CN111461297B - Solar irradiation quantity optimization prediction algorithm based on MPC and ELM neural network - Google Patents

Solar irradiation quantity optimization prediction algorithm based on MPC and ELM neural network Download PDF

Info

Publication number
CN111461297B
CN111461297B CN202010217224.3A CN202010217224A CN111461297B CN 111461297 B CN111461297 B CN 111461297B CN 202010217224 A CN202010217224 A CN 202010217224A CN 111461297 B CN111461297 B CN 111461297B
Authority
CN
China
Prior art keywords
prediction
data
time
moment
future
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.)
Active
Application number
CN202010217224.3A
Other languages
Chinese (zh)
Other versions
CN111461297A (en
Inventor
郭苏
章晗
丁菲
王嘉乐
张睿
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.)
Hohai University HHU
Original Assignee
Hohai University HHU
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 Hohai University HHU filed Critical Hohai University HHU
Priority to CN202010217224.3A priority Critical patent/CN111461297B/en
Publication of CN111461297A publication Critical patent/CN111461297A/en
Application granted granted Critical
Publication of CN111461297B publication Critical patent/CN111461297B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G01W1/10—Devices for predicting weather conditions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • 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
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Environmental & Geological Engineering (AREA)
  • Atmospheric Sciences (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Ecology (AREA)
  • Environmental Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Testing Resistance To Weather, Investigating Materials By Mechanical Methods (AREA)

Abstract

The invention discloses a solar irradiance optimization prediction algorithm based on MPC and an ELM neural network, which comprises the following steps of 1, establishing an ELM neural network model based on MPC; step 2, collecting historical data; step 3, normalization processing; step 4, training to obtain an optimal ELM neural network model; step 5, setting rolling prediction parameters; step 6, MPC rolling prediction, including step 61, collecting historical data at the t-m moment; step 62, initial prediction; step 63, collecting data at time t; step 64, calculating a process prediction error; step 65, correcting the initial prediction data; step 66, rolling prediction. The method determines the input variable by selecting meteorological factors directly related to the solar energy irradiation quantity, such as ambient temperature, relative humidity and cloud cover; secondly, performing irradiation prediction by adopting an ELM neural network model; and finally, improving the precision of the solar irradiation amount prediction data by using the rolling optimization idea of the MPC.

