Ultra-short-term prediction method and system for power generation of massive distributed photovoltaic power stations

Abstract

A method for ultra-short-term prediction of power generation of massive distributed photovoltaic power stations, comprising: S1. pre-processed data, S2. training model and S3. predicted power. A system for ultra-short-term prediction of power generation of massive distributed photovoltaic power stations, a meteorological data publishing platform and distributed photovoltaic sites transmit real-time meteorological and power data to a data pre-processing module, which is stored in a data storage module after pre-processing. The data storage module transmits the data required for training to a model training module, and a CNN-LSTM model group is obtained after training and is passed to a power prediction module. The power prediction module receives the prediction request transmitted by the distributed photovoltaic site for processing, and finally sends the prediction response to the designated photovoltaic site. The technical method provided by the present invention does not require high-resolution NWP and real-time irradiation information, but only based on the historical power data of each distributed photovoltaic site and low-cost meteorological platform data, thus realizing accurate ultra-short-term power prediction of a single distributed photovoltaic site.

Description

Technical Field

[0001] The present invention belongs to the technical field of renewable energy power generation, and in particular relates to an ultra-short-term prediction system for power generation of massive distributed photovoltaic power stations based on historical photovoltaic power data. Background Art

[0002] In recent years, distributed photovoltaic power generation has received attention from national policies and has developed rapidly due to its characteristics of adapting to local conditions, flexible placement, on-site utilization, and wide distribution. As of the end of 2021, its cumulative installed capacity has reached 306 million KW, a year-on-year increase of 26.7%. However, photovoltaic power generation output is affected by meteorological factors such as solar radiation intensity, atmospheric temperature, atmospheric humidity, wind speed and direction, and has the disadvantages of randomness and volatility. Especially when its penetration rate is too high, it will seriously affect the safe and stable operation of the power grid. At the same time, distributed photovoltaics are in the medium and low voltage distribution network with a low level of "four remote controls", and have been in the "blind spot" of new energy regulation for a long time, which has brought huge challenges to the operation of the power grid. Therefore, in order to ensure the safe and reliable operation of the distribution network, it is necessary to regulate the distributed photovoltaics in the distribution network, and the premise for achieving effective regulation is to first realize the ultra-short-term prediction of distributed photovoltaic output.

[0003] At present, there are various methods for ultra-short-term prediction of distributed photovoltaic power station power, but the prediction accuracy is still not high. Most prediction methods are to grid distributed photovoltaic sites, and need to obtain high-cost high-resolution numerical weather forecasts (NWP) and measured irradiance, which cannot meet the prediction requirements of massive distributed photovoltaic power stations. Therefore, there is a need for an ultra-short-term power prediction system that can accurately predict the power generation of a single site of massive distributed photovoltaic power stations and does not require high meteorological data costs. Summary of the invention

[0004] The present invention aims to overcome the above-mentioned shortcomings of the prior art and provide a system and method for ultra-short-term prediction of power generation of massive distributed photovoltaic power stations based on historical photovoltaic power data, so as to perform ultra-short-term power prediction for a single site of massive distributed photovoltaic power stations in the absence of high-resolution NWP.

[0005] The solution adopted by the present invention to solve the technical problem is:

[0006] A method for ultra-short-term prediction of power generation of massive distributed photovoltaic power stations, characterized in that it comprises the following steps: S1. preprocessing data, S2. training model and S3. predicting power, the three steps are executed in sequence when the system is started and are executed synchronously during the stable operation of the system;

[0007] S1. Preprocess data;

[0008] Step S1-1: According to the real-time power data of each distributed photovoltaic site uploaded synchronously by each distributed photovoltaic site, the longitude and latitude of the i-th photovoltaic site is recorded as (x i ,y i ), data is uploaded once every Δt, and the power of the kth upload is P ik , temperature is T ik ;, wind speed is WS ik ;

[0009] Step S1-2: According to the real-time meteorological data released by the meteorological data release platform, the i-th photovoltaic site is taken as the origin, the east direction and the north direction are recorded as the positive directions of the east-west direction and the north-south direction respectively, and the wind direction is decomposed into the east-west direction and the north-south direction respectively. If the k-th wind direction is the same as the positive direction, it is recorded as 1, otherwise it is -1, and no wind direction is recorded as 0. The east-west wind direction and the north-south wind direction at this time are recorded as (WX ik , WY ik );

