WO2020228568A1

WO2020228568A1 - Method for training power generation amount prediction model of photovoltaic power station, power generation amount prediction method and device of photovoltaic power station, training system, prediction system and storage medium

Info

Publication number: WO2020228568A1
Application number: PCT/CN2020/088709
Authority: WO
Inventors: 周希波; 李慧
Original assignee: 京东方科技集团股份有限公司
Priority date: 2019-05-14
Filing date: 2020-05-06
Publication date: 2020-11-19
Also published as: CN111950752A

Abstract

A method for training a power generation amount prediction model of a photovoltaic power station, a power generation amount prediction method and device of the photovoltaic power station, a training system, a prediction system and a storage medium. The power generation amount prediction method of the photovoltaic power station comprises: obtaining effective environment data, wherein the effective environment data comprises monitoring values of multiple preset environment features of the photovoltaic power station in one day; preprocessing the effective environment data to obtain an effective environment data feature vector; and inputting the effective environment data feature vector into a pre-constructed power generation amount prediction model, and outputting a prediction result corresponding to the effective environment data, wherein the prediction result is the effective power generation amount of the photovoltaic power station in one day.

Description

Method for training power generation forecast model of photovoltaic power station, method and device for power generation forecast of photovoltaic power station, training system, prediction system and storage medium

Cross references to related applications

This application claims the priority of the Chinese patent application No. 201910401500.9 filed on May 14, 2019, and the contents of the above-mentioned Chinese patent application are quoted here in full as a part of this application.

Technical field

The embodiments of the present disclosure relate to a method for training a photovoltaic power station power generation prediction model, a photovoltaic power station power generation prediction method and device, a training system, a prediction system, and a storage medium.

Background technique

The application of photovoltaic power generation in energy power systems is becoming more and more popular. The accurate prediction of photovoltaic power generation output has an important impact on grid security, power generation control and dispatch, and the operation and maintenance management of photovoltaic power plants.

Summary of the invention

At least one embodiment of the present disclosure provides a method for training a photovoltaic power station power generation prediction model, which includes:

Acquiring historical power generation data per day of the photovoltaic power station for multiple days and historical environmental data obtained by historical environmental data per day of the multiple days;

Preprocessing the historical environmental data; and

A training set and a prediction set are constructed based on the historical power generation data and the preprocessed historical environmental data, and multiple regression algorithms are used for training respectively to construct the relationship between the historical environmental data and the historical power generation data The mapping relationship is used as the power generation prediction model, and the output result of the power generation prediction model is obtained by fusing the prediction results obtained according to the multiple regression algorithms.

At least one embodiment of the present disclosure provides a method for predicting power generation of a photovoltaic power station, the method including:

Acquiring effective environmental data, the effective environmental data including one day's monitoring values of multiple preset environmental characteristics of the photovoltaic power station;

Preprocessing the effective environmental data to obtain a feature vector of effective environmental data; and

Inputting the effective environmental data feature vector into a pre-built power generation prediction model, and outputting a prediction result corresponding to the effective environmental data, the prediction result being the effective power generation of the photovoltaic power station for the day,

Wherein, the power generation prediction model is trained through the following operations:

Acquiring multiple days of historical power generation data of the photovoltaic power station and daily historical environmental data of the multiple days;

Preprocessing the historical environmental data; and

A training set and a prediction set are constructed based on the historical power generation data and the preprocessed historical environmental data, and multiple regression algorithms are used for training respectively to construct the relationship between the historical environmental data and the historical power generation data The mapping relationship is used as the power generation prediction model, and the output result of the power generation prediction model is obtained by fusing the prediction results obtained according to the multiple regression algorithms,

The preprocessing of the historical environmental data includes:

Feature coding the daily historical environmental data;

Normalize the daily historical environmental data after feature encoding; and

Convert the normalized daily historical environmental data into linear independent historical environmental data feature vectors,

The daily historical environmental data includes monitoring values of multiple environmental characteristics; and

The feature encoding of the daily historical environmental data includes:

Based on the monitoring values of the multiple environmental features included in the daily historical environmental data, respectively calculate a statistical feature corresponding to each of the multiple environmental features, where the statistical feature includes a maximum value feature , At least one of the minimum characteristic, the average characteristic, and the standard deviation characteristic;

Using the monitored value of each environmental feature in the plurality of environmental features and the corresponding statistical feature as a vector of each environmental feature in the plurality of environmental features; and

Filtering the environmental features whose sparsity is greater than or equal to a sparsity threshold among the plurality of environmental features, and

The plurality of preset environmental characteristics include other environmental characteristics in the plurality of environmental characteristics except the filtered environmental characteristics.

At least one embodiment of the present disclosure provides a device for predicting power generation of a photovoltaic power station, the device including:

An obtaining unit configured to obtain effective environmental data, the effective environmental data including one day's monitoring values of multiple preset environmental characteristics of the photovoltaic power station;

A preprocessing unit, configured to preprocess the effective environmental data to obtain a feature vector of effective environmental data; and

A prediction unit configured to input the effective environmental data feature vector into the pre-built power generation prediction model, and output a prediction result corresponding to the effective environmental data, where the prediction result is the day of the photovoltaic power station Effective power generation,

Acquiring daily historical power generation data of the photovoltaic power station for multiple days and daily historical environmental data of the multiple days;

Preprocessing the historical environmental data; and

The preprocessing of the historical environmental data includes:

Feature coding the daily historical environmental data;

Normalize the daily historical environmental data after feature encoding; and

The feature encoding of the daily historical environmental data includes:

Filtering the environmental features whose sparsity is greater than or equal to the sparsity threshold among the plurality of environmental features, and

At least one embodiment of the present disclosure provides a training system, which includes:

Processor; and

Memory, configured to store one or more computer programs;

Wherein, when the one or more computer programs are executed by the processor, the processor is caused to execute the method for training a photovoltaic power generation forecast model according to any embodiment of the present disclosure.

