Photovoltaic output prediction method, device and storage medium based on multi-source data

Abstract

The present invention relates to a photovoltaic output prediction method, device, and storage medium based on multi-source data. The present invention first calculates the environmental feature weight value, geographical feature weight value, and component feature weight value of the quantitative data sensitivity based on the measured multi-source data, and then distributes the weight ratio in combination with experience, obtains the updated fusion weight through weight distribution, and finally uses different models to predict from three dimensions of environmental features, geographical features, and component features, and weights the predicted results with the updated fusion weight to obtain the final predicted photovoltaic output. By dynamically updating the fusion weight, the influence of various factors on photovoltaic output can be better quantified, thereby improving the prediction accuracy. Compared with the prior art, the present invention has the advantages of being able to fully reflect the relationship between variables and accurately quantify the influence of photovoltaic output.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic output prediction, and in particular to a photovoltaic output prediction method, device and storage medium based on multi-source data. Background Art

[0002] There are many PV output prediction algorithms, including multiple linear regression, BP neural network, support vector machine (SVM) and grey theory algorithm. Multiple linear regression uses multiple related factors as independent variables. This method can reduce the difficulty of modeling and shorten the modeling time. Support vector machine (SVM) is a supervised learning method used for classification and regression analysis. SVM tries to find a hyperplane to separate data of different categories. In PV output prediction, SVM can be used to predict the output of PV power stations. Grey system theory is a mathematical method for processing small samples and incomplete information. It clarifies some known information, and obtains the regularity of the movement of things by generating, transforming and mining this known information. In PV output prediction, the grey theory algorithm can handle situations with small data volume and incomplete information.

[0003] Chinese patent application publication number CN114781723A discloses a short-term photovoltaic output forecasting method based on multi-model fusion. It aims to solve the problem of how to achieve more accurate photovoltaic power generation forecasting under the condition of high proportion of new energy access. By constructing multiple features and integrating the advantages of tree model and neural network model, it provides accurate short-term photovoltaic output forecasting analysis for power system managers, so as to overcome the volatility and intermittency of photovoltaic power generation and improve the effect of new energy consumption.

[0004] However, the above application does not fully consider the problem of the dynamic changes in the degree of influence of various factors affecting photovoltaic output on photovoltaic output in different photovoltaic power stations. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a photovoltaic output prediction method, device, and storage medium based on multi-source data, so as to realize photovoltaic output prediction under the premise of taking into account the change of fusion weight, so as to solve or partially solve the problem of dynamic changes in the degree of influence of various factors affecting photovoltaic output on photovoltaic output.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] One aspect of the present invention provides a photovoltaic output prediction method based on multi-source data, comprising the following steps:

[0008] Obtain the measured multi-source data of the target PV power station, and obtain the environmental feature prediction value, geographical feature prediction value and component feature prediction value based on the pre-built prediction model, update the fusion weight of the target PV power station, and fuse to obtain the final PV output prediction value.

[0009] The updating process of the fusion weights includes the following steps:

[0010] Based on the measured multi-source data, the environmental feature weight value is calculated using the photovoltaic panel temperature deviation of the target photovoltaic power station, the geographical feature weight value is calculated using a pre-built geographical feature association degree model, and the component feature weight value is calculated using a pre-built component body feature association degree model;

[0011] An experience-based allocation weight ratio is obtained to implement weight allocation of weight values ​​and obtain the fusion weight.

[0012] As a preferred technical solution, the process of calculating the environmental characteristic weight value using the photovoltaic panel temperature deviation of the target photovoltaic power station includes the following steps:

[0013] Based on the photovoltaic panel temperature, ambient temperature, actual power generation and theoretical power generation of the target photovoltaic power station, the environmental feature weight value is calculated by constructing an environmental feature correlation degree model.

[0014] As a preferred technical solution, the environmental feature correlation degree model is:

[0015] ,

[0016] in, is the weight value of environmental characteristics, , They are the ambient temperature and the photovoltaic panel temperature, in °C. , They are actual power generation and theoretical power generation respectively.

[0017] As a preferred technical solution, the process of calculating the geographical feature weight value using the pre-built geographical feature association degree model includes the following steps:

[0018] Construct a sunny day model;

[0019] Based on the geographical feature parameters in the measured multi-source data, the sunny day model is used to calculate theoretical uplink direct and scattered radiation as theoretical irradiance of photovoltaic panels;

[0020] Based on the deviation between the actually measured irradiance of the photovoltaic panel in the actually measured multi-source data and the theoretical irradiance of the photovoltaic panel, a geographic feature weight value is calculated using a pre-constructed geographic feature association degree model.

