Seite - 340 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1282 whereω is theconnectionweightsbetweeninput layerandhiddenlayer;n is the input layerneuron number,andLis thehiddenlayerneuronnumber,and, β=[βi1,βi2, · · · ,βim]L×m (8) whereβ is theconnectionweightsbetweenhiddenlayerandoutput layerandmis theoutput layer neuronnumber,and, X=[xi1,xi2, · · · ,xiQ]n×Q (9) Y= [ yi1,yi2, · · · ,yiQ ] m×Q (10) whereXis the inputvectorandYis thecorrespondingoutputvector,and, H= ⎡⎢⎢⎢⎢⎣ g(ω1x1+b1) g(ω2x1+b2) · · · g(ωlx1+bl) g(ω1x2+b1) g(ω2x2+b2) · · · g(ω1x2+b1) ... ... ... g(ω1xQ+b1) g(ω2xQ+b2) · · · g(ωlxQ+bl) ⎤⎥⎥⎥⎥⎦ (11) whereHis thehidden layeroutputmatrix,b is thebiaswhich isgeneratedrandomly in theprocessof network initialization,andg(x) is theactivationfunctionof theELM. 2.3.AntColonyClusteringAlgorithm Whenprocessing the largenumberof samples, the traditional clustering learningalgorithmoften has thedisadvantagesofslowclusteringspeed, fallingeasily into localoptimal,andit isdifficult to obtain theoptimalclusteringresult.At thesametime, theclusteringalgorithminvolves theselection of thenumberofclusteringK,whichdirectlyaffects theclusteringresult.Usingantcolonyclustering topre-process the loadsamplescanreduce thenumberof inputsamplesonthepremiseof including all sample features,andalsocaneffectivelysimplify thenetworkstructureandreduce thecalculation effort. Theflowchartof theantcolonyclusteringalgorithmisshowninFigure2. Start Initialize the parameters Calculate the Transition probability Of ant Update cluster centers and pheromones Terminal condition End Yes No Figure2.Theflowchartof theantcolonyclusteringalgorithm. 340