CN107256437B - Photovoltaic power station ultra-short-term irradiance prediction method and system - Google Patents

Photovoltaic power station ultra-short-term irradiance prediction method and system Download PDF

Info

Publication number
CN107256437B
CN107256437B CN201710340400.0A CN201710340400A CN107256437B CN 107256437 B CN107256437 B CN 107256437B CN 201710340400 A CN201710340400 A CN 201710340400A CN 107256437 B CN107256437 B CN 107256437B
Authority
CN
China
Prior art keywords
irradiance
data
predicted
time
variance
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
CN201710340400.0A
Other languages
Chinese (zh)
Other versions
CN107256437A (en
Inventor
朱长胜
蒿峰
文志刚
郭琦
郭抒翔
云峰
海威
贺旭伟
牛新
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.)
BEIJING ZHONGKE FURUI ELECTRIC TECHNOLOGY CO LTD
Inner Mongolia Power Group Co ltd
Original Assignee
BEIJING ZHONGKE FURUI ELECTRIC TECHNOLOGY CO LTD
Inner Mongolia Power Group 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 BEIJING ZHONGKE FURUI ELECTRIC TECHNOLOGY CO LTD, Inner Mongolia Power Group Co ltd filed Critical BEIJING ZHONGKE FURUI ELECTRIC TECHNOLOGY CO LTD
Priority to CN201710340400.0A priority Critical patent/CN107256437B/en
Publication of CN107256437A publication Critical patent/CN107256437A/en
Application granted granted Critical
Publication of CN107256437B publication Critical patent/CN107256437B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Marketing (AREA)
  • General Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Public Health (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Water Supply & Treatment (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Photovoltaic Devices (AREA)

Abstract

The invention relates to a method and a system for predicting ultra-short-term irradiance of a photovoltaic power station, wherein the method comprises the following steps: obtaining predicted irradiance according to the trained support vector regression model; calculating the similarity, and sequencing the historical numerical weather forecast data according to the similarity; calculating the variance between the predicted irradiance and the actual irradiance of the historical numerical weather forecast, and meanwhile, carrying out weighted accumulation on the variance; calculating the weight occupied by the numerical weather forecast predicted irradiance and the weight occupied by the predicted irradiance of the support vector regression model; calculating to obtain a predicted time tpThe invention also relates to a prediction system for predicting irradiance of a photovoltaic power station in an ultra-short term, which comprises: the system comprises a training support vector regression model module, a similarity calculation module, a variance calculation module, a weight calculation module and an ultra-short-term irradiance prediction module. The method can obviously improve the prediction precision in the prediction, simultaneously the calculation efficiency and the performance meet the prediction requirements, and the requirement of the real-time scheduling of the photovoltaic power generation is completely met.

