To construct a new power system with “new energy” as the main body, it is urgent to make more accurate prediction of new energy power and apply the prediction results to practical application from the perspective of overall power grid planning, so as to accept high proportion of new energy for the large power grid and make the power system operate securely and economiclly. Based on the actual demand of power dispatching as the guidance, the effectiveness of problem solving as the measurement standard, and the “open, efficient and practical” concept and mechanism innovation to achieve scientific and technological innovation, this paper proposes an innovation mechanism and implementation plan of the innovation platform for the accurate prediction of new energy power at multiple temporal and spatial scales. The paper discusses the top-level design, the basic idea of innovation mechanism, the architecture design and the main functions of innovation platform, and tries to explore a new path to improve the prediction level of new energy power by relying on technology and management innovation from the perspective of the power dispatchers.

To achieve the goals of carbon peaking and carbon neutrality, the development of renewable energies becomes the trend in building a power system with low-carbon emissions and clean energy sources. Analyses of power forecasting errors and improvements in the prediction accuracy of renewable energies are crucial for building new power systems. In this paper, by taking day-ahead prediction of renewable energy power of several southern provinces of China in 2021 as examples, the power prediction characteristics of renewable energies is systematically analyzed and corresponding strategies are proposed to cope with forecasting errors. Firstly, first, existing forecast technologies and the current status of their applications at home and abroad are summarized. Secondly, the common characteristics of the current day-ahead renewable energy power forecasting errors are sorted out. Finally, ideas for coping with renewable energy power forecasting errors are proposed, including the investigation of evaluation metrics for renewable energy power forecasting, research on the ensemble of forecast models, exploration of value ecology cultivation for renewable energy power predictions, etc.

Based on the self-dependent climate system model, the global-regional double coupling model system for Southern Power Grid region is developed, and the ensemble hindcast experiments from 1991 to 2020 are conducted. The bias correction method suitable for the rainfall prediction of China Southern Power Grid region is developed and a seasonal rainfall prediction system suitable for the Southern Power Grid region has been established. The results show that the global-regional double coupling technique can improve the modelling and prediction capability of the regional rainfall spatial distribution and dominant modes of summer rainfall in the Southern Power Grid region, and the bias correction scheme can significantly improve the prediction skills of summer rainfall anomalies in the region. The 30-year average pattern correlation coefficient can reach 0.27, with prediction score (PS) skill reaching 74.31, and the regional average temporal correlation coefficient is about 0.28. The system also shows good prediction skills for extreme flood and drought years. In general, the system can generate reliable seasonal rainfall forecast products, and hence provide good technology support for the safe operation and deployment management in China Southern Power Grid.

With the continuous increase of the proportion of new energy grid connection in China, improving the new energy consumption level has become the key to breaking through the bottleneck of new energy development. Carrying out the research in the cross-regional consumption strategy at different time scales is an effective way. In this paper, two cross-regional power grids are taken as research cases, and the operating characteristics and influencing factors of new energy consumption in different regions are analyzed in detail. Further, a cross-regional new energy consumption strategy is proposed in different time scales, including medium-long term, day ahead and day in. Among them, the medium-long term consumption strategy aims to achieve the optimization of medium-long term new energy consumption efficiency, and focuses on the factors affecting the standardization of equipment operation and maintenance, power generation plan management, and trading power allocation. The day ahead consumption strategy focuses on the factors affecting the peak shaving capacity of power grid in different operation periods and the transmission quality of DC power supply lines. The day in consumption strategy focuses on the factors affecting grid architecture and voltage security of each node. Finally, the feasibility and effectiveness of relevant strategies are verified by the simulation model. The results show that the proposed cross-regional new energy consumption strategies at different time scales can effectively solve the regional new energy consumption problem, reduce the phenomenon of wind and photovoltaic abandonment, and it has a certain practicality and reliability.

High-accuracy short-term wind power prediction is critical to ensure the power system security. A short-term wind power prediction method based on the information in neighboring wind farms and CNN-BiLSTM is proposed in this paper. In the deep learning prediction modeling process, in addition to using the numerical weather prediction (NWP) of the target wind farm as the input feature, the highly correlated features in neighboring wind farms are also introduced. Firstly, the composite correlation between each neighboring wind farm and the target wind farm in the region is constructed based on the relativity and distance among wind sequences and power sequences, and the highly similar neighboring wind farms are selected as the information source according to the correlation ranking. Then, CEEMDAN frequency domain signal decomposition and time series feature expansion are used to construct a high-dimensional feature set, and floating search feature selection algorithm is introduced to optimize the strongly correlated features. Finally, based on the selected core features, the power prediction model based on CNN-BiLSTM deep learning neural network is built. The results show that the proposed method can effectively improve the prediction accuracy compared with the traditional prediction methods which does not use information from neighboring wind farms.

