電力負(fù)荷預(yù)測(cè)已成為電力工程的重要研究?jī)?nèi)容之一,其是一個(gè)復(fù)雜的多變量、多維度的估計(jì)問題。智能電表及傳感器的應(yīng)用,提供了不同形式分布式能源的大量歷史數(shù)據(jù),進(jìn)一步增加了問題的復(fù)雜性。傳統(tǒng)的負(fù)荷預(yù)測(cè)方法無法準(zhǔn)確跟蹤負(fù)荷隨機(jī)變化,準(zhǔn)確性較差,但基于人工智能的機(jī)器學(xué)習(xí)算法因具有數(shù)據(jù)建模的靈活性和精確性,具有提高負(fù)荷預(yù)測(cè)準(zhǔn)確性的潛力。本文給出了人工神經(jīng)網(wǎng)絡(luò)（ANN）短期預(yù)測(cè)方法和支持向量機(jī)（SVM）的短期負(fù)荷預(yù)測(cè)方法。利用支持向量機(jī)的核函數(shù)對(duì)負(fù)荷預(yù)測(cè)中的復(fù)雜非線性關(guān)系進(jìn)行建模,建立了多項(xiàng)式核、無線基函數(shù)核和皮爾遜函數(shù)核函數(shù)的三種支持向量機(jī)模型。本文比較了支持向量機(jī)模型采用不同核函數(shù)時(shí)的性能,明確了適用于負(fù)荷預(yù)測(cè)的最佳核函數(shù)。通過軟件WEKA的仿真結(jié)果驗(yàn)證了所提方法的性能,確立了支持向量機(jī)參數(shù)、核的優(yōu)化以及神經(jīng)網(wǎng)絡(luò)模型的參數(shù)�；谏鲜鋈蝿�(wù),本文收集了巴基斯坦配電網(wǎng)6周的歷史數(shù)據(jù),并對(duì)該數(shù)據(jù)進(jìn)行預(yù)處理,刪除了丟失值和異常值,使數(shù)據(jù)規(guī)范化以獲得更好的性能;再利用所得數(shù)據(jù)對(duì)支持向量機(jī)模型和神經(jīng)網(wǎng)絡(luò)模型進(jìn)行檢驗(yàn),并基于錯(cuò)誤度量指標(biāo)比較了基于不同核函數(shù)的支持向量機(jī)模型與ANN模型的負(fù)荷預(yù)測(cè)優(yōu)劣。研究結(jié)果表明,基... 

【文章來源】：華北電力大學(xué)(北京)北京市 211工程院校 教育部直屬院校

【文章頁(yè)數(shù)】：78 頁(yè)

【學(xué)位級(jí)別】：碩士

【文章目錄】：
摘要
ABSTRACT
CHAPTER 1 INTRODUCTION
    1.1 MACHINE LEARNING AND DATA ANALYTICS FOR POWER GRIDS
    1.2 RESEARCH STATUS OF LOAD FORECASTING BY ML METHODS
    1.3 MOTIVATION TOWARDS TOPIC
    1.4 THESIS OBJECTIVE
    1.5 SOME NATURAL QUESTIONS
    1.6 SOLUTION OVERVIEW
    1.7 CONTRIBUTION OF THIS THESIS
    1.8 THESIS OUTLINE
CHAPTER 2 LOAD FORECASTING AND MACHINE LEARNING-LITERATURE REVIEW
    2.1 IMPORTANCE OF FORECASTING IN UTILITIES
    2.2 LOAD FORECASTING
        2.2.1 Very Short-Term Load Forecasting（VSTLF）:
        2.2.2 Short-Term Load Forecasting （STLF）:
        2.2.3 Medium Term Load Forecasting （MTLF）:
        2.2.4 Long-Term Load Forecasting（LTLF）:
    2.3 SHORT-TERM LOAD FORECASTING
    2.4 STATISTICAL METHODE
    2.5 MACHINE LEARNING METHODS
    2.6 SUMMART
CHAPTER 3 RESEARCH MODELING
    3.1 DATA COLLECTION
    3.2 ATTRIBUTE SELECTION
    3.3 DATA PREPARATION
    3.4 LIMITATIONS
    3.5 PREPROCESSING OF THE DATASET
    3.6 NORMALIZATION
    3.7 DATA FORECASTING MODELS:
    3.8 MODEL PARAMETERS FOR SVR AND ANN
    3.9 ERROR METRICS
        3.9.1 Popular Error Metrics
        3.9.2 Mean Absolute Error （MAE）
        3.9.3 Mean Absolute Percentage Error （MAPE）
        3.9.4 Root Mean Squared Error （RMSE）
        3.9.5 Mean Squared Error （MSE）
    3.10 WEKA
    3.11 SUMMARY
CHAPTER 4 LOAD FORECASTING MODEL BASED ON SVM
    4.1 SUPPORT VECTOR MACHINES
    4.2 SUPPORT VECTOR CLASSIfiCATION
    4.3 SVC FOR LINEARLY SFPARABLE SET
    4.4 SVC FOR NON-LINEARLY SEPARABLE SETS
    4.5 SUPPORT VECTOR REGRESSION
    4.6 KERNEL FUNCTIONS
    4.7 PEARSON VII KERNEL
    4.8 EXPERIMENTATION FOR KERNEL SELECTION
    4.9 SUMMARY
CHAPTER 5 COMPARISON OF LOAD FORECASTING MODELS
    5.1 ARTIFICIAL NEURAL NETWORK
        5.1.1 Artificial Neurons
        5.1.2 Layers of Neurons
        5.1.3 Backpropagation
        5.1.4 Multi-Layer Perceptron
    5.2 EVALUATION OF THE MODELS
        5.2.1 ANN Model
        5.2.2 SVR Model
    5.3 PERFORMANCE COMPARISON
    5.4 SUMMARY
CHAPTER 6 CONCLUSIONS AND FUTURE WORKS
    6.1 CONCLUSIONS
    6.2 FUTURE WORKS
ACKNOWLEDGEMENT
CHAPTER 7 REFERENCES



本文編號(hào)：2952708