Description

Photovoltaic power station ultra-short-term irradiance prediction method and system
Technical Field
The invention belongs to the field of photovoltaic prediction, and particularly relates to a method and a system for predicting ultra-short-term irradiance of a photovoltaic power station.
Background
Solar photovoltaic power generation, the most practical technology in solar energy utilization, has become a hot spot for competitive research and application in countries around the world. However, the inherent characteristics of high dependence on weather conditions, high randomness and volatility and difficult prediction of photovoltaic power generation limit the large-scale application of photovoltaic power generation.
The output power of photovoltaic power generation depends on the amount of solar radiation received by the photovoltaic panel to a large extent, and is easily influenced by weather factors. The photovoltaic panel inclined plane irradiance measured by an environmental monitor installed in a photovoltaic power station can not consider the fluctuation and the randomness of the irradiance, the prediction precision is low, and the prediction effect is worse when the weather condition changes violently or the prediction time scale is longer. Particularly, in the prior art, when irradiance in the next several hours is predicted based on the bevel irradiance measurement historical value, weather change factors in the next several hours cannot be reflected, so that the ultra-short-term irradiance prediction of the photovoltaic power station is inaccurate.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: when irradiance in the future hours is predicted based on the historical value of inclined plane irradiance measurement, weather change factors in the future hours cannot be reflected, and therefore ultra-short-term irradiance prediction of a photovoltaic power station is inaccurate.
In order to solve the technical problem, the invention provides a method for predicting ultra-short-term irradiance of a photovoltaic power station, which comprises the following steps:
s1, using the read current timet0Training a support vector regression model according to actual irradiance data of the photovoltaic power station in a certain time period before the actual irradiance data, obtaining the trained support vector regression model, and simultaneously utilizing the trained model to predict the time tpPredicting the irradiance to obtain a first predicted irradiance;
s2, reading the predicted time tpAnd tpFront and rear tp-Ts、tp+TsFirst data of numerical weather forecast at time, and current time t0Calculating the similarity between the first data and the second data at each moment, and sequencing the second data of the numerical weather forecast according to the similarity value to obtain third data of the numerical weather forecast;
s3, reading the actual irradiance and the first predicted irradiance of the photovoltaic power station corresponding to each moment in the third data, respectively calculating a first variance between the second predicted irradiance and the actual irradiance of the first data and a second variance between the first predicted irradiance and the actual irradiance, and respectively performing weighted accumulation on the first variance and the second variance by using a forgetting factor;
s4, calculating a first weight occupied by the second predicted irradiance and a second weight occupied by the first predicted irradiance according to a first weighted cumulative variance and a second weighted cumulative variance obtained after weighted accumulation;
s5, calculating to obtain a predicted time t according to the first predicted irradiance, the second predicted irradiance, the first weight and the second weightpThe ultra-short-term predicted irradiance of the photovoltaic power station.
The invention has the beneficial effects that: in ultra-short-term irradiance prediction, the method provided by the invention overcomes the problem that the existing algorithm only predicts an actually-measured irradiance sequence and has poor capability of mastering the variation trend of irradiance, and meanwhile, the method is based on the predicted irradiance of a Support Vector Regression (SVR) model and combines trend data provided by numerical weather forecast to invent a sample screening and irradiance correction algorithm based on similarity and forgetting factors.
Further, in S1, if the current time t is read0And if the photovoltaic power station actual irradiance data in a certain period before the photovoltaic power station has lost or illegal data, using the actual irradiance data adjacent to the lost or illegal data as substitute data, and training a support vector regression model by using the substitute data.
The further beneficial effects are as follows: when the missing or illegal data exists, the actual irradiance data adjacent to the missing or illegal data is used as the substitute data, so that the data is compact, the phenomenon of breakage is avoided, and meanwhile, the improvement of the prediction precision in the subsequent steps is also ensured.
Further, the S1 includes:
s11, reading the current time t0Actual irradiance data of the photovoltaic power station at a time period before the actual irradiance data;
s12, dividing the actual irradiance data into multiple continuous groups of training sample data, and training a support vector regression model by using each group of training sample data;
s13, using the trained support vector regression model to predict the time tpThe irradiance of is predicted to obtain a first predicted irradiance.
The further beneficial effects are as follows: the read data are sequentially divided into a plurality of groups, so that the data have hierarchy, errors generated in the data are greatly reduced, and the accuracy of the data read in the subsequent process can be greatly improved.
Further, in the step S12, the read data are sequentially divided into multiple continuous sets of training sample data, and a support vector regression model is trained by using each set of training sample data, where m continuous actual irradiance data in each set of training sample data are input to the training support vector regression model, and the training support vector regression model is trainedThe output of the model is the w th subsequent to the m continuous actual irradiance datastepWhere m is the number of phase spaces, wstepIs the w-thstepStep prediction step (w)step＝1...Nts)，NtsIn order to be a step of prediction,
Figure GDA0002648684060000031
predicting the time length as TfpThe time scale of ultra-short term irradiance prediction is Ts。
