CN114492945A - Short-term photovoltaic power prediction method, medium and equipment in electric power market background - Google Patents

Abstract

A short-term photovoltaic power prediction method, medium and equipment under the background of the electricity market of the present invention, wherein the method includes the following steps: based on the NWP value, the weather is divided into four weather models: sunny, cloudy, showers and full rain; To measure the weather type of the day, select the most recent similar daily PV power historical data and historical NWP value according to the weather type of the day to be measured; input the classified data set into each basic learner in the first-layer prediction model of the Stacking integrated model; obtain each Based on the prediction result of the base learner, construct a new training set and input it into the second-layer prediction model; obtain the prediction result of the second-layer meta-learner, which is the final photovoltaic power generation power prediction value. The invention adopts the combined mechanism model and the data-driven algorithm, which can effectively reduce the dependence of prediction on data. In the case of unsatisfactory data quality and quantity, the accuracy, speed and reliability of prediction can be taken into account, and the acquisition in actual production can be greatly reduced. The cost of data has significant economic benefits.

Description

Short-term photovoltaic power forecasting method, medium and equipment in the context of electricity market

technical field

The invention relates to the technical field of photovoltaic power generation, in particular to a short-term photovoltaic power prediction method, medium and equipment under the background of the electricity market.

Background technique

Accurate photovoltaic power prediction plays a crucial role in power system scheduling. In response to this requirement, the present invention conducts research on short-term photovoltaic power prediction methods in the context of the power market.

The data-driven photovoltaic power generation forecasting method has simple modeling, mature algorithm and fast calculation speed, but it is completely based on numerical calculation, highly dependent on the quality and quantity of data, and does not consider the internal mechanism of photovoltaic power generation, so the reliability of the prediction results is poor. The mechanism-driven method can well reflect the internal power generation principle and coupling effect of photovoltaics, and can analyze the physical nature of photovoltaic power generation. Disadvantages such as strong time-varying. Data-driven forecasting methods, which are widely used in engineering, are highly dependent on data quality and quantity, and obtaining comprehensive and qualified data for new energy power generation systems is often costly.

In [1], an improved clear sky power model based on online update was proposed to make ultra-short-term forecasts for photovoltaic power under small fluctuation weather, and the prediction accuracy for small fluctuation weather at the scale of 3 to 4 hours was improved. Reference [2] proposed a combined model of photovoltaic output forecasting based on adaptive fuzzy time series, which uses adaptive algorithm to process historical power data, then goes through clustering, domain definition and division, fuzzification data, and finally combines fuzzy time series. The method predicts and defuzzifies the results, and simulates the data with a time interval of 15 minutes on the actual photovoltaic experimental system. The mean absolute error (MAE) value and the mean absolute percentage error (EMAPE) value reach 1.038MW and 13.34 respectively. %. Reference [3] uses the ExtraTreesRegressor method to evaluate the importance of features according to the correlation between photovoltaic output and external environmental factors. After cleaning the data and extracting feature quantities, the SVM based on the artificial fish swarm method is used to classify the meteorological data. Compared with the traditional SVM model, the prediction accuracy has been greatly improved.

Patent [4] discloses a photovoltaic power generation prediction system based on T-S fuzzy neural network, which realizes the organic combination of fuzzy inference system and neural network learning system, and introduces meteorological factors, which effectively improves the prediction accuracy and reliability. Patent [5] Clusters the daily historical processing data of photovoltaic power plants into K clusters, and constructs an improved generalized weather map that corresponds to one or more digital labels for meteorological professional weather, which overcomes the shortcomings of low prediction accuracy of photovoltaic power generation under non-clear conditions. . The patent [6] uses the weather scoring mechanism to classify the weather to classify the power generation data, and obtains the probability density function estimation of each type of data through the kernel density function, so as to give the distribution law of the power generation data in the statistical sense. Patent [7] discloses a photovoltaic power generation prediction system, which uses a mechanism model to obtain predicted power generation by detecting real-time weather conditions and the serial number, latitude and longitude, placement angle and photovoltaic material data of photovoltaic power generation equipment.

