ارائه الگوريتم جديد براي پيش‌بيني سرعت باد مبتني بر مدل پنهان ماركوف

A Novel Algorithm for Wind Forecasting Based on Hidden Markov Model

در اين مقاله ضمن ارائه مباني نظري مدل پنهان ماركوف، ساختار مناسب آن براي مدل‌سازيِ سري زماني باد پيشنهاد و اجرا شده است. مدل پيشنهادي در شناسايي رژيم‌هاي حاكم در سري‌هاي زماني باد سطح زمين در فرودگاه امام خميني آزمايش و براي اجراي آن از داده جمع‌آوري‌شده طي چهار سال متوالي استفاده شده است. ضمن ارائه آزمون ايستاييِ زماني براي مدل ماركوف مرتبه اول، اين آزمون براي مدل پنهان ماركوف توسعه داده شده است و نتايج آزمون ايستايي دو روش مقايسه شده‌اند. نتايج نشان مي‌دهد كه آزمون ايستايي زماني روي داده سرعت باد در مدل پيشنهادي نسبت به مدل ماركوف مرتبه اول در 70 تا 85 درصد موارد بهبود يافته است كه اين افزايش ايستايي زماني به معني به‌دست‌آوردن دقتِ بيشتر در پيش‌بيني سرعت باد با استفاده از مدل پنهان ماركوف است. اثر تغيير تعداد رژيم‌ها از دو به سه و چهار، در ماه‌هاي مختلف سال بررسي و نتايج آن با نتايج اجراي مدل ماركوف مرتبه اول مقايسه شده است. نتايج نشان از اين دارد كه با تشخيص و تفكيك رژيم با مدل پيشنهادي، در پيش‌بيني ارائه‌شده پراكندگي احتمالات كمتر مي‌شود. درنهايت، با به‌دست‌آوردن پيش‌بيني سرعت باد با روش پيشنهادي و همچنين روش ماركوف مرتبه اول و مقايسه با مقادير واقعي ثبت‌شده و محاسبه ريشه مجموع مربعات خطا براي هر دو روش، نشان داده شده است كه روش پيشنهادي نتايج بهتري توليد مي‌كند.

Meteorological time series are used as important input for risk forecasting and related warning systems. Wind is one of the most important atmospheric parameters because of its extensive effects in many industries and fields of human life. Many researches have been carried out to improve forecasting of the wind with the aim of improving output of wind farms, issuing warning for public, detection of wind shear and turbulence in the airports and so on. Generally, there are two main groups of meteorological forecasting methods, one is based on physical relation of atmospheric parameters, and the other is based on historical data. For a long time, time series of wind have been used for forecasting the wind speed. ARMA (Auto-Regressive Moving Average) and Markov model are two important groups of time series analyzing methods. In this paper, the capability of HMM (Hidden Markov Model) is described and used for identification and classification of wind time series. Based on theoretical concept of HMM, a proper method is proposed, and utilized for simulation with real data. The proposed method is based on constructing a multinomial–HMM on wind direction time series. The whole range of possible wind direction (360 degrees) is divided into 16 groups and then categorized to different regimes. Wind forecasting is then carried out based on these separated categories. Temporal stationary test which is well known for Markov chain, is extended for the proposed method and used for its efficiency evaluation. Efficiency of the proposed model is investigated by using real data of IKIA (Imam Khomeini International Airport). A part of the collected data including wind speed and direction is used for constructing of the proposed model and another part is used for its evaluation. The achieved results show that there is improvement in temporal stationary for HMM vs simple Markov model, in 70 to 80 percent of cases. History of the observations in IKIA shows that there are two major wind directions in the area which are related to the local condition: from mountain to the desert in the day times from north-west and from the opposite direction at nights. These are the only important directions in the area in summer when there are no important meteorological phenomena, while in winter one major direction would be added from south-west because of the large scale meteorological systems. Increasing the number of regimes has also significant improvement in temporal stationary in winter times, while there is no important improvement in summer times. This has a good harmony with long term recorded data.