Study on the evaluation model of the contribution of power industry expansion considering the influencing factors of new energy consumption

2.2.1 Method for extracting characteristics of electricity consumption behavior of industrial expansion users

The characteristic extraction of electricity consumption behavior of expanding customers involves the characteristic values of capacity utilization hours and load operation ratio.

This paper puts forward the concept of capacity utilization hours according to the definition of generator set utilization hours, aiming at representing the utilization efficiency and power consumption changes of new energy consumption equipment in a certain industry. The utilization hours of industrial expansion capacity of new energy consumption in the industry are the embodiment of electricity consumption. Because of holidays, the difference between the applied capacity and the actual capacity, equipment maintenance and the characteristics of daily electricity in different industries, the utilization hours of stock capacity in different industries are very different.

The utilization hours of industrial capacity can be expressed as: t j , z = P j , z B j(1)

Among them, t_j,z is utilize hours for the business expansion of users in each cycle, j is the periodic ordinal number, z is the periodic attribute of data statistics, with values of month, quarter and year. P_j,z is the power consumption in the jth cycle with the statistical period z.

The load operation ratio of new energy consumption is: W j , z = t j , z t j + m , z(2)

Among them, W_j,z is the proportion of load put into operation for new energy consumption in each cycle; t_j+n,z is the stabilized capacity utilization hours; m is the duration of the stabilization process.

After stabilization, the utilization hours of industrial capacity expansion are: t j + m , z = ∑ i = 1 m t j + m + i , z / m(3)

Due to the stable process of electricity consumption after the business expansion, there is a continuous process of slight fluctuation due to the influence of some related factors. The monthly capacity utilization hours after stabilization can be taken as the average of the monthly utilization hours after stabilization, that is, in Formula (3). m can take 3 5.

In the judgment of the stability process, there are two types of criteria to be selected. If one of them is met, it is considered that the expansion of the business has reached a stable process.

(1) Judge by utilization of electricity consumption change ΔP_j,z Δ P j , z = ( P j , z - P j - 1 , z ) / P j - 1 , z(4)

Among them, P_j-1,z indicates that the statistical period is z under the condition of the j - 1 electricity consumption in a cycle.

(2) Judge by utilization of electricity consumption P_j,z

If P_h,z = f (P_j,z), and P_l,z shows an approximate decreasing trend (when the business expansion type is capacity expansion type) or an approximate increasing trend (when the business expansion type is capacity reduction type), h < z ⩽ m, the h cycle is considered to reach a steady state.

(1) Fitting of electric growth curve for industrial expansion

The growth process of power consumption of users after business expansion is consistent with the characteristics of growth curve, so this paper adopts Logistic model to fit the growth of power consumption components of each business expansion one after another [17], and the expression is: P t = b · t j , z · W j , z 1 + ae - β t(5)

Among them, b, a, β are undetermined parameters; t Is a time parameter; e is the logarithm with irrational number e as the base. P _t is the fitting result of the growth curve of the electricity expansion component.

(2) Analysis and prediction of power expansion components in each industry.

Remember that the monthly electricity growth curve in the user’s historical period is P_N,t, t = 1, 2, . . . , T _k , T _k is the number of months included in the historical period. Monthly electricity consumption P_N,t divided by the number of days in the current month, the daily average power consumption of each month is obtained P _t .

Write down the curve of the user’s net electricity consumption in the historical period and the period to be predicted as follows B _t , whose expression is: B t = B R + ∑ j = 1 M K + M Q R j B A , j γ ( t - t j ) P t(6)

Among them, B _R is the stock capacity; B_A,j is the first occurrence for the user j fluctuation value of secondary business expansion business capacity; R _j is a symbolic function, and for the expansion of capacity expansion, R _j  = 1; For capacity reduction industry expansion, R _j  = - 1; t _j is the user j month in which the secondary expansion occurred; M _K , M _Q is the number of business expansion of users in the historical period and the period to be predicted; γ is a unit step sequence; t = 1, . . . , T _K  + T _Q , T _Q is the number of months included in the time period to be forecasted. For monthly load forecasting, T _Q should not be too long, take T _Q  = 2 ∼3 is appropriate.

The power expansion components of each industry are separated, and the individual power forecast curve of users is obtained [18], with the following steps.

For the time interval t = 1, . . . , T _K  + T _Q , fitting to P₁ ∼ P _t , get the daily average storage power forecast curve p ˆ 0 , t .

According to relevant data, the influence of industry expansion on electricity consumption is lagging behind, and there is no direct connection between the industry expansion in the current month and the electricity consumption in the current month. Therefore, this paper considers reducing the industry expansion that has a practical impact on the electricity consumption in the current month.

Assume that the number of the month of stable power is m, application capacity of the tth monthly for new energy consumption and new clothes business is B ˜ t , after the business happens, the monthly electricity impact ratio is W₁, W₂, . . . , W _m , then the application capacity of new clothes business in the month t in the month of i (i ⩾ t) business expansion is: D i = { B ˜ t W 1 i = t B ˜ t ( W i - t + 1 - W i - t ) t < i ⩽ t + m - 1(9)

Accumulate the business expansion of typical users in this month by different business expansion and installation services, that is, the business expansion and installation increment that has a practical impact on the electricity consumption in this month.