[0021] As a preferred technical solution, the construction process of the geographic feature association degree model includes:

[0022] Based on the theoretical maximum irradiance of the photovoltaic panel and the longitude and latitude information and time information corresponding to the historical photovoltaic output, the sunny day model is used to calculate the theoretical irradiance of the photovoltaic panel;

[0023] The training of the geographic feature correlation degree model is implemented with the goal of minimizing the deviation between the measured irradiance of the photovoltaic panel and the theoretical irradiance of the photovoltaic panel, wherein the geographic feature correlation degree model is constructed based on a multi-layer perceptual neural network.

[0024] As a preferred technical solution, the process of calculating the component feature weight value using the pre-built component ontology feature association degree model includes:

[0025] Based on the measured irradiance of the photovoltaic panels of the target photovoltaic power station, the actual conversion efficiency of electricity and the theoretical conversion efficiency of electricity, the component characteristic weight value is calculated using the following formula: :

[0026] ,

[0027] in, is the measured irradiance of the photovoltaic panel, in w / m 2 , For electricity The corresponding conversion efficiency in the state is is the theoretical conversion efficiency of electricity.

[0028] As a preferred technical solution, the process of obtaining the environmental characteristic prediction value, the geographical characteristic prediction value and the component characteristic prediction value respectively based on the pre-built prediction model includes:

[0029] Inputting the environmental characteristic parameters in the measured multi-source data into a pre-trained environmental characteristic prediction model based on a multi-layer perception neural network to obtain an environmental characteristic prediction value;

[0030] Inputting the geographic feature parameters in the measured multi-source data into a pre-trained geographic feature prediction model based on a multi-layer perceptron neural network to obtain a geographic feature prediction value;

[0031] The component feature parameters in the measured multi-source data are input into a pre-trained component feature prediction model based on a multi-layer perceptron neural network to obtain component feature prediction values.

[0032] As a preferred technical solution, the environmental characteristic parameters include ambient temperature, humidity, wind speed, wind direction, scattered irradiance, direct irradiance and total irradiance; the geographical characteristic parameters include longitude and latitude and altitude; and the component characteristic parameters include current and voltage.

[0033] Another aspect of the present invention provides an electronic device, comprising: one or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing the aforementioned photovoltaic output prediction method based on multi-source data.

[0034] Another aspect of the present invention provides a computer-readable storage medium comprising one or more programs for execution by one or more processors of an electronic device, wherein the one or more programs include instructions for executing the aforementioned photovoltaic output prediction method based on multi-source data.

[0035] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0036] (1) Fully consider the dynamic changes of the influence of various factors affecting photovoltaic output on photovoltaic output, and improve the accuracy of output prediction: the present invention first calculates the environmental feature weight value, geographical feature weight value and component feature weight value for quantifying data sensitivity based on the measured multi-source data, and then distributes the weight ratio based on experience, and obtains the updated fusion weight through weight distribution. Finally, different models are used to predict from the three dimensions of environmental features, geographical features and component features, and the prediction results are weighted with the updated fusion weight to obtain the final predicted photovoltaic output. By dynamically updating the fusion weight, the influence of various factors on photovoltaic output can be better quantified, thereby improving the prediction accuracy.

[0037] (2) Ability to fully reflect the relationship between variables: The present invention uses a multi-layer perceptron neural network (MLP) to predict the irradiance of photovoltaic panels. In addition, a feature prediction model based on MLP is used to predict the photovoltaic output corresponding to the three types of features. The network or model based on MLP can better reflect the relationship between variables.

[0038] (3) Accurately quantify the impact of photovoltaic output: The present invention adopts a temperature deviation method to calculate the impact of photovoltaic output on environmental parameters, a sunny day model to calculate the correlation between photovoltaic output and geographical characteristic parameters, and a photoelectric conversion method to calculate the correlation between photovoltaic output and component body parameters. This can accurately quantify the impact of photovoltaic output while reducing the amount of calculation.