[1] Ma Yuan, Zhang Xuemin, Zhen Zhao, et al. Ultra-short-term photovoltaic power prediction method based on modified clear sky model [J]. Automation of Electric Power Systems, 2021, 45(11): 44-51;

[2] Yang Zhichao, Zhu Feng, Zhang Chenglong, et al. Short-term power prediction of photovoltaic power generation based on adaptive fuzzy time series method [J]. Journal of Nanjing Institute of Technology (Natural Science Edition), 2014, 12(1): 6-13;

[3] Wang Xiaoyang, Luo Duo, Sun Yunlin, et al. Research on the combined prediction method of daily power generation of photovoltaic microgrid based on ABC-SVM and PSO-RF [J]. Journal of Solar Energy, 2020, 41(03): 177-183 ;

[4] Lu Yuzheng, Wang Jun, Zhang Yaoming, Li Junjiao. A photovoltaic power generation prediction system based on T-S fuzzy neural network [P]. Jiangsu: CN103106544A, 2013-05-15;

[5] Zhang Dongliang, Yan Jian, Zong Zhaodan, Ren Xiaoda, Li Guoxin, Liu Jianhua. A photovoltaic power generation prediction method based on K-means clustering to improve generalized weather [P]. Jiangsu Province: CN106022538B, 2020-04-07;

[6] Ge Weichun, Pan Xiao, Li Jiajue, Zhang Tieyan, Ma Shaohua. Prediction method of photovoltaic power generation based on confidence interval of shape parameter [P]. Liaoning Province: CN108256690B, 2021-10-08;

[7] Liu Lele, Huang Le, Lin Dong. A prediction method of photovoltaic power generation prediction system [P]. Jiangsu Province: CN106909985B, 2021-02-09.

SUMMARY OF THE INVENTION

The short-term photovoltaic power prediction method, medium and device under the background of the electricity market proposed by the present invention can at least solve one of the problems of the background art.

To achieve the above object, the present invention has adopted the following technical solutions:

A short-term photovoltaic power forecasting method in the context of the electricity market, comprising the following steps,

Based on the NWP value, the weather is divided into four weather models: sunny, cloudy, shower and full rain;

Determine the weather type of the day to be measured, and select the most recent similar day's historical PV power data and historical NWP value according to the weather type of the day to be measured;

Input the classified data set into each basic learner in the one-layer prediction model of the Stacking ensemble model;

Obtain the prediction results of each base learner, construct a new training set and input the two-layer prediction model;

Obtain the prediction result of the two-layer meta-learner, which is the final photovoltaic power generation power prediction value.

Further, based on the NWP value, the weather is divided into four weather models: sunny day, cloudy day, shower and full rain; the details include:

The cloud amount C and the rainfall p in the NWP are selected as the weather classification factors. According to the average cloud amount during the day, it is divided into a sunny model and a cloudy model, and according to the daytime rainfall duration, it is divided into a shower model and a full rain model.

阴天模型为

阵雨模型为

全雨模型为

c₁为晴天和阴天模型的分型阈值，t₁为阵雨和全雨模型的分型阈值；根据短期天气预报国家标准及气象学原理，c₁＝0.7，t₁＝4。The sunny model is

The cloudy model is

The shower model is

The full rain model is

c ₁ is the classification threshold of sunny and cloudy models, t ₁ is the classification threshold of shower and full rain models; according to the national standard for short-term weather forecast and the principle of meteorology, c ₁ =0.7, t ₁ =4.

Further, the one-layer prediction model of the Stacking integrated model includes the physical characteristic model of photovoltaic power generation as shown in formula (1):

P=ηSI[1-0.005(t+25)] (1)

In the formula, P is the photovoltaic power generation, η is the conversion efficiency of the photovoltaic panel, S is the effective area of the photovoltaic panel, I is the irradiance, and t is the operating temperature of the photovoltaic panel.