Description

Solar irradiation quantity optimization prediction algorithm based on MPC and ELM neural network
Technical Field
The invention relates to a solar irradiation amount prediction algorithm, in particular to a solar irradiation amount optimization prediction algorithm based on an MPC (MPC) and ELM (empirical mode decomposition) neural network.
Background
With the gradual exhaustion of fossil energy and the increasing increase of ecological environment pollution, the exploration and development of renewable energy sources are reluctant to the world nowadays. Solar energy is one of the most major renewable energy sources worldwide, and photovoltaic power generation has become the major form of renewable energy power generation worldwide in recent years. Solar energy has intermittence, fluctuation and difficult predictability, and the instability of the photovoltaic power generation energy source causes the instability of the power generation power, thereby influencing the stability of the power grid power in China. Therefore, the large-scale photovoltaic power generation grid connection brings challenges to the dispatching management and the safe operation of the power grid system. An accurate solar irradiance information acquisition and prediction method is explored and found, and is the basis for accurately predicting photovoltaic power.
At present, a great deal of work is carried out on solar irradiation prediction at home and abroad, and common prediction methods can be classified into 2 types.
The first method comprises the following steps: based on detailed numerical weather forecast (NWP), the ultra-short-term solar irradiation amount is predicted by utilizing an observed numerical weather information and physical calculation model of irradiation amount, and the currently used prediction methods mainly comprise the following steps:
1. and (3) considering the matching degree of the load and the power supply time sequence, combining big data analysis, selecting an application market of the distributed photovoltaic power generation in the catering industry, and carrying out benefit analysis and sensitivity analysis.
2. The solar radiation dose is estimated by using a short-time prediction technology of albedo (albedo) and a satellite cloud picture.
3. The method comprises the steps of shooting a foundation cloud picture through a sky imager, fully acquiring cloud data, carrying out a series of image processing on the foundation cloud picture, and predicting solar irradiance after five minutes by matching with historical irradiance data.
4. The relation between the solar radiation intensity, the atmospheric temperature, the relative humidity, the wind condition, the weather type and the photovoltaic power generation power is visually analyzed through historical power generation data and meteorological data, and then correlation analysis is carried out on meteorological elements and photovoltaic output power by utilizing a correlation coefficient method.
5. And screening a data sequence with similar meteorological characteristics with the prediction time period from a large amount of data through a data mining technology, and predicting the photovoltaic power generation power by adopting a grey correlation degree theory.
However, although the first method has high prediction accuracy, it requires complicated satellite observation information and analysis methods, and is difficult to implement at present in China.
The second method comprises the following steps: through modeling historical data, a change rule of the irradiation amount is simulated, and then the future irradiation amount is predicted, and the specific prediction methods used mainly include the following methods:
1. artificial intelligence methods, such As Neural Networks (ANN), have been widely used in solar irradiance prediction. Although various environmental factors can be comprehensively considered, the method generally has the defects of low training speed, easy falling into local optimal solution and the like. For example, the BP neural network method has the defects that the optimal parameters of the network structure are difficult to determine, the convergence time is too long, and the method is easy to fall into the difficulty of local optimization.
2. Solar irradiance short-term prediction model in photovoltaic power generation based on Nash-Sut-Cliffe. The model firstly utilizes a Nash-Sut-Cliffe equation to establish a solar irradiance intensity model in photovoltaic power generation, determines the sunlight duration, radiation and minimum maximum temperature difference of the outdoor environment as main influence factors and conditions of a solar irradiance prediction model, and has low prediction precision.
3. LSSVM model: because the randomness of solar radiation is very large, the accuracy of a single Least Square Support Vector Machine (LSSVM) model established by the traditional method is not high.