Further, the method between S2 and S3 further includes: sorting the second data according to the similarity value, and calculating the data of each moment in the third data and the predicted moment t in the first data by using a Sigmoid functionpA forgetting factor between data.
The further beneficial effects are as follows: after the similarity is sequenced, the Sigmoid function is used for calculation, so that the subsequent calculation and the measurement accuracy are gradually improved, and the phenomenon of missing and missing of data is avoided.
Further, in S2, the method includes calculating the first data and the current time t0And any historical moment t in a certain period of time beforehAnd front and back th-Ts、th+TsSimilarity of data at time.
The further beneficial effects are as follows: selecting any historical moment t in the time periodhAnd front and back th-Ts、th+TsThe data at the moment participates in the calculation of the similarity and the forgetting factor, and not all data are adopted, so that the data are selected at intervals, the error between the data can be reduced, and the subsequent prediction precision is greatly improved.
Further, the S2 further includes: reading the predicted time tpFront and rear tp-Ts、tp+TsNumerical weather forecast at that moment;
the history time thAnd t before and after ith-Ts、th+TsSecond data point of timeRespectively associated with the predicted time tpAnd the predicted time tpFront and rear tp-Ts、tp+TsAnd the first data at the moment correspond to each other one by one, variance calculation is carried out according to the first data and the second data in the numerical weather forecast, and meanwhile, the obtained variance is weighted and accumulated.
Further, in S4, the first weight occupied by the second predicted irradiance is calculated according to the following formula:
Figure GDA0002648684060000041
therein, se2For the second weighted cumulative variance, se1The variance is accumulated for the first weight.
Further, in S5, the predicted time t is calculated according to the following formulapThe ultra-short-term predicted irradiance of the photovoltaic power station is as follows:
ModGhi＝weights×Gtire+(1.0-weights)×GhiSVR
where weights are the first weight, GtireFor the second predicted irradiance, GhiSVRIs a first predicted irradiance.
The invention also relates to a photovoltaic power station ultra-short term irradiance prediction system, which comprises: the system comprises a training support vector regression model module, a similarity calculation module, a variance calculation module, a weight calculation module and an ultra-short-term irradiance prediction module;
the training support vector regression model module is used for utilizing the read current time t0Training a support vector regression model according to actual irradiance data of the photovoltaic power station in a certain time period before the actual irradiance data, obtaining the trained support vector regression model, and simultaneously utilizing the trained model to predict the time tpPredicting the irradiance to obtain a first predicted irradiance;
the similarity calculation module is used for reading the predicted time tpFirst data of the numerical weather forecast, current time t0And numerical weather forecast of a time period beforeSecond data, calculating the similarity between the first data and the second data at each moment, and sequencing the second data of the numerical weather forecast according to the similarity value to obtain third data of the numerical weather forecast;
the variance calculation module is used for reading the actual irradiance and the first predicted irradiance of the photovoltaic power station corresponding to each moment in the third data, calculating a first variance between the second predicted irradiance and the actual irradiance of the first data and a second variance between the first predicted irradiance and the actual irradiance respectively, and meanwhile performing weighted accumulation on the first variance and the second variance by using a forgetting factor;
the weight calculation module is used for calculating a first weight occupied by the second predicted irradiance and a second weight occupied by the first predicted irradiance according to a first weighted cumulative variance and a second weighted cumulative variance obtained after weighted accumulation;
the ultra-short-term irradiance prediction module is used for calculating a prediction time t according to the first predicted irradiance, the second predicted irradiance, the first weight and the second weightpThe ultra-short-term predicted irradiance of the photovoltaic power station.
The invention has the beneficial effects that: in ultra-short-term irradiance prediction, the method provided by the invention overcomes the problem that the existing algorithm only predicts an actually-measured irradiance sequence and has poor capability of mastering the variation trend of irradiance, and meanwhile, the method is based on the predicted irradiance of a Support Vector Regression (SVR) model and combines trend data provided by numerical weather forecast to invent a sample screening and irradiance correction algorithm based on similarity and forgetting factors.
Drawings
FIG. 1 is a flow chart of a method for predicting ultra-short term irradiance of a photovoltaic power plant in accordance with the present invention;
FIG. 2 is a schematic diagram of a method for predicting ultra-short term irradiance of a photovoltaic power plant in accordance with the present invention;
FIG. 3 is a schematic diagram of an ultra-short term irradiance prediction system of a photovoltaic power plant according to the present invention
Detailed Description
The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.
Example 1
As shown in fig. 1 and fig. 2, in an embodiment of the present invention, a method for predicting ultra-short term irradiance of a photovoltaic power plant includes the following steps:
s1, using the read current time t0Training a support vector regression model according to actual irradiance data of the photovoltaic power station in a certain time period before the actual irradiance data, obtaining the trained support vector regression model, and simultaneously utilizing the trained model to predict the time tpPredicting the irradiance to obtain a first predicted irradiance;