At least one embodiment of the present disclosure provides a prediction system, which includes:

Processor; and

Memory, configured to store one or more computer programs;

Wherein, when the one or more computer programs are executed by the processor, the processor is caused to execute the method for predicting power generation of a photovoltaic power station according to any embodiment of the present disclosure.

At least one embodiment of the present disclosure provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the training of the photovoltaic power generation prediction model according to any embodiment of the present disclosure is realized. method.

At least one embodiment of the present disclosure provides a non-transitory computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the method for predicting power generation of a photovoltaic power station according to any embodiment of the present disclosure is implemented.

Description of the drawings

In order to explain the technical solutions of the embodiments of the present disclosure more clearly, the drawings of the embodiments will be briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present disclosure, rather than limit the present disclosure.

Fig. 1 shows a schematic flow chart of a method for predicting power generation of a photovoltaic power station provided by at least one embodiment of the present disclosure;

Fig. 2 shows a schematic flow chart of a method for constructing a power generation prediction model provided by at least one embodiment of the present disclosure;

Fig. 3 shows a schematic structural diagram of a photovoltaic power station generating capacity prediction device provided by at least one embodiment of the present disclosure;

FIG. 4 shows a schematic structural diagram of a preprocessing unit provided by at least one embodiment of the present disclosure;

Figure 5 shows a schematic structural diagram of a system provided by at least one embodiment of the present disclosure; and

Fig. 6 shows a schematic diagram of a computer system suitable for implementing a method for training a photovoltaic power station power generation prediction model, a photovoltaic power generation prediction method or device, a training system, and a computer system for the prediction system according to an embodiment of the disclosure.

Detailed ways

In order to make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions of the embodiments of the present disclosure will be described clearly and completely in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present disclosure, rather than all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the protection scope of the present disclosure.

In addition to the hardware structure of photovoltaic power plants, photovoltaic panel arrays, power conversion components and other hardware structures, the power generation of photovoltaic power plants is also affected by factors such as sunlight, weather, and precipitation. However, the records of monitoring environmental information hardware and photovoltaic power plant hardware may have inconsistencies in granularity, or records may be missing due to hardware equipment shutdown or operation and maintenance. The existence of these problems makes accurate prediction of photovoltaic power generation capacity face severe challenges.

Please refer to FIG. 1, which shows a schematic flowchart of a method for predicting power generation of a photovoltaic power station according to at least one embodiment of the present disclosure.

As shown in Figure 1, the method includes:

Step 110: Obtain valid environmental data.

Step 120, preprocess the effective environmental data to obtain the effective environmental data feature vector;

Step 130: Input the effective environmental data feature vector into the pre-built power generation prediction model, and output a prediction result corresponding to the effective environmental data. The prediction result is the daily effective power generation of the photovoltaic power station. The power generation prediction model can be trained by the method of training a photovoltaic power station power generation prediction model described below, which will not be repeated here.

In the embodiment of the present disclosure, the effective environmental data includes one-day monitoring values of multiple preset environmental characteristics of the photovoltaic power station. For example, effective environmental data refers to the monitoring values of multiple preset environmental characteristics of photovoltaic power plants obtained according to the granularity period during the working hours of the day. The granularity period is defined as 1 hour, 30 minutes, etc. The daily working hours can be determined according to the sunshine hours in different regions. For example, in Beijing, China, the daily working hours can be expressed from 7:00 to 18:00 Beijing time. For example, when the granularity period is 1 hour, the aforementioned effective environmental data may include hourly monitoring values of multiple preset environmental characteristics of the photovoltaic power station obtained between 7:00 and 18:00.

The aforementioned preset environmental characteristics may include multiple of the following environmental characteristics, such as wind direction, wind speed, temperature, humidity, atmospheric pressure, precipitation, total radiation, direct radiation, diffuse radiation, exposure, sunshine hours, etc. The monitoring values of these environmental characteristics can be recorded by the power station environmental monitor. Hereinafter, the preset environment characteristics will be described with reference to the method for training a photovoltaic power station power generation prediction model according to at least one embodiment of the present disclosure.

The forecast target is the daily effective power generation of photovoltaic power stations. For example, the difference between the historical cumulative positive active power generation of the power station recorded from the zero point of the next day to the zero point of the day.

In some technical solutions, the monitoring of environmental data is performed 24 hours a day without interruption, and the effective working time of photovoltaic power plants in actual work is related to the length of sunshine. For example, the actual effective working hours of photovoltaic power plants in Beijing, China may be 7:00-18:00 daily. Therefore, when predicting the effective power generation of photovoltaic power plants, effective environmental data is one of the important factors that affect the accuracy of the prediction results. In the embodiments of the present disclosure, the idea of using effective environmental data to predict the daily positive active power generation of a photovoltaic power station is proposed, where the effective environmental data collection granularity may be a fixed period, such as 1 hour.

After obtaining valid environmental data, it is necessary to preprocess the valid environmental data, which may include the following steps:

Characterize effective environmental data;

Normalize the effective environmental data after feature encoding;

Convert the normalized effective environmental data into linear independent effective environmental data feature vectors.

Among them, the feature encoding of effective environmental data includes the following steps:

Based on the daily monitoring values of multiple preset environmental features of the photovoltaic power station included in the valid environmental data, the statistical feature corresponding to each of the multiple preset environmental features is calculated respectively, and the statistical feature includes the maximum value feature , At least one of the minimum characteristic, the average characteristic, and the standard deviation characteristic; and

Use the monitored value of each of the plurality of preset environmental features and the corresponding statistical feature as a vector of each of the plurality of preset environmental features.

