CN112508734A - Method and device for predicting power generation capacity of power enterprise based on convolutional neural network - Google Patents
Method and device for predicting power generation capacity of power enterprise based on convolutional neural network Download PDFInfo
- Publication number
- CN112508734A CN112508734A CN202011363422.7A CN202011363422A CN112508734A CN 112508734 A CN112508734 A CN 112508734A CN 202011363422 A CN202011363422 A CN 202011363422A CN 112508734 A CN112508734 A CN 112508734A
- Authority
- CN
- China
- Prior art keywords
- neural network
- convolutional neural
- data
- enterprise
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 134
- 238000010248 power generation Methods 0.000 title claims abstract description 60
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 126
- 230000005611 electricity Effects 0.000 claims abstract description 104
- 238000012360 testing method Methods 0.000 claims abstract description 43
- 239000011159 matrix material Substances 0.000 claims description 40
- 238000011176 pooling Methods 0.000 claims description 18
- 238000004519 manufacturing process Methods 0.000 claims description 16
- 230000009467 reduction Effects 0.000 claims description 10
- 238000005457 optimization Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 2
- 239000013598 vector Substances 0.000 description 18
- 210000002569 neuron Anatomy 0.000 description 8
- 238000011161 development Methods 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000012271 agricultural production Methods 0.000 description 2
- 230000003631 expected effect Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Abstract
The invention provides a method and a device for predicting the power generation capacity of a power enterprise based on a convolutional neural network, wherein the prediction method comprises the following steps: acquiring historical electricity selling data and historical loads of an electric power enterprise through an electric power sensor, establishing a training set based on the acquired data, and establishing a test set based on historical enterprise income data of the electric power enterprise; inputting the training set into a preset convolutional neural network to obtain a training result of the convolutional neural network, and training the convolutional neural network based on the training result and the test set; and inputting the expected electricity selling data of the year into the trained convolutional neural network, determining an expected load through the convolutional neural network, and determining the electricity generating capacity of the year based on the expected load. A nonlinear relation between the historical electricity selling condition and the enterprise income is established based on the convolutional neural network, so that the obstacles of different index data dimensions can be overcome, the relation between the historical electricity selling condition and the enterprise income can be accurately described, and the electric power enterprise can conveniently plan electric power resources reasonably according to the prediction result.
Description
Technical Field
The invention belongs to the field of power generation amount prediction, and particularly relates to a method and a device for predicting power generation amount of a power enterprise based on a convolutional neural network.
Background
In the power industry, the income situation of the next year is generally required to be predicted according to historical data of the past year, so that power resources of the next year are planned, the income situation of a power enterprise is influenced by factors such as user type change, national overall economic development situation and the like, the relevance between the electricity selling quantity and the enterprise income is difficult to establish due to the fact that data dimensions are not uniform, and the accuracy of enterprise income prediction is reduced.
Disclosure of Invention
In order to solve the defects in the prior art, the invention provides a method for predicting the power generation capacity of a power enterprise based on a convolutional neural network, which comprises the following steps:
historical electricity selling data and historical loads of an electric power enterprise are obtained through an electric power sensor, a training set is established based on the obtained data,
establishing a test set based on historical enterprise income data of the power enterprise;
inputting the training set into a preset convolutional neural network to obtain a training result of the convolutional neural network, and training the convolutional neural network based on the training result and the test set;
and inputting the expected electricity selling data of the year into the trained convolutional neural network, determining an expected load through the convolutional neural network, and determining the electricity generating capacity of the year based on the expected load.
Optionally, the obtaining historical electricity selling data and historical load of the power enterprise through the power sensor, and establishing a training set based on the obtained data includes:
acquiring historical electricity selling data and historical loads of an electric power enterprise through an electric power sensor, wherein the historical electricity selling data comprises historical electricity selling quantity of the electric power enterprise in a preset unit time, and the historical loads comprise daytime load density and nighttime load density;
calculating the increase rate of the electricity sold in unit time according to the historical electricity sold;
and adding the electricity selling quantity increase rate, historical electricity selling data and historical load into a training set as training data.
Optionally, the establishing a test set based on historical enterprise revenue data of the power enterprise includes:
acquiring pre-stored historical enterprise income data of the power enterprise, wherein the historical enterprise income data comprises the electricity charge increase rate and the power generation amount of the power enterprise in a preset unit time;
acquiring a domestic production total value and a domestic production total value increase rate in a preset unit time;
calculating the ratio of the electric charge growth rate to the domestic total production value growth rate to obtain the electric power consumption elasticity coefficient of the electric power enterprise;
calculating the ratio of the generated energy to the total domestic production value to obtain the kilowatt-hour power generation value of the power enterprise;
and adding the power consumption elasticity coefficient and the kilowatt-hour power generation value into a test set as test data.