in the embodiment 1, the current time t is read first0And the actual irradiance data of the photovoltaic power station in a certain time period before the actual irradiance data, then training a support vector regression model according to the read data to obtain the trained support vector regression model, such as: reading actual irradiance data of the photovoltaic power station at the current moment 8:00 and 3 days before 8:00, training a Support Vector Regression (SVR) model according to the actual irradiance data of the photovoltaic power station at the current moment 8:00 and in a time period before 3 days, using the newly established support vector regression model and the first data as input, and predicting the current moment t and the actual irradiance data of the photovoltaic power station at the current moment 8:00 and in a certain time period beforepIs predicted, such as: and predicting the irradiance at the prediction time of 8:15 minutes to obtain a first predicted irradiance.
S2, reading the predicted time tpAnd the predicted time tpFront and rear tp-Ts、tp+TsFirst data of numerical weather forecast at time, and current time t0And second data of numerical weather forecast of a time period before the first data, andcalculating the similarity between the first data and the second data at each moment, and sequencing the second data of the numerical weather forecast according to the similarity to obtain third data of the numerical weather forecast; in the present embodiment, the latest 31-day period is used as the current time t0A previous certain time period;
in the present embodiment 1, the predicted time t is read firstpAnd the predicted time tpFront and rear tp-Ts、tp+TsFirst data of a numerical weather forecast for a moment, such as: the time of the data of the numerical weather forecast selected for the prediction time of 8:30 is 8:15 points, 8:30 points and 8: for 45 minutes, the current time t is also read0And second data of numerical weather forecast for a time period prior to the first data, such as: reading the data of the numerical weather forecast of the current time 8:00 and the time period 31 days before the current time 8:00, and after reading the data of the numerical weather forecast, calculating the similarity between the first data and the second data at each time, such as: for the forecast time of 8:30 points, selecting the data of the numerical weather forecast of 8:15 points, 8:30 points and 8:45 points and the data of the numerical weather forecast of the current time of 8:00 and the time period of 31 days before 8:00 to carry out one-to-many data similarity calculation, for example: the first data of 8:15 points corresponds to the second data of 9:15 points on the previous day, the first data of 8:30 points corresponds to the second data of 9:30 points on the previous day, and the ratio of 8: the first data of 45 points corresponds to 9: respectively calculating the variances of the weather elements of the corresponding first data with the point of 8:15, the point of 8:30 and the point of 8 and 45 and the second data value of the previous day for the weather forecast according to the second data with the point of 8:15, the point of 8:30 and the point of 8, performing standard deviation normalization processing and weighting processing on the variances of the obtained weather elements, and then performing standard deviation normalization processing and weighting processing on the processed variances of the weather elements to obtain 8:15 min, 8:30 min and 8: accumulating the 45-point data to obtain a predicted time 8: 30-point first data and a previous day 9: the similarity of the second data is divided into 30 points. And after the similarity is obtained, sequencing the second data of the numerical weather forecast according to the similarity value to obtain third data of the numerical weather forecast. In addition, the numerical weather forecast participation similarity and the numerical weather forecast participation left within 2 hours before and after the historical day and the forecast time are selected in the implementationAnd 4, the calculation of the forgetting factor, and the calculation of the similarity and the forgetting factor are not participated in by other numerical weather forecasts.
S3, reading the actual irradiance and the first predicted irradiance of the photovoltaic power station corresponding to each moment in the third data, respectively calculating a first variance between the second predicted irradiance and the actual irradiance of the first data and a second variance between the first predicted irradiance and the actual irradiance, and respectively performing weighted accumulation on the first variance and the second variance by using a forgetting factor;
in this embodiment 1, the actual irradiance of the photovoltaic power station and the predicted irradiance of the support vector regression model corresponding to each moment in the sorted second data, that is, the third data, are read first, after the reading is completed, the variance between the predicted irradiance and the actual irradiance of the first data and the variance between the predicted irradiance and the actual irradiance of the support vector regression model are calculated, and then after the variance is calculated, the variance between the predicted irradiance and the actual irradiance of the first data and the variance between the predicted irradiance and the actual irradiance of the support vector regression model are weighted and accumulated by using a forgetting factor.
In addition, in this embodiment 1, in addition to calculating irradiance, the method further includes: the variance of the meteorological elements such as ambient temperature, wind speed, humidity, total cloud amount, high cloud amount, medium cloud amount, low cloud amount, air pressure and wind direction is calculated, and the variance of 3 same meteorological elements is accumulated.
S4, calculating a first weight occupied by the second predicted irradiance and a second weight occupied by the first predicted irradiance according to a first weighted cumulative variance and a second weighted cumulative variance obtained after weighted accumulation;
in this embodiment 1, after performing weighted accumulation in step S3, a weighted accumulated variance between the predicted irradiance and the actual irradiance of the first data and a weighted accumulated variance between the predicted irradiance and the actual irradiance of the support vector regression model are obtained, and the weight occupied by the predicted irradiance of the first data and the weight occupied by the predicted irradiance of the support vector regression model are calculated according to the obtained weighted accumulated variance.
S5, according toThe first predicted irradiance, the second predicted irradiance, the first weight and the second weight are calculated to obtain a predicted time tpThe ultra-short-term predicted irradiance of the photovoltaic power station.
