A photovoltaic power generation prediction method based on big data

Abstract

The present application discloses a photovoltaic power generation prediction method and architecture based on big data, the method comprising the following steps: receiving power generation data from an inverter on a big data platform; grouping the received data according to the device and time, and calculating the power generation and discrete rate of the inverter per unit time; acquiring meteorological data and matching it with the geographic location of the inverter; removing outliers and normalizing the matching data of the inverter to obtain corresponding feature data; constructing a power generation prediction module, and performing model training and power generation prediction through an algorithm based on the obtained feature data. The architecture is used to implement the above method. Beneficial effects of the present application: During data synchronization, the cumulative power generation is calculated in the inverter to provide more accurate power generation data; and the power generation per unit time can be corrected by the cumulative power generation per unit time and the average power, thereby improving the quality of power generation data.

Description

Technical Field

[0001] The present application relates to the field of new energy power generation technology, and in particular to a photovoltaic power generation prediction method based on big data. Background Art

[0002] In the prior art, a Chinese invention patent application with publication number CN117172363A provides a method for predicting photovoltaic power generation, which mainly includes the following contents: collecting data from photovoltaic power generation equipment through an information collection module to obtain collected data, and the collected data includes actual weather data and actual power generation; predicting the weather at the location of the photovoltaic power generation equipment through a meteorological data module of a meteorological station to obtain predicted weather data; predicting the power generation of the photovoltaic power generation equipment according to the predicted weather data and the collected data through a prediction module to obtain predicted power generation; determining a comparison result between the actual power generation and the predicted power generation through an error statistics module; and displaying the comparison result through a display module.

[0003] In the above method, the collected data obtained by using the information collection module to collect data from the photovoltaic equipment includes weather data and actual power generation data. However, this power generation data is the actual power at a certain point in time, and there is a certain error with the actual power generation. Assuming that the frequency of the acquisition module to collect data from the photovoltaic equipment is once every t minutes, then the actual power generation (unit W) collected is P1, ..., Pn. Then, the power generation during this period calculated based on the actual power generation is (P1+P2+...+Pn)×t / 60 / 1000 (kWh).

[0004] If within time t, the actual power generation is P k If the actual power does not change, then the power generation calculated using the actual power is equal to the actual power generation. But in fact, the power generation changes instantaneously and cannot remain unchanged within the collection time interval. Therefore, some errors will be introduced in the data collection process. In addition, in actual application scenarios, the existing technology considers fewer factors for power generation prediction, generally only considering weather factors, which will reduce the training accuracy of the training model, thereby further reducing the accuracy of the power generation prediction results. Summary of the invention

[0005] One of the objectives of the present application is to provide a photovoltaic power generation prediction method based on big data that can solve at least one of the defects in the above-mentioned background technology.

[0006] Another object of the present application is to provide a photovoltaic power generation prediction framework that can solve at least one of the defects in the above-mentioned background technology.

[0007] In order to achieve at least one of the above-mentioned purposes, the technical solution adopted in the present application is: a photovoltaic power generation prediction method based on big data, comprising the following steps: receiving power generation data from inverters on a big data platform; grouping and processing the received power generation data of all inverters according to equipment and time, and calculating the power generation and discrete rate data of each inverter per unit time to obtain a first calculation result; obtaining meteorological data and matching it with the geographical location of the inverter, and obtaining matching data of weather, alarm, discrete rate and power generation of the inverter at various historical time points based on the power generation data of the inverter and the first calculation result; removing outliers and normalizing the matching data of the inverter to obtain characteristic data and power generation data of each inverter per unit time, wherein the characteristic data includes weather characteristic data, alarm characteristic data and string discrete rate characteristic data; constructing a power generation prediction model, and performing model training and power generation prediction through an algorithm based on the obtained characteristic data.

