Modified Harmony Search Algorihms And Its Application To Low-carbon Energy Forecasting

Harmony Search Algorithm is a new meta-heuristic optimization algorithm mimicking the improvisation process of musicians. The HS algorithm behaves excellent effectiveness and robustness when applied to several optimization problems and presents lots of advantages compared to other heuristic optimization algorithms. Its searching ability depends on the initialization of Harmony Memory(HM), selection of parameters and new harmony generation pattern. However, it will get into trouble in performing local search for numerical applications. In order to improve the fine-tuning characteristic of HS algorithm, lots of modified HS algorithms are investigated here to enhance the performance of HS algorithm. And these modified algorithms are applied to forecast the power generation, CO2emissions and energy assumption which belongs to the domain of low-carbon energy forecasting. Forecasting the power generation, CO2emission and energy assumption could play important role for energy planning and environmental strategy decision.The main results in this dissertation can be summarized as follows:(1) To improve the performance of HS algorithm and eliminate the drawbacks lies with fixed parameter settings, a novel Multi-Hrmony Memory Adaptive Harmony Search Algorithm(MHMAHS) is proposed. Inspired by the crossover in music domain, multi-HM are employed in MHMAHS--every HM evolves separately, then after certain iterations different HM shares the best harmony. Adaptive parameters are adopted in MHMAHS. Golden selection strategy is employed to define the boundary of pitch adjusting region.By applying HS, MHMAHS, Imroved HS(IHS) and Global HS(GHS) on different benchmark problems, simulation results show that the performance of MHMAHS is better than HS, IHS and GHS.(2) A novel MHMAHS algorithm-based joint parameters optimization combination model(MHS-JPOCM) is proposed to forecast the annual power generation:the single forecasting model adopts a power function form instead of the traditional fixed form, and the exponential parameter in power functions can be adjusted under certain criteria; the exponential parameter and the combination weights, called joint parameters, are adjusted simultaneously; the optimal values of joint parameters are determined by using the MHMAHS algorithm. The annual power generation data for typical countries with different trends are forecasted to test the effect and accuracy of the proposed method. Compared with four single models and four combination models forecasting results, the forecasting results of the proposed methods is the best.(3) Just one new harmony is generated in basic HS algorithm that means the efficiency is poor. A novel modified method named Multi-universe Quantum Harmony Search Algorithm(MUQHS) is presented:quantum harmony is employed in HM to improve the information of harmony; the quantum angle is adjusted adaptively to improve the searching efficiency; adopting multi-universe theory to generate more new harmonies in one iteration to enhance the local searching performance.The simulation results show that the proposed method is efficient, especially to high order problems.(4) The discounted mean square forecast error (DMSFE) combination model is an effective forecasting method. In most researches the discounting factor β is the same for different individual forecasting model and different forecasting period, while it is seemly more reasonable to adopt different P value for different individual model and different period. In this proposed work, the novel intelligence optimization method--MUQHS algorithm is adopted to determine the optimal P values for each individual model and each forecasting period. The empirical analysis shows that the MUQHS based optimization DMSFE combination method performs much better than the original method with arbitrarily choosing parameter β value.(5) A novel Hybrid Harmony Search Algorithm with Catastrophe(HHSC) is presented. Catastrophe theory is employed to help the algorithm jumping out local optimum. Adaptive parameter inspired by Bat algorithm is employed. The simulation results show that the proposed method is efficient, especially its ability to jump out from local optimum.(6) The background values of GM(1,1) forecasting model are assigned as0.5in most researches. And this is one resource of the forecasting error. A modified GM(1,1) forecasting model which adopting HHSC algorithm to optimize the background values is investigated to forecast the annual energy assumption. The empirical analysis shows the proposed method has better prediction accuracy than the original GM(1,1) forecasting model.