4. ELM neural network: the prior irradiance prediction model has the defects that the input space dimension is too high, the structure is too complex, great difficulty is brought to learning and training, and the prediction accuracy is influenced by the loss of meteorological factors and other information related to irradiance. Yellow-to-wide bin and the like provide an Extreme Learning Machine (ELM) on the basis of a single-hidden layer forward neural Network (SLFNS) Learning algorithm, which is a Machine Learning system or method constructed based on a feed-Forward Neural Network (FNN), and is considered as a special FNN during research or an improvement on the FNN and a back propagation algorithm thereof. The method is characterized in that the weight of the hidden layer node is randomly or artificially given and does not need to be updated, and the learning process only calculates the output weight, so as to improve a Back Propagation (BP) algorithm to improve the learning efficiency and simplify the setting of learning parameters. ELM has been demonstrated to have extremely fast learning rates in the fields of regression analysis, classification, prediction, etc. Compared with the traditional neural network, the method has many excellent characteristics, such as high learning speed and good generalization capability, and can overcome the problems of local minimum, over-fitting, inappropriate selection of learning rate and the like of the traditional gradient algorithm. Later, some researches propose an improved ELM neural network algorithm, but all the researches lack a feedback correction link, so that model prediction errors are increased.
Predictive control, or Model Predictive Control (MPC), is one of the only advanced control methods that have been successfully applied in industrial control, and is a control algorithm based on a predictive process model that determines future inputs and outputs from historical information of the process. Predictive control is an algorithm for optimal control, and calculates future control actions based on a compensation function or a performance function. The optimization process of the predictive control is not completed off-line once, and is repeatedly performed on-line within a limited moving time interval. The moving time interval is called a limited time domain, which is the biggest difference from the traditional optimal control, which uses a performance function to judge the global optimization. For a complex system with dynamic characteristic change and uncertain factors, the optimal performance does not need to be judged in a global scope, so the rolling optimization method is very suitable for the complex system. Predictive control is also an algorithm for feedback control. If the model and the process are matched wrongly or the control performance problem is caused by uncertain factors of the system, the predictive control can compensate errors or correct model parameters according to online identification, and the method has strong anti-jamming capability and robustness. Meanwhile, the MPC can conveniently take various constraint conditions into account, has no specific requirements on the form of a prediction model, and is suitable for the solar irradiation resource prediction problem containing various factors such as the randomness of solar irradiation resources, the intermittence and the volatility of photovoltaic power generation power and the like.
Therefore, if the MPC is combined with the ELM neural network, the model prediction error of the ELM neural network can be greatly reduced.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide an optimized prediction algorithm of solar irradiance based on MPC and ELM neural networks, which first determines input variables by selecting meteorological factors directly related to solar irradiance, such as ambient temperature, relative humidity, cloud cover; secondly, performing irradiation prediction by adopting an ELM neural network model; and finally, improving the precision of the solar irradiation amount prediction data by using the rolling optimization idea of the MPC.
In order to solve the technical problems, the invention adopts the technical scheme that:
a solar irradiation quantity optimization prediction algorithm based on an MPC and an ELM neural network comprises the following steps:
step 1, establishing an ELM neural network model based on MPC: the built ELM neural network model comprises an input layer, a hidden layer and an output layer. The input layer comprises four input nodes of historical irradiation amount at t-m moment, ambient temperature at t-m moment, relative humidity at t-m moment and cloud cover at t-m moment, and the output layer is provided with n output nodes which are respectively: the future t moment irradiation dose, the future t + m moment irradiation dose, the future t +2m moment irradiation dose … … future t + m (n-1) moment irradiation dose. Wherein n is more than or equal to 1. m is the prediction interval duration.
Step 2, collecting historical data: and selecting the solar irradiation amount of the previous five years and the corresponding environmental factor data as historical data samples. The environmental factor data includes, among other things, ambient temperature, relative humidity, and cloud cover.