[0010] Step S1-3: Record the pre-processed data of the kth uploaded data as D k , and transmit it to the data storage module. The specific expression is as follows:

[0011]

[0012] Where: D ik is the data information of the i-th distributed photovoltaic site at the k-th moment;

[0013] S2. Training model;

[0014] Step S2-1: Determine whether the model needs to be updated, if yes, proceed to step S2-2, if no, continue to wait;

[0015] Step S2-2: Obtain historical data D from the data storage module = {D m-c ,D m-c+1 ...D k ...D m}, m is the total number of current time points, c is the number of data sets used to train the model, and n training models need to be trained for n photovoltaic sites;

[0016] Step S2-3: Train the i-th photovoltaic site model according to its longitude and latitude (x i ,y i ) Update the geographical features and wind direction and speed information of the historical data D to obtain the updated data set D' = {D' m-c, D' m-c+1 ...D' k ...D' m}, the specific expression is as follows:

[0017]

[0018] Where: D' ik represents the updated data information of the i-th distributed photovoltaic site at the k-th time, D' k Indicates the updated data information of all photovoltaic sites at the kth moment;

[0019] Step S2-4: Construct a training input data set In and a training output data set Out, input them into the training model for training to obtain the model M i ,in:

[0020]

[0021] Where: P ik＝ [P ik P i(k+1) ... P i(k+Δk) ] is the predicted power sequence, Δk is the number of time points in the prediction interval, P ik is the k-th power of the i-th PV station;

[0022] Step S2-5: Repeat steps S2-2 to S2-4 for n times to obtain n models, and the model training is now complete;

[0023] S3. Prediction power;

[0024] Step S3-1: monitor the prediction request of the distributed photovoltaic site, if a request is received, proceed to step S3-2, if not, continue to monitor;

[0025] Step S3-2: Record the sequence number of the photovoltaic site to be predicted as j, perform power prediction for the jth photovoltaic site, and obtain the data D' at the most recent moment m , the specific expression is as follows:

[0026]

[0027] Where: D' m Represents the updated data information of the jth photovoltaic station at the most recent moment of all photovoltaic stations;

[0028] Step S3-3: D' m Input model M j , and obtain the predicted power sequence P of the jth station j , where P j Contains the photovoltaic power forecast value of the jth site in the future from Δt to ΔkΔt;

[0029] Step S3-4: The predicted sequence P jThe prediction request is sent to the j-th photovoltaic site through the communication network; the processing of this prediction request is completed and the process returns to step S3-1.

[0030] Preferably, a system for implementing the above-mentioned method for ultra-short-term prediction of power generation of a massive distributed photovoltaic power station comprises a meteorological data publishing platform, distributed photovoltaic sites, a data preprocessing module, a data storage module, a model training module, and a power prediction module;

[0031] The data preprocessing module receives real-time meteorological data from the meteorological data publishing platform and real-time power data from distributed photovoltaic sites, and transmits the processed real-time comprehensive information matrix to the data storage module;

[0032] The data storage module receives the real-time comprehensive information matrix from the data preprocessing module and stores it in the data storage device, receives the prediction data request from the power prediction module, returns the comprehensive information matrix from time t-cΔt to time t to the power prediction module, and transmits the data required for training to the model training module;

[0033] The model training module receives the data required for training transmitted by the data storage module, and transmits the trained CNN-LSTM model group to the power prediction module. The model used in the training is the CNN-LSTM (Convolution al Neural Networks-Long Short Term Memory) model;

[0034] The power prediction module includes a request analysis module, a power calculation module and a CNN-LSTM model group; the request analysis module receives the prediction request of the distributed photovoltaic site, transmits the data request required for the prediction to the data storage module, and transmits the photovoltaic site serial number to the CNN-LSTM model group; the CNN-LSTM model group receives the photovoltaic site serial number transmitted by the request analysis module, and transmits the model required for the prediction to the power calculation module; the power calculation module receives the comprehensive information matrix from time t-cΔt to time t transmitted by the data storage module and the model required for the prediction transmitted by CNN-LSTM, and sends the calculated power prediction sequence as a prediction response to the designated distributed photovoltaic site.