Optionally, the inputting the training set into a preset convolutional neural network to obtain a training result of the convolutional neural network includes:
inputting training data in the training set into a convolutional neural network;
performing feature extraction on input training data through a convolutional layer of a convolutional neural network, and outputting a feature map;
carrying out dimensionality reduction on the characteristic graph through a pooling layer of the convolutional neural network to obtain a training matrix subjected to dimensionality reduction;
and inputting the training matrix into a preset classifier, and outputting a training result through the classifier.
Optionally, the training the convolutional neural network based on the training result and the test set includes:
calculating the average value of the data in the test set;
and calculating the error of the training result of the convolutional neural network relative to the calculated average value, and finishing the training of the convolutional neural network if the error does not exceed a preset threshold value.
Specifically, the calculating an error of the training result of the convolutional neural network with respect to the calculated average value includes:
calculating an error L between a training result f (x) of the convolutional neural network and the calculated average value y based on a formula I:
L＝(y-f(x))2a first formula;
wherein x is data input into the convolutional neural network, and the value ranges of x, y and f (x) are positive numbers.
Optionally, the training of the convolutional neural network further includes a process of optimizing the convolutional neural network, where the process of optimizing includes:
adjusting various parameters in a convolutional neural network based on a gradient descent algorithm, the gradient descent algorithm comprising the formula:
wherein, thetaj' As optimized parameter, θjFor the parameters before optimization, j is the index of the parameter, alpha is the preset learning rate, L (theta)j) For a predetermined parameter thetajThe objective function of (1); thetaj’、θjThe value range of alpha is a real number, and the value range of j is a positive integer.
Optionally, the inputting of the expected electricity selling data of the current year into the trained convolutional neural network, determining an expected load through the convolutional neural network, and determining the electricity generation amount of the current year based on the expected load includes:
determining expected electricity selling data of the year according to the income condition of the enterprise in the previous year;
inputting the predicted electricity selling data and the preset load data into the trained convolutional neural network, and outputting the income of the enterprise in the current year through the convolutional neural network;
if the prediction result does not reach the preset income threshold value, adjusting the input preset load data until the preset income threshold value is reached;
and if the prediction result reaches the preset income limit value, the adjusted preset load data is used as the expected load, and the equipment for power generation is distributed according to the expected load to obtain the predicted power generation amount of the year.
The invention also provides a prediction device of the power generation capacity of the power enterprise of the convolutional neural network based on the same thought, and the prediction device comprises:
training set unit: the system comprises a power sensor, a data acquisition module and a data processing module, wherein the power sensor is used for acquiring historical electricity selling data and historical loads of an electric power enterprise and establishing a training set based on the acquired data;
a test set unit: the method comprises the steps of establishing a test set based on historical enterprise income data of the power enterprise;
a training unit: the test set is used for inputting the training set into a preset convolutional neural network to obtain a training result of the convolutional neural network, and the convolutional neural network is trained based on the training result and the test set;
a prediction unit: and the expected power selling data of the year is input into the trained convolutional neural network, the expected load is determined through the convolutional neural network, and the power generation amount of the year is determined based on the expected load.
Optionally, the prediction unit is configured to:
determining expected electricity selling data of the year according to the income condition of the enterprise in the previous year;
inputting the predicted electricity selling data and the preset load data into the trained convolutional neural network, and outputting the income of the enterprise in the current year through the convolutional neural network;
if the prediction result does not reach the preset income threshold value, adjusting the input preset load data until the preset income threshold value is reached;
and if the prediction result reaches the preset income limit value, the adjusted preset load data is used as the expected load, and the equipment for power generation is distributed according to the expected load to obtain the predicted power generation amount of the year.
The technical scheme provided by the invention has the beneficial effects that:
the nonlinear relation between the electricity selling condition and the enterprise income is established based on the convolutional neural network, the obstacles of different index data dimensions can be overcome, the relation between the two is accurately described, the prediction accuracy of the enterprise income is improved, and the electric power enterprise can plan electric power resources reasonably according to the prediction result.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
Fig. 1 is a schematic flow chart of a method for predicting power generation capacity of an electric power enterprise based on a convolutional neural network according to the present invention;
FIG. 2 is a neural network architecture of a classifier;
fig. 3 is a structural block diagram of the prediction device for power generation amount of the power enterprise based on the convolutional neural network.
Detailed Description
To make the structure and advantages of the present invention clearer, the structure of the present invention will be further described with reference to the accompanying drawings.
Example one
As shown in fig. 1, the present invention provides a method for predicting power generation of an electric power enterprise based on a convolutional neural network, including:
s1: historical electricity selling data and historical loads of an electric power enterprise are obtained through an electric power sensor, and a training set is established based on the obtained data.
