This paper deals with the problem of the electricity consumption forecasting method. An MPSO-BP (modified particle swarm optimization-back propagation) neural network model is constructed based on the history data of a mineral company of Anshan in China. The simulation showed that the convergence of the algorithm and forecasting accuracy using the obtained model are better than those of other traditional ones, such as BP, PSO, fuzzy neural network and so on. Then we predict the electricity consumption of each month in 2017 based on the MPSO-BP neural network model.
Mineral companies consume large quantities of electricity in the processing of coal every day. The electricity consumption predicting system is always an important part of planning and operating of the power. Because of the complicated change of the electrical power system, it is difficult to establish an exact predicting model [1] . Many companies have changed the traditional methods to predict the electricity consumption, but the accuracy is not high. Traditional BP neural network training algorithms are mostly based on the gradient. The speed of network learning process convergence is slow and falls into the local minimum value easily. It is also difficult to decide the number of neurons in the hidden layer. In terms of the electric power loading randomness, it lacks the ability of precise to screen data processing. The original particle swarm optimization (PSO) has many advantages such as the simple algorithm, easily implement and less parameters. However, it has some disadvantages like is not sensitive to the environmental changes and falls into non-optimal regions easily [2][3][4][5] .
In this paper, PSO-BP algorithm is modified to train the neural network parameters, realize the optimizing of the network and achieve the automatically optimized parameters of BP neural network. The algorithm is applied to predict the electricity consumption prediction by using Matlab. In addition, our method is used to compare with methods of BP, PSO, Elman, FNN, and ANFIS [6][7][8][9][10] , the results show that our algorithm has a higher convergence speed, and it provides a higher accuracy for predicting the electricity consumption.
In the PSO algorithm, each individual is called a particle, and each particle represents a potential solution. In the D-dimensional search space, each particle is a point in space and group forms by m particles. z i =(z i1 ,z i2 ,…z iD ) and v i =(v i1 ,v i2 ,…v id ,…,v iD ) are the position vector and the speed vector of i ( i=1,2,…,m) particle, p i =(p i1 ,p i2 ,…p id ,…,p iD ) is the best position of the search particle, p g =(p g1 ,p g2 ,…p gd ,…,p gD )is the best position of all particles. The velocity and position updating equations:
) Where: k is the iteration index; r 1 and r 2 are random numbers among [0, 1]; c 1 and c 2 are the acceleratory coefficient [7] .
Modified ideas: 1. Keeping flight diversity of later stage and different flight speed in the same direction. 2. Dividing particles into two categories: high-speed particles satisfy the global search requirements and avoid premature and local optimum; and low-speed particles satisfy the refined search requirements, and avoid exceeding optimal solution. The modified equations are as follows:
where: 0 id v ( ) is a base part of the particle i in D-dimensional velocity; ( )
z is a part of the particle i in D-dimensional search position; ω is inertia weight; P id is the best position particle achieved based on its own experience; P gd is the best particle position based on overall swarm’s experience.
a(n) is the coefficient variation of speed, N i1 is the maximum speed, N i2 is the minimum speed, a(n) changes the search speed according to the equation (4).
Where: σ is the positive coefficient; k max is the upper limit iteration index;ω 0 is the upper limit ofω (k); k is the iteration index.
The original PSO algorithm cannot keep the convergence of global optimum, and the probability of getting an optimal solution is small. The modified PSO algorithm introduces a speed variable coefficients a(m) and inertia factor ω. It keeps diversity of particle swarm, and it can obtain the global optimum and improve the convergence speed and accuracy.
The following equation can be used to judge the fitness of particles:
where d i is the actual output, t k is the target output, m is the number of output nodes, and n is the number of training set samples.
- Initialization: generating the positions and velocities randomly to decide the local best position (p id ) and the global best position (p gd ). The equation ( 3) and ( 4) decide the initial parameters: σ,ω,ω 0 ,c 1 ,c 2 ,r 1 ,r 2 . 2. Evaluation：calculating the particle fitness function f according to the equation ( 6)，Comparing the best position of each particle with the experience position of each particle, and replacing the current position as the best position if the current position is better than the best position, otherwise, the current position remains unchanged. 3. Update extreme value: comparing the current position of each particle in the group with all the best position experienced. If the position of the particle is better, it will be setting to the best position in the current; otherwise, the position will stay unchanged. 4. Update the inertia weight：The inertia weight is updated according to the equation (5). 5. Update the position and velocity of the particle: the position and velocity of the particle changed by using equation ( 3) and (4). 6. Check: If the curren
This content is AI-processed based on open access ArXiv data.