensured, and the final prediction accuracy is improved.
Further, in the modification of the long-term and short-term memory network based on the quantile regression method, a power prediction model is constructed, wherein the power prediction model corresponds to the quantile tau, and a loss function of the power prediction model corresponding to the quantile tau is specifically expressed as:
wherein: l (·) represents a loss function; w (τ) = (W) f (τ),W i (τ),W c (τ),W o (τ)) represents a weight matrix for a long and short term memory network at quantiles τ, where W f (. Cndot.) a weight matrix representing the forgetting gate, W i (. Cndot.) represents the weight matrix of the input gates, W c (. Cndot.) represents the weight matrix of the memory element, W o (. Cndot.) represents the weight matrix of the output gates; b (τ) = (b) f (τ),b i (τ),b c (τ),b o (τ)) represents the bias matrix of the long and short term memory network at quantiles τ, where b f (. Cndot.) the bias matrix representing the forgetting gate, b i (. Cndot.) represents the bias matrix of the input gate, b c (. Cndot.) represents the bias matrix of the memory element, b o (-) represents the bias matrix of the output gate; ρ τ Is a checking function; y is Y t A response variable representing the time t; x is X t The input variable at time t is indicated.
Further, in the modification of the long-term and short-term memory network based on the quantile regression method, a calculation formula of a parameter vector beta (tau) corresponding to the quantile tau in the quantile regression method is as follows:
wherein, the optimal value of the parameter vector beta (tau) corresponding to the quantile tau is determined by minimizing the function; x' is the transposed matrix of the input variables; y is a response variable.
Further, the obtaining photovoltaic power generation data includes: acquiring photovoltaic power generation data, dividing the photovoltaic power generation data according to time in proportion, and acquiring corresponding combined similar moment data under all time for being used as input of a power prediction model; the photovoltaic power generation data comprise photovoltaic power generation output power, historical meteorological data and numerical weather forecast NWP data corresponding to time and all times.
In this embodiment, since the photovoltaic power output is closely related to the weather factor, the photovoltaic power output on similar days has a great similarity. Therefore, the historical data (i.e., similar day data) with high correlation with the day data to be predicted can be selected for model training to improve the accuracy of photovoltaic power generation power prediction.
Further, the obtaining the corresponding combined similar time data at all times includes: and obtaining the similar moment j of the moment i and the combined similar moment data corresponding to the similar moment j by calculating the minimum Euclidean distance between the photovoltaic power generation data corresponding to the moment i and different moments.
According to the embodiment, the traditional similarity day selection algorithm is improved, the fluctuation characteristics of the photovoltaic data in reality can be reflected more effectively by acquiring the combined similarity time data, the similarity of each period is further guaranteed, the severe fluctuation scene is tracked, and therefore the prediction accuracy of the model is improved.
The embodiment utilizes combination of similar day and fractional number regression to improve the traditional long-short-period memory network, and further obtains a power prediction model for photovoltaic power prediction. The similarity of each period is guaranteed by combining the similarity days, namely the similarity of the data at similar moments is combined, so that the prediction accuracy is improved; and the quantile regression realizes the point output and probability output of photovoltaic power prediction, ensures the accuracy of a photovoltaic power predicted value and a predicted interval, and improves the final prediction accuracy.
In order to further explain the technical gist of the technical scheme, the technical advantages are highlighted, and referring to fig. 2, the embodiment provides a photovoltaic power prediction application framework based on combination of similar days and fractional number regression (QR), and the framework can effectively track severe fluctuation scenes, improve prediction precision, realize accurate point prediction and probability interval prediction, and has good reliability and comprehensive performance. Specifically:
the photovoltaic power generation data are acquired and imported firstly, and include photovoltaic power generation output power, historical meteorological data, numerical weather forecast NWP data and the like under the condition that the photovoltaic power generation output power, the historical meteorological data and the numerical weather forecast NWP data correspond to each other in time, such as solar irradiation intensity, temperature, humidity, air pressure, precipitation and the like of the photovoltaic power generation data.
And the obtained photovoltaic power generation data can be subjected to data preprocessing. The data preprocessing is divided into abnormal data processing and normalization processing. The abnormal data processing needs to remove abnormal data such as negative numbers and the like, and correct the abnormal data by utilizing the data average value near the abnormal data. Assuming that there is abnormal dataThe corrected data value is then:
