Strategies for multi-step-ahead available parking spaces forecasting based on wavelet transform

A new methodology for multi-step-ahead forecasting was proposed herein which combined the wavelet transform(WT), artificial neural network(ANN) and forecasting strategies based on the changing characteristics of available parking spaces(APS). First, several APS time series were decomposed and reconstituted by the wavelet transform. Then, using an artificial neural network, the following five strategies for multi-step-ahead time series forecasting were used to forecast the reconstructed time series: recursive strategy, direct strategy, multi-input multi-output(MIMO) strategy, DIRMO strategy(a combination of the direct and MIMO strategies), and newly proposed recursive multi-input multi-output(RECMO) strategy which is a combination of the recursive and MIMO strategies. Finally, integrating the predicted results with the reconstructed time series produced the final forecasted available parking spaces. Three findings appear to be consistently supported by the experimental results. First, applying the wavelet transform to multi-step ahead available parking spaces forecasting can effectively improve the forecasting accuracy. Second, the forecasting resulted from the DIRMO and RECMO strategies is more accurate than that of the other strategies. Finally, the RECMO strategy requires less model training time than the DIRMO strategy and consumes the least amount of training time among five forecasting strategies.

A hybrid decomposition-boosting model for short-term multi-step solar radiation forecasting with NARX neural network

Due to global energy depletion,solar energy technology has been widely used in the world.The output power of the solar energy systems is affected by solar radiation.Accurate short-term forecasting of solar radiation can ensure the safety of photovoltaic grids and improve the utilization efficiency of the solar energy systems.In the study,a new decomposition-boosting model using artificial intelligence is proposed to realize the solar radiation multi-step prediction.The proposed model includes four parts:signal decomposition(EWT),neural network(NARX),Adaboost and ARIMA.Three real solar radiation datasets from Changde,China were used to validate the efficiency of the proposed model.To verify the robustness of the multi-step prediction model,this experiment compared nine models and made 1,3,and 5 steps ahead predictions for the time series.It is verified that the proposed model has the best performance among all models.

Accurate Multi-Site Daily-Ahead Multi-Step PM_(2.5)Concentrations Forecasting Using Space-Shared CNN-LSTM

Accurate multi-step PM_(2.5)(particulate matter with diameters≤2.5 um)concentration prediction is critical for humankinds’health and air populationmanagement because it could provide strong evidence for decisionmaking.However,it is very challenging due to its randomness and variability.This paper proposed a novel method based on convolutional neural network(CNN)and long-short-term memory(LSTM)with a space-shared mechanism,named space-shared CNN-LSTM(SCNN-LSTM)for multi-site dailyahead multi-step PM_(2.5)forecasting with self-historical series.The proposed SCNN-LSTM contains multi-channel inputs,each channel corresponding to one-site historical PM_(2.5)concentration series.In which,CNN and LSTM are used to extract each site’s rich hidden feature representations in a stack mode.Especially,CNN is to extract the hidden short-time gap PM_(2.5)concentration patterns;LSTM is to mine the hidden features with long-time dependency.Each channel extracted features aremerged as the comprehensive features for future multi-step PM_(2.5)concentration forecasting.Besides,the space-shared mechanism is implemented by multi-loss functions to achieve space information sharing.Therefore,the final features are the fusion of short-time gap,long-time dependency,and space information,which enables forecasting more accurately.To validate the proposed method’s effectiveness,the authors designed,trained,and compared it with various leading methods in terms of RMSE,MAE,MAPE,and R^(2)on four real-word PM_(2.5)data sets in Seoul,South Korea.The massive experiments proved that the proposed method could accurately forecast multi-site multi-step PM_(2.5)concentration only using self-historical PM_(2.5)concentration time series and running once.Specifically,the proposed method obtained averaged RMSE of 8.05,MAE of 5.04,MAPE of 23.96%,and R^(2)of 0.7 for four-site daily ahead 10-hourPM_(2.5)concentration forecasting.