For example, the one-day monitoring values of multiple preset environmental characteristics of the photovoltaic power station included in the effective environmental data are obtained according to the granularity period during the working hours of the day. For example, each preset environmental feature in the effective environmental data is flattened into a 12-dimensional vector according to the hourly monitoring value of a single day, and the statistical feature corresponding to each preset environmental feature is counted. The statistical feature may include the maximum value feature , Minimum feature, average feature, standard deviation feature, and then integrate each environmental feature with the statistical feature, and get the form of single-day hourly monitoring value, maximum feature, minimum feature, average feature, and standard deviation feature The 16-dimensional vector is used as the vector of each environment feature. Finally, after normalizing each column of environmental features, that is, the vector of each environmental feature, the principal component analysis method is used to convert multiple environmental features into a set of linear independent effective environmental data feature vectors.

For further explanation, the preprocessing process of the effective environmental data in the embodiments of the present disclosure is illustrated as follows. It is assumed that the effective environmental data obtained at 8:00 am is as shown in the following table:

Environmental characteristics refer to wind direction, wind speed, temperature, humidity, atmospheric pressure, precipitation, total radiation, direct radiation, diffuse radiation, exposure, sunshine hours, etc. Each environmental characteristic is that the environmental monitor obtains real data according to hourly monitoring, that is, the granularity period is 1 hour.

Among them, the environmental characteristics at other time points that are not monitored are filled with zeros to construct effective environmental data.

Then, statistical analysis is performed on each environmental feature to obtain a statistical feature corresponding to each environmental feature, and the statistical feature is the maximum, minimum, average, and standard deviation of each environmental feature. Then, these statistical features are integrated with the corresponding environmental features to obtain the vector of each environmental feature.

Then, after normalizing the environmental feature vectors, the principal component analysis method is used to convert all environmental features into a set of linear independent effective environmental data feature vectors.

The effective environmental data feature vector is input to the pre-built power generation prediction model, and the prediction result corresponding to the effective environmental data is output. The prediction result is the daily effective power generation of the photovoltaic power station. The method of training the power generation prediction model will be described in detail below.

The power generation prediction model provided in the embodiments of the present disclosure may be trained using multiple regression algorithms, for example, Bayesian Ridge regression algorithm, support vector regression algorithm, gradient boosting tree algorithm, and extreme random forest algorithm. That is, the power generation forecasting model includes sub-models constructed according to Bayesian Ridge regression algorithm, support vector regression algorithm, gradient boosting tree algorithm and extreme random forest algorithm. It should be understood that the aforementioned Bayesian ridge regression algorithm, support vector regression algorithm, gradient boosting tree algorithm, and extreme random forest algorithm are only exemplary, and the embodiments of the present disclosure are not limited thereto.

The output result of the power generation prediction model is obtained by fusing the prediction results obtained according to multiple regression algorithms. For example, the output result of the power generation prediction model may be a weighted average of the prediction results obtained according to multiple regression algorithms, which is not limited in the embodiment of the present disclosure.

The embodiment of the present disclosure unifies the environmental monitoring statistical period with the statistical period of the photovoltaic power station power generation, and performs refined preprocessing on the environmental monitoring data, so that the statistical results are more accurate.

Please refer to FIG. 2, which shows a schematic flowchart of a method for training a photovoltaic power station power generation prediction model provided by at least one embodiment of the present disclosure. As shown in Figure 2, the method includes:

Step 201: Obtain daily historical power generation data of the photovoltaic power station for multiple days and historical environmental data for each day of the multiple days;

Step 202, preprocessing historical environmental data;

Step 203: Construct a training set and a prediction set based on historical power generation data and preprocessed historical environmental data, and use multiple regression algorithms for training respectively to construct a mapping relationship between historical environmental data and historical power generation data as power generation The prediction model, the output result of the power generation prediction model is obtained by fusing the prediction results obtained according to the above-mentioned multiple regression algorithms.

For example, the multiple regression algorithms may include at least two of the following algorithms: Bayesian Ridge Regression Algorithm, Support Vector Regression Algorithm, Gradient Boosting Tree Algorithm, and Extreme Random Forest Algorithm. However, it should be understood that the embodiments of the present disclosure do not deal with this. limit.

For example, in at least one embodiment, preprocessing historical environmental data may include:

Characterize daily historical environmental data;

Normalize the daily historical environmental data after feature encoding; and

Convert the normalized daily historical environmental data into linear independent historical environmental data feature vectors.

For example, the aforementioned daily historical environmental data includes daily monitoring values of multiple environmental characteristics. For example, the multiple environmental characteristics may include wind direction, wind speed, temperature, humidity, atmospheric pressure, precipitation, total radiation, direct radiation, diffuse radiation, exposure, and sunshine hours. However, it should be understood that the embodiments of the present disclosure are not limited to this.

For example, the aforementioned daily historical environmental data may be historical environmental data obtained during daily working hours according to a granular cycle. In the case where the aforementioned daily historical environmental data includes the daily monitoring values of multiple environmental characteristics, the aforementioned daily historical environmental data may include the monitoring values of multiple environmental characteristics obtained during daily working hours according to the granularity period. In addition, it should be understood that the granularity period for obtaining historical environmental data should be the same as the granularity period for obtaining valid environmental data described above.

In the embodiment of the present disclosure, it is assumed that the historical power generation data of the photovoltaic power station in the preset time range and the historical environmental data obtained during the daily working hours in the preset time range are used as the sample data set, and the sample data set is proportional to Divided into training set and prediction set, the division ratio is for example 7:3. The preset time range may be 120 days, 30 days, etc., for example.