wherein x is i-1 And x i+ 1 represents two normal data values before and after the time i. The calculation formula of the normalization process is as follows:
wherein x is i As a result of the normalization of the data,for the modified raw data, x max And x min Representing the maximum and minimum values, respectively, of a selected column in the dataset.
And secondly, dividing the photovoltaic power generation data into a training data set and a test set data according to the time proportion, and calculating and acquiring corresponding combined similar moment data under all the time as the input of a subsequent prediction model. In the training set data, the input variable of the model can be historical meteorological data and photovoltaic output at similar moments corresponding to the historical meteorological data at training moments, and the output variable is photovoltaic power at the training moments; in order to restore the prediction scenario, for the test set data, the model input variables are historical meteorological data and photovoltaic power at similar moments corresponding to the test moment NWP data. Obtaining a similar time j of the time i and combined similar time data corresponding to the similar time j by calculating the minimum Euclidean distance between the time i and the photovoltaic power generation data corresponding to different times, namely:
wherein, the liquid crystal display device comprises a liquid crystal display device,and->The x-th meteorological feature at time i and time j is represented, and k represents the total number of meteorological features.
The embodiment combines the combination similarity theory into the photovoltaic power prediction method, so as to further improve the prediction accuracy of the photovoltaic power prediction model. The weather data of each moment is selected to be corresponding similar moment data, the weather data of similar moments are combined to form weather data of similar days, and then the selected data can be used as model input for prediction. The prediction effect based on the improved method and the prediction effect based on the traditional similarity day can be seen in fig. 3 and 4. Obviously, compared with the traditional similarity day theory, the similarity of each period is further guaranteed by combining the similarity days, the severe fluctuation scene can be effectively tracked, and the prediction accuracy is improved.
Then, the long-term memory network can be improved based on a quantile regression method, namely, the loss function of the long-term memory network is improved, and a power prediction model is constructed.
In the quantile regression model, the parameter vector β (τ) corresponding to the quantile τ may be calculated as follows:
where X' is the transposed matrix of the input variables and Y is the response variable. And further an optimal value of the quantile τ is determined by determining the minimum value of the parameter vector β (τ). Therefore, after determining the quantile τ, quantile regression may be introduced into the long-short-term memory network, and within time t e [1, t ], the loss function of the model corresponding to the quantile τ is:
wherein: l (·) represents a loss function; w (τ) = (W) f (τ),W i (τ),W c (τ),W o (τ)) represents a weight matrix for a long and short term memory network at quantiles τ, where W f (. Cndot.) a weight matrix representing the forgetting gate, W i (. Cndot.) represents the weight matrix of the input gates, W c (. Cndot.) represents the weight matrix of the memory element, W o (. Cndot.) represents the weight matrix of the output gates; b (τ) = (b) f (τ),b i (τ),b c (τ),b o (τ)) represents the bias matrix of the long and short term memory network at quantiles τ, where b f (. Cndot.) means bias moment of forgetting doorArray, b i (. Cndot.) represents the bias matrix of the input gate, b c (. Cndot.) represents the bias matrix of the memory element, b o (-) represents the bias matrix of the output gate; ρ τ Is a checking function; y is Y t A response variable representing the time t; x is X t The input variable at time t is indicated. Wherein, the checking function can be specifically expressed as:
note that, when the quantile τ is 0.5, the calculation of the loss function is degenerated to a least-squares regression, and the prediction effect at this time is the same as that of the LSTM-based point prediction model.
And finally, training a plurality of long-short term memory network models under the condition of different quantiles, and outputting photovoltaic power prediction results under different quantiles. The output data is subjected to inverse normalization processing, so that the prediction model 1 with the quantile value τ1=0.5 finally generates a point prediction value of photovoltaic power, and the prediction model 2 with the quantile value τ2 epsilon (0, 0.5) and the prediction model 3 with the quantile value τ3 epsilon (0.5, 1) respectively generate upper and lower limits of a prediction interval.
It should be noted that, as a typical deep learning algorithm, the long-short term memory network LSTM can refer to fig. 5 for its unit structure, and on the basis of effectively acquiring the input-output time sequence relationship, the problems of gradient elimination and explosion in model training are effectively solved.
It should be noted that, in addition to the machine learning implemented by using the long-short-term memory network, other machine learning methods, such as a neural network, a cycle gate unit, and the like, may also be used in the present embodiment, and power prediction may also be implemented.