Combined forecast method of HMM and LS-SVM about electronic equipment state based on MAGA

For the deficiency that the traditional single forecast methods could not forecast electronic equipment states, a combined forecast method based on the hidden Markov model（HMM） and least square support vector machine（LS-SVM） is presented. The multi-agent genetic algorithm（MAGA） is used to estimate parameters of HMM to overcome the problem that the Baum-Welch algorithm is easy to fall into local optimal solution. The state condition probability is introduced into the HMM modeling process to reduce the effect of uncertain factors. MAGA is used to estimate parameters of LS-SVM. Moreover, pruning algorithms are used to estimate parameters to get the sparse approximation of LS-SVM so as to increase the ranging performance. On the basis of these, the combined forecast model of electronic equipment states is established. The example results show the superiority of the combined forecast model in terms of forecast precision,calculation speed and stability.

Stock Price Forecasting: An Echo State Network Approach

Forecasting stock prices using deep learning models suffers from pro-blems such as low accuracy,slow convergence,and complex network structures.This study developed an echo state network(ESN)model to mitigate such pro-blems.We compared our ESN with a long short-term memory(LSTM)network by forecasting the stock data of Kweichow Moutai,a leading enterprise in China’s liquor industry.By analyzing data for 120,240,and 300 days,we generated fore-cast data for the next 40,80,and 100 days,respectively,using both ESN and LSTM.In terms of accuracy,ESN had the unique advantage of capturing non-linear data.Mean absolute error(MAE)was used to present the accuracy results.The MAEs of the data forecast by ESN were 0.024,0.024,and 0.025,which were,respectively,0.065,0.007,and 0.009 less than those of LSTM.In terms of con-vergence,ESN has a reservoir state-space structure,which makes it perform faster than other models.Root-mean-square error(RMSE)was used to present the con-vergence time.In our experiment,the RMSEs of ESN were 0.22,0.27,and 0.26,which were,respectively,0.08,0.01,and 0.12 less than those of LSTM.In terms of network structure,ESN consists only of input,reservoir,and output spaces,making it a much simpler model than the others.The proposed ESN was found to be an effective model that,compared to others,converges faster,forecasts more accurately,and builds time-series analyses more easily.

On the Time Series Forecasting of Road Traffic Accidents in Ondo State of Nigeria

This paper focuses on time series forecasting of monthly occurrence of fatal road accidents in Ondo State of Nigeria. Its aim, however, is to use time series analysis to analyze the data obtained from Federal Road Safety Corps （FRSC）, Ondo State Command; which was considered in two cases： the total cases reported （TCR） and the number of deaths resulted from accidents （NOD）. Various smoothing models for time series were used to analyze the two cases. Based on the models, predictions were made and the results show a steady increase as a result of long-term effects on road accidents for the two cases. It was found also that simple exponential smoothing model is the appropriate model for both TCR and NOD.

基于平方根UPF的电力系统鲁棒预测状态估计

针对辅助预测状态估计器在迭代计算中会出现状态预测误差协方差矩阵不正定,导致估计精度差甚至发散的问题,提出了基于平方根UPF的电力系统鲁棒辅助预测状态估计。该方法采用两种数学方法:矩阵Cholesky分解因子更新和矩阵QR分解,引入平方根技术动态更新状态预测误差协方差矩阵以保持状态预测误差协方差矩阵的正定性。运用MATLAB进行仿真模拟测试,结果表明:IEEE 30节点系统非高斯噪声测试中,平方根UPF电压相角的均方根误差平均值为UPF相应测试值的0.09%,平方根UPF电压幅值的均方根误差平均值为UPF相应测试值的0.14%;IEEE 57节点系统非高斯噪声测试中,平方根UPF电压相角的均方根误差平均值为UPF相应测试值的0.67%,平方根UPF电压幅值的均方根误差平均值为UPF相应测试值的0.57%。所提出的平方根UPF对解决辅助预测状态估计中状态预测误差协方差矩阵不正定的问题具有很好的效果,具有更高估计精度和鲁棒性。