[0035] The advantages of the present invention are: without the need for high-resolution NWP and real-time irradiation information, accurate ultra-short-term power prediction for a single distributed photovoltaic site can be achieved based only on the historical power data of each distributed photovoltaic site and low-cost meteorological platform data. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is the structure diagram of the ultra-short-term power prediction system of distributed photovoltaic power station.

[0037] Figure 2It is a flow chart of the model training and update module of the ultra-short-term power prediction system of distributed photovoltaic power stations.

[0038] Figure 3 It is a flow chart of the power prediction module of the ultra-short-term power prediction system of distributed photovoltaic power stations.

[0039] Figure 4 This is the photovoltaic power prediction effect diagram for the second site. DETAILED DESCRIPTION

[0040] The present invention is further described below in conjunction with the embodiments and the accompanying drawings in accordance with the steps of the invention content.

[0041] Taking five distributed photovoltaic sites in Jiangsu Province as an example, after the system starts running, the five sites upload real-time power data to the power prediction platform at a sampling interval of 15 minutes. At the same time, the power prediction platform receives meteorological data from the meteorological data publishing platform. The meteorological data includes wind direction, wind speed and temperature information. Taking the first upload of the first photovoltaic site as an example, the pre-processed data is expressed as follows:

[0042]

[0043] Subsequently, D1 is stored in the data storage device, and data is received, preprocessed and stored every 15 minutes.

[0044] Then, the model is trained and updated, and the frequency of model update is set to once a day. When model training or updating is required, the model training and updating process is entered. Taking the update of model M2 of the second photovoltaic site as an example, 100 sets of data D = {D 101, D 102 ...D k ...D 200}, first update the data to get D'={D' 101, D' 102 ...D' k ...D' 200}, the specific expression is as follows:

[0045]

[0046] Where: D' k Indicates the updated data information of all photovoltaic sites at the kth moment;

[0047] Construct a training input data set In and a training output data set Out, input them into a deep learning model based on a convolutional neural network for training to obtain a model M2, where:

[0048]

[0049] Where: P 2k＝ [P 2k P 2(k+1) ... P 2(k+16) ] is the predicted power sequence, Δk is the number of time points in the prediction interval, P 2k is the k-th power of the second PV station.

[0050] At this point, the training and updating process of the second model is completed.

[0051] When the power prediction module receives the prediction request of the second photovoltaic site, it obtains the most recent data D'5 from the data storage, and the expression is as follows:

[0052]

[0053] Where: D'5 represents the updated data information for the second photovoltaic station at the most recent moment, i.e., the fifth time data is uploaded by all photovoltaic stations;

[0054] Input D'5 into the deep learning model M2 based on convolutional neural network to obtain the predicted power sequence P2 of the second site, where P2 contains the predicted photovoltaic power value of the second site in the next 15 minutes to 4 hours. Draw the photovoltaic power prediction effect diagram of the second site as shown in Figure 4 shown.

[0055] Finally, the prediction sequence P2 is sent to the second PV station, completing the processing of this prediction request. The power prediction module then continues to monitor and waits for the next incoming prediction request.

[0056] The solution provided by the present invention does not require high-resolution NWP and real-time irradiation information, but only based on the historical power data of each distributed photovoltaic site and low-cost meteorological platform data, thus achieving accurate ultra-short-term power prediction for a single distributed photovoltaic site.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application rather than to limit its protection scope. Although the present application has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that after reading the present application, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the protection scope of the claims to be approved.

