Seite - 378 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1138 the proposedmethod, the load curve samples are divided into certain categories usingK-means clusteringalgorithm. Sampleswith similar characteristics formacertain category,which formthe input ofGRUneural networks for the corresponding customers. K-means clusteringalgorithm is asimpleandavailablemethodforclusteringthroughunsupervised learningwithfastconvergence and lessparameters. Theonlyparameter,K, numberof clusteringcategory, canbedeterminedby Elbowmethodwith the turningpointof loss functioncurve. Suppose the inputsample isS=x1,x2,...,xm. Thealgorithmisshownas follows. 1. Randomly initializeK clusteringcentroidsc1,c2,...,cK. 2. For i=1,2, ...,m, labeleachsamplexiwiththeclusteringcentroidclosest toxi,gettingKcategories notedbyGk. labeli=argmin 1≤k≤K ‖xi−ck‖, i=1,2,...,m (11) 3. Fork=1,2, ...,K, average thesamplesassignedtoGk toupdateck. ck= 1 |Gk| ∑i∈Gk xi,k=1,2,...,K (12) 4. Repeat Steps 2 and3until the changeof clustering centroidor the loss functionof clustering less thanaset threshold. The loss function isgivenbyEquation(13),wherexj is thesamples in categoriesGk, j=1,2,...,nk andnk is thenumberofsamples incategoriesGk. J(c1,c2,...,cK)= 1 2 K ∑ k=1 nk ∑ j=1 ‖xj−ck‖ (13) Moreover, the factors of date, weather and temperature should be added into input with quantization. First, thepowerconsumptionshouldbedifferentbetweenweekdaysandweekends. Theofficialholidaysarealsoan important factor, sowequantify thedate indexasshowninTable1, wheretheindexofofficialholidaysis1nomatterwhatdayit is. Similarly, theweatherandtemperature arequantifiedaccordingto their innerrelations,asshowninTables2and3. Table1.Quantizationfor the factorsofdate. Date (D) Mon. Tues. Wed. Thur. Fri. Sat. Sun. OfficialHolidays Index 0 0.02 0.04 0.06 0.08 0.6 0.8 1 Table2.Quantizationfor the factorsofweather. Weather (W) Sunny Cloud Overcast LightRain Shower HeavyRain Typhoon Snow Index 0 0.1 0.2 0.4 0.5 0.6 0.8 1 Table3.Quantizationfor the factorsof temperature. Temperature (T/◦C) T≤0 0<T≤10 10<T≤20 20<T≤30 30<T≤40 T≥40 Index 0 0.2 0.4 0.6 0.8 1 2.4. TheProposedFrameworkBasedonGRUNeuralNetworks TheschematicdiagramofproposedframeworkbasedonGRUneuralnetworks forshort-term loadforecasting is showninFigure5. The individual customersareclustered intoa fewcategories formoreaccurate forecasting. Thesamplesarerecordedfromthecategorieswhere thecustomer to 378