Power Demand Prediction Research Articles

Prediction of power demand for hybrid vehicles operating in fixed-route service

The efficiency of energy management strategies for hybrid electric vehicles depends significantly on the accuracy of the prediction of the load power which has to be provided by the hybrid drive. For vehicles in fixed-route service, measurements gained during vehicle operation can be used for the design of a predictor for the online calculation of the expected speed profile which is directly related to the load power profile. The paper describes, how the relevant information contained in measurements of the vehicle speed and position is processed and compressed in order to obtain a prediction algorithm that can cope with the limited memory space and computing power of a vehicle controller. The characteristics of the resulting algorithm are illustrated with real-life data.

Prediction of power demand for ship motion control system of sea mining ship fitted with tubular winning system

Prediction of power demand for ship motion control system of sea mining ship fitted with tubular winning system One of the crucial systems of sea mining ship for poly-metallic concretions is its motion control system (SMC). Power of such system depends on sea environment characteristics and main dimensions of the ship. It can be expected that it will have important influence on total power of the ship's power plant and in effect on the mining ship's dimensions. In this paper is presented one of the possible ways of preliminary estimation of design power of SMC system for sea mining ship. Since details of design solution of such system for the ship in question are unknown (ships of the kind have not been built so far) the presented results should be considered to be the first estimation of the order of the power demand.

Prediction of the long-term electric power demand under the influence of A/C systems

This paper examines the long-term power demand behaviour under the influence of air-conditioning (A/C) systems in Kuwait. An artificial neural network- (ANN-) based simulation model has been developed to forecast the long-term power demand considering different A/C-system quantity import scenarios. Beside the A/C factor, four socio-economic factors are utilized as inputs for the simulation model, including gross national product, population, number of buildings, and historic peak power demand. The baseline scenario shows that the peak power demand will reach 27 440 MW by the year 2025, with an average annual power growth rate of 5.9 per cent. The A/C import quantity scenario shows that an average A/C import quantity change of 1 per cent is proportionate to an ∼1.1 per cent change in power demand.

Short term electric load prediction using Fuzzy BP

The privatization of electricity industry in various parts of the world has increased the significance of the load prediction problem and in particular there is a need to understand and predict the demand for power with greater accuracy, even in case of imprecise input data. Prediction of power demand is essential for an efficient operation of any utility company. In this paper, a fuzzy version of neural network, namely Fuzzy back propagation network (Fuzzy BP) has been developed for short-term electric load prediction. The load is predicted using fuzzy back propagation algorithm. This model is capable of handling imprecise information in input data. The proposed architecture consists of a module with 51 inputs and 24 outputs. The inputs are fuzzified and the outputs are crisp values representing the predicted load. The proposed method is implemented in MATLAB. The simulation results are presented for each day (24 hours) of the week. Besides this, a multi layer perceptron (MLP) was also implemented separately and the load was predicted using back propagation algorithm. The results obtained from Fuzzy BP were found to be satisfactory when compared to those of MLP network.

Dynamical Model Reconstruction and Accurate Prediction of Power-Pool Time Series

The emergence of the power pool as a popular institution for trading of power in different countries has led to increased interest in the prediction of power demand and price. In this paper, the authors investigate whether the time series of power-pool demand and price can be modeled as the output of a low-dimensional chaotic dynamical system by using delay embedding and estimation of the embedding dimension, attractor-dimension or correlation-dimension calculation, Lyapunov-spectrum and Lyapunov-dimension calculation, stationarity and nonlinearity tests, as well as prediction analysis. Different dimension estimates are consistent and show close similarity, thus increasing the credibility of the fractal-dimension estimates. The Lyapunov spectrum consistently shows one positive Lyapunov exponent and one zero exponent with the rest being negative, pointing to the existence of chaos. The authors then propose a least squares genetic programming (LS-GP) to reconstruct the nonlinear dynamics from the power-pool time series. Compared to some standard predictors including the radial basis function (RBF) neural network and the local state-space predictor, the proposed method does not only achieve good prediction of the power-pool time series but also accurately predicts the peaks in the power price and demand based on the data sets used in the present study.