Step 3, normalization treatment: the collected historical data samples are normalized and all mapped to the (0,1) range.
Step 4, training: and (3) training the ELM neural network model established in the step (1) by using the normalized historical data sample in the step (3) to obtain an optimal connection weight and an optimal ELM neural network model with the optimal connection weight.
And 5, setting rolling prediction parameters: and determining a rolling prediction time period t to (t + N-1) according to the sunshine condition of a prediction place, wherein N is the number of prediction time points and is more than or equal to m (N-1). The prediction interval duration m is set according to the prediction purpose requirement.
Step 6, MPC rolling prediction, comprising the following steps:
step 61, collecting historical data at the t-m moment: and collecting historical irradiation amount at the t-m moment, ambient temperature at the t-m moment, relative humidity at the t-m moment and cloud cover at the t-m moment.
Step 62, initial prediction: inputting the collected t-m time historical data into an optimal ELM neural network model to obtain the future t time irradiation amount yr(t), future irradiation amount y at time t + mr(t + m) and the future irradiation amount y at t +2mr(t +2m), … …, future t + m (n-1) time exposure yr(t+m(n-1))。
Step 63, collecting data at time t: collecting t moment irradiation amount y*(t), ambient temperature at time t, relative humidity at time t and cloud cover at time t.
Step 64, calculating a process prediction error (error) (t), wherein the specific calculation formula is as follows:
error(t)＝y*(t)-yr(t)
step 65, correcting the initial prediction data: correcting the other initial prediction data except the irradiation amount at the future time t obtained in the step 62 to form corrected irradiation amount prediction data, wherein the specific correction result is as follows:
yrev(t+m)＝yr(t+m)+α·error(t)
yrev(t+2m)＝yr(t+2m)+α·error(t)
……
yrev(t+m(n-1))＝yr(t+m(n-1))+α·error(t)
wherein, yrev(t+m)、yrev(t+2m)、yrev(t + m (n-1)) correcting the irradiation dose at a time t + m, at a time t +2m and at a time t + m (n-1) in the future, respectively. And alpha is an error correction coefficient.
Step 66, rolling prediction: and (4) taking the data at the time t acquired in the step 63 as historical data, namely, taking t as t + m, and repeating the steps 62 to 65 to obtain corrected irradiation dose prediction data at n-1 future times. And the rest is repeated until the rolling prediction end time t is equal to t + N.
In step 3, the normalization processing formula is:
Figure BDA0002424786060000041
in the formula xi(t)、
Figure BDA0002424786060000042
Respectively representing the values, x, before and after the normalization of the data at time timin、ximaxRespectively representing the minimum and maximum values in the sample data.
Further comprising step 7, predictive assessment: the corrected irradiance prediction data formed in step 65 is evaluated using the mean absolute percent error MAPE and the root mean square error RMSE. The calculation formula of MAPE and RMSE is as follows:
Figure BDA0002424786060000043
Figure BDA0002424786060000044
in the formula, Y (t) and YmAnd (t) the measured values and the corrected irradiation dose prediction data are respectively, and N is the number of the prediction time points.
In step 2, the data of the previous 1 year in the historical data sample is used as a prediction set. And taking the data of the previous 2-5 years in the historical data sample as a training set and a testing set.
In step 5, the determined rolling prediction time period is 6:00-18:00, and the prediction interval duration m is 1h, that is, N is 13.
The invention has the following beneficial effects: the method comprises the steps of firstly predicting the solar irradiance of a time sequence through an ELM neural network, then continuously performing rolling prediction by using a predicted value of a latest time point by using the rolling optimization idea of MPC and assisting with feedback correction to realize the optimization processing of solar irradiance prediction data. The algorithm utilizes the advantages of the MPC and the ELM, improves the prediction precision and has important significance for accurately predicting the photovoltaic power generation capacity. Test results show that the algorithm can remarkably improve the prediction precision of the solar irradiance and provide a basis for the prediction of the photovoltaic power generation interval.
Drawings
FIG. 1 shows a flow chart of the solar irradiance optimization prediction algorithm based on MPC and ELM neural networks.
Figure 2 shows a comparison of the predicted results of the three models in this example at 3/11 of 2019.
Fig. 3 shows a comparison of the predicted results of the three models in this example at 3/20 of 2019.
Fig. 4 shows a comparison of the predicted results of the three models in this example at 31/3/2019.