In this embodiment, the predicted irradiance of the first data, the predicted irradiance of the support vector regression model, the weight occupied by the predicted irradiance of the first data and the weight occupied by the predicted irradiance of the support vector regression model, which are obtained by the calculation in the steps S1 to S4, are obtained by calculating, and the predicted time t is obtained by the calculationpThe ultra-short-term irradiance prediction of the photovoltaic power station.
It should be noted that, in embodiment 1 of the present invention, the method is to predict the time length TfpThe time scale (prediction step size) is TsUltra-short term irradiance prediction, dividing N altogethertsStep by step
Figure GDA0002648684060000091
Making a prediction, whereinstepStep (w)step＝1...Nts) The prediction method is the steps of S1 to S5 described above.
In ultra-short-term irradiance prediction, the method is used for overcoming the defect that the existing algorithm only predicts an actually-measured irradiance sequence and has poor controllability on irradiance change trend, and the sample screening and irradiance correction algorithm based on similarity and forgetting factors is invented on the basis of the predicted irradiance of a Support Vector Regression (SVR) model and by combining trend data provided by numerical weather forecast. Practical application shows that the prediction precision of the algorithm can be obviously improved in prediction after 1 hour, meanwhile, the calculation efficiency and performance of the algorithm meet the requirement of 15-minute rolling prediction of a photovoltaic power station, and the requirement of real-time scheduling of photovoltaic power generation is completely met.
Preferably, the current time t in S10Actual irradiance data of the photovoltaic power plant for a time period prior to the actual irradiance data comprises: if the lost or illegal data exist in the actual irradiance data of the photovoltaic power station, the actual irradiance data adjacent to the lost or illegal data are used as substitute data, and a support vector regression model is trained on the substitute data.
In this embodiment 1, missing or illegal data exists, and actual irradiance data adjacent to the missing or illegal data is used as substitute data, so that the data is compact, a fracture phenomenon does not occur, and meanwhile, the improvement of prediction accuracy in subsequent steps is ensured.
Preferably, in the present embodiment 1, the step S1 includes:
s11, reading the current time t0Actual irradiance data of the photovoltaic power station at a time period before the actual irradiance data;
s12, dividing the actual irradiance data into multiple groups of training sample data, inputting each group of training sample as m continuous actual irradiance data, and outputting the training sample as the w th following of the m continuous actual irradiance datastepActual irradiance data, where m is the number of phase spaces, which can be set, and the predicted time length is TfpThe time scale (prediction step size) is TsUltra-short term irradiance prediction, dividing N altogethertsStep by step
Figure GDA0002648684060000092
Making a prediction of wstepIs the w-thstepStep (w)step＝1...Nts) A prediction step;
s13, using the trained support vector regression model to predict the time tpThe irradiance of is predicted to obtain a first predicted irradiance.
In this embodiment 1, the read data is sequentially divided into multiple continuous groups, so that the data is hierarchical, errors generated in the data are greatly reduced, and the accuracy of the data read in the subsequent process is greatly improved.
Preferably, the embodiment 1 further involves sequentially dividing the read data into a plurality of consecutive groups in step S12, and training the support vector regression model using the grouped data, and simultaneously training the support vector regression model to output w th data following the m consecutive actual irradiance datastepActual irradiance data, where m is the number of phase spaces, which can be set, and in this embodiment, the value is 4。
Preferably, in this embodiment 1, the method further includes, between S2 and S3: sorting the second data according to the similarity value, and calculating the data of each moment in the third data and the predicted moment t by using a Sigmoid functionpA forgetting factor in between.
In this embodiment, after the acquaintance is sequenced, the subsequent calculation accuracy is gradually improved, and the phenomenon of missing data is avoided. Wherein, a Sigmoid function is used as a forgetting function, and a forgetting factor W is calculated, wherein the Sigmoid function has the following form:
Figure GDA0002648684060000101
the value of the inclination is related to the number of the effective numerical weather forecast history records, the value of the embodiment is 0.00807949, and i is the serial number of the third data sequence.
Preferably, in the embodiment 1, the step S2 includes calculating the first data and the current time t0And any historical moment t in a certain period of time beforehAnd front and back th-Ts、th+TsSimilarity of data at time.
Preferably, the calculation of the similarity and the calculation of the forgetting factor are performed at the current time t0And selecting any historical time t in the time period from the second data of the numerical weather forecast of the previous time periodhAnd front and back th-Ts、th+TsThe data at the moment participate in the calculation of the similarity and the forgetting factor.
In embodiment 1 of the present invention, any historical time t in the time period is selectedhAnd front and back th-Ts、th+TsData at a moment participates in the calculation of similarity and forgetting factor, not all data are adopted, and the data are selected at intervals, so that the error between the data can be reduced, and the subsequent prediction precision is greatly improved, for example: at the above said presentIn the time period 31 days before the time 8:00, the calculation of the data participation similarity and the forgetting factor in the time period is selected, for example, the predicted time is 8: and if the time is 30 minutes, first data of 8:15 minutes, 8:30 minutes and 8:45 minutes are selected, and the similarity calculation method for calculating the time 9:30 minutes before the day and the predicted time comprises the following steps: selecting second data of 9:15 points, 9:30 and 9:45 of the previous day, respectively calculating the variance of each meteorological element of the weather forecast with the value of 8:15 points of the first data and 9:15 points of the previous day of the second data, the variance of each meteorological element of the weather forecast with the value of 8:30 points of the first data and 9:45 points of the previous day of the second data, carrying out standard deviation normalization processing and weighting processing on the calculated variance of each meteorological element, and accumulating the processed data of each meteorological element to obtain the similarity value of 9:30 points of the first data forecast time 8:30 and 9:30 points of the previous day of the second data.