It is assumed that the historical power generation of the photovoltaic power station within 30 days and the historical environmental data recorded by the power station environment monitor every day according to the granularity cycle within the 30 days are acquired. Among them, the historical power generation data correspond to the daily positive active power generation of each photovoltaic power station. The historical power generation can be read from the electricity meter of the photovoltaic power station. Assume that Y represents historical power generation data, such as Y={y ₁ , y ₂ ,..., y _n }, where n represents the maximum value of the preset time range.

Correspondingly, historical environmental data is read from environmental monitors installed around the photovoltaic power station. Suppose X represents historical environmental data, such as X={x ₁ , x ₂ ,..., x _n }, where n represents the maximum value of the preset time range, and each x can be as shown in Table 1 below:

Table 1

For example, in at least one embodiment, the feature encoding of daily historical environmental data includes:

Based on the monitoring values of multiple environmental features included in daily historical environmental data, respectively calculate the statistical features corresponding to each of the multiple environmental features. The statistical features can include maximum, minimum, and average values. , At least one of the standard deviation features;

Using the monitored value and corresponding statistical feature of each of the multiple environmental features as the vector of each of the multiple environmental features; and

Filter the environmental features whose sparsity is greater than or equal to the sparsity threshold among the multiple environmental features.

For example, statistical analysis is performed on each environmental feature in Table 1 to obtain a statistical feature corresponding to each environmental feature, and the statistical feature is the maximum, minimum, average, and standard deviation of each environmental feature. Then, these statistical features are integrated with the corresponding environmental features to obtain Table 2, where each column in Table 2 forms a vector for each environmental feature. That is, for each environmental feature, a 16-dimensional vector formed by a single day's hourly monitoring value, maximum feature, minimum feature, average feature, and standard deviation feature is used as the vector of each environmental feature.

Table 2

For example, in at least one embodiment, filtering environmental features whose sparsity is greater than or equal to a sparsity threshold includes:

The ratio between the number of data equal to zero in the vector of each environmental feature and the total number of data in the vector of each environmental feature is taken as the sparsity of each environmental feature.

The sparsity threshold ε can be set according to requirements. For example, you can judge whether the sparsity is greater than or equal to the sparsity threshold according to the following formula:

For example, in at least one embodiment, normalizing effective environment data after feature encoding includes normalizing a vector of environment features whose sparsity is less than a sparsity threshold.

For example, where the sparsity of one or more environmental features in wind direction, wind speed, temperature, humidity, atmospheric pressure, precipitation, total radiation, direct radiation, diffuse radiation, exposure, and sunshine hours is greater than or equal to the sparsity threshold In this case, the one or more environmental features are filtered, and the vectors of the remaining environmental features are normalized.

It should be understood that the preset environmental characteristics used in the method for predicting power generation of a photovoltaic power plant according to at least one embodiment of the present disclosure are environmental characteristics other than the filtered environmental characteristics among the multiple environmental characteristics included in the historical environmental data.

For example, in at least one embodiment, converting the normalized daily historical environmental data into linear independent historical environmental data feature vectors includes:

Principal component analysis is used to convert the normalized daily historical environmental data into linear independent historical environmental data feature vectors.

Filter the environmental features whose sparsity is greater than or equal to the sparsity threshold, and after normalizing the remaining environmental features, use principal component analysis to convert all environmental features into a set of linear independent historical environmental data feature vectors

among them,

Expressed as t _1,1 ,t _1,2 ,...t _1,16 , where t _1,i is the value of the first environmental feature corresponding to different monitoring moments, for example, when i is 1-12, it is used to indicate The monitoring value read every hour in the time range of 7:00-18:00 for each environmental feature. When the value of i is 13-16, it is used to represent the maximum, minimum, and average of each environmental feature Value and standard deviation.

Then, the historical environment data feature vectors are input as training samples to multiple regression algorithms for training. For example, the multiple regression algorithms may include but are not limited to at least two of Bayesian Ridge regression algorithm, support vector regression algorithm, gradient boosting tree algorithm, and extreme random forest algorithm.

Bayesian Ridge Regression Algorithm is a biased estimation regression method for collinearity data analysis.

Assuming that the sample set is X={x ₁ , x ₂ ,..., x _n }, the parameter W obeys Gaussian distribution

The accuracy of the data noise is β, β ^-1 corresponds to the variance of the Gaussian distribution of the sample set X, α ^-1 corresponds to the variance of the Gaussian distribution of the parameter W, and h _W (X)-Y obeys the Gaussian distribution

Then the logarithmic posterior probability distribution formula of the linear model is:

Among them, θ is the posterior probability under the condition of the known sample set X, T is the target value vector of the sample data, const is a constant independent of the parameter W, and Y is the historical power generation data, such as Y={y ₁ , y ₂ ,...,y _n }. The learning process of the Bayesian Ridge Regression algorithm is: multiply the posterior probability of the previous training set by the likelihood estimate of the new test sample point to obtain the posterior probability of the new training set. During the training process, X is drawn from the full set of samples. As the sample size increases, the predicted value h _W (X) gradually converges and stabilizes, and finally the stable predicted value h _W (X) is output.

Support vector regression algorithm is also a regression method. The support vector regression algorithm uses support vector ideas and Lagrangian multiplier methods to perform regression analysis on data when fitting.

Assuming that the sample set is X={x ₁ , x ₂ ,..., x _n }, the support vector regression algorithm constructs a constrained optimization problem:

The support vector regression model makes the objective function reach the minimum value under constraint conditions through iteration. Assuming that the fitting function between the sample set and the predicted value is y=h(x)=W ^T X, the error between the predicted value and the true value in the test set is constructed as the objective function, and the objective function reaches the extreme under the above constraints through iteration. Small value. The sequence minimum optimization algorithm can be used to find the corresponding α,α ^{* when} the above formula is minimum. The algorithm selects two variables through a two-layer loop to maximize the decrease of the objective function, and then updates the regression model parameters. When the termination condition is met within the accuracy range ε, the algorithm ends. Output h(x) under the optimal parameters as the prediction result.