[0039] (4) The structure of the MLP multi-layer neural network is simple, and the cost of retraining and maintenance is much lower than that of a deep neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of the process of photovoltaic output prediction method based on multi-source data in the embodiment;

[0041] Figure 2Schematic diagram of the current-voltage of the photovoltaic module under different irradiance conditions in the embodiment;

[0042] Figure 3 A schematic diagram of photovoltaic module power-voltage under different irradiance conditions in the embodiment;

[0043] Figure 4 The corresponding curve diagram of the slope irradiance and photovoltaic power generation power in the embodiment;

[0044] Figure 5 Schematic diagram of the current-voltage of the photovoltaic module under different temperature conditions in the embodiment;

[0045] Figure 6 A schematic diagram of photovoltaic module power-voltage under different temperature conditions in the embodiment;

[0046] Figure 7 The corresponding curve diagram of the ambient temperature, the module backplane temperature and the photovoltaic power generation power in the embodiment;

[0047] Figure 8 Schematic diagram of actual prediction results in the embodiment. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0049] Example 1

[0050] Regarding the problems existing in the above prior art, see Figure 1 This embodiment provides a photovoltaic output prediction method based on multi-source data, including the following steps:

[0051] Step S1, collecting the measured multi-source data of the photovoltaic power station to achieve data collection.

[0052] Among them, the measured multi-source data includes various types of data of photovoltaic sites, including environmental parameters, geographical feature parameters and component body parameters. If it is in the model training stage, it also includes the corresponding output data.

[0053] Environmental parameters include ambient temperature, humidity, wind speed, wind direction, diffuse irradiance, direct irradiance, and total irradiance.

[0054] Geographic feature parameters include latitude, longitude and altitude.

[0055] Component body parameters include current and voltage.

[0056] Preferably, after the data is collected, outlier removal and filtering processing are also performed.

[0057] Step S2, based on the measured multi-source data, calculate the environmental feature weight value, geographical feature weight value and component feature weight value used to quantify the influence of various factors on photovoltaic output (hereinafter referred to as the correlation degree) to achieve data preprocessing. This step specifically includes the following sub-steps:

[0058] Step S201, using the temperature deviation method to calculate the correlation between photovoltaic output and environmental parameters, the specific process is as follows:

[0059] According to the photovoltaic panel temperature in the environmental parameters With ambient temperature , calculate the temperature deviation , in °C, collects the corresponding actual power generation .

[0060] The actual power generation corresponding to the temperature difference of the photovoltaic panels and the theoretical power generation corresponding to the standard photovoltaic output (at 25°C) , build the environmental characteristics correlation degree model .

[0061] Calculate and update the environmental feature weight value based on the environmental feature association degree model .

[0062] Step S202, using the sunny day model to calculate the correlation between photovoltaic output and geographical characteristic parameters, the specific process is as follows:

[0063] Collect historical data, determine the clear sky model parameters and build a clear sky model.

[0064] Obtain the geographical characteristic parameters of the photovoltaic power station and input them into the sunny day model to calculate the theoretical uplink direct and diffuse radiation and approximate the theoretical irradiance of the photovoltaic panel. (Unit: w / m 2 ).

[0065] Measured irradiance of photovoltaic panels based on geographical characteristic parameters of photovoltaic power stations (Unit: w / m 2 ), and the calculated theoretical irradiance of the photovoltaic panel , input the deviation of the two into the geographic feature correlation degree model to obtain the geographic feature correlation degree model .

[0066] The geographic feature correlation model is constructed based on a multi-layer neural network (i.e., multi-layer perceptron neural network, MLP). The MLP uses irradiance deviation as the regression term, and the network loss function is the theoretical irradiance of the photovoltaic panel. Measured irradiance of photovoltaic panels The deviation is determined by training the MLP with the collected historical data set to realize the training of the model.

[0067] Step S203, using a photoelectric conversion method to calculate the correlation between photovoltaic output and the module body parameters, the specific process is as follows:

[0068] Obtain the measured photovoltaic panel output power and photovoltaic panel irradiance .

[0069] Construct a component feature correlation model: Based on the measured irradiance of the photovoltaic panel To power Conversion efficiency Theoretical conversion efficiency Deviation construction component ontology feature correlation degree model .

[0070] Based on the measured photovoltaic panel output power and photovoltaic panel irradiance (Unit: w / m 2 ), calculate and update the component feature weight value .

[0071] Step S3, calculating weight parameters (ie, fusion weights).

[0072] According to the correlation between the photovoltaic output and various types of data calculated in step S2, the initial weight coefficient of each parameter is calculated: .