And respectively calculating the similarity of the second data and the predicted time of the first data according to the method for the second data, and sequencing the second data according to the similarity to obtain third data. And sequentially using a Sigmoid function as a forgetting function to calculate a forgetting factor W for the sorted third data, wherein the form of the Sigmoid function is as follows:
Figure GDA0002648684060000111
the value of the inclination is related to the number of the effective numerical weather forecast history records, the value of the embodiment is 0.00807949, and i is the serial number of the third data sequence.
Preferably, the S2 further includes: reading the predicted time tpFront and rear tp-Ts、tp+TsNumerical weather forecast at that moment;
the history time thAnd t before and after ith-Ts、th+TsThe second data of the time respectively correspond to the predicted time tpAnd the predicted time tpFront and rear tp-Ts、tp+TsThe first data of the time correspond to each other one by one, and weather factors are forecasted according to numerical weatherAnd calculating the variance of each meteorological factor by the elements, and performing weighted accumulation on the variances.
In the present embodiment 1, the read predicted time t is selectedpFront and rear tp-Ts、tp+TsNumerical weather forecast for a moment of time, such as: the data of the numerical weather forecast with the forecast time of 8:30 and the data of the numerical weather forecasts of 8:15 and 8:45 respectively correspond to the data of the three time points of the previous day of 9:15, 9:30 and 9:45 one by one, and the calculation method can obtain the data by calculating according to the calculation formula.
Preferably, in S4, the first weight occupied by the second predicted irradiance is calculated according to the following formula:
Figure GDA0002648684060000121
therein, se2For the second weighted cumulative variance, se1The variance is accumulated for the first weight.
In this embodiment, the weight of the predicted irradiance of the first data calculated in S4 is:
Figure GDA0002648684060000122
therein, se2Is a weighted cumulative variance, se, of predicted irradiance and actual irradiance of a support vector regression model1Is the weighted cumulative variance of the predicted irradiance and the actual irradiance of the numerical weather forecast, and weight is the predicted time tpNumerical weather forecast predicts the weight taken up by irradiance.
In this embodiment 1, the accumulated variance of each meteorological element is subjected to a comprehensive weighting process according to different factors affecting irradiance, where the weight of irradiance is 0.8, the weight of temperature is 0.04, the weight of wind speed is 0.01, the weight of humidity is 0.01, the weight of high cloud cover is 0.02, the weight of medium cloud cover is 0.02, the weight of low traffic cover is 0.02, the weight of solar zenith angle cosine is 0.04, and the weight of solar incident angle cosine of the photovoltaic panel is 0.04.
Preferably, in S5, the predicted time t is calculated according to the following formulapThe ultra-short-term predicted irradiance of the photovoltaic power station is as follows:
ModGhi＝weights×Gtire+(1.0-weights)×GhiSVR
wherein ModGhi is the predicted time tpUltra-short term irradiance of a photovoltaic power station, weights being a first weight, GtireFor the second predicted irradiance, GhiSVRIs a first predicted irradiance.
It should be noted that in embodiment 1, weight is the first weight, that is, the predicted time tpThe predicted irradiance of the numerical weather forecast is weighted, GtireFor the second predicted irradiance, that is to say the predicted time tpPredicted irradiance, Ghi, of a numerical weather forecastSVRFor the first predicted irradiance, that is to say the predicted time tpThe support vector regression model predicts irradiance.
Example 2
As shown in fig. 3, the present embodiment 2 further relates to a system for predicting ultra-short term irradiance of a photovoltaic power station, the system comprising: the system comprises a training support vector regression model module, a similarity calculation module, a variance calculation module, a weight calculation module and an ultra-short-term irradiance prediction module;
the training support vector regression model module is used for utilizing the read current time t0Training a support vector regression model according to actual irradiance data of the photovoltaic power station in a certain time period before the actual irradiance data, obtaining the trained support vector regression model, and simultaneously utilizing the trained model to predict the time tpPredicting the irradiance to obtain a first predicted irradiance;
the similarity calculation module is used for reading the predicted time tpFirst data of the numerical weather forecast, current time t0And the second data of the numerical weather forecast in a certain time period before the first data, calculating the similarity between the first data and the second data at each moment, and sequencing the second data of the numerical weather forecast according to the similarity value to obtain the numerical weather forecast(iii) forecasted third data;
the variance calculation module is used for reading the actual irradiance and the first predicted irradiance of the photovoltaic power station corresponding to each moment in the third data, calculating a first variance between the second predicted irradiance and the actual irradiance of the first data and a second variance between the first predicted irradiance and the actual irradiance respectively, and meanwhile performing weighted accumulation on the first variance and the second variance by using a forgetting factor;
the weight calculation module is used for calculating a first weight occupied by the second predicted irradiance and a second weight occupied by the first predicted irradiance according to a first weighted cumulative variance and a second weighted cumulative variance obtained after weighted accumulation;
the ultra-short-term irradiance prediction module is used for calculating a prediction time t according to the first predicted irradiance, the second predicted irradiance, the first weight and the second weightpThe ultra-short-term predicted irradiance of the photovoltaic power station.
In this embodiment 2 the system refers to the contents of all the mentioned methods in embodiment 1 above.
In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (8)