[0008] Preferably, the power generation data received by the big data platform includes equipment number, time, rated power, real-time power, cumulative power generation for the day, historical cumulative power generation, alarm code and current data of each string; the process of grouping the received data is as follows: taking k hours as the unit time, the collected data of each inverter is summarized; the average power per unit time and the maximum and minimum cumulative power generation are calculated; the power generation per unit time is calculated = the maximum cumulative power generation per unit time - the minimum cumulative power generation per unit time; if the calculated power generation per unit time is less than or equal to 0, the power generation per unit time is replaced by the average power per unit time; the number of alarms per unit time and the content with the most alarms are summarized, and the currents of each string of the inverter are summarized and summed; the discrete rate of the calculated string = the variance of the current of each string / the average value of the current of each string.

[0009] Preferably, the inverter's cumulative power generation is synchronized once a day, and the calculation of the inverter's daily cumulative power generation includes the following process: starting at time point T every day, if the inverter has power generation, the real-time power per second is accumulated and converted into power generation; when the power generation is greater than the set threshold, the power generation is accumulated to the cumulative power generation of the day, and the accumulated power is assigned to 0; when the power generation is less than the set threshold, the power accumulation continues until the power generation is greater than the set threshold; the power generation is calculated one second before time point T of the next day to obtain the cumulative power generation of the day.

[0010] Preferably, data reception of the big data platform includes the following process: data of the inverter is collected once every t time interval by the collection device; the collected data is sent to the IOT Internet of Things platform in the form of a message queue, and is stored in a relational database after data cleaning and message format conversion; the data stored in the relational database is synchronized to the big data platform at a fixed time every day.

[0011] Preferably, the meteorological data includes light radiation, temperature, humidity and wind speed; the meteorological data is suitable for downloading from the weather forecast center, and after the meteorological data download is completed, it is uploaded to the big data platform after format conversion for data matching; the data matching process is as follows: match the meteorological data with the nearest address location according to the latitude and longitude data of the inverter; splice the matched meteorological data with the inverter data to obtain the weather, alarm, discrete rate and power generation data of the inverter at various historical time points.

[0012] Preferably, the process of processing the matching data of the inverter is as follows: eliminate data with power generation of 0, abnormal power generation per unit time, and irradiance of null or 0; normalize weather characteristics, alarm characteristics of the inverter itself, and string discrete rate characteristics to obtain the characteristic data and power generation data of each inverter per unit time.

[0013] Preferably, model training includes the following processes: splitting the characteristic data of the inverter into training data and calibration data; training the power generation prediction model using an intelligent algorithm on the training data; and evaluating the prediction accuracy of the prediction results of the trained power generation prediction model using the calibration data and the calculated power generation data.

[0014] Preferably, the power generation prediction result is compared with the calibration data to evaluate the accuracy of the prediction result; the accuracy is calculated as follows:

[0015] ;

[0016] Among them, precision represents the accuracy, y represents the predicted value of power generation, y' represents the actual value of power generation, E represents the actual startup capacity, and n is the number of samples.

[0017] A photovoltaic power generation prediction architecture is used to implement the above-mentioned photovoltaic power generation prediction method based on big data, including a cloud, a feature calculation architecture and multiple inverters; the feature calculation architecture is suitable for receiving the power generation data of the inverter to perform feature calculation, and sending the calculation results to the cloud; the cloud is suitable for training the power generation prediction model and predicting the power generation through an electricity prediction algorithm according to the results of the feature calculation.

[0018] Preferably, a plurality of the inverters are arranged in a plurality of substations according to a set number; the number of the feature computing architectures is one and is arranged in the cloud, and the feature computing architecture is suitable for receiving the power generation data of the inverters in all the substations; or, the number of the feature computing architectures corresponds to the number of substations, and each of the feature computing architectures is arranged in a corresponding substation, and each of the feature computing architectures receives the power generation data of all the inverters in the corresponding substation.

[0019] Compared with the prior art, the beneficial effects of this application are:

[0020] During the data synchronization process, the cumulative power generation is calculated in the inverter to provide more accurate power generation data; and the power generation per unit time can be corrected by the cumulative power generation per unit time and the average power, thereby improving the quality of power generation data. During the model training process, the alarm and string discrete rate characteristics of the equipment itself are added to improve the stability and accuracy of the model prediction. Compared with traditional methods, this application improves the accuracy of power generation prediction by improving the quality of basic data and adding key features of equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a schematic diagram of the overall workflow of power generation prediction in this application.