The gradient boosting tree (GBRT) algorithm is an iterative learning model based on the CART regression tree.

Assuming that the training sample is X={x ₁ , x ₂ ,..., x _n }, the maximum number of iterations is T, and the loss function is L. GBRT first initializes the weak learner:

Among them, y _i is the value of historical power generation data Y, such as Y={y ₁ , y ₂ ,..., y _n }.

In each round of iteration, the goal of the algorithm is to find a CART regression tree that minimizes the loss function of this round. For each iteration, calculate the negative gradient for the sample:

Use (X,r _t ) to fit the CART regression tree, the corresponding leaf node area is R _tj ,j=1, 2,...,k, calculate the best fit value:

According to the fitted value, update the strong learner:

After T rounds of iteration, the final regression model is obtained:

The extreme random forest algorithm is to select all training samples X={x ₁ , x ₂ ,..., x _n }, randomly select features according to a specified percentage, build K CART trees, and then perform regression prediction on the test samples, and take K trees The average of the predicted results is used as the final predicted result. That is, output f _t (X) as the final prediction result.

Extreme random forest uses all samples every time it generates a decision tree, and the decision tree split is completely random. When the feature attribute is in category form, different categories are randomly clustered together as branches; when the feature attribute is in numerical form, any number between the maximum and minimum values of the feature attribute is randomly selected as the bifurcation value to establish Branch. When the decision tree is split, all feature attributes in the node are traversed, the branches of all feature attributes are obtained according to the above method, the mean square error is calculated, and the feature with the largest mean square error is finally selected for splitting.

When predicting the power generation of a photovoltaic power station, a proportional voting model is used to fuse the prediction results of the sub-models constructed by the above various algorithms to obtain the final prediction results.

The proportional voting model is used to fuse the prediction results of the sub-models constructed by the above various algorithms to obtain the final prediction results, including: weighted average of the prediction results obtained by each regression algorithm, and the weighted average of the prediction results as the final forecast result. It should be understood that the weighted average of the prediction results obtained by each regression algorithm is only an example of fusing the prediction results obtained by each regression algorithm. In other embodiments, the prediction results obtained by various regression algorithms can also be fused in other ways, which is not limited in the embodiments of the present disclosure.

The power generation prediction model constructed in the embodiments of the present disclosure is trained using multiple regression algorithms, which can effectively prevent overfitting problems, and obtain the final prediction results by fusing the prediction results obtained by multiple regression algorithms, thereby effectively Improve forecast accuracy.

The embodiments of the present disclosure can overcome the problem of granularity inconsistency between the power generation of photovoltaic power plants and environmental data. They use training feature granularity to be small measurement units, such as hours, minutes, etc., and predict targets to be large measurement units, such as days. The embodiment of the present disclosure also obtains multiple prediction results through multiple different regression models, and obtains the final prediction result according to a proportional voting method, which can obtain better prediction accuracy without causing over-fitting problems.

It should be noted that although the operations of the method of the present disclosure are described in a specific order in the drawings, this does not require or imply that these operations must be performed in the specific order, or that all the operations shown must be performed to achieve the desired result. . Conversely, the steps depicted in the flowchart can change the order of execution. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step for execution, and/or one step may be decomposed into multiple steps for execution.

Referring to Fig. 3, Fig. 3 shows a schematic structural diagram of a photovoltaic power station generating capacity prediction device 300 provided by at least one embodiment of the present disclosure. As shown in Figure 3, the photovoltaic power station generating capacity prediction device 300 includes:

The obtaining unit 301 is configured to obtain valid environmental data. The effective environmental data includes one day's monitoring values of multiple preset environmental characteristics of the photovoltaic power station.

The preprocessing unit 302 is configured to preprocess valid environmental data to obtain a feature vector of valid environmental data;

The prediction unit 303 is configured to input the effective environmental data feature vector into a pre-built power generation prediction model, and output a prediction result corresponding to the effective environmental data. The prediction result is the effective power generation of the photovoltaic power station in a day. The power generation prediction model can be trained by the method of training the photovoltaic power station power generation prediction model described above, which will not be repeated here.

In the embodiments of the present disclosure, the effective environmental data refers to the monitoring values of multiple preset environmental characteristics of the photovoltaic power station obtained according to the granularity period during the working hours of a day. The granularity period is defined as 1 hour, 30 minutes, etc. The daily working hours can be determined according to the sunshine hours in different regions. For example, in Beijing, China, the daily working hours can be expressed from 7:00 to 18:00 Beijing time. For example, when the granularity period is 1 hour, the aforementioned effective environmental data may include hourly monitoring values of multiple preset environmental characteristics of the photovoltaic power station obtained between 7:00 and 18:00.

The preset environmental characteristics in the aforementioned effective environmental data may include multiple environmental characteristics, such as wind direction, wind speed, temperature, humidity, atmospheric pressure, precipitation, total radiation, direct radiation, diffuse radiation, exposure, and sunshine hours Wait. The monitoring values of these environmental characteristics can be recorded by the power station environmental monitor.

For the obtaining unit 301, the preprocessing unit 302, and the prediction unit 303, please refer to the detailed description of step 110, step 120 and step 130 above, which will not be repeated in the embodiment of the present disclosure.