The photovoltaic power (PV) power between similar days has great similarity. To further improve the accuracy of PV power prediction, this paper improves the conventional similar day selection method and proposes a combined PV power prediction method based on the theory of multi-temporal similarity day. The method firstly establishes the time-sensitive meteorological feature quantity, and then selects similar time periods in different time periods by using the similarity combination index to construct multiple combined similar days, and the combined similar days further guarantee the similarity of each time period, and constructs the PSO-LightGBM-BiLSTM time-varying weight combined prediction model to improve the prediction model accuracy and make up for the shortcomings of a single model. Finally, the PSO-BiLSTM model is used to correct the PV power of the day to be predicted. In this paper, the feasibility and superiority of the proposed method are verified by using the data of a PV plant in Guangdong, China as an example.

Aiming at the problem of low prediction accuracy caused by strong intermittency and high random fluctuation of wind power, this paper combines data decomposition technology, artificial intelligence-based prediction model and error correction technology, and proposes a novel combined method for short-term wind power prediction (EEMD-GRU-MC) which integrates ensemble empirical mode decomposition (EEMD), gated recurrent unit (GRU) and Markov chain (MC). Firstly, EEMD algorithm is used to decompose the historical wind power sequence into a group of relatively stationary sub-sequence to reduce the influence of random fluctuation component and disorder noise on the prediction model. Then, GRU model is employed to predict each sub-sequence, and the predicted values of each sub-sequence are superimposed to get the preliminary prediction results. Finally, in order to further improve the prediction accuracy, MC is used to predict the future state of the residual, and the prediction results of the EEMD-GRU model are further modified. The short-term power prediction of a wind farm in Yunnan Province is taken as an example to verify the proposed method. A large number of numerical examples show that compared with ARIMA, LSTM, GRU, EEMD-LSTM and EEMD-GRU models, the combined prediction method proposed in this paper has stronger prediction accuracy and generalization ability, and the average absolute prediction error in different seasons is less than 2%, showing a good short-term wind power prediction prospect.

Arming at the current inaccurate problem of multi-feature power load prediction accuracy, in order to fully exploit the feature information such as time-series information and weather information in power load data, a variational mode decomposition (VMD)-long short-term memory (LSTM) neural network-light gradient boosting machine (LightGBM) based prediction model is proposed to optimize the problems of non-linearity, non-stationary and long memory of load data, and solve the problem of poor extraction of feature information for multi-feature prediction. The method first decomposes the characteristic mode components representing different scales with VMD, which reduces instability of the original sequence, while the residual amount of the decomposition represents the strongly nonlinear part of the load data, which is predicted by the characteristic strong algorithm. Each modal component is predicted by the single feature of LSTM, and then each component is added to the multi-feature using LightGBM for load prediction. Experimental comparison with current multi-characteristic power load forecasting models through example analysis is conducted, the proposed method MAE value is only 23% ~ 73% , and MAPE value can reach 0.37%, with better prediction accuracy.

Aiming at the problems of univariate processing method of wind power and insufficient fitting ability of prediction model, a combined short-term wind power prediction method based on multivariate phase space reconstruction (MPSR) and whale optimization algorithm optimized deep extreme learning machine (WOA-DELM) is proposed in this paper. Firstly, the meteorological factors associated with wind power are screened out using Pearson correlation coefficients and formed into multivariate time series with wind power series; secondly, the optimal embedding dimension and time delay of each time series are determined using C-C method to achieve multivariate phase space; then, the dataset established by multivariate phase space reconstruction is input into the DELM model, while WOA is used to optimize the weight parameters of DELM to obtain the WOA-DELM prediction model, which is used to predict the short-term wind power and finally obtain the prediction results. The mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE) are used as evaluation indicators, combined with example analysis and compared with the traditional model. The results show that the three evaluation indicators obtained by the proposed prediction model are 0.412 0 MW, 0.492 1 MW and 1.782 2%, respectively, which are better than other models and have better stability and prediction performance.

The movement of the cloud causes strong fluctuations in solar irradiance, which in turn causes the randomness and volatility of photovoltaic power generation, and has a serious impact on safe and stable operation of the power system. Aiming at the above problems, the B-Informer combined interval prediction method based on cloud features of ground-based cloud images is proposed in this paper. Firstly, image processing technology is used to obtain cloud features of cloud images that affect solar irradiance, including the percentage of cloud with correction coefficient, and RGB value of optical flow cloud image. Secondly, the cloud features and historical meteorological data are combined to form the input sequence features, and the informer prediction model based on sparse attention mechanism is constructed. And then, the Bootstrap method is used to increase the sample diversity and to generate the prediction interval, which further improves the prediction accuracy of the model for long time series. Finally, the historical operation data and ground-based cloud images of a power station in Colorado are taken as an example, compared the prediction results with LUBE and other existing methods at a given confidence level. The mean prediction interval width of the proposed method is reduced by up to 27%, which verifies the effectiveness.