基于改进Att-LSTNet与无迹粒子滤波融合的主动配电网预测辅助状态估计

针对传统的无迹粒子滤波(unscented particle filter,UPF)存在不准确的新息向量及未知的量测噪声协方差矩阵导致估计精度低的问题,提出一种改进Att-LSTNet与UPF融合的主动配电网预测辅助状态估计(forecasting-aided state estimation,FASE)方法。首先,采用引力搜索算法(gravitational search algorithm,GSA)对支持向量回归(support vector regression,SVR)的关键参数进行优化处理,利用历史数据建立GSA-SVR模型,并将其引入至Att-LSTNet模型的输出层,构建一种增强预测模型。然后,利用UPF中的新息向量来训练该模型,并结合孤立森林算法和箱线图法对原始新息向量进行监控和修正。最后,针对量测噪声协方差矩阵未知的情况,结合修正后的新息向量和UPF计算出未知量测噪声协方差矩阵,并进行状态估计。基于IEEE33与IEEE118节点标准配电系统的算例结果表明,所提出的方法在估计精度、泛化能力和鲁棒性等方面具有优越性。

基于负荷预测和无迹粒子滤波的配电网动态状态估计

随着汽车充电成为新型重要负荷,为确保此时配电网运行与控制安全,对其进行实时准确的态势感知,提出一种基于卷积神经网络和门控循环单元的短期负荷预测与无迹粒子滤波算法自适应混合的配电网动态状态估计方法。结合使用卷积神经网络和门控循环单元进行短期负荷预测,将预测得到的有功与无功功率进行潮流计算,再与无迹粒子滤波量测估计值自适应加权得到电压幅值和相角状态估计结果。以IEEE33节点配电网为例,验证了所提状态估计方法的准确性与面对不良数据时的鲁棒性。

基于TCN-BiLSTM-Attention-ESN的光伏功率预测

针对光伏发电功率随机性强、难以准确预测的问题,提出一种基于时间卷积网络(TCN)、双向长短期记忆网络(BiLSTM)和回声状态网络(ESN)的组合预测方法。首先,使用自适应噪声完备集合经验模态分解(CEEMDAN)将功率数据分解为一系列相对平稳、不同波动模式的子功率序列;再将分解重构后的功率序列和其他特征序列输入到TCN-BiLSTM-Attention-ESN组合模型中,其中TCN-BiLSTM-Attention用于提取光伏序列波动特征并构建时空特征向量;最后,将所提取的时空特征向量输入ESN获得预测结果。采用新疆某光伏电站的光伏功率数据进行验证,结果表明与时下先进的预测方法相比,所提方法具有更高的预测精度,有助于提升光伏发电占比,保障电力系统平衡和运行安全。

基于PLESN和LESQRN概率预测模型的短期电力负荷预测

针对现有电力负荷预测不能很好反映负荷数据的周期性和趋势性以及残差的波动性特征提出一种考虑周期性建模的泄露积分型回声状态网络点预测模型和泄露积分型回声状态分位数回归网络概率预测模型组合的短期电力负荷预测方法.首先为了捕捉负荷的多重特征定义了周期性和趋势性损失函数辅助优化点预测模型然后为克服残差的波动问题利用概率预测模型对点预测值与真实值的残差进行建模预测最后整合同时刻的点预测值与残差预测区间得到概率预测模型结果.实际算例结果表明与其他模型相比所提模型不仅有效抑制尖端振荡现象而且能够生成可靠的概率密度分布.