[0073] Different types of parameters are allocated according to the experience-based allocation weight ratio obtained in advance based on expert experience. The weight ratio of environmental parameters is recorded as , the weight ratio of geographical feature parameters is recorded as , the weight ratio of the component body parameters is recorded as The sum of weight coefficients , to ensure the final weight coefficient and .

[0074] Step S4, model prediction.

[0075] Using neural networks to build environmental characteristics prediction model , a multi-layer neural network with actual power generation as regression term The loss function of the network is the deviation between the actual power generation and the theoretical power generation. The multi-layer neural network is trained with the collected historical environmental data set. is the environmental parameter.

[0076] Using Neural Networks to Build Geographical Feature Prediction Models , a multi-layer neural network with actual power generation as regression term The loss function of the network is the deviation between the actual power generation and the theoretical power generation. The multi-layer neural network is trained with the collected historical data set of geographical features. It is the geographical feature parameter.

[0077] Using neural network to build component ontology prediction model , a multi-layer neural network with actual power generation as regression term The loss function of the network is the deviation between the actual power generation and the theoretical power generation. The multi-layer neural network is trained with the collected component feature historical data set, where It is the component body parameter.

[0078] Step S5, using the weight parameters in step S3 to fuse the prediction results:

[0079] .

[0080] Step S6: Output the final photovoltaic output prediction value , realizing short-term prediction of photovoltaic power.

[0081] In order to verify the effectiveness of this method, this method was applied to a photovoltaic power plant in northern Zhejiang, and one year of data from the photovoltaic power plant was collected to train the model and network used in the model training stage. The specific data included longitude and latitude, altitude, irradiation intensity, ambient temperature, photovoltaic panel temperature, horizontal total radiation, direct radiation, diffuse radiation, current, voltage, wind speed, zenith angle, relative humidity, photovoltaic output, etc. Among them, ambient temperature, humidity, wind speed, wind direction, diffuse irradiance, direct irradiance, and total irradiance are environmental data, longitude and latitude, and altitude are geographical data, and current and voltage are component data.

[0082] See also Figure 2 , Figure 3 The current-voltage (IU) and power-voltage (PU) curves of the photovoltaic module are shown in Figure 1. When the output voltage is constant, the power generation of the photovoltaic module increases with each level of solar irradiance increase. The change of the output voltage at the optimal working point is particularly obvious. When the irradiance is maximum and the photovoltaic module is working at the optimal working point, the output power of the photovoltaic module reaches the maximum value. Figure 4 From the mid-level irradiance and photovoltaic power generation curve, it can be seen that the changes in solar irradiance and photovoltaic power generation are basically positively correlated, and the peaks and troughs of the solar irradiance change curve and the peaks and troughs of the photovoltaic power generation curve are mostly consistent, and the overall change trends are basically consistent. It can be judged that solar irradiance has an important impact on photovoltaic power generation.

[0083] Depend on Figure 5 , Figure 6 It can be seen from the IU curve and PU curve in that when the output voltage of the photovoltaic module is below 300V, the temperature change has little effect on photovoltaic power generation. When the photovoltaic module works normally, the output voltage is generally above 300V. At this time, the temperature change has a greater impact on photovoltaic power generation. When the temperature of the photovoltaic module rises, the rate of decrease of the output current of the photovoltaic module will increase. Under the condition of constant solar irradiance and optimal working point voltage, the higher the daily limit temperature, the lower the photovoltaic power generation. Figure 7 From the PV module temperature and PV power generation curves, we can see that during the period when the PV module has power output, the change trend of the PV power generation curve is roughly consistent with the change trend of the PV module temperature curve. It can be judged that the PV module temperature has a great influence on PV power generation.

[0084] After data preprocessing, the environmental weight function, geographic weight function and component weight function were constructed respectively, and a multi-layer neural network was used to fit the three types of weight functions. , , This method was compared with a certain prediction software in this power plant, and the results are as follows Figure 8 The figure shows a 7-day power generation forecast, where the two lines represent the true value of the photovoltaic power station power generation and the forecast result of this method respectively. The dotted line is the forecast result of a certain software. The experimental results verify that the short-term forecast accuracy of the certain software is 90%, and the accuracy of this method is 94%, which has higher accuracy.

[0085] Example 2

[0086] Based on Example 1, this embodiment provides an electronic device, including: one or more processors and a memory, wherein one or more programs are stored in the memory, and the one or more programs include instructions for executing the photovoltaic output prediction method based on multi-source data in Example 1.

