Load-forecasting method for IES based on LSTM and dynamic similar days with multi-features

To fully exploit the rich characteristic variation laws of an integrated energy system(IES)and further improve the short-term load-forecasting accuracy,a load-forecasting method is proposed for an IES based on LSTM and dynamic similar days with multi-features.Feature expansion was performed to construct a comprehensive load day covering the load and meteorological information with coarse and fine time granularity,far and near time periods.The Gaussian mixture model(GMM)was used to divide the scene of the comprehensive load day,and gray correlation analysis was used to match the scene with the coarse time granularity characteristics of the day to be forecasted.Five typical days with the highest correlation with the day to be predicted in the scene were selected to construct a“dynamic similar day”by weighting.The key features of adjacent days and dynamic similar days were used to forecast multi-loads with fine time granularity using LSTM.Comparing the static features as input and the selection method of similar days based on non-extended single features,the effectiveness of the proposed prediction method was verified.

Short Term Load Forecasting Using Subset Threshold Auto Regressive Model

The subset threshold auto regressive (SSTAR) model, which is capable of reproducing the limit cycle behavior of nonlinear time series, is introduced. The algorithm for fitting the sampled data with SSTAR model is proposed and applied to model and forecast power load. Numerical example verifies that desirable accuracy of short term load forecasting can be achieved by using the SSTAR model.

Data-Driven Load Forecasting Using Machine Learning and Meteorological Data

Electrical load forecasting is very crucial for electrical power systems’planning and operation.Both electrical buildings’load demand and meteorological datasets may contain hidden patterns that are required to be investigated and studied to show their potential impact on load forecasting.The meteorological data are analyzed in this study through different data mining techniques aiming to predict the electrical load demand of a factory located in Riyadh,Saudi Arabia.The factory load and meteorological data used in this study are recorded hourly between 2016 and 2017.These data are provided by King Abdullah City for Atomic and Renewable Energy and Saudi Electricity Company at a site located in Riyadh.After applying the data pre-processing techniques to prepare the data,different machine learning algorithms,namely Artificial Neural Network and Support Vector Regression(SVR),are applied and compared to predict the factory load.In addition,for the sake of selecting the optimal set of features,13 different combinations of features are investigated in this study.The outcomes of this study emphasize selecting the optimal set of features as more features may add complexity to the learning process.Finally,the SVR algorithm with six features provides the most accurate prediction values to predict the factory load.

Wavelet time series MPARIMA modeling for power system short term load forecasting

The wavelet power system short term load forecasting(STLF) uses a mulriple periodical autoregressive integrated moving average(MPARIMA) model to model the mulriple near periodicity, nonstationarity and nonlinearity existed in power system short term quarter hour load time series, and can therefore accurately forecast the quarter hour loads of weekdays and weekends, and provide more accurate results than the conventional techniques, such as artificial neural networks and autoregressive moving average(ARMA) models test results. Obtained with a power system networks in a city in Northeastern part of China confirm the validity of the approach proposed.

A Weighted Combination Forecasting Model for Power Load Based on Forecasting Model Selection and Fuzzy Scale Joint Evaluation

To solve the medium and long term power load forecasting problem,the combination forecasting method is further expanded and a weighted combination forecasting model for power load is put forward.This model is divided into two stages which are forecasting model selection and weighted combination forecasting.Based on Markov chain conversion and cloud model,the forecasting model selection is implanted and several outstanding models are selected for the combination forecasting.For the weighted combination forecasting,a fuzzy scale joint evaluation method is proposed to determine the weight of selected forecasting model.The percentage error and mean absolute percentage error of weighted combination forecasting result of the power consumption in a certain area of China are 0.7439%and 0.3198%,respectively,while the maximum values of these two indexes of single forecasting models are 5.2278%and 1.9497%.It shows that the forecasting indexes of proposed model are improved significantly compared with the single forecasting models.

Autonomous Kernel Based Models for Short-Term Load Forecasting

The application of support vector machines to forecasting problems is becoming popular, lately. Several comparisons between neural networks trained with error backpropagation and support vector machines have shown advantage for the latter in different domains of application. However, some difficulties still deteriorate the performance of the support vector machines. The main one is related to the setting of the hyperparameters involved in their training. Techniques based on meta-heuristics have been employed to determine appropriate values for those hyperparameters. However, because of the high noneonvexity of this estimation problem, which makes the search for a good solution very hard, an approach based on Bayesian inference, called relevance vector machine, has been proposed more recently. The present paper aims at investigating the suitability of this new approach to the short-term load forecasting problem.