Accurate identification of wind power ramp events is of great significance for maintaining power grid security and stability. In order to improve the detection accuracy of wind power ramp events, this paper proposes an identification method based on parameter adaptive swinging door and bump event selection. First, the original power data is filtered to eliminate unreasonable data and noise effects. Then, the parameter adaptive swinging door algorithm is proposed to compress data on the basis of retaining power fluctuation trend. Third, a trend division rule is proposed for the compressed data to select bump events, and the data set with an upward or downward ramp trend is established. Finally, the identification of wind power ramp events is completed according to the several existing definitions. The calculation results show that, compared with the original swinging door algorithm, the method proposed in this paper can identify more wind power ramp events, and the improvement of recognition accuracies based on definition 1, 2 and 3 are 23.28%～53.56%, 13.70%～41.51% and 12.49%～41.52%, respectively.

In the cold wave weather, it is quite difficult to forecast the ice coating of wind turbines in mountain areas, and the prediction deviation of wind power is large. The accurate temperature prediction of wind turbines is the key to improve the prediction accuracy of the ice coating period. Due to the heterogeneity and volatility of air temperature, it is hard for a single model to adapt to the demand for air temperature prediction of different wind farms. This paper presents a multi-mode fusion fan temperature prediction method (MMTP), which not only improves the universality of prediction, but also realizes the complementary advantages of different models by using the combined weight method of AHP-independence-entropy method, and several artificial intelligence algorithms. The prediction results show that: in the 48 h forecast time, the MMTP significantly improves the temperature correction effect compared with the single mode, and the temperature forecast error is decreased by 8%. Besides, the fan icing period forecast is more accurate after temperature correction using this method within the 48 h forecast period, and the forecast error of the glacial period is reduced by 3-6 hours.

Ultra-short-term wind power predictions have a guiding effect on the operation control and energy scheduling of the unit. In order to weaken the effects of the wind speed of the numerical weather prediction (NWP) on ultra-short-term prediction accuracy, an ultra-short-term NWP wind speed correction method considering multiple error scenarios set division is proposed. The bidirectional short-term memory network (BILSTM) predicts the prediction error of the NWP wind speed in the next 4 hours, and the error scenes set is divided into the wind speed error prediction value based on error scene different BILSTM networks are trained to match error and wind speed forcast error prediction and correct the wind speed. Based on the revision results, some modes have been used to predict the ultra-short-term wind power. The method is applied to a wind farm in Inner Mongolia in China for example. The results show that the method of this article effectively reduces the NWP wind speed error. On the basis of the original data, compared with the non -fixed wind speed of NWP, the RMSE value has reduced 1.859, and the MAE value has decreased by 1.464, and MAPE reduces 26.01%. Among them, the BP neural network ultra-short-term power prediction accuracy increased by 7.5%, and the GRU deep network is increased by 8.7%, the multi-linear regression model is increased by 9.6%, the results verify the effectiveness of the method.

Because the prediction error of photovoltaic point can not be avoided, interval prediction can be used to describe the uncertainty of photovoltaic more accurately, which can provide guidance for the decision-making of power system, but the existing research methods can not fully mine the physical change process of photovoltaic power. A prediction framework of intraday photovoltaic output interval considering the spatiotemporal-conditional dependence of prediction error is proposed. Firstly, the prediction error considering time dependence is obtained by appearance similarity update （ASU) model, then the prediction error considering spatial dependence is obtained by long short-term memory （LSTM) model and spatial correlation analysis, and the prediction output is modified. Finally, the interval prediction under different confidences is obtained according to the conditional dependence of the error. The effect of the whole framework has been verified in a photovoltaic electric field in Xinjiang, and its root mean square error can be reduced by more than 3%. At the same time, the interval prediction effect considering the spatiotemporal-conditional dependence of the updated prediction error has been improved, which verifies the effectiveness and feasibility of the proposed method.

Since wind power point forecast errors are unavoidable, probabilistic forecasts can fully describe the uncertainty of wind power, and then provide further guidance for the dispatching department’s decision-making. The current wind power probabilistic prediction methods are still incomplete in mining its physical change process. Therefore, this paper constructs a new short-term wind power probabilistic prediction framework that considers the spatiotemporal dependence of errors by mining the spatiotemporal characteristics of historical wind power data and numerical weather prediction (NWP）. Firstly, the point prediction results are obtained through the gated recurrent unit (GRU）; then, a multi-position NWP is introduced, and a multi-level error scene division method considering the characteristics of space-time dependence is proposed; finally, the Bootstrap sampling method is used to reconstruct the error to form a new adaptive modeling a sample set of short-term wind power probabilistic prediction at different confidence levels is carried out. The experimental results show that the effect of the overall framework has been verified in a wind farm in Northeast China under the probabilistic prediction considering the spatial and temporal dependencies. Compared with the same confidence level, the prediction accuracy is effectively and significantly improved, and the evaluation index PICP is improved by 0.53% and 0.44%, 0.32%, PINAW shrinks by 2.40%, 2.14%, 0.06%, which proves the feasibility and effectiveness of the proposed method.