Referring to FIG. 4, in at least one embodiment, the preprocessing unit 302 may further include:

The feature encoding subunit 3021 is configured to perform feature encoding on effective environment data;

The normalization subunit 3022 is configured to normalize the effective environment data after feature encoding;

The dimensionality reduction processing subunit 3023 is configured to convert the normalized effective environment data into a linear independent effective environment data feature vector.

For example, in at least one embodiment, the feature encoding subunit 3021 is further configured to: based on one day's monitoring values of multiple preset environmental features of the photovoltaic power station included in the effective environmental data, respectively calculate the multiple preset environmental features The statistical feature corresponding to each preset environmental feature; and filtering the environmental feature or statistical feature whose sparsity is greater than or equal to the sparsity threshold, the statistical feature includes at least one of the maximum feature, the minimum feature, the average feature, and the standard deviation feature One; and using the monitored value of each of the plurality of preset environmental features and the corresponding statistical feature as the vector of each preset environmental feature.

For example, in at least one embodiment, the normalization subunit 3022 is further configured to normalize the vector of each preset environment feature.

For example, in at least one embodiment, the dimensionality reduction processing subunit 3023 is further configured to adopt a principal component analysis method to convert the normalized effective environmental data into a linear independent effective environmental data feature vector.

Please refer to FIG. 5, which shows a schematic structural diagram of a system provided by at least one embodiment of the present disclosure. The system shown in Figure 5 may be a predictive system. As shown in Figure 5, the prediction system may include:

The power station environment monitoring device 401 is configured to monitor the photovoltaic power station to obtain environmental data of the photovoltaic power station;

The memory 402 is configured to store one or more computer programs; and

Processor 403.

When one or more computer programs stored in the memory 402 are executed by the processor 403, the processor 403 is caused to execute the method for predicting the power generation amount of the photovoltaic power plant described in the foregoing embodiment.

The power station environment monitoring device 401 may be connected to the memory 402 and the processor 403 by wired or wireless connection. Wired and wireless connections mainly implement communication functions for data transmission and control signaling transmission. The power station environment monitoring device 401 may include various sensors for detecting various environmental characteristics, for example.

However, it should be understood that, in some embodiments, the photovoltaic power station prediction system may not include the power station environment monitoring device 401. For example, the environmental data can be manually input or the environmental data can be stored in the memory in advance.

In some embodiments, the power station environment monitoring device 401 may be configured to monitor multiple preset environmental characteristics of the photovoltaic power station to obtain one day's monitoring value of the multiple preset environmental characteristics of the photovoltaic power station. For example, the power station environment monitoring device 401 may be configured to monitor multiple preset environmental characteristics of the photovoltaic power station, so as to obtain the monitoring values of the multiple preset environmental characteristics of the photovoltaic power station according to the granular period during the working hours of the day.

The memory 402 and the processor 403 may be implemented as the same device, or may exist independently. The processor 403 may be implemented as a functional structure for implementing the method for predicting the power generation amount of a photovoltaic power plant described in the foregoing embodiment.

It should be understood that the system shown in FIG. 5 may also be a training system, which is used to implement the method for training a photovoltaic power station power generation prediction model described in the foregoing embodiment.

For example, when one or more computer programs stored in the memory 402 are executed by the processor 403, the processor 403 is caused to execute the method for training a photovoltaic power station power generation prediction model described in the foregoing embodiment.

In addition, the power station environment monitoring device 401 may be configured to monitor the photovoltaic power station to obtain daily historical environmental data of the photovoltaic power station for multiple days. For example, the power station environment monitoring device 401 may be configured to obtain historical environmental data according to the granular period during daily working hours of multiple days.

It should be understood that the units or modules recorded in the photovoltaic power station power generation forecasting device 300 correspond to the steps in the method described with reference to FIG. 1. Therefore, the operations and features described above for the method in FIG. 1 are also applicable to the device 300 and the units contained therein, and will not be repeated here. The apparatus 300 may be implemented in a browser or other security application of an electronic device in advance, or it may be loaded into the browser of the electronic device or its security application by downloading or the like. Corresponding units in the photovoltaic power station generating capacity prediction device 300 can cooperate with units in electronic equipment to implement the solutions of the embodiments of the present disclosure.

In the several modules or units mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above may be embodied in one module or unit. Conversely, the features and functions of a module or unit described above can be further divided into multiple modules or units to be embodied.

6, which shows the structure of a computer system 500 suitable for implementing a method for training a photovoltaic power station power generation prediction model, a photovoltaic power generation prediction method or device, a training system or a prediction system according to an embodiment of the present disclosure. Schematic.

As shown in FIG. 6, the computer system 500 includes a central processing unit (CPU) 501, which can be based on a program stored in a read only memory (ROM) 502 or a program loaded from a storage portion 508 into a random access memory (RAM) 503 And perform various appropriate actions and processing. In the RAM 503, various programs and data required for the operation of the system 500 are also stored. The CPU 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. An input/output (I/O) interface 505 is also connected to the bus 504.

The following components are connected to the I/O interface 505: an input part 506 including a keyboard, a mouse, etc.; an output part 507 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage part 508 including a hard disk, etc. ; And a communication section 509 including a network interface card such as a LAN card, a modem, etc. The communication section 509 performs communication processing via a network such as the Internet. The driver 510 is also connected to the I/O interface 505 as needed. A removable medium 511, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 510 as needed, so that the computer program read therefrom is installed into the storage portion 508 as needed.

In particular, according to an embodiment of the present disclosure, the process described above with reference to FIGS. 1 and 2 may be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a machine-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program may be downloaded and installed from the network through the communication part 509, and/or installed from the removable medium 511. When the computer program is executed by the central processing unit (CPU) 501, the above-mentioned functions defined in the system of the present disclosure are executed.