Hybrid partial least squares and neural network approach for short-term electrical load forecasting

Intelligent systems and methods such as the neural network （NN） are usually used in electric power systems for short-term electrical load forecasting. However, a vast amount of electrical load data is often redundant, and linearly or nonlinearly correlated with each other. Highly correlated input data can result in erroneous prediction results given out by an NN model. Besides this, the determination of the topological structure of an NN model has always been a problem for designers. This paper presents a new artificial intelligence hybrid procedure for next day electric load forecasting based on partial least squares （PLS） and NN. PLS is used for the compression of data input space, and helps to determine the structure of the NN model. The hybrid PLS-NN model can be used to predict hourly electric load on weekdays and weekends. The advantage of this methodology is that the hybrid model can provide faster convergence and more precise prediction results in comparison with abductive networks algorithm. Extensive testing on the electrical load data of the Puget power utility in the USA confirms the validity of the proposed approach.

Electric Load Forecasting for Shanghai Urban Area

In this paper electric load is forecast for the classified power consumers of Shanghai urban area for the scheduled years in short term and in long term respectively. The monthly load in 1999 is forecast on the basis of the data during 1992～1998, and the approximate load in 2010 is forecast on the basis of the data during 1990～1998.

End-to-end data-driven modeling framework for automated and trustworthy short-term building energy load forecasting

Conventional automated machine learning(AutoML)technologies fall short in preprocessing low-quality raw data and adapting to varying indoor and outdoor environments,leading to accuracy reduction in forecasting short-term building energy loads.Moreover,their predictions are not transparent because of their black box nature.Hence,the building field currently lacks an AutoML framework capable of data quality enhancement,environment self-adaptation,and model interpretation.To address this research gap,an improved AutoML-based end-to-end data-driven modeling framework is proposed.Bayesian optimization is applied by this framework to find an optimal data preprocessing process for quality improvement of raw data.It bridges the gap where conventional AutoML technologies cannot automatically handle missing data and outliers.A sliding window-based model retraining strategy is utilized to achieve environment self-adaptation,contributing to the accuracy enhancement of AutoML technologies.Moreover,a local interpretable model-agnostic explanations-based approach is developed to interpret predictions made by the improved framework.It overcomes the poor interpretability of conventional AutoML technologies.The performance of the improved framework in forecasting one-hour ahead cooling loads is evaluated using two-year operational data from a real building.It is discovered that the accuracy of the improved framework increases by 4.24%–8.79%compared with four conventional frameworks for buildings with not only high-quality but also low-quality operational data.Furthermore,it is demonstrated that the developed model interpretation approach can effectively explain the predictions of the improved framework.The improved framework offers a novel perspective on creating accurate and reliable AutoML frameworks tailored to building energy load prediction tasks and other similar tasks.

Demand of Electric Power and Its Forecasting in Iron and Steel Complex

A systematic study on the electrical load forecasting for large-scale iron and steel companies was made. After analyzing the electrical load＇s characteristics, an algorithm framework for the load forecasting in iron and steel complex was formulated based on model combination and scheme filtration. The algorithm features data quality self- adaptation, convenient forecasting model extension, easy practical application, etc. , and has been successfully applied in Baoshan Iron and Steel Co Ltd, Shanghai, China, resulting in great economic benefit.

Novel grey forecast model and its application

The advancement of grey system theory provides an effective analytic tool for power system load fore-cast. All kinds of presently available grey forecast models can be well used to deal with the short-term load fore-cast. However, they make big errors for medium or long-term load forecasts, and the load that does not satisfythe approximate exponential increasing law in particular. A novel grey forecast model that is capable of distin-guishing the increasing law of load is adopted to forecast electric power consumption (EPC) of Shanghai. Theresults show that this model can be used to greatly improve the forecast precision of EPC for a secondary industryor the whole society.

Day-Ahead Probabilistic Load Flow Analysis Considering Wind Power Forecast Error Correlation

Short-term power flow analysis has a significant influence on day-ahead generation schedule. This paper proposes a time series model and prediction error distribution model of wind power output. With the consideration of wind speed and wind power output forecast error’s correlation, the probabilistic distributions of transmission line flows during tomorrow’s 96 time intervals are obtained using cumulants combined Gram-Charlier expansion method. The probability density function and cumulative distribution function of transmission lines on each time interval could provide scheduling planners with more accurate and comprehensive information. Simulation in IEEE 39-bus system demonstrates effectiveness of the proposed model and algorithm.