1. The method for predicting the ultra-short-term irradiance of the photovoltaic power station is characterized by comprising the following steps of:
s1, using the read current time t0Training a support vector regression model according to actual irradiance data of the photovoltaic power station in a certain time period before the actual irradiance data, obtaining the trained support vector regression model, and simultaneously utilizing the trained model to predict the time tpPredicting the irradiance to obtain a first predicted irradiance;
s2, reading the predicted time tpAnd tpFront and rear tp-Ts、tp+TsFirst data of numerical weather forecast at time, and current time t0Calculating the similarity between the first data and the second data at each moment, and sequencing the second data of the numerical weather forecast according to the similarity value to obtain third data of the numerical weather forecast; wherein, TsValues included 15 min;
s3, reading the actual irradiance and the first predicted irradiance of the photovoltaic power station corresponding to each moment in the third data, respectively calculating a first variance between the second predicted irradiance and the actual irradiance of the first data and a second variance between the first predicted irradiance and the actual irradiance of the third data, and simultaneously respectively performing weighted accumulation on the first variance and the second variance by using a forgetting factor;
s4, calculating a first weight occupied by the second predicted irradiance and a second weight occupied by the first predicted irradiance according to a first weighted cumulative variance and a second weighted cumulative variance obtained after weighted accumulation; wherein, the first weight occupied by the second predicted irradiance is calculated according to the following formula:
Figure FDA0002648684050000011
therein, se2For the second weighted cumulative variance, se1Accumulating the variance for the first weight;
s5, calculating to obtain prediction according to the first predicted irradiance, the second predicted irradiance, the first weight and the second weightTime tpThe irradiance of the photovoltaic power station is predicted in an ultra-short period; wherein the predicted time t is calculated according to the following formulapThe ultra-short-term predicted irradiance of the photovoltaic power station is as follows:
ModGhi＝weights×Gtire+(1.0-weights)×GhiSVR
wherein ModGhi is the predicted time tpUltra-short term irradiance of a photovoltaic power station, weights being a first weight, GtireFor the second predicted irradiance, GhiSVRIs a first predicted irradiance.
2. The prediction method according to claim 1, wherein in step S1, if the current time t is read0And if the photovoltaic power station actual irradiance data in a certain period before the photovoltaic power station has lost or illegal data, using the actual irradiance data adjacent to the lost or illegal data as substitute data, and training a support vector regression model by using the substitute data.
3. The prediction method according to claim 2, wherein the S1 includes:
s11, reading the current time t0Actual irradiance data of the photovoltaic power station at a time period before the actual irradiance data;
s12, dividing the actual irradiance data into multiple continuous groups of training sample data, and training a support vector regression model by using each group of training sample data;
s13, using the trained support vector regression model to predict the time tpThe irradiance of is predicted to obtain a first predicted irradiance.
4. The prediction method according to claim 3, wherein the step S12 sequentially divides the read data into a plurality of consecutive groups of training sample data, and trains the SVM regression model using each group of training sample data, wherein m consecutive actual irradiance data in each group of training sample data are inputs of the SVM regression model, and an output of the SVM regression model is the output of the SVM regression modelW-th following the m continuous actual irradiance datastepWhere m is the number of phase spaces, wstepIs the w-thstepStep prediction step (w)step＝1...Nts)，NtsIn order to be a step of prediction,
Figure FDA0002648684050000021
Tfpto predict the length of time, TsTime scale for ultra-short term irradiance prediction.
5. The prediction method according to any one of claims 1 to 4, further comprising between S2 and S3: sorting the second data according to the similarity value, and calculating the data of each moment in the third data and the predicted moment t in the first data by using a Sigmoid functionpA forgetting factor between data.
6. The prediction method according to claim 5, wherein said step S2 includes calculating said first data and said current time t0And any historical moment t in a certain period of time beforehAnd front and back th-Ts、th+TsSimilarity of data at time.
7. The prediction method according to claim 6, wherein the step of S2 further comprises: reading the predicted time tpFront and rear tp-Ts、tp+TsNumerical weather forecast at that moment;
the history time thAnd t before and after ith-Ts、th+TsThe second data of the time respectively correspond to the predicted time tpAnd the predicted time tpFront and rear tp-Ts、tp+TsAnd the first data at the moment correspond to each other one by one, variance calculation is carried out according to the first data and the second data in the numerical weather forecast, and meanwhile, the obtained variance is weighted and accumulated.
8. A prediction system using the method of predicting ultra-short term irradiance of a photovoltaic power plant of any of claims 1 to 7, the system comprising: the system comprises a training support vector regression model module, a similarity calculation module, a variance calculation module, a weight calculation module and an ultra-short-term irradiance prediction module;
the training support vector regression model module is used for utilizing the read current time t0Training a support vector regression model according to actual irradiance data of the photovoltaic power station in a certain time period before the actual irradiance data, obtaining the trained support vector regression model, and simultaneously utilizing the trained model to predict the time tpPredicting the irradiance to obtain a first predicted irradiance;
the similarity calculation module is used for reading the predicted time tpAnd tpFront and rear tp-Ts、tp+TsFirst data of the numerical weather forecast, current time t0Calculating the similarity between the first data and the second data at each moment, and sequencing the second data of the numerical weather forecast according to the similarity value to obtain third data of the numerical weather forecast; wherein, TsValues included 15 min;
the variance calculation module is used for reading the actual irradiance and the first predicted irradiance of the photovoltaic power station corresponding to each moment in the third data, calculating a first variance between the second predicted irradiance and the actual irradiance of the first data and a second variance between the first predicted irradiance and the actual irradiance of the third data respectively, and meanwhile performing weighted accumulation on the first variance and the second variance by using a forgetting factor;
the weight calculation module is used for calculating a first weight occupied by the second predicted irradiance and a second weight occupied by the first predicted irradiance according to a first weighted cumulative variance and a second weighted cumulative variance obtained after weighted accumulation; wherein, the first weight occupied by the second predicted irradiance is calculated according to the following formula:
Figure FDA0002648684050000041
therein, se2For the second weighted cumulative variance, se1Accumulating the variance for the first weight;
the ultra-short-term irradiance prediction module is used for calculating a prediction time t according to the first predicted irradiance, the second predicted irradiance, the first weight and the second weightpThe irradiance of the photovoltaic power station is predicted in an ultra-short period; wherein the predicted time t is calculated according to the following formulapThe ultra-short-term predicted irradiance of the photovoltaic power station is as follows:
ModGhi＝weights×Gtire+(1.0-weights)×GhiSVR
wherein ModGhi is the predicted time tpUltra-short term irradiance of a photovoltaic power station, weights being a first weight, GtireFor the second predicted irradiance, GhiSVRIs a first predicted irradiance.
CN201710340400.0A 2017-05-15 2017-05-15 Photovoltaic power station ultra-short-term irradiance prediction method and system Active CN107256437B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710340400.0A CN107256437B (en) 2017-05-15 2017-05-15 Photovoltaic power station ultra-short-term irradiance prediction method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710340400.0A CN107256437B (en) 2017-05-15 2017-05-15 Photovoltaic power station ultra-short-term irradiance prediction method and system