It should be noted that the computer-readable medium shown in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or a combination of any of the above. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable Programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or in combination with an instruction execution system, apparatus, or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as a part of a carrier wave, and a computer-readable program code is carried therein. This propagated data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. The computer-readable signal medium may also be any computer-readable medium other than the computer-readable storage medium. The computer-readable medium may send, propagate, or transmit the program for use by or in combination with the instruction execution system, apparatus, or device . The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions, and operations of the system, method, and computer program product according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, program segment, or part of the code, and the aforementioned module, program segment, or part of the code contains one or more for realizing the specified logic function Executable instructions. It should also be noted that, in some alternative implementations, the functions marked in the block may also occur in a different order from the order marked in the drawings. For example, two blocks shown in succession can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and/or flowchart, and the combination of the blocks in the block diagram and/or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations Or it can be realized by a combination of dedicated hardware and computer instructions.

The units or modules involved in the embodiments described in the present disclosure can be implemented in software or hardware. The described unit or module may also be provided in the processor. For example, it may be described as: a processor includes an acquisition unit, a preprocessing unit, and a prediction unit. Wherein, the names of these units or modules do not constitute a limitation on the units or modules themselves under certain circumstances. For example, the acquiring unit can also be described as "a unit for acquiring effective environmental data."

As another aspect, the present disclosure also provides a computer-readable storage medium. The computer-readable storage medium may be included in the electronic device described in the above embodiment; or it may exist alone without being assembled into the In electronic equipment. The computer readable storage medium stores one or more programs, and when the foregoing programs are used by one or more processors to execute the method for predicting the power generation amount of a photovoltaic power station described in the present disclosure.

The foregoing descriptions are merely exemplary implementations of the present disclosure, and are not used to limit the protection scope of the present disclosure, which is determined by the appended claims.

Claims

A method for training a photovoltaic power station's power generation forecast model, including:

Acquiring daily historical power generation data of the photovoltaic power station for multiple days and daily historical environmental data of the multiple days;

Preprocessing the historical environmental data; and

A training set and a prediction set are constructed based on the historical power generation data and the preprocessed historical environmental data, and multiple regression algorithms are used for training respectively to construct the relationship between the historical environmental data and the historical power generation data The mapping relationship is used as the power generation prediction model, and the output result of the power generation prediction model is obtained by fusing the prediction results obtained according to the multiple regression algorithms.
The method according to claim 1, wherein the multiple regression algorithms comprise at least two selected from the group consisting of Bayesian Ridge regression algorithm, support vector regression algorithm, gradient boosting tree algorithm, and extreme random forest algorithm.
The method according to claim 1 or 2, wherein said preprocessing said historical environmental data comprises:

Feature coding the daily historical environmental data;

Normalize the daily historical environmental data after feature encoding; and

The normalized daily historical environmental data is converted into linear independent historical environmental data feature vectors.
The method of claim 3, wherein:

The daily historical environmental data includes monitoring values of multiple environmental characteristics; and

The feature encoding of the daily historical environmental data includes:

Based on the monitoring values of the multiple environmental features included in the daily historical environmental data, respectively calculate a statistical feature corresponding to each of the multiple environmental features, where the statistical feature includes a maximum value feature , At least one of the minimum characteristic, the average characteristic, and the standard deviation characteristic;

Using the monitored value of each environmental feature in the plurality of environmental features and the corresponding statistical feature as a vector of each environmental feature in the plurality of environmental features; and

Filter the environmental features whose sparsity is greater than or equal to the sparsity threshold among the multiple environmental features.
The method according to claim 4, wherein the filtering of the environmental features whose sparsity is greater than or equal to a sparsity threshold among the plurality of environmental features comprises:

The ratio between the number of data equal to zero in the vector of each of the plurality of environmental features and the total number of data in the vector of each of the plurality of environmental features is used as the number The sparsity of each of the environmental features.
The method according to claim 4 or 5, wherein the normalizing the daily historical environmental data after feature encoding includes:

Normalize the vector of the environmental feature whose sparsity is less than the sparsity threshold among the plurality of environmental features.
The method according to any one of claims 3-6, wherein the converting the normalized daily historical environmental data into a linear independent historical environmental data feature vector comprises:

A principal component analysis method is used to convert the normalized daily historical environmental data into the linear independent historical environmental data feature vector.
The method according to any one of claims 4-7, wherein the multiple environmental characteristics include wind direction, wind speed, temperature, humidity, atmospheric pressure, precipitation, total radiation, direct radiation, diffuse radiation, exposure, At least two of the sunshine hours.
The method according to any one of claims 1-8, wherein:

The daily historical environmental data is historical environmental data obtained during daily working hours according to the granularity cycle.
The method according to claim 9, wherein the particle size period is 1 hour or 30 minutes.
A method for predicting power generation of photovoltaic power stations, including:

Acquiring effective environmental data, the effective environmental data including one-day monitoring values of multiple preset environmental characteristics of the photovoltaic power station;

Preprocessing the effective environmental data to obtain a feature vector of effective environmental data; and

Inputting the effective environmental data feature vector into a pre-built power generation prediction model, and outputting a prediction result corresponding to the effective environmental data, the prediction result being the effective power generation of the photovoltaic power station for the day,

Wherein, the power generation prediction model is trained through the following operations:

Acquiring daily historical power generation data of the photovoltaic power station for multiple days and daily historical environmental data of the multiple days;

Preprocessing the historical environmental data; and

A training set and a prediction set are constructed based on the historical power generation data and the preprocessed historical environmental data, and multiple regression algorithms are used for training respectively to construct the relationship between the historical environmental data and the historical power generation data The mapping relationship is used as the power generation prediction model, and the output result of the power generation prediction model is obtained by fusing the prediction results obtained according to the multiple regression algorithms,