Referring to fig. 6, the present embodiment provides a photovoltaic power prediction system, configured to implement a photovoltaic power prediction method, which specifically includes:
the data acquisition module is used for acquiring photovoltaic power generation data;
the model construction module is used for improving the long-term and short-term memory network based on a quantile regression method and constructing a power prediction model;
the model execution module is used for inputting the photovoltaic power generation data into the power prediction model so that the power prediction model predicts a plurality of predicted output values;
and the photovoltaic power prediction module is used for acquiring a photovoltaic power predicted value and a predicted interval based on the plurality of predicted output values.
The system provided by the embodiment is simple in architecture, easy to build, easy to apply and popularize, and capable of improving a traditional long-period and short-period memory network by means of a quantile regression method, obtaining a power prediction model for photovoltaic power prediction and finishing photovoltaic power point prediction and interval probability prediction. The prediction output can reflect the probability range and fluctuation characteristics of power generation, and can also quantify the uncertainty of power generation, ensure the accuracy of the photovoltaic power prediction value and the prediction interval, and improve the final prediction accuracy.
Further, the model construction module is used for improving the long-term and short-term memory network based on a quantile regression method, and constructing a power prediction model, specifically: the power prediction model corresponds to the quantile tau, and in the time t epsilon [1, T ], the loss function of the power prediction model corresponding to the quantile tau is specifically expressed as:
wherein: l (·) represents a loss function; w (τ) = (W) f (τ),W i (τ),W c (τ),W o (τ)) represents a weight matrix for a long and short term memory network at quantiles τ, where W f (. Cndot.) a weight matrix representing the forgetting gate, W i (. Cndot.) represents the weight matrix of the input gates, W c (. Cndot.) represents the weight matrix of the memory element, W o (. Cndot.) represents the weight matrix of the output gates; b (τ) = (b) f (τ),b i (τ),b c (τ),b o (τ)) represents the bias matrix of the long and short term memory network at quantiles τ, where b f (. Cndot.) the bias matrix representing the forgetting gate, b i (. Cndot.) represents the bias matrix of the input gate, b c (. Cndot.) representationBias matrix of memory link, b o (-) represents the bias matrix of the output gate; ρ τ Is a checking function; y is Y t A response variable representing the time t; x is X t The input variable at time t is indicated.
Further, in the model building module, a calculation formula of a parameter vector β (τ) corresponding to the quantile τ in the quantile regression method is as follows:
wherein, the optimal value of the parameter vector beta (tau) corresponding to the quantile tau is determined by minimizing the function; x' is the transposed matrix of the input variables; y is a response variable.
Further, the method is used for acquiring photovoltaic power generation data, and specifically comprises the following steps: acquiring photovoltaic power generation data, dividing the photovoltaic power generation data according to time in proportion, and acquiring corresponding combined similar moment data under all time for being used as input of a power prediction model; the photovoltaic power generation data comprise photovoltaic power generation output power, historical meteorological data and numerical weather forecast NWP data corresponding to time and all times.
Further, in the data obtaining module, the obtaining the corresponding combined similar time data at all times includes: and obtaining the similar moment j of the moment i and the combined similar moment data corresponding to the similar moment j by calculating the minimum Euclidean distance between the photovoltaic power generation data corresponding to the moment i and different moments.
According to the embodiment, the traditional similarity day selection algorithm is improved, the fluctuation characteristics of the photovoltaic data in reality can be reflected more effectively by acquiring the combined similarity time data, the similarity of each period is further guaranteed, the severe fluctuation scene is tracked, and therefore the prediction accuracy of the model is improved.
The embodiment utilizes combination of similar day and fractional number regression to improve the traditional long-short-period memory network, and further obtains a power prediction model for photovoltaic power prediction. The similarity of each period is guaranteed by combining the similarity days, namely the similarity of the data at similar moments is combined, so that the prediction accuracy is improved; and the quantile regression realizes the point output and probability output of photovoltaic power prediction, ensures the accuracy of a photovoltaic power predicted value and a predicted interval, and improves the final prediction accuracy.
While the foregoing is directed to the preferred embodiments of the present invention, it will be appreciated by those skilled in the art that changes and modifications may be made without departing from the principles of the invention, such changes and modifications are also intended to be within the scope of the invention.