Claims

1. A method for ultra-short-term prediction of power generation of massive distributed photovoltaic power stations, characterized in that: The method comprises the following steps: S1. preprocessing data, S2. training model and S3. predicting power, the three steps are executed sequentially when the system is started and are executed synchronously during the stable operation of the system; S1. Preprocess data; Step S1-1: According to the real-time power data of each distributed photovoltaic site uploaded synchronously by each distributed photovoltaic site, the longitude and latitude of the i-th photovoltaic site is recorded as (x i ,y i ), data is uploaded once every Δt, and the power of the kth upload is P ik , temperature is T ik ; Wind speed is WS ik ; Step S1-2: According to the real-time meteorological data released by the meteorological data release platform, the i-th photovoltaic site is taken as the origin, the east direction and the north direction are recorded as the positive directions of the east-west direction and the north-south direction respectively, and the wind direction is decomposed into the east-west direction and the north-south direction respectively. If the k-th wind direction is the same as the positive direction, it is recorded as 1, otherwise it is -1, and no wind direction is recorded as 0. The east-west wind direction and the north-south wind direction at this time are recorded as (WX ik , WY ik ); Step S1-3: Record the pre-processed data of the kth uploaded data as D k , and transfer it to the data storage module. The specific expression is as follows: Where: D ik is the data information of the i-th distributed photovoltaic site at the k-th moment; S2. Training model; Step S2-1: Determine whether the model needs to be updated, if yes, proceed to step S2-2, if no, continue to wait; Step S2-2: Obtain historical data D from the data storage module = {D m-c, D m-c+1 ...D k ...D m }, m is the total number of current time points, c is the number of data sets used to train the model, and n training models need to be trained for n photovoltaic sites; Step S2-3: Train the i-th photovoltaic site model according to its longitude and latitude (x i ,y i ) Update the geographical features and wind direction and speed information of the historical data D to obtain the updated data set D' = {D' m-c, D' m-c+1 ...D' k ...D' m }, the specific expression is as follows: Where: D' ik represents the updated data information of the i-th distributed photovoltaic site at the k-th time, D' k Indicates the updated data information of all photovoltaic sites at the kth moment; Step S2-4: Construct a training input data set In and a training output data set Out, input them into the training model for training to obtain the model M i ,in: Where: P ik =[P ik P i(k+1) ... P i(k+Δk) ] is the predicted power sequence, Δk is the number of time points in the prediction interval, P ik is the k-th power of the i-th PV station; Step S2-5: Repeat steps S2-2 to S2-4 for n times to obtain n models, and the model training is now complete; S3. Prediction power; Step S3-1: monitor the prediction request of the distributed photovoltaic site, if a request is received, proceed to step S3-2, if not, continue to monitor; Step S3-2: Record the sequence number of the photovoltaic site to be predicted as j, perform power prediction for the jth photovoltaic site, and obtain the data D' at the most recent moment m , the specific expression is as follows: Where: D' m Represents the updated data information of the jth photovoltaic station at the most recent moment of all photovoltaic stations; Step S3-3: D' m Input model M j , and obtain the predicted power sequence P of the jth station j , where P j Contains the photovoltaic power forecast value of the jth site in the future from Δt to ΔkΔt; Step S3-4: The predicted sequence P j The prediction request is sent to the j-th photovoltaic site through the communication network; the processing of this prediction request is completed and the process returns to step S3-1.

2. A system for implementing the method for ultra-short-term prediction of power generation of a massive distributed photovoltaic power station as claimed in claim 1, characterized in that: It includes meteorological data publishing platform, distributed photovoltaic sites, data preprocessing module, data storage module, model training module and power prediction module; The data preprocessing module receives real-time meteorological data from the meteorological data publishing platform and real-time power data from distributed photovoltaic sites, and transmits the processed real-time comprehensive information matrix to the data storage module; The data storage module receives the real-time comprehensive information matrix from the data preprocessing module and stores it in the data storage device, receives the prediction data request from the power prediction module, returns the comprehensive information matrix from time t-cΔt to time t to the power prediction module, and transmits the data required for training to the model training module; The model training module receives the data required for training transmitted by the data storage module, and transmits the trained CNN-LSTM model group to the power prediction module. The model used in the training is the CNN-LSTM (Convolution al Neural Networks-Long Short Term Memory) model; The power prediction module includes a request analysis module, a power calculation module and a CNN-LSTM model group; the request analysis module receives the prediction request of the distributed photovoltaic site, transmits the data request required for the prediction to the data storage module, and transmits the photovoltaic site serial number to the CNN-LSTM model group; the CNN-LSTM model group receives the photovoltaic site serial number transmitted by the request analysis module, and transmits the model required for the prediction to the power calculation module; the power calculation module receives the comprehensive information matrix from time t-cΔt to time t transmitted by the data storage module and the model required for the prediction transmitted by CNN-LSTM, and sends the calculated power prediction sequence as a prediction response to the designated distributed photovoltaic site.

Images