The preprocessing of the historical environmental data includes:

Feature coding the daily historical environmental data;

Normalize the daily historical environmental data after feature encoding; and

Convert the normalized daily historical environmental data into linear independent historical environmental data feature vectors,

The daily historical environmental data includes monitoring values of multiple environmental characteristics; and

The feature encoding of the daily historical environmental data includes:

Based on the monitoring values of the multiple environmental features included in the daily historical environmental data, respectively calculate a statistical feature corresponding to each of the multiple environmental features, where the statistical feature includes a maximum value feature , At least one of the minimum characteristic, the average characteristic, and the standard deviation characteristic;

Using the monitored value of each environmental feature in the plurality of environmental features and the corresponding statistical feature as a vector of each environmental feature in the plurality of environmental features; and

Filtering the environmental features whose sparsity is greater than or equal to the sparsity threshold among the plurality of environmental features, and

The plurality of preset environmental characteristics include other environmental characteristics in the plurality of environmental characteristics except the filtered environmental characteristics.
The method for predicting power generation of a photovoltaic power station according to claim 11, wherein said preprocessing said effective environmental data comprises:

Feature encoding the effective environmental data;

Normalize the effective environmental data after feature encoding; and

The normalized effective environmental data is converted into a linear independent effective environmental data feature vector.
The method for predicting power generation of a photovoltaic power station according to claim 12, wherein said performing feature coding on said effective environmental data comprises:

Based on the monitoring value of the day of the plurality of preset environmental characteristics of the photovoltaic power station included in the effective environmental data, respectively calculate the corresponding of each preset environmental feature of the plurality of preset environmental characteristics The statistical features of, the statistical features include at least one of the maximum feature, the minimum feature, the average feature, and the standard deviation feature; and

Use the monitored value of each preset environmental feature of the plurality of preset environmental features and the corresponding statistical feature as a vector of each preset environmental feature of the plurality of preset environmental features.
The method for predicting power generation of a photovoltaic power station according to claim 13, wherein said normalizing said effective environmental data after feature encoding comprises:

Normalize the vector of each preset environment feature.
The method for predicting power generation capacity of a photovoltaic power station according to any one of claims 12-14, wherein said converting the normalized effective environmental data into a linear independent effective environmental data feature vector comprises:

The principal component analysis method is used to convert the normalized effective environmental data into the linear independent effective environmental data feature vector.
The method for predicting the power generation capacity of a photovoltaic power station according to any one of claims 11-15, wherein the one-day monitoring values of the multiple preset environmental characteristics of the photovoltaic power station include The obtained monitoring values of the plurality of preset environmental characteristics of the photovoltaic power station.
A photovoltaic power station generating capacity prediction device, including:

An obtaining unit configured to obtain effective environmental data, the effective environmental data including one day's monitoring values of multiple preset environmental characteristics of the photovoltaic power station;

A preprocessing unit, configured to preprocess the effective environmental data to obtain a feature vector of effective environmental data; and

A prediction unit configured to input the effective environmental data feature vector into the pre-built power generation prediction model, and output a prediction result corresponding to the effective environmental data, where the prediction result is the day of the photovoltaic power station Effective power generation,

Wherein, the power generation prediction model is trained through the following operations:

Acquiring daily historical power generation data of the photovoltaic power station for multiple days and daily historical environmental data of the multiple days;

Preprocessing the historical environmental data; and

A training set and a prediction set are constructed based on the historical power generation data and the preprocessed historical environmental data, and multiple regression algorithms are used for training respectively to construct the relationship between the historical environmental data and the historical power generation data The mapping relationship is used as the power generation prediction model, and the output result of the power generation prediction model is obtained by fusing the prediction results obtained according to the multiple regression algorithms,

The preprocessing of the historical environmental data includes:

Feature coding the daily historical environmental data;

Normalize the daily historical environmental data after feature encoding; and

Convert the normalized daily historical environmental data into linear independent historical environmental data feature vectors,

The daily historical environmental data includes monitoring values of multiple environmental characteristics; and

The feature encoding of the daily historical environmental data includes:

Based on the monitoring values of the multiple environmental features included in the daily historical environmental data, respectively calculate a statistical feature corresponding to each of the multiple environmental features, where the statistical feature includes a maximum value feature , At least one of the minimum characteristic, the average characteristic, and the standard deviation characteristic;

Using the monitored value of each environmental feature in the plurality of environmental features and the corresponding statistical feature as a vector of each environmental feature in the plurality of environmental features; and

Filtering the environmental features whose sparsity is greater than or equal to the sparsity threshold among the plurality of environmental features, and

The plurality of preset environmental characteristics include other environmental characteristics in the plurality of environmental characteristics except the filtered environmental characteristics.
A training system including:

Processor; and

Memory, configured to store one or more computer programs;

Wherein, when the one or more computer programs are executed by the processor, the processor is caused to execute the method according to any one of claims 1-10.
The training system according to claim 18, further comprising: a power station environment monitoring device,

The power station environment monitoring device is configured to monitor the photovoltaic power station to obtain historical environmental data of the photovoltaic power station for each day in the multiple days.
A forecasting system including:

Processor; and

Memory, configured to store one or more computer programs;

Wherein, when the one or more computer programs are executed by the processor, the processor is caused to execute the method according to any one of claims 11-16.
The prediction system according to claim 20, further comprising: a power station environment monitoring device,

The power station environment monitoring device is configured to monitor the plurality of preset environmental characteristics of the photovoltaic power station to obtain a day's monitoring value of the plurality of preset environmental characteristics of the photovoltaic power station.
A non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method according to any one of claims 1-10 is realized.
A non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method according to any one of claims 11-16 is realized.