[0022] Figure 2 Schematic diagram of the workflow for power generation data collection in this application.

[0023] Figure 3 This is a schematic diagram of the workflow for data synchronization in this application.

[0024] Figure 4 This is a schematic diagram of the workflow for grouping synchronization data in this application.

[0025] Figure 5 Schematic diagram of the workflow for model training in this application.

[0026] Figure 6 A schematic diagram of the structure of one example of the photovoltaic power generation prediction architecture of the present application.

[0027] Figure 7 This is a schematic diagram of another example of the photovoltaic power generation prediction architecture of the present application. DETAILED DESCRIPTION

[0028] Below, the present application is further described in conjunction with specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0029] In the description of the present application, it should be noted that directional words, such as the terms "center", "lateral", "longitudinal", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", etc., indicating directions and positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of narrating the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and cannot be understood as limiting the specific scope of protection of the present application.

[0030] It should be noted that the terms "first", "second", etc. in the description and claims of the present application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0031] The terms "including" and "having" and any variations thereof in the specification and claims of the present application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0032] One aspect of the present application provides a photovoltaic power generation prediction method based on big data, such as Figure 1 As shown, one of the preferred embodiments includes the following steps:

[0033] Power generation data collection process: Receive power generation data from the inverter on the big data platform.

[0034] Power generation data preprocessing process: The received power generation data of all inverters are grouped and processed according to the device and time, and the power generation and discrete rate data of each inverter per unit time are calculated to obtain the first calculation result.

[0035] Meteorological data acquisition and matching process: obtain the required meteorological data through the interface or other means; obtain the meteorological data and match it with the geographic location of the inverter, and based on the power generation data of the inverter and the first calculation result mentioned above, obtain the matching data of the inverter's weather, alarm, discrete rate and power generation at various historical time points.

[0036] Secondary preprocessing of data: outliers are eliminated and normalized on the matching data of the inverter to obtain the characteristic data and power generation data of each inverter per unit time; the characteristic data includes weather characteristic data, alarm characteristic data and discrete rate characteristic data.

[0037] Model training & result prediction process: Build a power generation prediction module, and use algorithms to perform model training and power generation prediction based on the obtained feature data.

[0038] Forecast result evaluation process: After completing the power generation forecast, the forecast results are compared with the actual values ​​to evaluate the accuracy of the forecast results.

[0039] It is understandable that in the actual application scenario of the power station, the characteristics of the inverter equipment itself, such as characteristic data such as alarms and discrete rates, will affect the power generation of the inverter. However, the prior art does not take the above factors into consideration when predicting the power generation of the inverter, which will greatly reduce the accuracy of model training. In this embodiment, the power generation data of the inverter is obtained in the data acquisition preparation process and the power generation and discrete rate data of the inverter are calculated. Then, after matching the weather data according to the geographical location of the device and undergoing secondary preprocessing, the required weather, alarm, discrete rate and other characteristic data can be obtained. Finally, the power generation prediction model is trained according to the obtained characteristic data, which can effectively improve the accuracy of the model prediction. Compared with the traditional power generation prediction method, the present application can improve the quality of the basic data and increase the key features of the equipment through the refined processing in the acquisition, preprocessing and feature extraction process, thereby improving the accuracy of the power generation prediction. The construction of the power generation prediction model is a conventional technology of those skilled in the art, so it will not be elaborated in detail here. For ease of understanding, each of the above steps will be described in detail below.

[0040] 1. Regarding the power generation data collection process.

[0041] In this embodiment, the big data platform can determine the synchronization of inverter data according to actual needs, for example, it can be one day or half a day. For ease of understanding, in this embodiment, the data synchronization of the big data platform is preferably synchronized once a day. During the power generation data collection process, the inverter can calculate the cumulative power generation of the day and the historical cumulative power generation based on the actual power per second, and the accuracy of the data can be effectively improved through second-level calculations.