Further, the one-layer prediction model also includes a TCN network prediction model, and the construction steps of the TCN network prediction model are as follows,

In the hole convolution step, each TCN layer contains L convolution layers, and the dilated convolution calculation formula is:

In the formula: the hole coefficient d=(1,...,2 ^L ), k is the size of the convolution kernel;

Residual linking steps:

Relu represents the linear rectification function, which is used as the activation function of the neural network; DCConv represents the hole convolution layer;

Formulas (3) and (4) represent the activation function of TCN:

In the formula: W ⁽¹⁾ and W ⁽²⁾ are the weight matrices corresponding to the input, b is the bias vector, S _{(i, j)} represents the activation function of the i-th layer of the j-th block, and formula (3) represents the hole at time t The result of the convolution, formula (4) represents the result after adding the residual link.

Further, the prediction results of each base learner are obtained, a new training set is constructed and a two-layer prediction model is input, using the following algorithm:

The histogram-based algorithm of the LightGBM model is used to alleviate the impact of high-dimensional data on prediction; the LightGBM model is optimized by the Leaf-wise algorithm with depth limitation;

The objective function of the LightGBM model is as follows:

Obj(t)=L(t)+Ω(t)+c (5)

In the formula: Obj(t) is the optimization objective, Ω(t) is the regular function, which reflects the complexity of the model; t is the sampling time; c is the extra parameter to avoid overfitting and optimize the tree depth;

的比较来反映模型的拟合度；定义如下：L(t) represents the loss function, which describes the actual value _yi and predicted value of N sampling points

to reflect the fit of the model; the definition is as follows:

由剩余树的累加生成。Residual information of previous learners is transmitted by serial coupling of regression trees; final output

Generated by the accumulation of the remaining trees.

Further, the prediction results of each base learner are obtained, a new training set is constructed and input into the two-layer prediction model, and the following algorithm is also used:

The gradient-based unilateral sampling algorithm GOSS is adopted. When sampling, GOSS will retain all the large gradient samples that meet the conditions as the sampled data, and adopt random sampling for the samples with smaller gradients.

Further, the prediction results of each base learner are obtained, a new training set is constructed and a two-layer prediction model is input, and the following algorithm is also used:

The mutually exclusive feature binding method EFB is adopted. After GOSS sampling, it will use the method of binding mutually exclusive features to reduce the dimension of features to prevent dimension disaster and improve computational efficiency.

Further, the one-layer prediction model of the Stacking ensemble model divides the data set into multiple subsets through cross-validation, and fuses the evaluation results to reduce the variance of the model prediction results, improve the generalization ability of the model, and avoid overfitting. happened;

K-fold cross-validation is to divide the data set into K parts on average, of which K-1 is used as a training set, and the remaining 1 is a validation set; the model hyperparameters are obtained by training with the training sets of the above K cases.

On the other hand, the present invention also discloses a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, causes the processor to execute the steps of the above method.

In yet another aspect, the present invention also discloses a computer device, comprising a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method .

It can be seen from the above technical solutions that, in order to overcome the deficiencies of the prior art, the present invention proposes a photovoltaic power prediction method with a multi-model fusion Stacking integrated learning method taking into account the mechanism model. Considering the differences in data observation and training principles of different algorithms, give full play to the advantages of each model, build multiple machine learning algorithms and combine the photovoltaic prediction model of Stacking integrated learning embedded in the photovoltaic mechanism model. The basic learners of the model include photovoltaic physical model, LightGBM algorithm and temporal convolutional network algorithms.

The short-term photovoltaic power prediction method under the background of the power market proposed by the present invention can effectively reduce the dependence of prediction on data by using a joint mechanism model and a data-driven algorithm. In the case of unsatisfactory data quality and quantity, the prediction can be Accuracy, speed and reliability, and greatly reduce the cost of data acquisition in actual production, with significant economic benefits. At the same time, it can effectively improve the accuracy of short-term forecasting of new energy, and play an important role in the formulation of dispatch plans, power market transactions, load management, etc. quality of grid-connected services.

