With the increasing proportion of distributed photovoltaic (DPV) installations in county-level power grids, to improve the centralized operation and maintenance of the stations and to meet the needs of power grid dispatching, the output of the county-level regional DPV stations group needs to be predicted. In this paper, the weather prediction information is used to predict the output based on the model input average strategy. To eliminate the effect of the selected non-optimal training sample collection period on the prediction accuracy, an ensemble prediction method based on the minimum redundancy maximum relevance criterion and TabNet model is carried out. To reduce the influence of weather prediction errors on the power output prediction, a modified model based on error prediction is proposed. The ensemble prediction model is used to predict the day-ahead output, and a combination prediction model based on the proposed ensemble prediction model and the proposed modified model is established to predict the hour-ahead output. The experimental results verify the effectiveness of the proposed models. Compared with the corresponding reference models, the proposed ensemble prediction method reduces the normalized mean absolute errors (nMAEs) and the normalized root mean square errors (nRMSEs) of the day-ahead output prediction results by 2.86% and 5.51%, respectively. The combination prediction model reduces the nMAE and nRMSE of the hour-ahead output prediction results by 3.05% and 3.05%, respectively. Therefore, the prediction accuracy can be improved by the proposed models.

The meteorological environment is a determining factor in photovoltaic (PV) system feasibility (PVSF). To evaluate this impact more accurately, a quantitative analysis model based on multimeteorological factors and the Random Forest Regression model is proposed in this work. Firstly, an evaluation system is established to assess the impact. Then, to predict the indicators of the evaluation system, a parameter, i.e., performance ratio in sampling period, is defined. Secondly, a set of essential influences on the performance ratio in the sampling period is established through analyzing and reducing the discovered influences on the PV system performance. Finally, data from the Desert Knowledge Australia Solar Centre (DKASC) website are used to conduct the experiment. During the experiment, the sample set is cleaned using the model based on the cosine of the zenith angle. The functional relationship between the performance ratio in the sampling period and its essential influences is established through training a Random Forest Regression model with the data of the modeling system. The data of the test system are used to verify the forecast performance of the proposed model. Compared with the reference model, which is based on the traditional physical experiment, the results of the proposed model accord better with the measured values.

Forecasting can overcome the shortcomings of volatility and intermittent nature of photovoltaic (PV) power. The PV power output is affected by many factors. To improve the forecasting accuracy, the forecasting model needs considering as many relevant factors as possible, but this makes the input feature vector contain redundant information and leads to high complexity forecasting algorithm, low system stability, and poor application practicability. In this paper, we develop a novel optimization method based on Gamma test and Non-dominated Sorting Genetic Algorithm to optimize the input feature vector space and reduce the information redundancy. Meanwhile, to build targeted models for different weather conditions, a new weather variability and smoothed clearness index are deduced in this paper to measure the weather fluctuation. Based on them, the weather is classified into A and B microweather types. Two sets of input vector for the two types are optimized, and only half of the impact factors (21/51) are selected by this optimization method. Two sets of support vector machine, ELMAN artificial neural network and multivariate autoregressive models, are built. The method can greatly simplify the computational complexity as well as improve the accuracy and efficiency of the models.