FIG. 5 shows a comparison of the predicted results of the three models in this example at 8/4/2019.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific preferred embodiments.
In the description of the present invention, it is to be understood that the terms "left side", "right side", "upper part", "lower part", etc., indicate orientations or positional relationships based on those shown in the drawings, and are only for convenience of describing the present invention and simplifying the description, but do not indicate or imply that the referred device or element must have a specific orientation, be constructed in a specific orientation, and be operated, and that "first", "second", etc., do not represent an important degree of the component parts, and thus are not to be construed as limiting the present invention. The specific dimensions used in the present example are only for illustrating the technical solution and do not limit the scope of protection of the present invention.
As shown in FIG. 1, a solar irradiance optimization prediction algorithm based on MPC and ELM neural networks comprises the following steps.
Step 1, establishing an ELM neural network model based on MPC.
The built ELM neural network model comprises an input layer, a hidden layer and an output layer.
The input layer comprises four input nodes of historical irradiation at the t-m moment, ambient temperature at the t-m moment, relative humidity at the t-m moment and cloud cover at the t-m moment.
The output layer has n output nodes, which are respectively: the future t moment irradiation dose, the future t + m moment irradiation dose, the future t +2m moment irradiation dose … … future t + m (n-1) moment irradiation dose. Wherein n is more than or equal to 1, and m is the prediction interval duration.
Step 2, collecting historical data: and selecting the solar irradiation amount of the previous five years and the corresponding environmental factor data as historical data samples. The environmental factor data comprises meteorological factors such as environmental temperature, relative humidity and cloud cover which are directly related to the solar irradiation amount.
In addition, the data of the previous 1 year in the historical data sample is used as a prediction set. And taking the data of the previous 2-5 years in the historical data sample as a training set and a testing set.
In this example, the historical data samples are irradiation dose and environmental factor data from 2014 to 2019, and the irradiation dose samples are divided into a training set, a testing set and a prediction set. The data from 2014 to 2018 are training data and testing data, namely training sets and testing sets, and the data from 2019 are prediction data, namely prediction sets. And determining the optimal number of nodes of the hidden layer through a training set and a testing set, and determining the application of each node parameter to a prediction set.
Step 3, normalization treatment: the collected historical data samples are normalized and all mapped to the (0,1) range.
Different evaluation indexes often have different dimensions and dimension units, and the condition can influence the result of data analysis. And (3) carrying out normalization processing on the input attribute by using Matlab software, mapping the historical irradiation dose, the environmental temperature, the relative humidity and the cloud amount to a (0,1) range for processing, and enabling each index to be in the same order of magnitude, so that the operation is more convenient.
Normalization formula:
Figure BDA0002424786060000061
in the formula xi(t)、
Figure BDA0002424786060000062
Respectively representing the values, x, before and after the normalization of the data at time timin、ximaxRespectively representing the minimum and maximum values in the sample data.
Step 4, training: and (3) training the ELM neural network model established in the step (1) by using the normalized historical data sample in the step (3) to obtain an optimal connection weight and an optimal ELM neural network model with the optimal connection weight.
And 5, setting rolling prediction parameters: and determining a rolling prediction time period t to (t + N-1) according to the sunshine condition of a prediction place, wherein N is the number of prediction time points and is more than or equal to m (N-1). The prediction interval duration m is set according to the prediction purpose requirement.
In the present embodiment, the determined rolling prediction period is preferably 6:00 to 18:00, and the prediction interval duration m is 1h, that is, N is 13.
Step 6, MPC rolling prediction, comprising the following steps.
Step 61, collecting historical data at the t-m moment: and collecting historical irradiation amount at the t-m moment, ambient temperature at the t-m moment, relative humidity at the t-m moment and cloud cover at the t-m moment.
In this embodiment, it is preferable to collect historical exposure dose at the time of 5:00, ambient temperature at the time of 5:00, relative humidity at the time of 5:00, and cloud amount at the time of 5: 00.