配电网精细化拓扑运行状态DCNN在线辨识方法

构建以新能源为主体的新型电力系统是实现碳达峰、碳中和目标的重要手段。新型电力系统中新能源将成为主力电源,高渗透率接入的新能源将深刻改变电力系统的形态、特性和机理。提出了一种结合潮流方程和深度神经网络的融合方法求解与量测值最优匹配的拓扑和线路参数估计方法,通过分析海量信息数据,透过数据关系探究电网运行规律,用于配电网精细化拓扑辨识及线路参数估计。首先,利用线性回归方法对拓扑和线路参数进行初步估计,得到初步辨识参数,并对初始辨识参数进行降噪处理;然后,基于深度神经网络对量测数据进行特征筛选,将筛选出的特征类别与相应的拓扑结构一一对应,构建训练数据集,进行离线训练,最终得到训练后的模型,从而得到精准的拓扑结构。最后,在IEEE 33节点配电网中进行了仿真验证,证明了该方法的有效性和较强的工程实用性。

考虑小样本不平衡的主动配电网预测辅助鲁棒状态估计

为了解决主动配电网状态估计中数据集的小样本不平衡问题,提出了基于改进少数过采样技术(synthetic minorityoversamplingtechnique,SMOTE)的Prophet和粒子滤波(particlefilter,PF)的主动配电网预测辅助鲁棒状态估计方法(robust forecasting-aided state estimation, FASE),对主动配电网进行状态估计。首先,针对主动配电网小样本不平衡问题,基于主动配电网的数据特征构建哈希函数,提出利用哈希函数对Borderline-SMOTE+Tomek-Links算法进行优化的方法,处理主动配电网数据集。然后,针对主动配电网海量数据量、分布式能源的出力随机变化等特点,将Prophet预测模型用于主动配电网状态估计,提出了一种基于Prophet-PF的鲁棒FASE方法,达到快速、准确地估计主动配电网状态的目的。最后以IEEE 118节点标准配电网和DTU 7k 47实际配电系统为测试系统进行仿真,结果表明所提方法具有较高的精度和鲁棒性,为主动配电网状态估计提供相应参考。

中美人工智能医药领域专利发展趋势分析

目的为我国人工智能医药领域产业政策制定提供参考。方法检索HimmPat平台2018年至2023年中美两国人工智能医药领域的发明专利数据,通过灰色预测模型预测并比较两国在该领域的技术发展趋势,探讨两国技术类别分布,并解析该领域重点申请人组成。结果我国发明专利申请数量预测将于2028年与美国相当;智能诊疗是人工智能医药领域的研发和投资热点,我国前十大申请人中有5家公司、3所高等院校和2家大型医院,美国则均为公司。结论我国人工智能医药领域目前处于高速发展时期,尤其是发明专利申请数量有超越美国的潜力,需继续加强政策支持,注重人才培养和引进,加快技术成果产业化,提高知识产权保护水平。

利用深度学习预报美国东北部日降水分布

现阶段降水预报主要依靠数值天气预报模式。但受物理参数化、计算资源等因素的影响,基于数值模式的降水预报还存在非常大的不确定性。近年来,深度学习在天气预报领域显示出巨大优势和潜力。本文通过构建神经网络预报美国东北部日降水分布,探讨神经网络模型基于低分辨率气象场(ERA-Interim, 0.7°)预报高分辨率降水(CPC,0.25°)的能力,并比较3种主流网络框架(VGG,ResNet, GoogleNet)在该任务中的表现。结果表明,3种网络框架都对美国东北部日降水分布具有一定的预报能力(VGG框架表现最优),但三者的均方根误差(RMSE)均高于ERA-Interim 24-h(ERA24)的降水预报。3种神经网络的集合预报结果优于ERA24预报,且这三者与ERA24预报结果的集合平均能够显著提高ERA24对不同季节、不同强度降水的预报。

在线选择性集成即时学习风电功率自适应预测

风电功率预测可为风电的并网和优化调度提供有效的指导信息,在风能的开发利用中扮演着举足轻重的角色。然而,由于风电固有的间歇性和随机性给准确的风电功率预测带来巨大挑战。同时,由于受季节性、气候性、设备老化等因素影响,随着时间的推移,风电功率数据特征难免发生改变,这将直接导致离线的风电功率模型性能发生退化。为此,提出一种在线选择性集成即时学习(OSEJIT)自适应风电功率预测方法。首先,为了有效处理风电的非线性和时变性特征,通过相似度、学习器扰动以构建多样性JIT基模型库。其次,为了保证集成有效性,定义基于Friedman检验的多样性指标和基于预测精度的准确性指标以实现模型的在线选择。随后,在线预测阶段,根据模型近期的预测性能通过自适应加权集成的方式获得最终预测值。为了保证基模型库的更新,同时规避模型频繁重建导致计算资源耗费的问题,采用一种基于KL散度的过程状态识别方法以减少模型重建频率。所提方法的有效性和优越性在一个实际风电功率数据应用中获得了验证。