From the characteristics of the monthly electricity curve itself, it can be seen that the monthly electricity consumption of typical users has a dual trend of growth and fluctuation, and the monthly electricity consumption has a similar trend every year, and the electricity consumption in the same month shows an increasing trend at different time periods [20]. In addition, the monthly capacity of industrial expansion also has an important impact on electricity consumption. Based on the above considerations, the input variables selected in this paper are shown in Table 1.

There are seven input variables in the model, 1 6 are the horizontal and vertical historical electricity data; 7 is the actual business expansion calculated according to the business expansion capacity combined with the stable period and the monthly influence ratio. Figure 1 is the structure diagram of RBF network, which is mainly divided into input layer, hidden layer and output layer.

OPEN IN VIEWER

Fig. 1

RBF network structure diagram.

When RBF solves the regression problem, the accuracy of its solution and prediction are greatly affected by the kernel function. Therefore, the types of kernel functions and their corresponding kernel parameters are the key to the prediction model.

The types of kernel functions include global kernel function and local kernel function. Because the kernel function of RBF does not have to meet Mercer condition, combining the two kernel functions can improve the disadvantages of single kernel function. In this paper, binomial kernel function (global function) and Gaussian radial basis kernel function (local kernel function) are linearly combined, and the combined kernel function is obtained as follows:

K ( x , x j ) = ϖ ( μ 1 〈 x , x j 〉 + 1 ) 2 + ( 1 - ϖ ) exp ( - ∥ x , x j ∥ 2 / μ 2 2 )(10)

Among them, ϖ is the weight of a single kernel function, 0 ⩽ ϖ ⩽ 1, when ϖ = 0 or 1, the combined kernel function degenerates into a single kernel function; μ₁ is a binomial kernel parameter; μ₂ is the Gaussian radial basis kernel parameter.

Because RBF only needs to set kernel function parameters, the parameters that need to be optimized in the prediction model are as follows ϖ, μ₁, μ₂. In order to avoid subjectivity, this paper uses particle swarm optimization algorithm to find the optimal model parameters of RBF, and takes the mean square error obtained by K-fold cross training as the evaluation standard of RBF model parameters.

As a bridge between birds and real problems, Particle Swarm Optimization (PSO) simulates the optimization mechanism of birds and finds the optimal solution of RBF model parameter setting problem to solve practical problems. Particle swarm optimization will need to be optimized. The particle swarm algorithm will be the range of E RBF model parameters to be optimized as E-dimensional search space, there are n n particles in E-dimensional search space. The particle position and velocity of each particle consists of E dimension, representing a candidate solution to the solution space of RBF model parameters. Particle Swarm Optimization (PSO) uses velocity-position search and velocity vector to determine the displacement of particles in the search space. By substituting φ _j into the fitness function, the fitness value is obtained to judge the position searched by particles.

The particle representing the parameter solution of RBF model in particle swarm optimization algorithm updates its speed and position according to Formula (11) and Formula (12): V jE = V jE + θ 1 ϑ 1 ( σ jE - φ jE ) + θ 2 ϑ 2 ( σ gE - φ gE )(11) φ jE = φ jE + V jE(12)

Where the subscript j = 1, . . . , n is the number of particles; V _jE is j flight velocity vector of a particle E Dimension component; φ _jE is j particle position vector E Dimension component; σ _j  = (σ_j1, σ_j2, . . . , σ _jE ) is j optimal position searched by particles; σ _g  = (σ_g1, σ_g2, . . . , σ _gE ) is the optimal position searched for the whole particle swarm; Learning factor θ₁, θ₂ is non-negative constant, which directly affects the degree of particles flying to their own optimal position and global optimal position; ϑ₁, ϑ₂ is a random number between 0 and 1.

When training RBF network with particle swarm optimization, the connection weights and thresholds of all neurons in a specific structure should be encoded into individuals represented by real number strings. Neural network contains optimization parameters E, then each individual will be composed of a E dimension vector φ _j  = (φ_j1, φ_j2, . . . , φ _jE ) to represent, take the mean square error generated by the network on the training set as the objective function, and construct the following fitness function to calculate the individual fitness value: ψ = 1 2 M ∑ m = 1 M ∑ l = 1 L ( P ln - P ln ″ ) 2(13)

Among them, P_ln is the forecast result of monthly electricity expansion of the industry calculated and output for neural network; P ln ″ is the output for the target; M is the number of training samples; L is the number of output neurons.

The implementation steps of particle swarm optimization neural network are as follows:

Determine the structure of neural network; Set the group size of particle swarm as n, learning factor, inertia weight, within the allowable range, the initial position and velocity of each particle are given randomly, and the fitness function is defined according to the optimized goal.