Step 62, initial prediction: inputting the collected t-m time historical data into an optimal ELM neural network model to obtain the future t time irradiation amount yr(t), future irradiation amount y at time t + mr(t + m) and the future irradiation amount y at t +2mr(t +2m), … …, future t + m (n-1) time exposure yr(t+m(n-1))。
In this embodiment, n is preferably 4, and thus: inputting the collected historical data at the 5:00 moment into an optimal ELM neural network model to obtain the future radiation dose y at the 6:00 momentr(6:00) future irradiation dose y at 7:00 timer(7:00) future irradiation dose y at 8:00 timer(8:00) future 9:00 hours of irradiation dose yr(9:00)。
Step 63, collecting data at time t: collecting t moment irradiation amount y*(t) at tAmbient temperature, relative humidity at time t and cloud cover at time t.
In the embodiment, the irradiation amount y at the time of 6:00 is preferably collected*(6:00), ambient temperature at time 6:00, relative humidity at time 6:00, and cloud cover at time 6: 00.
Step 64, calculating a process prediction error (error) (t), wherein the specific calculation formula is as follows:
error(t)＝y*(t)-yr(t)
in this embodiment, taking 6:00 as an example, the following are:
error(6:00)＝y*(6:00)-yr(6:00)
step 65, correcting the initial prediction data: correcting the other initial prediction data except the irradiation amount at the future time t obtained in the step 62 to form corrected irradiation amount prediction data, wherein the specific correction result is as follows:
yrev(t+m)＝yr(t+m)+α·error(t)
yrev(t+2m)＝yr(t+2m)+α·error(t)
……
yrev(t+m(n-1))＝yr(t+m(n-1))+α·error(t)
wherein, yrev(t+m)、yrev(t+2m)、yrev(t + m (n-1)) correcting the irradiation dose at a time t + m, at a time t +2m and at a time t + m (n-1) in the future, respectively. α is an error correction factor for overcoming the undetectable interference, and is usually equal to 1.
In this example, the initial prediction data at time 7:00, 8:00, and 9:00 are corrected with a time 6:00 error (6:00) as follows:
yrev(7:00)＝yr(7:00)+α·error(6:00)
yrev(8:00)＝yr(8:00)+α·error(6:00)
yrev(9:00)＝yr(9:00)+α·error(6:00)
step 66, rolling prediction: and (4) taking the data at the time t acquired in the step 63 as historical data, namely, taking t as t + m, and repeating the steps 62 to 65 to obtain corrected irradiation dose prediction data at n-1 future times. And the rest is repeated until the rolling prediction end time t is equal to t + N.
In this example, the initial prediction data at time 8:00, 9:00, and 10:00 is then corrected by the time error (7:00), and so on, to achieve a time-by-time rolling prediction until 13 cycles to 18:00 are completed.
Step 7, prediction and evaluation: the corrected irradiance prediction data formed in step 65 is evaluated using the mean absolute percent error MAPE and the root mean square error RMSE. The calculation formula of MAPE and RMSE is as follows:
Figure BDA0002424786060000081
Figure BDA0002424786060000082
in the formula, Y (t) and YmAnd (t) the measured values and the corrected irradiation dose prediction data are respectively, and N is the number of the prediction time points.
In the application, 3 models are adopted to predict the solar irradiation quantities of different types respectively, so that the comparison shows that the model combining MPC and ELM has higher prediction precision, and the specific prediction effect analysis is as follows.
Model 1 is a conventional prediction model BP model.
Model 2 is the ELM model.
Model 3 is a model combining MPC and ELM of the present invention. And (3) taking historical irradiation data, ambient temperature, relative humidity and cloud cover as the input of the model, and predicting the solar irradiation amount of the predicted day time by time.
Taking Nanjing as an example, researching and analyzing the solar irradiation amount from 2014 to 2018, selecting examples of 11 days in 3 months in 2019, 20 days in 3 months in 2019, 31 days in 3 months in 2019 and 8 days in 4 months in 2019 for prediction, wherein the prediction time interval is 1 hour, and selecting 06: 00-18: 13 whole time points between 00 for the weather variable and the irradiation amount as the predicted time of day. The prediction results are shown in fig. 2 to 5.
And (4) selecting the average absolute percent error (MAPE) and the Root Mean Square Error (RMSE) to evaluate the corrected prediction result, wherein the error analysis time period is also 6:00-18: 00.
The error analysis results are shown in table 1:
Figure BDA0002424786060000083
the test results show that the solar irradiance optimization prediction algorithm based on the MPC and the ELM neural network can remarkably improve the solar irradiance prediction precision and provide a basis for photovoltaic power generation interval prediction.
Although the preferred embodiments of the present invention have been described in detail, the present invention is not limited to the details of the embodiments, and various equivalent modifications can be made within the technical spirit of the present invention, and the scope of the present invention is also within the scope of the present invention.

