Photovoltaic power generation is highly intermittent and random, and the output power of photovoltaic modules is affected by solar irradiance, temperature, and photoelectric conversion efficiency.
Influence of solar irradiance on photovoltaic power generation
According to the calculation formula of photovoltaic power generation (7-20), it can be known that different solar irradiance affects the output power of photovoltaic power generation. There is a positive correlation between solar irradiance and the output power of photovoltaic power generation, and the relationship between the two is shown in Figure 7-2. Taking the historical data of a photovoltaic power station in Qinghai as an example, the data collected every day is from 6:00 to 20:00. The change trend of solar irradiance generally reflects the change of photovoltaic power generation. Therefore, the solar irradiance is the prediction Important input variables for PV output power.

The influence of different weather types on photovoltaic power generation
Under different weather types, the air molecules, dust, and clouds in the atmosphere are different, and the corresponding scattering effects on solar radiation are also different. Therefore, different weather types have a certain impact on photovoltaic power generation, as shown in Figure 7-3. It can be clearly seen from the figure that the photovoltaic power station generates the most power in sunny days; in sunny to cloudy weather, when it is cloudy from 12:00 to 14:00, the clouds block the sun and the power generation is significantly reduced; in rainy and snowy weather, the photovoltaic The trend of power generation is not much different from that of a sunny day, but the value is much smaller.
The effect of temperature on photovoltaic power generation
The influence of irradiance and day type on photovoltaic power generation has been discussed above. Under the same conditions, photovoltaic power generation is also different at different temperatures. Therefore, temperature is also an influencing factor of photovoltaic power generation.
Elman neural network
Elman neural network is a typical global feed forward local recurrent network, which is a two-layer back propagation network. Elman neural network includes input layer, hidden layer, output layer, and a successor layer. The hidden layer is a neuron connected to the input vector, and its output is not only used as the input of the output layer, but also connected to other neurons in the hidden layer and fed back to the input of the hidden layer. The role of the successor layer is to memorize the output value of the hidden layer at the previous time. It can be regarded as a delay operator with dynamic memory ability.
Elman neural network short-term prediction model
According to the previous analysis, the power prediction of photovoltaic power plants mainly refers to the meteorological factors received by photovoltaic panels. The photovoltaic prediction model constructed in this chapter has three types of input parameters, which are historical power generation, solar radiation Illumination, temperature. At the same time, the prediction model constructed in this chapter is divided into three sub-models according to the weather type (sunny day, sunny to cloudy day, rain/snow day). In addition, the output power of photovoltaic panels is also very different at different times of the day. The typical daily power generation usually starts to rise from 8:00 in the morning, reaches a peak state at some time between 12:00 and 16:00, and then gradually decreases, and reaches a trough at 20:00.

For the short-term prediction model of photovoltaic power generation, the input layer has 41 input parameters, which are the generation power (13), the solar Illuminance (13), temperature (13), solar irradiance (1) and temperature (1) on the forecast day, and photovoltaic power generation is predicted one day in advance. The number of neurons in the output layer of the Elman neural network is 13, that is, the output power per hour from 8:00 to 20:00 on the forecast day. Generally, the more neurons in the hidden layer, the smaller the prediction error. However, too many neurons will prolong the training time of the prediction model and reduce the fault tolerance of the network. Therefore, the number of neurons in the hidden layer must be selected appropriately, first use the formula to estimate the approximate range, and then select it based on experience and many experiments.
Comparative analysis of Elman neural network and NSET modeling
The nonlinear state estimation (NSET) method was proposed by Singer et al. It is a non-parametric calibration method and is currently used in many aspects, including equipment monitoring, nuclear power plant sensor verification, photovoltaic and wind power prediction, etc. The nonlinear state is briefly introduced below. Principles of estimation. There are many options for this nonlinear operator. After reviewing the data and repeated verification, this paper selects the Euclidean distance between the two vectors, that is, selects the same historical data as the sample, only the algorithm changes, and the rest are the same. Neural network and NSET prediction model for comparison. Figure 7-14 shows the prediction results and errors. It is obvious from the figure that the prediction accuracy of Elman neural network is higher than that of NSET prediction, and the Elman neural network predicts photovoltaic power generation in short-term.

Comparative analysis of Elman neural network and BP neural network modeling
BP neural network is a feed-forward neural network, which is characterized by the forward transmission of signals, and the neurons in each layer are only affected by the neurons in the previous layer. Figure 7-15 is the structure diagram of BP neural network. In the training process of the BP network, too many hidden layers and hidden layer nodes will cause the convergence speed to slow down, the training time will become longer, and it is easy to make the model fall into the local minimum, which is the disadvantage of the BP neural network. Like NSET, select the same Historical data is used as a sample, only the algorithm changes, the rest are the same.