Operational Wave Now- and Forecast in the German Bight as a Basis for the Assessment of Wave-Induced Hydrodynamic Loads on Coastal Dikes

The knowledge of the wave-induced hydrodynamic loads on coastal dikes including their temporal and spatial resolution on the dike in combination with actual water levels is of crucial importance of any risk-based early warning system. As a basis for the assessment of the wave-induced hydrodynamic loads, an operational wave now-and forecast system is set up that consists of i) available field measurements from the federal and local authorities and ii) data from numerical simulation of waves in the German Bight using the SWAN wave model. In this study, results of the hindcast of deep water wave conditions during the winter storm on 5–6 December, 2013(German name ‘Xaver') are shown and compared with available measurements. Moreover field measurements of wave run-up from the local authorities at a sea dike on the German North Sea Island of Pellworm are presented and compared against calculated wave run-up using the Eur Otop(2016) approach.

Data-driven comparison of federated learning and model personalization for electric load forecasting

Residential short-term electric load forecasting is essential in modern decentralized power systems.Load forecasting methods mostly rely on neural networks and require access to private and sensitive electric load data for model training.Conventional neural network training aggregates all data on a centralized server to train one global model.However,the aggregation of user data introduces security and data privacy risks.In contrast,this study investigates the modern neural network training methods of federated learning and model personalization as potential solutions to security and data privacy problems.Within an extensive simulation approach,the investigated methods are compared to the conventional centralized method and a pre-trained baseline predictor to compare their respective performances.This study identifies that the underlying data structure of electric load data has a significant influence on the loss of a model.We therefore conclude that a comparison of loss distributions will in fact be considered a comparison of data structures,rather than a comparison of the model performance.As an alternative method of comparison of loss values,this study develops the"differential comparison".The method allows for the isolated comparison of model loss differences by only comparing the losses of two models generated by the same data sample to build a distribution of differences.The differential comparison method was then used to identify model personalization as the best performing model training method for load forecasting among all analyzed methods,with a superior performance in 59.1%of all cases.

基于VMD-改进最优加权法的短期负荷变权组合预测策略

为提升短期电力负荷预测精度,提出了一种变权组合预测策略。首先,为了降低负荷数据的不平稳度,使用变分模态分解(variational mode decomposition,VMD)将负荷数据分解成了高频、低频、残差3种特征模态分量。其次,充分计及负荷数据的时序特点,参考指数加权法原理设计自适应误差重要性量化函数,并结合组合模型在时间窗口内的历史负荷数据的均方预测误差设计改进最优加权法的目标函数和约束条件,以完成子模型的准确变权。最后,针对波动较强的高频分量选定极端梯度提升(XGBoost)和卷积神经网络-长短期记忆(CNN-LSTM)模型并使用改进最优加权法进行组合预测、低频分量使用多元线性回归(MLR)模型预测、残差分量使用LSTM模型预测,叠加各模态分量的预测结果,实现了短期负荷数据的准确预测。实验结果表明,使用策略组合模型的平均绝对百分比误差为4.18%。与使用传统组合策略的组合模型相比,平均绝对百分比预测误差平均降低了0.87%。

基于CNN-BiGRU-Attention的短期电力负荷预测

针对目前电力负荷数据随机性强,影响因素复杂,传统单一预测模型精度低的问题,结合卷积神经网络(Convolutional neural network,CNN)、双向门控循环单元(Bi-directional gated recurrent unit,BiGRU)以及注意力机制(Attention)在短期电力负荷预测上的不同优点,提出一种基于CNN-BiGRU-Attention的混合预测模型。该方法首先通过CNN对历史负荷和气象数据进行初步特征提取,然后利用BiGRU进一步挖掘特征数据间时序关联,再引入注意力机制,对BiGRU输出状态给与不同权重,强化关键特征,最后完成负荷预测。试验结果表明,该模型的平均绝对百分比误差(Mean absolute percentage error,MAPE)、均方根误差(Root mean square error,RMSE)、判定系数(R-square,R~2)分别为0.167%、0.057%、0.993,三项指标明显优于其他模型,具有更高的预测精度和稳定性,验证了模型在短期负荷预测中的优势。