Claims (5)

1. A solar irradiation quantity optimization prediction algorithm based on MPC and ELM neural networks is characterized in that: the method comprises the following steps:
step 1, establishing an ELM neural network model based on MPC: the built ELM neural network model comprises an input layer, a hidden layer and an output layer; the input layer comprises four input nodes of historical irradiation amount at t-m moment, ambient temperature at t-m moment, relative humidity at t-m moment and cloud cover at t-m moment, and the output layer is provided with n output nodes which are respectively: the future t moment irradiation dose, the future t + m moment irradiation dose, the future t +2m moment irradiation dose … … future t + m (n-1) moment irradiation dose; wherein n is more than or equal to 1; m is the prediction interval duration;
step 2, collecting historical data: selecting the solar irradiation amount of the previous five years and corresponding environmental factor data as historical data samples; wherein the environmental factor data comprises ambient temperature, relative humidity and cloud cover;
step 3, normalization treatment: normalizing the collected historical data samples, and mapping the samples to a (0,1) range;
step 4, training: training the ELM neural network model established in the step 1 by using the normalized historical data sample in the step 3 to obtain an optimal connection weight and an optimal ELM neural network model with the optimal connection weight;
and 5, setting rolling prediction parameters: determining a rolling prediction time period t to (t + N-1) according to the sunshine condition of a prediction place, wherein N is the number of prediction time points and is more than or equal to m (N-1); setting a prediction interval duration m according to the prediction purpose requirement;
step 6, MPC rolling prediction, comprising the following steps:
step 61, collecting historical data at the t-m moment: collecting historical irradiation amount at the t-m moment, ambient temperature at the t-m moment, relative humidity at the t-m moment and cloud cover at the t-m moment;
step 62, initial prediction: inputting the collected t-m time historical data into an optimal ELM neural network model to obtain the future t time irradiation amount yr(t), future irradiation amount y at time t + mr(t + m) and the future irradiation amount y at t +2mr(t +2m), … …, future t + m (n-1) time exposure yr(t+m(n-1))；
Step 63, collecting data at time t: collecting t moment irradiation amount y*(t), ambient temperature at time t, relative humidity at time t and cloud cover at time t;
step 64, calculating a process prediction error (error) (t), wherein the specific calculation formula is as follows:
error(t)＝y*(t)-yr(t)
step 65, correcting the initial prediction data: correcting the other initial prediction data except the irradiation amount at the future time t obtained in the step 62 to form corrected irradiation amount prediction data, wherein the specific correction result is as follows:
yrev(t+m)＝yr(t+m)+α·error(t)
yrev(t+2m)＝yr(t+2m)+α·error(t)
yrev(t+m(n-1))＝yr(t+m(n-1))+α·error(t)
wherein, yrev(t+m)、yrev(t+2m)、yrev(t + m (n-1)) is the corrected irradiation dose at the future time t + m, the corrected irradiation dose at the future time t +2m and the corrected irradiation dose at the future time t + m (n-1) respectively; alpha is an error correction coefficient;
step 66, rolling prediction: taking the data at the time t acquired in the step 63 as historical data, namely, enabling t to be t + m, and repeating the steps 62 to 65 to obtain corrected irradiation dose prediction data at n-1 future times; and the rest is repeated until the rolling prediction end time t is equal to t + N.
2. The solar irradiance optimizing prediction algorithm based on the MPC and ELM neural networks as claimed in claim 1, wherein: in step 3, the normalization processing formula is:
Figure FDA0002424786050000021
in the formula xi(t)、
Figure FDA0002424786050000022
Respectively representing the values, x, before and after the normalization of the data at time timin、ximaxRespectively representing the minimum and maximum values in the sample data.
3. The solar irradiance optimizing prediction algorithm based on the MPC and ELM neural networks as claimed in claim 1, wherein: further comprising step 7, predictive assessment: estimating the corrected irradiation dose prediction data formed in the step 65 by using the average absolute percentage error MAPE and the root mean square error RMSE; the calculation formula of MAPE and RMSE is as follows:
Figure FDA0002424786050000023
Figure FDA0002424786050000024
in the formula, Y (t) and YmAnd (t) the measured values and the corrected irradiation dose prediction data are respectively, and N is the number of the prediction time points.
4. The solar irradiance optimizing prediction algorithm based on the MPC and ELM neural networks as claimed in claim 1, wherein: in step 2, taking the data of the previous 1 year in the historical data sample as a prediction set; and taking the data of the previous 2-5 years in the historical data sample as a training set and a testing set.
5. The solar irradiance optimizing prediction algorithm based on the MPC and ELM neural networks as claimed in claim 1, wherein: in step 5, the determined rolling prediction time period is 6:00-18:00, and the prediction interval duration m is 1h, that is, N is 13.
CN202010217224.3A 2020-03-25 2020-03-25 Solar irradiation quantity optimization prediction algorithm based on MPC and ELM neural network Active CN111461297B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010217224.3A CN111461297B (en) 2020-03-25 2020-03-25 Solar irradiation quantity optimization prediction algorithm based on MPC and ELM neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010217224.3A CN111461297B (en) 2020-03-25 2020-03-25 Solar irradiation quantity optimization prediction algorithm based on MPC and ELM neural network