Claims (10)

1. A method of photovoltaic power prediction comprising the steps of:
acquiring photovoltaic power generation data;
improving the long-term and short-term memory network based on a quantile regression method, and constructing a power prediction model;
inputting the photovoltaic power generation data into a power prediction model so that the power prediction model predicts a plurality of predicted output values;
and obtaining a photovoltaic power predicted value and a predicted interval based on a plurality of predicted output values, so as to realize the prediction of the photovoltaic power.
2. The photovoltaic power prediction method according to claim 1, wherein in the modification of the long-short-term memory network based on the quantile regression method, a power prediction model is constructed, wherein the power prediction model corresponds to a quantile τ, and a power prediction model loss function corresponding to the quantile τ is specifically expressed as:
wherein: l (·) represents a loss function; w (τ) = (W) f (τ),W i (τ),W c (τ),W o (τ)) represents a weight matrix for a long and short term memory network at quantiles τ, where W f Weight matrix representing forgetting gate，W i (. Cndot.) represents the weight matrix of the input gates, W c (. Cndot.) represents the weight matrix of the memory element, W o (. Cndot.) represents the weight matrix of the output gates; b (τ) = (b) f (τ),b i (τ),b c (τ),b o (τ)) represents the bias matrix of the long and short term memory network at quantiles τ, where b f (. Cndot.) the bias matrix representing the forgetting gate, b i (. Cndot.) represents the bias matrix of the input gate, b c (. Cndot.) represents the bias matrix of the memory element, b o (-) represents the bias matrix of the output gate; ρ τ Is a checking function; y is Y t A response variable representing the time t; x is X t The input variable at time t is indicated.
3. The photovoltaic power prediction method according to claim 2, wherein in the modification of the long-short-term memory network based on the quantile regression method, a calculation formula of a parameter vector β (τ) corresponding to a quantile τ in the quantile regression method is as follows:
wherein, the optimal value of the parameter vector beta (tau) corresponding to the quantile tau is determined by minimizing the function; x' is the transposed matrix of the input variables; y is a response variable.
4. A photovoltaic power generation prediction method according to any one of claims 1 to 3, wherein the obtaining photovoltaic power generation data comprises:
acquiring photovoltaic power generation data, dividing the photovoltaic power generation data according to time in proportion, and acquiring corresponding combined similar moment data under all time for being used as input of a power prediction model; the photovoltaic power generation data comprise photovoltaic power generation output power, historical meteorological data and numerical weather forecast NWP data corresponding to time and all times.
5. The method for predicting photovoltaic power according to claim 4, wherein the obtaining the corresponding combined similar moment data at all times comprises:
and obtaining the similar moment j of the moment i and the combined similar moment data corresponding to the similar moment j by calculating the minimum Euclidean distance between the photovoltaic power generation data corresponding to the moment i and different moments.
6. A photovoltaic power generation system, comprising:
the data acquisition module is used for acquiring photovoltaic power generation data;
the model construction module is used for improving the long-term and short-term memory network based on a quantile regression method and constructing a power prediction model;
the model execution module is used for inputting the photovoltaic power generation data into the power prediction model so that the power prediction model predicts a plurality of predicted output values;
and the photovoltaic power prediction module is used for acquiring a photovoltaic power predicted value and a predicted interval based on the plurality of predicted output values.
7. The photovoltaic power prediction system according to claim 6, wherein the model building module is configured to improve the long-term and short-term memory network based on a quantile regression method, and build a power prediction model, specifically: the power prediction model corresponds to the quantile tau, and in the time t epsilon [1, T ], the loss function of the power prediction model corresponding to the quantile tau is specifically expressed as:
wherein: l (·) represents a loss function; w (τ) = (W) f (τ),W i (τ),W c (τ),W o (τ)) represents a weight matrix for a long and short term memory network at quantiles τ, where W f (. Cndot.) a weight matrix representing the forgetting gate, W i (. Cndot.) represents the weight matrix of the input gates, W c (. Cndot.) means memoryWeight matrix of links, W o (. Cndot.) represents the weight matrix of the output gates; b (τ) = (b) f (τ),b i (τ),b c (τ),b o (τ)) represents the bias matrix of the long and short term memory network at quantiles τ, where b f (. Cndot.) the bias matrix representing the forgetting gate, b i (. Cndot.) represents the bias matrix of the input gate, b c (. Cndot.) represents the bias matrix of the memory element, b o (-) represents the bias matrix of the output gate; ρ τ Is a checking function; y is Y t A response variable representing the time t; x is X t The input variable at time t is indicated.
8. The photovoltaic power prediction system according to claim 7, wherein in the model building module, a calculation formula of a parameter vector β (τ) corresponding to a quantile τ in the quantile regression method is:
wherein, the optimal value of the parameter vector beta (tau) corresponding to the quantile tau is determined by minimizing the function; x' is the transposed matrix of the input variables; y is a response variable.
9. The photovoltaic power prediction system according to any one of claims 6 to 8, wherein the data acquisition module is configured to acquire photovoltaic power generation data, specifically:
acquiring photovoltaic power generation data, dividing the photovoltaic power generation data according to time in proportion, and acquiring corresponding combined similar moment data under all time for being used as input of a power prediction model; the photovoltaic power generation data comprise photovoltaic power generation output power, historical meteorological data and numerical weather forecast NWP data corresponding to time and all times.
10. The photovoltaic power generation system of claim 9, wherein in the data acquisition module, the acquiring the corresponding combined similar time data at all times comprises:
and obtaining the similar moment j of the moment i and the combined similar moment data corresponding to the similar moment j by calculating the minimum Euclidean distance between the photovoltaic power generation data corresponding to the moment i and different moments.
CN202310635698.3A 2023-05-31 2023-05-31 Photovoltaic power prediction method and system Pending CN116663727A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310635698.3A CN116663727A (en) 2023-05-31 2023-05-31 Photovoltaic power prediction method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310635698.3A CN116663727A (en) 2023-05-31 2023-05-31 Photovoltaic power prediction method and system