[0042] For the calculation of the cumulative power generation for the day, Figure 2As shown, the specific calculation process includes the following: starting from the T time point of each day, if the inverter has power generation, the real-time power Pi (unit W) per second is accumulated, and the accumulated result is Ps. Then the accumulated result Ps is converted into the power generation e_pow, and the specific conversion formula is: e_pow=Ps / 3600 / 1000 (unit kWh). The threshold pow_t (unit kWh) is set according to the required accuracy. When the power generation e_pow is greater than the set threshold pow_t, the power generation is accumulated to the accumulated power generation e_today of the day (the initial value of each day is 0), and the accumulated power Ps is assigned to 0. When the power generation e_pow is less than the set threshold pow_t, the power generation Pi continues to be accumulated to Ps until the power generation e_pow is greater than the set threshold pow_t. Finally, the power generation is calculated one second before the T time point of the next day to obtain the accumulated power generation e_today of the day.

[0043] It should be known that the choice of the T time point can be made according to actual needs. In order to facilitate the calculation and statistics of data, the choice of the T time point in this embodiment can be preferably 0 o'clock every day, so the final result calculation of the cumulative power generation e_today for the day can be performed at 23:59:59 every day.

[0044] It is understandable that the traditional method does not take into account the fact that the inverter's power generation changes instantaneously when predicting power generation, which results in inaccurate collected data. In this embodiment, by setting an accuracy threshold, the influence of instantaneous power changes can be eliminated, thereby improving the calculation accuracy of the cumulative power generation for the day.

[0045] In this embodiment, Figure 3 As shown in the figure, after completing the calculation of the cumulative power generation for the day, the process of synchronizing the inverter data to the big data platform is as follows: the data of the inverter is collected once every t time interval through the collection device; the collected data is sent to the IOT Internet of Things platform in the form of a message queue, and is stored in a relational database after data cleaning and message format conversion; the data stored in the relational database is synchronized to the big data platform at a fixed time every day.

[0046] It should be known that in order to ensure the accuracy of the data, the interval time t of the collection device can be in minutes or seconds; in this embodiment, it is preferred to use the minute level, that is, data collection is performed every t minutes. The specific structure and working principle of the collection device are well-known technologies for those skilled in the art, so they will not be elaborated in detail here. The IOT Internet of Things platform is part of the big data platform and belongs to the prior art, so it will not be elaborated in detail here. The data information collected by the collection device includes device number, time (minute level), rated power, real-time power, cumulative power generation for the day, historical cumulative power generation, alarm code and current data of each string, etc.

[0047] 2. Preprocessing process of power generation data.

[0048] In this embodiment, Figure 4 As shown, the grouping process of the data synchronously sent by all inverters mainly includes two steps.

[0049] Step 1: Summarize the collected data of each inverter in units of k hours. Based on the summarized data, the average power per unit time and the maximum and minimum cumulative power generation can be calculated. At the same time, the number of alarms per unit time and the content with the most alarms can be summarized. At the same time, the current of each inverter string can be summarized and summed to prevent errors caused by data transmission delays.

[0050] Step 2: Calculate the power generation per unit time based on the cumulative power generation calculated in the previous step. The specific calculation formula is: Power generation per unit time = Maximum cumulative power generation per unit time - Minimum cumulative power generation per unit time. After completing the calculation of the power generation per unit time, if the value of the calculated power generation per unit time is less than or equal to 0, the average power per unit time calculated in the previous step can be used to replace the power generation per unit time. At the same time, the discrete rate of the string is calculated. The specific calculation formula is: Discrete rate of the string = Variance of the current of each string / Average value of the current of each string; the higher the discrete rate, the lower the power generation efficiency of the equipment. Compared with the traditional method, correcting the power generation per unit time by the cumulative power generation per unit time and the average power can improve the quality of the power generation data, so as to further improve the accuracy of the subsequent power generation forecast results.