Publications (2)

Publication Number Publication Date
CN111461297A CN111461297A (en) 2020-07-28
CN111461297B true CN111461297B (en) 2022-04-29

Family

ID=71682267

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010217224.3A Active CN111461297B (en) 2020-03-25 2020-03-25 Solar irradiation quantity optimization prediction algorithm based on MPC and ELM neural network

Country Status (1)

Country Link
CN (1) CN111461297B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112215478B (en) * 2020-09-27 2024-05-28 珠海博威电气股份有限公司 Power coordination control method and device for optical storage station and storage medium
CN113780640A (en) * 2021-08-27 2021-12-10 河北工业大学 TCN-Attention-based solar radiation prediction method

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108388956A (en) * 2018-01-18 2018-08-10 华北电力大学 Consider the photovoltaic power prediction technique of attenuation
CN109902874A (en) * 2019-02-28 2019-06-18 武汉大学 A kind of micro-capacitance sensor photovoltaic power generation short term prediction method based on deep learning

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108388956A (en) * 2018-01-18 2018-08-10 华北电力大学 Consider the photovoltaic power prediction technique of attenuation
CN109902874A (en) * 2019-02-28 2019-06-18 武汉大学 A kind of micro-capacitance sensor photovoltaic power generation short term prediction method based on deep learning

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
基于小波与最小资源分配网络的超短期风电功率预测研究;杨杰等;《电力系统保护与控制》;20180501;第46卷(第09期);全文 *
基于改进粒子群算法的光伏最大功率跟踪研究;李荣等;《电源技术》;20191130;第43卷(第11期);全文 *

Also Published As

Publication number Publication date
CN111461297A (en) 2020-07-28

Similar Documents

Publication Publication Date Title
CN111008728B (en) Prediction method for short-term output of distributed photovoltaic power generation system
CN109117951B (en) BP neural network-based probability load flow online calculation method
CN109002915B (en) Photovoltaic power station short-term power prediction method based on Kmeans-GRA-Elman model
CN113205207A (en) XGboost algorithm-based short-term power consumption load fluctuation prediction method and system
Kou et al. Photovoltaic power forecasting based on artificial neural network and meteorological data
Jebli et al. Deep learning based models for solar energy prediction
CN108053061A (en) A kind of solar energy irradiation level Forecasting Methodology based on improvement convolutional neural networks
CN110738344A (en) Distributed reactive power optimization method and device for load prediction of power system
CN113496311A (en) Photovoltaic power station generated power prediction method and system
CN110334870B (en) Photovoltaic power station short-term power prediction method based on gated cyclic unit network
CN110601250B (en) Method for evaluating stability of static voltage of access point based on new energy power station
Zhang et al. Wavelet-based neural network with genetic algorithm optimization for generation prediction of PV plants
CN111461297B (en) Solar irradiation quantity optimization prediction algorithm based on MPC and ELM neural network
CN110866633B (en) Micro-grid ultra-short-term load prediction method based on SVR support vector regression
CN114091317A (en) Photovoltaic power station power prediction method based on NWP irradiance correction and error prediction
CN111509768A (en) Scene simulation method and system of comprehensive energy system
Aliberti et al. Forecasting Short-term Solar Radiation for Photovoltaic Energy Predictions.
CN116029419A (en) Deep learning-based long-term new energy daily average generation power prediction method and system
CN112183877A (en) Photovoltaic power station fault intelligent diagnosis method based on transfer learning
CN116167508B (en) Short-term photovoltaic output rapid prediction method and system based on meteorological factor decomposition
CN112541633A (en) Ultrahigh-precision photovoltaic power generation power prediction method and system based on neural network
CN116632823A (en) Short-term photovoltaic power prediction method based on power conversion model and multi-layer perceptron
CN110796292A (en) Photovoltaic power short-term prediction method considering haze influence
CN116485004A (en) Wind power prediction method based on power heat map and convolutional neural network
Wang et al. Wind power day-ahead prediction based on LSSVM with fruit fly optimization algorithm

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
GR01 Patent grant
GR01 Patent grant