Publications (1)

Publication Number Publication Date
CN116663727A true CN116663727A (en) 2023-08-29

Family

ID=87709239

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310635698.3A Pending CN116663727A (en) 2023-05-31 2023-05-31 Photovoltaic power prediction method and system

Country Status (1)

Country Link
CN (1) CN116663727A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116862079A (en) * 2023-09-04 2023-10-10 应辉环境科技服务(烟台)有限公司 Enterprise pollutant emission prediction method and prediction system

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116862079A (en) * 2023-09-04 2023-10-10 应辉环境科技服务(烟台)有限公司 Enterprise pollutant emission prediction method and prediction system
CN116862079B (en) * 2023-09-04 2023-12-05 应辉环境科技服务(烟台)有限公司 Enterprise pollutant emission prediction method and prediction system

Similar Documents

Publication Publication Date Title
Han et al. Mid-to-long term wind and photovoltaic power generation prediction based on copula function and long short term memory network
CN112906987B (en) Photovoltaic power prediction method based on convolutional neural network and two-dimensional meteorological matrix
CN106548270A (en) A kind of photovoltaic plant power anomalous data identification method and device
CN112381282B (en) Photovoltaic power generation power prediction method based on width learning system
CN111695736A (en) Photovoltaic power generation short-term power prediction method based on multi-model fusion
CN116663727A (en) Photovoltaic power prediction method and system
CN114330934A (en) Model parameter self-adaptive GRU new energy short-term power generation power prediction method
CN116014722A (en) Sub-solar photovoltaic power generation prediction method and system based on seasonal decomposition and convolution network
CN115358437A (en) Power supply load prediction method based on convolutional neural network
CN116629416A (en) Photovoltaic power station power prediction method and device
CN117117819A (en) Photovoltaic power generation short-term power prediction method, system, equipment and medium
CN115238854A (en) Short-term load prediction method based on TCN-LSTM-AM
CN111967660B (en) Ultra-short-term photovoltaic prediction residual error correction method based on SVR
CN110555566B (en) B-spline quantile regression-based photoelectric probability density prediction method
Zheng et al. Short-term photovoltaic power prediction based on daily feature matrix and deep neural network
CN117293809A (en) Multi-time space scale new energy generation power prediction method based on large model
CN116707331A (en) Inverter output voltage high-precision adjusting method and system based on model prediction
CN116581755A (en) Power prediction method, device, equipment and storage medium
CN116611702A (en) Integrated learning photovoltaic power generation prediction method for building integrated energy management
CN114819382B (en) LSTM-based photovoltaic power prediction method
CN101929684A (en) Coal consumption calculating method of composite firing low heat value gas unit
CN115563848A (en) Distributed photovoltaic total radiation prediction method and system based on deep learning
CN113962432A (en) Wind power prediction method and system integrating three-dimensional convolution and light-weight convolution threshold unit
Huang et al. Probabilistic prediction intervals of wind speed based on explainable neural network
Lian et al. Trend-guided small hydropower system power prediction based on extreme learning machine

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination