Kalman Fusion algorithm in electricity price forecasting

Abstract

In this paper, Kalman Fusion algorithm is applied to combine outputs of three forecasting engines which are used to predict electricity price signal of the Spanish electricity market. Employed engines which are Adaptive Neuro-fuzzy Inference System (ANFIS), Artificial Neural Networks (ANN) and Autoregressive Moving Average (ARMA), are all powerful and popular kinds of time series models. After applying these algorithms on the preprocessed data of the Spanish electricity market, outputs of the aforementioned models are fused by Kalman fusion algorithm in order to exploit the advantages of these forecasting engines simultaneously, as a result of which different patterns existing among price time series can be forecasted more accurately. In comparison with single forecasting methods utilized in this paper to forecast electricity price signal, results of the proposed model based on Kalman Fusion algorithm prove that this approach in effective to enhance accuracy of prediction.

Electricity price forecasting using a new data fusion algorithm

“…Table 1 also compares performance of the modified OWA with those of six other fusion algorithms: average strategy (AS) [23], weighted strategy (WS) [23] and Kalman fusion [24], weighted average fusion (WAF) [35], fixed OWA (offline learning OWA) and original gradient decent‐based OWA (online learning). AS averages input forecasts obtained from the agents.…”

“…It is obvious that the AS is a particular case of the WS where all the agents have the same weight. Kalman fusion is a non‐linear fusion algorithm [24]. The WAF algorithm presented in [35] is an online learning algorithm which assigns weights based on the variance of error of each agent.…”

“…In [23], data fusion framework fusion is applied in wind speed prediction. A Kalman fusion algorithm is also used in electricity price forecasting in [24].…”