Description of drawings

1 is a flow chart of a mechanism model and a data-driven combined photovoltaic power prediction according to an embodiment of the present invention;

Figure 2 is a schematic diagram of a time convolutional network structure;

Figure 3 is a schematic diagram of residual link;

Figure 4 is a schematic diagram of the histogram algorithm;

Fig. 5 is a Leaf-wise growth pattern diagram;

Figure 6 is the schematic diagram of the Stacking integration model;

FIG. 7 is a schematic diagram of cross-validation in this embodiment.

Detailed ways

In order to make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments These are some embodiments of the present invention, but not all embodiments.

The short-term photovoltaic power prediction method in the context of the electricity market described in this embodiment is as follows:

1.1 Mechanism model and data-driven combined photovoltaic power generation forecast

1.1.1 Weather classification based on NWP

分为晴天模型和阴天模型，按照白天降雨时长t分为阵雨模型和全雨模型。The photovoltaic power generation power is affected by the received solar irradiance on the day, and the received solar irradiance is affected by the weather type, and the photovoltaic power generation power fluctuates differently in different weathers. The cloud amount C and the rainfall p in the NWP were selected as weather classification factors. According to the average cloud cover during the day

It is divided into a sunny model and a cloudy model, and is divided into a shower model and a full rain model according to the rainfall duration t during the day.

Table 1 Weather classification model

In Table 1, c ₁ is the classification threshold for sunny and cloudy models, and t ₁ is the classification threshold for shower and full rain models. According to the national standard for short-term weather forecast and the principle of meteorology, c ₁ =0.7, t ₁ =4.

1.1.2 Stacking multi-model fusion short-term photovoltaic power generation prediction considering the mechanism model

Based on the above principle introduction, the photovoltaic physical power generation model is embedded in the first-level prediction model of the Stacking integrated model, and it is set as one of the basic learners to construct a Stacking multi-model fusion photovoltaic power generation prediction model that takes into account the mechanism model. The realization process of short-term photovoltaic power generation power prediction is given below, and the flow chart is shown in Figure 1.

Proceed as follows:

(1) Based on the NWP value, the weather is divided into four weather models: sunny, cloudy, shower and full rain.

(2) Judge the weather type of the day to be measured, and select the recent similar day's PV power historical data and historical NWP value according to the weather type of the to-be-measured day.

(3) Input the classified data set into each basic learner in the one-layer prediction model of the Stacking ensemble model.

(4) Obtain the prediction results of each base learner, construct a new training set and input the two-layer prediction model.

(5) Obtain the prediction result of the two-layer meta-learner, which is the final photovoltaic power generation power prediction value.

1.2 Physical model of photovoltaic power generation

The physical model of photovoltaic power generation predicts the generated power based on the characteristics of solar radiation and photoelectric conversion. The physical characteristic model of photovoltaic power generation is shown in formula (1).

P=ηSI[1-0.005(t+25)] (1)

In the formula, P is the photovoltaic power generation, η is the conversion efficiency of the photovoltaic panel, S is the effective area of the photovoltaic panel, I is the irradiance, and t is the operating temperature of the photovoltaic panel.

1.3 TCN network prediction model

The main structure of TCN can be divided into causal convolution suitable for sequences and atrous convolution plus residual module model suitable for historical data memory. Due to the causal relationship between its convolutional layers, it can memorize more historical data, which is suitable for the data of photovoltaic power plants.

(1) Hole convolution

The TCN structure is shown in Figure 2. Each TCN layer contains L convolutional layers. The dilated convolution calculation formula is:

In the formula: the hole coefficient d=(1,...,2 ^L ), and k is the size of the convolution kernel.

(2) Residual link

Figure 3 is a residual link diagram. Dropout means that in the process of neuron propagation, the activation value of a neuron stops working with a certain probability, thereby enhancing the generalization of the model. Relu represents the linear rectification function, which is used as the activation function of the neural network; DC Conv represents the atrous convolutional layer.

Formulas (3) and (4) represent the activation function of TCN:

In the formula: W ⁽¹⁾ and W ⁽²⁾ are the weight matrices corresponding to the input, b is the bias vector, S _{(i, j)} represents the activation function of the i-th layer of the j-th block, and formula (3) represents the hole at time t The result of the convolution, formula (4) represents the result after adding the residual link.

