International Core Journal of Engineering 2020-26 | Page 167

4) Establish a predictive model. And constantly adjust its
own parameters to deal with different situations.
neuron of the output layer is w hj . The input received by the h
neuron is:
B. Load Curve
The curve of the distribution network load and time is
called the load curve. It reflects the change in load over time
over a period of time. Through the load curve, the law of load
change in each period can be grasped. Therefore, the plan for
the operation, maintenance and overhaul of the distribution
network equipment can be formulated, and the trend of the
equipment and its load in the distribution network can be
estimated to formulate the construction plan of the
distribution network. 
ℎ
=∑
B. Genetic Algorithm Optimization BP Neural Network
Design
Genetic algorithm [16] is an optimization method that
simulates the genetic and evolutionary methods of the
biological world. Through the three steps of selection,
crossover and mutation, new individuals with better fitness
are added from the old group to join the new group. The
selection operation uses the roulette method to select a
function with good fitness from the old group to form a new
group; the crossover operation crosses the random position
by selecting two individuals from the old group; the mutation
operation selects an individual by random selection from the
old group. The set probability mutations form new
individuals.
A. Overview of BP neural network
BP network is a kind of multi-layer feedforward neural
network [15] . Its name stems from the fact that during the
network training process, the algorithm for adjusting the
weight of the network is the back propagation of the error,
which is the BP learning algorithm. Because of its simple
structure, many adjustable parameters, many training
algorithms, and good operability, BP neural network has been
widely used. As one of the most successful algorithms in
artificial neural networks, it is widely used in data prediction,
data classification, linear fitting and many other aspects.
However, BP neural network also has some inherent defects.
For example, the learning convergence speed is too slow, it
can not guarantee convergence to the global minimum point,
and the network structure is not easy to determine.
Because BP neural network has local minimum
determination, the genetic algorithm is used to optimize the
weight and threshold of the original neural network, which
makes the neural network more powerful, and the predicted
data is more accurate and more practical.
The optimization algorithm is mainly divided into three
parts: BP neural network framework design, genetic
algorithm optimization weight and threshold, neural network
training and prediction. This article has two input parameters
and one output parameter. Therefore, the structure of the
neural network is 2-5-1, corresponding to two nodes of the
input layer, 5 nodes of the hidden layer, and 1 node of the
output layer, a total of 2*5 5*1=15 weights, 5+1=6 Threshold.
Therefore, the coding length of the genetic algorithm is
15+6=21.
y L
E M
T
Z KM E K
¦
K
A total of 1440 groups of urban power load data of a city
on June 16 were used as test data, of which 1300 groups were
randomly selected as training samples and 140 groups were
used as test samples.

b 1
b 2
b h
v 1h
v ih
b q
D K
v dh
(2)
Because the choice of network structure, initial
connection weight and threshold value has great influence on
network training, but it can not be accurately obtained. For
these characteristics, this paper intends to use genetic
algorithm to optimize neural network.
III. N EURAL N ETWORK O PTIMIZATION
w 1j w 2j w hj w qj
(1)
Where b h is the output of the h neuron of the hidden layer.

y j
ℎ
The input received by the jth neuron in the output layer is:
Fig.1. 24-hour electricity load record in a city
y 1
=∑
G
Y LK [ L
¦
L

IV. P REDICTIVE M ODEL E STABLISHMENT
x 1
x i
x d
A. Data preprocessing
Since the units of input data are different, the range of
some data may be particularly large, resulting in slow
convergence of neural networks and long training time. At
the same time, since the value range of the activation function
of the neural network output layer is limited, it is necessary
to map the target data of the network training to the value
range of the activation function.
澳
Fig.2. Neural network diagram
Figure 2 shows a multi-layer feedforward network
structure with d input elements, 1 output neuron, and q hidden
layer neurons. The threshold of the j neuron in the output
layer is represented by θ j , and the threshold of the h neuron
of the hidden layer is represented by γ h . The connection
weight between the i neuron of the input layer and the h
neuron of the hidden layer is v ih , and the connection weight
between the h neuron of the hidden layer neuron and the j
The data needs to be processed before training the neural
network. Normalization is an important preprocessing
145