一种时频尺度下的多元短期电力负荷组合预测方法

随机因素的增加导致电力负荷数据成分日渐复杂,使短期负荷预测的难度逐渐增大。针对该问题,提出一种时频尺度下的时间卷积网络与多元线性回归相融合的组合预测模型。利用自适应噪声完备集合经验模态分解(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)在时频域上将负荷数据分解为若干个频率特征不同的本征模态分量,在模糊熵准则下聚类为随机项和趋势项。采用皮尔逊系数从诸多影响因素中筛选出与电力负荷高度相关的特征,鉴于小时间尺度分析更易于挖掘局部细节特征,分别构建了随机项与趋势项的细颗粒度特征集。利用具有强非线性处理能力的时间卷积网络(temporal convolutional network,TCN)去预测随机项,利用结构简单及线性拟合效果好的多元线性回归(multiplelinearregression,MLR)去预测趋势项,将二者的预测结果进行叠加重构后获得最终预测值。在新加坡和比利时两组数据集上的实验结果证明:所提模型具有较高的预测精度、较好的泛化性能及鲁棒性。

基于水电储能调节的风光水发电联合优化调度策略

为缓解新能源装机容量扩大引起的弃风弃光现象,在已有梯级水电上下电站之间加入储能泵站,提出风光水储短期优化调度策略。构建以风光水储系统负荷跟踪误差最小、梯级水电站发电量最大和梯级水电站发电耗水量最小的多目标优化调度模型;提出基于季节性自回归移动平均(seasonal auto-regressive lntegrated moving average, SARIMA)模型和Copula函数的风光出力预测模型作为优化调度模型的边界条件,通过SARIMA预测模型将风光出力历史数据分解为季节性分量、趋势分量以及随机噪声余项进行全天96个调度时段风光出力预测,并叠加上基于Copula函数生成风光出力预测误差,然后通过拉丁超立方采样以及K-means聚类进行场景生成和缩减得到5个风光出力场景。选取风光典型日出力数据为例进行算例分析,算例结果表明:所提预测模型较SARIMA模型可以显著提高预测准确度,模型预测风光出力均方根误差从33.34、229.49 MW分别下降至0.697、9.534 MW;所提优化调度策略可以在全年丰、平、枯水期有效减少弃风弃光现象,并可将过剩新能源中的50%转化为上级水库储存水能。

基于STL-XGBoost-NBEATSx的小时天然气负荷预测

小时天然气负荷预测受外部特征因素与预测方法的影响,为提高其预测精度并解决其他深度学习类模型或组合模型可解释性差、训练时间过长的问题,在引入“小时影响度”这一新特征因素的同时提出一种基于极端梯度提升树(extreme gradient boosting tress,XGBoost)模型与可解释性神经网络模型NBEATSx组合预测的方法;以XGBoost模型作为特征筛选器对特征集数据进行筛选,再将筛选降维后的数据集输入到NBEATSx中训练,提高NBEATSx的训练速度与预测精度;将负荷数据与特征数据经STL(seasonal and trend decomposition using Loess)算法分解为趋势分量、季节分量与残差分量,再分别输入到XGBoost中进行预测,减弱原始数据中的噪音影响;将优化后的NBEATSx与XGBoost模型通过方差倒数法进行组合,得出STL-XGBoost-NBEATSx组合模型的预测结果。结果表明:“小时影响度”这一新特征是小时负荷预测的重要影响因素,STL-XGBoost-NBEATSx模型训练速度有所提高,具有良好的可解释性与更高的预测准确性,模型预测结果的平均绝对百分比误差、均方误差、平均绝对误差分别比其余单一模型平均降低54.20%、63.97%、49.72%,比其余组合模型平均降低24.85%、34.39%、23.41%,模型的决定系数为0.935,能够很好地拟合观测数据。

基于模态分解及GRU-XGBoost短期电力负荷预测

精确的短期电力负荷预测能有效提高电力系统运营水平。针对电力负荷数据受多种因素影响,波动性和随机性强等问题,提出了一种基于模态分解及混合模型的负荷预测方法。首先,采用主成分分析法(principal component analysis,PCA)对负荷特征向量进行处理,去掉冗余信息,再用完全自适应噪声集合经验模态分解(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)将历史负荷分解为简化的几个子序列;其次,选择引入样本熵(sample entropy,SE)来计算子序列熵值,将相近的子序列重构得到随机、细节、低频和趋势分量后选用不同结构门控循环单元(gate recurrent unit,GRU)对不同分量类型进行预测,再使用极致梯度提升模型(extreme gradient boosting,XGBoost)对各分量残差进行拟合,各重组序列的预测值为GRU预测值与XBGoost拟合值之和,重组各序列得到最终预测值。选取3年时电力负荷数据进行实验,结果表明,所提模型的均方根误差(root mean square error,RMSE)、平均绝对百分比误差(mean absolutepercentage error,MAPE)和平均绝对误差(mean absolute error,MAE)分别为370.676 MW、99.07%和246.89 MW,与单一模型和混合模型相比,实现了评价指标的明显减少。