Publications (2)

Publication Number Publication Date
CN107256437A CN107256437A (en) 2017-10-17
CN107256437B true CN107256437B (en) 2020-11-27

Family

ID=60027611

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710340400.0A Active CN107256437B (en) 2017-05-15 2017-05-15 Photovoltaic power station ultra-short-term irradiance prediction method and system

Country Status (1)

Country Link
CN (1) CN107256437B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108776843A (en) * 2018-04-12 2018-11-09 天津大学 Photovoltaic ultra-short term power forecasting method based on Boltzmann machine
CN109884896B (en) * 2019-03-12 2022-02-11 河海大学常州校区 Photovoltaic tracking system optimized tracking method based on similar day irradiation prediction
CN110956270A (en) * 2019-10-18 2020-04-03 广州丰石科技有限公司 XGBOOST algorithm-based solar irradiance prediction method
CN116167531A (en) * 2023-04-26 2023-05-26 天津福天科技有限公司 Photovoltaic power generation prediction method based on digital twin

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103390116A (en) * 2013-08-07 2013-11-13 华北电力大学（保定） Method for predicting electricity generation power of photovoltaic power station in step-by-step way
CN103559561A (en) * 2013-11-13 2014-02-05 上海电气集团股份有限公司 Super-short-term prediction method of photovoltaic power station irradiance
CN105373857A (en) * 2015-11-27 2016-03-02 许昌许继软件技术有限公司 Photovoltaic power station irradiance prediction method
CN106485093A (en) * 2016-11-10 2017-03-08 哈尔滨工程大学 Based on the solar irradiance time series synthetic method for improving Markov chain

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10482197B2 (en) * 2007-02-12 2019-11-19 Locus Energy, Inc. Solar irradiance modeling augmented with atmospheric water vapor data

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103390116A (en) * 2013-08-07 2013-11-13 华北电力大学（保定） Method for predicting electricity generation power of photovoltaic power station in step-by-step way
CN103559561A (en) * 2013-11-13 2014-02-05 上海电气集团股份有限公司 Super-short-term prediction method of photovoltaic power station irradiance
CN105373857A (en) * 2015-11-27 2016-03-02 许昌许继软件技术有限公司 Photovoltaic power station irradiance prediction method
CN106485093A (en) * 2016-11-10 2017-03-08 哈尔滨工程大学 Based on the solar irradiance time series synthetic method for improving Markov chain

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
光伏功率预测技术;龚莺飞等;《电力***自动化》;20160225;140-151页 *
基于天气类型聚类识别的光伏***短期;代倩等;《中国电机工程学报》;20111205;28-35页 *

Also Published As

Publication number Publication date
CN107256437A (en) 2017-10-17

Similar Documents

Publication Publication Date Title
Raza et al. An ensemble framework for day-ahead forecast of PV output power in smart grids
CN107256437B (en) Photovoltaic power station ultra-short-term irradiance prediction method and system
Wang et al. A review of wind power forecasting models
Abuella et al. Solar power forecasting using support vector regression
CN103955749A (en) Sun direct radiation value prediction method and system
Alanazi et al. Long-term solar generation forecasting
CN111488896B (en) Distribution line time-varying fault probability calculation method based on multi-source data mining
CN103473621A (en) Wind power station short-term power prediction method
KR102284253B1 (en) Power generation forecasting system of solar power plant using weather information and forecasting method of power generation using it
CN113722375B (en) Double-layer hybrid photovoltaic prediction method based on multi-algorithm optimization and computer equipment
CN102147839A (en) Method for forecasting photovoltaic power generation quantity
Jiang et al. Ultra-short-term prediction of photovoltaic output based on an LSTM-ARMA combined model driven by EEMD
Wang et al. Solar power prediction with data source weighted nearest neighbors
CN112307672A (en) BP neural network short-term wind power prediction method based on cuckoo algorithm optimization
CN116595394A (en) Training method of wind speed correction model, wind speed prediction method, wind speed prediction equipment and medium
CN115759467A (en) Time-division integrated learning photovoltaic prediction method for error correction
Zhao et al. A review on proliferation of artificial intelligence in wind energy forecasting and instrumentation management
Browell et al. Use of turbine-level data for improved wind power forecasting
Bai et al. Hybrid model based on K-means++ algorithm, optimal similar day approach, and long short-term memory neural network for short-term photovoltaic power prediction
Wang Forecasting short-term wind speed using support vector machine with particle swarm optimization
Cabezon et al. Comparison of methods for power curve modelling
CN113610285A (en) Power prediction method for distributed wind power
Agga et al. Convolutional neural network (cnn) extended architectures for photovoltaic power production forecasting
CN117060407A (en) Wind power cluster power prediction method and system based on similar day division
CN116484998A (en) Distributed photovoltaic power station power prediction method and system based on meteorological similar day

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