[0051] It should be noted that in order to facilitate the subsequent matching of weather and equipment data, it is necessary to ensure that the data collection time of the two is basically consistent, and weather data is usually accurate to the hour, so in this embodiment, the unit time for inverter data aggregation is k hours, and the value of k can be selected according to actual needs, such as k=1, k=2, etc. There are many types of inverter alarm contents, such as grid overvoltage, voltage undervoltage, no grid, DC bus voltage unevenness, etc.

[0052] It should also be known that during data transmission, message delays or retransmissions may occasionally occur, which will cause the cumulative power generation at later times to be less than the cumulative power generation at earlier times, and thus cause the calculated value of power generation per unit time to be less than 0.

[0053] 3. Regarding the process of meteorological data acquisition and matching.

[0054] It is important to know that meteorological data is an important factor affecting photovoltaic power generation, including light radiation, temperature, humidity and wind speed, etc. Therefore, it is necessary to obtain meteorological data of the geographical location of the power station to analyze the relationship between meteorological data and actual power generation, and optimize power generation forecasts based on the relationship between the two.

[0055] In this embodiment, the meteorological data can be downloaded from the weather forecast center through an interface or other means, and after the meteorological data is downloaded, it is uploaded to the big data platform for data matching after format conversion. The specific data matching process is as follows: match the meteorological data with the nearest address location according to the longitude and latitude data of the inverter; splice the matched meteorological data with the inverter data to obtain the weather, alarm, discrete rate and power generation data of the inverter at various historical time points.

[0056] For ease of understanding, the following will use ERA5 meteorological data as an example for detailed explanation; of course, data from other sources can also be used for meteorological data.

[0057] ERA5 is a global high-resolution atmospheric reanalysis dataset developed and maintained by the European Centre for Medium-Range Weather Forecasts, providing continuous, consistent and high-quality data on a variety of meteorological and climate variables. ERA5 provides a wide range of meteorological and climate change information, including temperature, humidity, wind speed, precipitation, transport, surface radiation, surface temperature, etc. These data are updated at fixed time intervals (hourly or monthly), and are distributed at a spatial resolution (0.25 degrees of latitude and longitude grid, from several kilometers to tens of kilometers).

[0058] When acquiring meteorological data, the required meteorological data can be downloaded from the official website of ERA5 manually or through Python code in the format of nc. The downloaded data can be parsed into CSV format through the Python program, and the specific parsing process is a well-known technology for those skilled in the art. After completing the format conversion, the meteorological data can be uploaded to the table of the big data platform through the tunnel upload tool. There are many types of big data platforms. Maxcompute is preferably used in this embodiment. The upload process is in string format, and then converted to the corresponding partition of the table when necessary. Conversion type. It should be noted that when performing data conversion, Greenwich Mean Time needs to be converted to Beijing time in different partitions.

[0059] Match the ERA5 data with the nearest address according to the geographic longitude and latitude data of the inverter device. Assuming that the geographical location of a power station is located at 121.4392776 east longitude and 29.3070666 north latitude, since the spatial resolution of ERA5 data is 0.25 degrees, the ERA5 data location closest to the current power station is 121.5 east longitude and 29.25 north latitude. Then splice the weather data of 121.5 east longitude and 29.25 north latitude with the data of the inverter device at 121.4392776 east longitude and 29.3070666 north latitude to obtain the weather, alarm, discrete rate and power generation data of the inverter device at various historical time points.

[0060] 4. Secondary data preprocessing process.

[0061] In this embodiment, the process of processing the matching data of the inverter is as follows: remove the data with power generation of 0, remove the abnormal value data of power generation per unit time, and remove the data with irradiance ssrd being empty or 0. Normalize the weather characteristics, the alarm characteristics of the inverter itself, and the string discrete rate characteristics to obtain the weather characteristic data, equipment characteristic data, string discrete rate characteristic data, and power generation data of each inverter per unit time.

[0062] For a power station with multiple inverters, the power generation data per unit time needs to be aggregated to the power station dimension. This step is not required if the power generation forecast is for a single inverter.