1.4LightGBM

1.4.1 Histogram Algorithm

The LightGBM model adopts a histogram-based algorithm to alleviate the influence of high-dimensional data on prediction, improve the calculation speed, and avoid the phenomenon of over-fitting of the prediction model. The basic idea of a histogram consists of converting consecutive floating-point eigenvalues into k integers, getting k to give "bins", and constructing a histogram of width k. The structure is shown in Figure 4.

1.4.2 Leaf-wise algorithm with depth limit

During the optimization process, LGBM uses the Leaf-wise algorithm to find suitable leaves, then splits, and repeats this cycle. As shown in Figure 5. LGBM adds a depth limit on Leaf-wise to prevent overfitting. The objective function of LGBM is as follows:

Obj(t)=L(t)+Ω(t)+c (5)

In the formula: Obj(t) is the optimization objective, and Ω(t) represents the regular function, which reflects the complexity of the model. t represents the sampling time. c stands for extra parameters to avoid overfitting and optimize tree depth.

的比较来反映模型的拟合度。定义如下：L(t) represents the loss function, which describes the actual value _yi and predicted value of N sampling points

to reflect the fit of the model. Defined as follows:

由剩余树的累加生成。Residual information from previous learners is transmitted by serially coupling regression trees. final output

Generated by the accumulation of the remaining trees.

1.4.3 Unilateral Gradient Sampling Algorithm

Gradient-based one-sided sampling algorithm (GOSS) is used. When sampling, GOSS will retain all the large gradient samples that meet the conditions as the sampled data, and randomly sample the samples with smaller gradients, which not only retains the insufficient training data, but can get more in the next training. At the same time, it will not make a huge change in the sample distribution.

1.4.4 Mutually Exclusive Feature Binding Algorithm

The method of mutually exclusive feature binding (EFB) is adopted. After GOSS sampling, it will use the method of binding mutually exclusive features to reduce the dimension of features to prevent dimensional disaster and improve computing efficiency. Because most high-dimensional data are sparse data, most of the features in the feature space are mutually exclusive, and the mutually exclusive features can be bound together to form new features to reduce the feature dimension.

1.5Stacking Integrated Learning Framework

1.5.1Stacking Integration Model

The Stacking ensemble model divides the original data set into several sub-data sets, which are input to each basic learner of the first-layer prediction model. And output the final result, the structure is shown in Figure 6. The Stacking ensemble model generalizes the output results of multiple models, learns the information about the combination of features, and effectively improves the overall prediction accuracy.

1.5.2 Cross-Validation

Cross-validation is often used to evaluate the predictive performance of a model. The one-layer prediction model of the Stacking ensemble model divides the data set into multiple subsets through cross-validation, and fuses the evaluation results to reduce the variance of the model prediction results, improve the generalization ability of the model, and avoid the occurrence of overfitting. K-fold cross-validation is to divide the data set into K parts on average, of which K-1 is used as the training set and the remaining 1 is the validation set. The model hyperparameters are obtained by training with the training set of the above K cases. Taking 4 fold as an example, the process is shown in Figure 7.

It can be seen from the above that the mechanism of the present invention and the data-driven prediction mode can be combined with the domain prior knowledge model and the data-driven deep learning algorithm. Apply the power knowledge model, optimize model parameters, and improve model adaptability; improve the efficiency of data-driven machine learning and data mining, and reduce the risk of machine learning generalization without increasing the number of training samples. The effective combination of the mechanism model and the data-driven model realizes the organic integration of rules and experience, which can better integrate the advantages of the two models, and can achieve better synthesis by using fewer data samples and a more simplified mechanism model. performance, ensure good prediction accuracy and efficiency, and effectively improve the reliability of prediction results.

In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the processor causes the processor to execute the method for short-term photovoltaic power prediction in the context of the above electricity market. step.

In yet another aspect, the present invention also discloses a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to execute the above-mentioned electricity market background The steps of the short-term photovoltaic power prediction method below.

It is understandable that the system provided by the embodiment of the present invention corresponds to the method provided by the embodiment of the present invention, and reference may be made to the corresponding part of the above-mentioned method for explanation, examples and beneficial effects of related content.