[0087] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0088] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0089] Example 3

[0090] This embodiment provides a computer-readable storage medium, including one or more programs for execution by one or more processors of an electronic device, wherein the one or more programs include instructions for executing the photovoltaic output prediction method based on multi-source data in Embodiment 1.

[0091] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0092] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A photovoltaic output prediction method based on multi-source data, characterized in that: The steps include: Obtain the measured multi-source data of the target PV power station, and obtain the environmental feature prediction value, geographical feature prediction value and component feature prediction value based on the pre-built prediction model, update the fusion weight of the target PV power station, and fuse to obtain the final PV output prediction value. The updating process of the fusion weights includes the following steps: Based on the measured multi-source data, the environmental feature weight value is calculated using the photovoltaic panel temperature deviation of the target photovoltaic power station, the geographical feature weight value is calculated using a pre-built geographical feature association degree model, and the component feature weight value is calculated using a pre-built component body feature association degree model; Obtaining the distribution weight ratio based on experience, realizing the weight distribution of the weight value, and obtaining the fusion weight, The process of calculating the environmental characteristic weight value using the photovoltaic panel temperature deviation of the target photovoltaic power station includes the following steps: Based on the photovoltaic panel temperature, ambient temperature, actual power generation and theoretical power generation of the target photovoltaic power station, the environmental feature weight value is calculated by constructing an environmental feature correlation degree model, and the environmental feature correlation degree model is: ， in, is the weight value of environmental characteristics, , They are the ambient temperature and the photovoltaic panel temperature, in °C. , are the actual power generation and theoretical power generation respectively. The process of calculating the weight value of geographic features using the pre-built geographic feature association degree model includes the following steps: Construct a sunny day model; Based on the geographical feature parameters in the measured multi-source data, the sunny day model is used to calculate the theoretical maximum direct and diffuse radiation as the theoretical irradiance of the photovoltaic panel; Based on the deviation between the measured irradiance of the photovoltaic panel in the measured multi-source data and the theoretical irradiance of the photovoltaic panel, the geographical feature weight value is calculated using a pre-constructed geographical feature association degree model, The construction process of the geographic feature association degree model includes: Based on the theoretical maximum irradiance of the photovoltaic panel and the longitude and latitude information and time information corresponding to the historical photovoltaic output, the sunny day model is used to calculate the theoretical irradiance of the photovoltaic panel; The goal is to minimize the deviation between the measured irradiance of photovoltaic panels and the theoretical irradiance of photovoltaic panels, and to train the geographical feature correlation degree model, wherein the geographical feature correlation degree model is constructed based on a multi-layer perceptual neural network. Using the pre-built component ontology feature association degree model, the process of calculating the component feature weight value includes: Based on the measured irradiance of the photovoltaic panels of the target photovoltaic power station, the actual conversion efficiency of electricity and the theoretical conversion efficiency of electricity, the component feature weight value is calculated using the following formula: : ， in, is the measured irradiance of the photovoltaic panel, in w / m 2 , For electricity The corresponding conversion efficiency in the state is is the theoretical conversion efficiency of electricity.

2. A photovoltaic output prediction method based on multi-source data according to claim 1, characterized in that: Based on the pre-built prediction model, the process of obtaining the predicted values ​​of environmental characteristics, geographical characteristics and component characteristics includes: Inputting the environmental characteristic parameters in the measured multi-source data into a pre-trained environmental characteristic prediction model based on a multi-layer perception neural network to obtain an environmental characteristic prediction value; Inputting the geographic feature parameters in the measured multi-source data into a pre-trained geographic feature prediction model based on a multi-layer perceptron neural network to obtain a geographic feature prediction value; The component feature parameters in the measured multi-source data are input into a pre-trained component feature prediction model based on a multi-layer perceptron neural network to obtain component feature prediction values.

3. The photovoltaic output prediction method based on multi-source data according to claim 2 is characterized in that: The environmental characteristic parameters include ambient temperature, humidity, wind speed, wind direction, scattered irradiance, direct irradiance and total irradiance; the geographical characteristic parameters include longitude and latitude and altitude; the component characteristic parameters include current and voltage.

4. An electronic device, characterized in that: include: One or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing the photovoltaic output prediction method based on multi-source data as described in any one of claims 1-3.

5. A computer-readable storage medium, characterized in that: It comprises one or more programs executed by one or more processors of an electronic device, wherein the one or more programs comprise instructions for executing the photovoltaic output prediction method based on multi-source data as claimed in any one of claims 1 to 3.

Images