[0063] It should be known that the abnormal value of power generation per unit time can be judged by comparing the power generation per unit time with the value of the component capacity multiplied by X, where X is a coefficient, usually greater than or equal to 1; if the power generation per unit time is greater than the component capacity × X, it means that the power generation per unit time is abnormal. Irradiance refers to the radiation flux per unit area of ​​the irradiated surface.

[0064] 5. Regarding the model training and result prediction process.

[0065] In this embodiment, Figure 5 As shown, the model training includes the following process: splitting the characteristic data of the inverter into training data and calibration data. Using intelligent algorithms to train the power generation prediction model on the training data. The prediction accuracy of the prediction results of the trained power generation prediction model is evaluated through calibration data and calculated power generation data.

[0066] It should be known that the characteristic data of the inverter include relative humidity, cloud cover, surface thermal radiation, top net solar radiation, temperature, wind speed along longitude, wind speed along latitude, land surface solar radiation, number of alarms, alarm content code, inverter string discrete rate, etc.; the above characteristic data are included in the training data and calibration data. There are many types of intelligent algorithms, such as linear regression, GBDT, XGBoost, LSTM and random forest; the specific principles of algorithms such as linear regression, GBDT, XGBoost, LSTM and random forest are well known to those skilled in the art, so they will not be elaborated here. Algorithms for power generation prediction include but are not limited to the above-mentioned algorithm, as long as the algorithm has a supervised regression model, it can be used. For the use of intelligent algorithms, one of them can be adopted, or it can be a combination of multiple algorithms.

[0067] It should be noted that the power generation prediction model is based on the aforementioned data collection accuracy, and the power generation prediction results are in hours; of course, the short-term (hourly) prediction results can also be summarized to days to evaluate the daily, monthly and annual power generation prediction deviations.

[0068] 6. Evaluation process for prediction results.

[0069] It is important to know that since the characteristic data of the inverter is the real value, after the power generation prediction model completes the prediction, the prediction result can be compared with the calibration data as the real value. If the difference between the two is within the specified error range, it means that the prediction accuracy of the power generation prediction model meets the requirements. Otherwise, continue to optimize the model until the error requirements are met. For ease of understanding, the comparison between the power generation prediction results and the calibration data can be expressed by the accuracy rate.

[0070] It should be known that there are many ways to express accuracy. The traditional way can be expressed by the average value MAE of the absolute value of the prediction error, which measures the average deviation between the predicted value and the true value, but is not sensitive to outliers; the specific calculation formula of MAE is as follows:

[0071] .

[0072] Among them, y represents the predicted value of power generation, y' represents the actual value of power generation, that is, the power generation data calculated based on the power generation data of the inverter, and n is the number of samples, that is, the number of daily prediction points.

[0073] Of course, the traditional method of accuracy can also be expressed by the mean square of the prediction error, MSE, which gives greater weight to the error and is therefore more sensitive to outliers than MAE; the calculation formula of MSE is as follows:

[0074] .

[0075] In this embodiment, based on the advantages of the traditional calculation formula, the calculation formula for the accuracy rate in this embodiment is as follows:

[0076] .

[0077] Among them, precision represents the accuracy, and E represents the actual startup capacity.

[0078] It is understandable that this embodiment introduces the real startup capacity, which can also be called the component handover capacity, that is, the actual maximum power generation power, when calculating the accuracy rate on the basis of traditional parameters, so that the calculation result can be more in line with reality and the accuracy of the calculation result can be improved.

[0079] Another aspect of the present application provides a photovoltaic power generation prediction architecture for implementing the above photovoltaic power generation prediction method based on big data, such as Figure 6 and Figure 7 As shown, one preferred embodiment includes a cloud, a feature calculation architecture, and multiple inverters. The feature calculation architecture is connected to each inverter in communication, so that the power generation data of each inverter can be received to perform feature calculation, and then the calculation result can be sent to the cloud. The cloud can receive the calculation result of the feature calculation architecture, and train the power generation prediction model and predict the power generation through the power prediction algorithm according to the result of the feature calculation.