基于支持向量机的电力系统状态估计多类型数据异常检测

为了解决异常数据严重影响电力系统状态估计性能的问题,提出了一种基于支持向量机(SVM)的电力系统预测辅助状态估计(FASE)多类型数据异常检测方法。首先,针对传统FASE的预测准确率欠佳的问题,提出了基于极限学习机的FASE方法,并利用SVM并基于预测数据、量测数据与估计值,实现了对坏数据、负荷突变和单相接地等多种类型的数据异常检测。其次,针对惩罚因子和核函数参数会影响分类精度的问题,提出采用灰狼算法对SVM参数进行优化,在兼顾计算速度的同时提高了数据异常检测的准确率。最后,在IEEE 33和丹麦DTU 7K 47节点主动配电网系统上进行仿真测试,所提方法在正常工况下提升26.08%与26.76%,计算速度提升46.05%,在数据异常情况下准确率综合提升32.04%与29.27%,结果表明,所提方法具备较强的通用性与实时性,可以有效地检测电力系统中各种类型的数据异常,并提高状态估计的性能。

Short-term prediction of photovoltaic power generation based on LMD-EE-ESN with error correction

Considering the instability of the output power of photovoltaic(PV)generation system,to improve the power regulation ability of PV power during grid-connected operation,based on the quantitative analysis of meteorological conditions,a short-term prediction method of PV power based on LMD-EE-ESN with iterative error correction was proposed.Firstly,through the fuzzy clustering processing of meteorological conditions,taking the power curves of PV power generation in sunny,rainy or snowy,cloudy,and changeable weather as the reference,the local mean decomposition(LMD)was carried out respectively,and their energy entropy(EE)was taken as the meteorological characteristics.Then,the historical generation power series was decomposed by LMD algorithm,and the hierarchical prediction of the power curve was realized by echo state network(ESN)prediction algorithm combined with meteorological characteristics.Finally,the iterative error theory was applied to the correction of power prediction results.The analysis of the historical data in the PV power generation system shows that this method avoids the influence of meteorological conditions in the short-term prediction of PV output power,and improves the accuracy of power prediction on the condition of hierarchical prediction and iterative error correction.

民用建筑逐时负荷的理论分析

从复杂多变的实际用电负荷曲线分析入手,基于电能损耗,提出用特定时段内的平均负荷代替实际负荷的等效分析思路,并利用中心极限定理得出不同类型、不同情况下设备平均功率近似趋于平稳所需要的稳态时间。进一步讨论稳态时间对分段时间宽度的影响,定性分析逐时负荷在不同时间、不同配电点的波动,为下一步逐时负荷计算的数据分析提供理论依据。

基于ESN模型的制造商库存需求预测

准确的库存需求预测对制造商提升库存管理能力具有重要意义。首先,针对制造商库存需求非平稳性、非线性和高波动的特点,使用奇异谱分析(SSA)和变分模态分解(VMD)实现特征工程,将原始数据分解为趋势信号和多种不同周期的信号,作为模型的输入。然后,引入回声状态网络(ESN),提出SSA-VMD-ESN模型,对库存需求进行预测。最后,将所提模型应用到Kaggle平台真实的制造商库存需求数据集中,与6种模型对比,并单独使用9个产品实验,结果表明所提出模型各项误差指标均属最优。由此证明,研究所提出模型可以有效提高库存需求预测的准确率,对库存管理具有较强的应用价值。