Embodiments of the present application further provide an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus,

memory for storing computer programs;

The processor is configured to implement the short-term photovoltaic power prediction method under the background of the electricity market when executing the program stored in the memory, and the method includes:

Based on the NWP value, the weather is divided into four weather models: sunny, cloudy, shower and full rain;

Determine the weather type of the day to be measured, and select the most recent similar day's historical PV power data and historical NWP value according to the weather type of the day to be measured;

Input the classified data set into each basic learner in the one-layer prediction model of the Stacking ensemble model;

Obtain the prediction results of each base learner, construct a new training set and input the two-layer prediction model;

Obtain the prediction result of the two-layer meta-learner, which is the final photovoltaic power generation power prediction value.

The communication bus mentioned in the above electronic device may be a Peripheral Component Interconnect (English: Peripheral Component Interconnect, abbreviated: PCI) bus or an Extended Industry Standard Architecture (English: Extended Industry Standard Architecture, abbreviated: EISA) bus or the like. The communication bus can be divided into an address bus, a data bus, a control bus, and the like.

The communication interface is used for communication between the above electronic device and other devices.

The memory may include random access memory (English: Random Access Memory, short: RAM), or may include non-volatile memory (English: Non-Volatile Memory, short: NVM), for example, at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

The above-mentioned processor may be a general-purpose processor, including a central processing unit (English: Central Processing Unit, referred to as: CPU), a network processor (English: Network Processor, referred to as: NP), etc.; or a digital signal processor (English: Digital Signal Processing, referred to as DSP), application specific integrated circuit (English: ApplicationSpecific Integrated Circuit, referred to as: ASIC), Field Programmable Gate Array (English: Field-Programmable Gate Array, referred to as: FPGA) or other programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components.

In yet another embodiment provided by the present application, a computer program product including instructions is also provided, which, when running on a computer, causes the computer to execute the short-term photovoltaic power forecasting method in the context of any electricity market in the above-mentioned embodiments .

In the above-mentioned embodiments, it may be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented in software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer may be a general purpose computer, special purpose computer, computer network, or other programmable device. The computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be downloaded from a website site, computer, server, or data center Transmission to another website site, computer, server, or data center is by wire (eg, coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (eg, infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, a data center, or the like that includes an integration of one or more available media. The usable media may be magnetic media (eg, floppy disks, hard disks, magnetic tapes), optical media (eg, DVDs), or semiconductor media (eg, Solid State Disk (SSD)), among others.

It should be noted that, in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any relationship between these entities or operations. any such actual relationship or sequence exists. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass a non-exclusive inclusion such that a process, method, article or device that includes a list of elements includes not only those elements, but also includes not explicitly listed or other elements inherent to such a process, method, article or apparatus. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in a process, method, article or apparatus that includes the element.

Each embodiment in this specification is described in a related manner, and the same and similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for related parts, please refer to the partial descriptions of the method embodiments.

The above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: The recorded technical solutions are modified, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. A short-term photovoltaic power forecasting method under the background of the electricity market, characterized in that, comprising the following steps, Based on the NWP value, the weather is divided into four weather models: sunny, cloudy, shower and full rain; Determine the weather type of the day to be measured, and select the most recent similar day's historical PV power data and historical NWP value according to the weather type of the day to be measured; Input the classified data set into each basic learner in the one-layer prediction model of the Stacking ensemble model; Obtain the prediction results of each base learner, construct a new training set and input the two-layer prediction model; Obtain the prediction result of the two-layer meta-learner, which is the final photovoltaic power generation power prediction value. 2. The short-term photovoltaic power prediction method under the background of the electricity market according to claim 1 is characterized in that: based on the NWP value, the weather is divided into four weather models: sunny day, cloudy day, shower and full rain; specifically including: The cloud amount C and the rainfall p in the NWP are selected as the weather classification factors. According to the average cloud amount during the day, it is divided into a sunny model and a cloudy model, and according to the daytime rainfall duration, it is divided into a shower model and a full rain model. The sunny model is