[0080] It should be known that there are many ways to set up the feature calculation architecture, and different settings can be applied to different scenarios. For ease of understanding, two specific examples will be used below for detailed explanation.

[0081] Example 1: Centralized architecture. Figure 6 As shown, multiple inverters can be set in multiple areas according to the set number. The feature computing architecture can be set in the cloud, that is, the number of feature computing architectures is one. All inverters in all areas communicate with the feature computing architecture located in the cloud, so that the feature computing architecture can simultaneously receive the power generation data of all inverters in the power station.

[0082] Example 2: Distributed architecture. Figure 7 As shown, multiple inverters can be set in multiple areas according to the set number. A separate area controller is set in each area, and the number of feature calculation architectures is equal to the number of areas, and is correspondingly set in the area controller of each area. Each feature calculation architecture can communicate with all inverters in the corresponding area. After the feature calculation architecture of each area completes the calculation of the power generation data of all inverters in the corresponding area, the calculation data can be aggregated to the cloud for training the power generation prediction algorithm and result prediction.

[0083] It is understandable that, for Example 1, the entire power generation prediction process is completed in the cloud, and the data is highly centralized. However, when the number of inverters is very large, the computing power requirements of the cloud are very high; that is, when the number of inverters is large, the workload of information interaction is very large, and the reliability requirements of the cloud are relatively high, and the delay of signal interaction is also relatively large. For Example 2, some functions of its power generation prediction process can be transferred to the substation controller, so that when the number of inverters is very large, the computing power requirements of the cloud platform can be appropriately reduced; that is, the substation controller only needs to pass the results of edge computing to the cloud platform, so the amount of information transmitted to the cloud platform can be greatly reduced, so that the workload of information interaction will also be greatly reduced, which can effectively improve the reliability of the cloud platform and reduce the delay of signal interaction.

[0084] Based on the above analysis process, it can be known that when the number of inverters is small, the setting of the feature calculation architecture can adopt the above-mentioned example 1 or the above-mentioned example 2; when the number of inverters is relatively large, the setting of the feature calculation architecture preferably adopts the above-mentioned example 2.

[0085] It should be known that the number of inverters contained in each substation can be selected according to actual needs; accordingly, a corresponding number of substations can be formed according to the number of inverters. For example, the number of inverters in each substation can be set to N, and a total of M substations can be formed. For the above example one, the feature calculation architecture can receive M×N power generation data, and the power generation data is generally current and voltage; for the above example two, each feature calculation architecture can receive N power generation data, and finally the cloud can receive M groups of feature data that have been calculated.

[0086] The above describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited by the above embodiments, and the above embodiments and the specification only describe the principles of the present application. The present application may have various changes and improvements without departing from the spirit and scope of the present application, and these changes and improvements fall within the scope of the present application for which protection is sought. The scope of protection claimed by the present application is defined by the attached claims and their equivalents.