A Fuzzy Neural Approach for Forecasting Peak Power Demands

Accurate prediction of future electrical power demands greatly facilitates the task of power generation reliably and economically. In this paper, the new soft computing technologies, namely neuro-fuzzy techniques are adopted to improve precision of medium to long-term load forecasting. Performance of the proposed methodology is verified using simulations of some data pertaining to the Riyadh power system. This approach is compared with time series and neural networks design methods. It is demonstrated that the proposed methodology surpasses other methods by producing very accurate peak load forecasts.

An expert system for unit commitment and power demand prediction using fuzzy logic and neural networks

Abstract: The paper discusses the implementation of a fuzzy logic and artificial neural networks approach to providing a structural framework for the representation, manipulation and utilisation of data and information concerning prediction of power demand and generation commitments. An algorithm has been implemented and trained to predict the power demand at each load point on an hourly basis. The neural network is then implemented to supply the brute force necessary to accommodate the large amount of sensory data to provide the initial evaluation of the generation units to be committed. Results of the fuzzy model show a reasonable correspondence with the actual power demand. A standard deviation error for an hourly based prediction is limited to 4.4. Further refinement of the fuzzy model may produce further improvements. Implementation of artificial neural networks for scheduling an hourly unit commitment based on load demands is also discussed The backpropagation technique based on the I/O mapping method has been chosen for structuring the neural network. Geographically related load points and generating units are clustered into groups. Grouping has significantly reduced the number of inputs and outputs to the neural network and, hence, reduced the system complexity. As a result, both training requirements and running real time interaction are significantly improved. The expert system would replace and utilise the requirement for skilled dispatchers in scheduling the generators. It is anticipated that this facility is more accurate, dynamic, adaptive and more efficient than a skilled dispatcher. The overall cost of power generation is expected to be less if the new facility is used. Initial results have reflected a satisfactory correlation between predicted and actual results, with a standard deviation error of 1.71% and 1.96% in the base load units of HTPS and ATPS respectively.

The Hausdorff dimension as a tool of step-by-step prediction

Forecasts of the random processes occurring in power systems should be characterized by very high accuracy—inaccurate forecasts always cause serious economic losses for electrical energy producers and distributors. The processes of power demand, electrical energy demand, and development of power networks exhibit self-similar features. This peculiarity enables the Hausdorff dimension to be applied to formulate prediction models of the above-mentioned quantities. The models are characterized by simplicity and high accuracy. This work deals with models of step prediction of random linear and periodic functions characterized by self-similarity. The definition of the Hausdorff dimension and the principles of generating self-similar geometrical forms and formulating the prediction models for self-similar processes are presented. Examples of prediction models and their application in electrical power engineering are given. The problem of predicting power demands in state and local electrical power systems is described in detail. The reader is also provided with an annotated list of English-language papers issued in 1918–1980 concerning the prediction of power demands, as well as the most important American and Polish works of the last few years.

Prediction of railway power demand

The short-term prediction problem is considered for power demand at railway energy supply stations of the Dutch national railway company. A model and an adaptive prediction algorithm are presented. The algorithm is based on a self-tuning predictor for the output of a Gaussian system.

Identification of stochastic electric load models from physical data

The three step identification process of model development, parameter estimation, and performance analysis is illustrated through the identification of models for the prediction of electric power demand. Each step is carefully supported by numerical results based on physical data. Three types of progressively more complex but more accurate load models are identified which describe 1) time periodicity, 2) time periodicity plus load autocorrelation, and 3) time periodicity plus load autocorrelation plus dynamic temperature effects. Accurate predictions up to one week are demonstrated. General guidelines are extrapolated from this identification example when possible.