The cloudy model is

The shower model is

The full rain model is

c ₁ is the classification threshold of sunny and cloudy models, t ₁ is the classification threshold of shower and full rain models; according to the national standard for short-term weather forecast and the principle of meteorology, c ₁ =0.7, t ₁ =4. 3. The short-term photovoltaic power prediction method under the background of the electricity market according to claim 1 is characterized in that: the one-layer prediction model of the Stacking integrated model includes the physical characteristic model of photovoltaic power generation as shown in formula (1): P=ηSI[1-0.005(t+25)] (1) In the formula, P is the photovoltaic power generation, η is the conversion efficiency of the photovoltaic panel, S is the effective area of the photovoltaic panel, I is the irradiance, and t is the operating temperature of the photovoltaic panel. 4. The short-term photovoltaic power forecasting method under the background of the electricity market according to claim 3, wherein the one-layer forecasting model further comprises a TCN network forecasting model, and the steps of constructing the TCN network forecasting model are as follows: In the hole convolution step, each TCN layer contains L convolution layers, and the dilated convolution calculation formula is:

In the formula: the hole coefficient d=(1,...,2 ^L ), k is the size of the convolution kernel; Residual linking steps: Relu represents the linear rectification function, which is used as the activation function of the neural network; DC Conv represents the hole convolution layer; Formulas (3) and (4) represent the activation function of TCN:

In the formula: W ⁽¹⁾ and W ⁽²⁾ are the weight matrices corresponding to the input, b is the bias vector, S _{(i, j)} represents the activation function of the i-th layer of the j-th block, and formula (3) represents the hole at time t The result of the convolution, formula (4) represents the result after adding the residual link. 5. The short-term photovoltaic power prediction method under the background of the electricity market according to claim 4, wherein, Obtain the prediction results of each base learner, construct a new training set and input the two-layer prediction model, using the following algorithm: The histogram-based algorithm of the LightGBM model is used to alleviate the influence of high-dimensional data on prediction; the LightGBM model is optimized by the Leaf-wise algorithm with depth limitation; The objective function of the LightGBM model is as follows: Obj(t)=L(t)+Ω(t)+c (5) In the formula: Obj(t) is the optimization objective, Ω(t) represents the regular function, which reflects the complexity of the model; t represents the sampling time; c represents the extra parameter to avoid overfitting and optimize the tree depth; L(t) represents the loss function, which describes the actual value _yi and predicted value of N sampling points

to reflect the fit of the model; the definition is as follows:

Residual information of previous learners is transmitted by serial coupling of regression trees; final output

Generated by the accumulation of the remaining trees. 6. The short-term photovoltaic power prediction method under the background of the electricity market according to claim 5, characterized in that: Obtain the prediction results of each base learner, construct a new training set and input the two-layer prediction model, and also use the following algorithm: The gradient-based unilateral sampling algorithm GOSS is adopted. When sampling, GOSS will retain all the large gradient samples that meet the conditions as the sampled data, and adopt random sampling for the samples with smaller gradients. 7. The short-term photovoltaic power prediction method under the background of the electricity market according to claim 6, characterized in that: Obtain the prediction results of each base learner, construct a new training set and input the two-layer prediction model, and also use the following algorithm, The mutually exclusive feature binding method EFB is adopted. After GOSS sampling, it will use the method of binding mutually exclusive features to reduce the dimension of features to prevent dimension disaster and improve computational efficiency. 8. The short-term photovoltaic power prediction method under the background of the electricity market according to claim 7, characterized in that: The one-layer prediction model of the Stacking ensemble model divides the data set into multiple subsets through cross-validation, and fuses the evaluation results to reduce the variance of the model prediction results, improve the generalization ability of the model, and avoid the occurrence of overfitting; K-fold cross-validation is to divide the data set into K parts on average, of which K-1 is used as a training set, and the remaining 1 is a validation set; the model hyperparameters are obtained by training with the training sets of the above K cases. 9. A computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of the method according to any one of claims 1 to 8. 10. A computer device comprising a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the execution as claimed in any one of claims 1 to 8. steps of the method described.

Images