Claims

1. A photovoltaic power generation prediction method based on big data, characterized in that: The steps include: Receive power generation data from the inverter on the big data platform; The received power generation data of all inverters are grouped and processed according to the device and time, and the power generation and discrete rate data of each inverter per unit time are calculated to obtain a first calculation result; Acquire meteorological data and match it with the geographic location of the inverter, and obtain matching data of weather, alarm, discrete rate and power generation of the inverter at various historical time points based on the power generation data of the inverter and the first calculation result; Perform outlier elimination and normalization processing on the matching data of the inverter to obtain characteristic data and power generation data of each inverter in unit time; the characteristic data includes weather characteristic data, alarm characteristic data and discrete rate characteristic data; Constructing a power generation prediction model, and performing model training and power generation prediction through an algorithm based on the obtained characteristic data; The power generation data received by the big data platform includes equipment number, time, rated power, real-time power, cumulative power generation on the day, historical cumulative power generation, alarm code and current data of each string; the process of grouping the received data is as follows: Summarize the collected data of each inverter with k hours as the unit time; calculate the average power per unit time and the maximum and minimum values ​​of the accumulated power generation; Calculate the power generation per unit time = maximum value of cumulative power generation per unit time - minimum value of cumulative power generation per unit time; If the calculated power generation per unit time is less than or equal to 0, the power generation per unit time is replaced by the average power per unit time; Summarize the number of alarms per unit time and the content with the highest number of alarms, and summarize and sum the currents of each string of the inverter; calculate the discrete rate of the string = the variance of the current of each string / the average value of the current of each string; The inverter's cumulative power generation is synchronized once a day. The calculation of the inverter's daily cumulative power generation includes the following process: Starting from time T every day, if the inverter has power generation, the real-time power per second is accumulated and converted into power generation; When the power generation is greater than the set threshold, the power generation is accumulated to the cumulative power generation of the day, and the accumulated power is assigned to 0; when the power generation is less than the set threshold, the power accumulation continues until the power generation is greater than the set threshold; The power generation is calculated one second before the T time point of the next day to obtain the cumulative power generation for the day.

2. The photovoltaic power generation prediction method based on big data according to claim 1, characterized in that: The data reception of the big data platform includes the following processes: The data of the inverter is collected once every t time interval by the collection device; The collected data is sent to the IoT platform in the form of a message queue and stored in a relational database after data cleaning and message format conversion. The data stored in the relational database is synchronized to the big data platform at a fixed time every day.

3. The photovoltaic power generation prediction method based on big data according to claim 1, characterized in that: Meteorological data include solar radiation, temperature, humidity and wind speed; Meteorological data is suitable for downloading from the weather forecast center. After the meteorological data is downloaded, it is uploaded to the big data platform for data matching after format conversion. The data matching process is as follows: Match the nearest weather data with the address location according to the longitude and latitude data of the inverter; The matched meteorological data is spliced ​​with the inverter data to obtain the weather, alarm, discrete rate and power generation data of the inverter at various historical time points.

4. The photovoltaic power generation prediction method based on big data according to claim 3, characterized in that: The process of processing the inverter matching data is as follows: Eliminate data with power generation of 0, abnormal power generation per unit time, and irradiance of null or 0; The weather characteristics, the alarm characteristics of the inverter itself, and the characteristics of the string discrete rate are normalized to obtain the characteristic data and power generation data of each inverter per unit time.

5. The photovoltaic power generation prediction method based on big data according to any one of claims 1 to 4, characterized in that: Model training includes the following steps: Splitting the characteristic data of the inverter into training data and calibration data; Use intelligent algorithms to train the power generation prediction model on the training data; The prediction accuracy of the power generation prediction model after training is evaluated through calibration data and calculated power generation data.

6. The photovoltaic power generation prediction method based on big data according to claim 5, characterized in that: The power generation forecast results are compared with the calibration data to evaluate the accuracy of the forecast results; the accuracy calculation formula is as follows: ； Among them, precision represents the accuracy, y represents the predicted value of power generation, represents the true value of power generation, E represents the actual startup capacity, and n is the number of samples.

7. A photovoltaic power generation prediction framework, used to implement the photovoltaic power generation prediction method based on big data according to any one of claims 1 to 6, characterized in that: Includes cloud,feature computing architecture and multiple inverters; The feature calculation architecture is adapted to receive power generation data of the inverter to perform feature calculation, and send the calculation result to the cloud; The cloud is suitable for training the power generation prediction model and predicting the power generation through the power prediction algorithm according to the result of feature calculation.

8. The photovoltaic power generation prediction framework according to claim 7, characterized in that: The plurality of inverters are arranged in a plurality of stations according to a set number; The number of the feature calculation architecture is one and is set in the cloud, and the feature calculation architecture is suitable for receiving power generation data of the inverters in all the substations; Alternatively, the number of the feature calculation architectures corresponds to the number of the substations, and each of the feature calculation architectures is disposed in a corresponding substation, and each of the feature calculation architectures receives power generation data of all the inverters in the corresponding substation.

Images