In this embodiment, the historical electricity selling data includes the electricity selling amount and the number of users of each type in a preset unit time of the electric power enterprise, and the historical load includes daytime load density and nighttime load density. The types of users include industrial, commercial, agricultural production, and residential class 4 users. Calculating the increase rate of the electricity sold in unit time according to the electricity sold, calculating the increase rate of the users in unit time according to the number of the users of each type, and adding the increase rate of the electricity sold and the increase rate of the users into a training set as training data. In this embodiment, the increase rate of the electricity sold per month with respect to the electricity sold in the previous month is calculated by taking a month as a unit time, that is, the electricity sold increase rate is (n-th month electricity sold/n-1-th month electricity sold) x 100%, the user increase rate is (n-th month user number/n-1-th month user number) x 100%, and the calculated electricity sold increase rate is added to the training set in the form of a row vector of 1 × 12, that is, 12-dimensional training data.
In the training set, besides the electricity sales amount and the electricity sales amount increase rate, other indexes reflecting the electricity development are specifically as follows:
(1) the total monthly power consumption of the enterprise: the method comprises the steps that the total power supply of each month of power enterprises to industrial users and commercial users is represented, index data of the monthly power consumption total of the enterprises in the current year are obtained, and 1 x 12 row vectors, namely 12-dimensional training data are formed;
(2) the total monthly electricity consumption of residents: expressing the total electricity consumption of residents in each month and the region to obtain index data of the total monthly electricity consumption of the residents in the year, and forming a 1 x 12 row vector, namely 12-dimensional training data;
(3) daytime load density: representing the ratio of the average load to the area in an area from 6 points in the day to 6 points at night, reflecting the daytime economic index in the area, and forming a 1 x 365 row vector, namely 365-dimensional training data by taking the day as a unit;
(4) load density at night: the average load-area ratio in the area from 6 nights to 6 daytime in the next day is represented, the nighttime economic index in the area is reflected, and a 1 x 365 row vector, namely 365-dimensional training data is formed by taking the day as a unit.
S2: and establishing a test set based on historical enterprise income data of the power enterprise.
The test set of the convolutional neural network comprises two parts of data of power consumption elasticity coefficient and kilowatt-hour power generation value. The method comprises the steps of firstly, obtaining pre-stored historical enterprise income data of the electric power enterprise, wherein the historical enterprise income data comprises the electric power charge increase rate and the generated energy of the electric power enterprise in a preset unit time. A Gross Domestic Product (GDP) and a rate of increase in the Gross Domestic Product (GDP) are obtained in a preset unit time, and in this embodiment, the GDP and the rate of increase in the Gross Domestic Product are both known in advance. And calculating the ratio of the electricity charge growth rate to the domestic total production value growth rate in a preset time period to obtain the electric power consumption elasticity coefficient of the electric power enterprise. In the present embodiment, the ratio of the electricity rate increase in a certain year to the total domestic production rate increase in the same year is calculated in units of years, and the electricity consumption elasticity coefficient reflects the relationship between electricity consumption and national economic development. And calculating the ratio of the generated energy to the total domestic production value to obtain the kilowatt-hour power generation value of the power enterprise, wherein the kilowatt-hour power generation value reflects the GDP generated by each kilowatt-hour. In this embodiment, the annual power consumption elasticity coefficient and the kilowatt-hour power generation value are added to the test set as test data on an annual basis.
The data in the training set reflect the condition of power development, and the data in the testing set reflect the condition of economic benefit created by power enterprises.
S3: and inputting the training set into a preset convolutional neural network to obtain a training result of the convolutional neural network, and training the convolutional neural network based on the training result and the test set.
In the embodiment, the convolutional neural network comprises four parts, namely an input layer, a convolutional layer, a pooling layer and a classifier.
The training data of 12 dimensions in the training set are spliced into a first matrix of 4 x 12, the training data of 365 dimensions are spliced into a second matrix of 2 x 365, and the first matrix and the second matrix are input into the convolutional neural network through an input layer, namely 4 types of training data including the electricity selling growth rate of 12 months in a year, the number of newly added users in each month, the total electricity consumption of an enterprise and the total electricity consumption of a resident month obtained in S1, and 2 types of training data including the night economic index and the daytime economic index of 365 days in the year are respectively input into the convolutional neural network.
And performing feature extraction on the input training data through a convolutional layer of the convolutional neural network, and outputting a feature map. In this embodiment, the size of the convolution kernel used by the first matrix composed of 12-dimensional vectors is set to 3 × 3, the number is 3, and the sliding step s is 1; the convolution kernel used by the second matrix composed of 365-dimensional vectors is set to 1 × 5, the number of the convolution kernels is 1, and the sliding step s is 12. The length H 'and the width w' of the output feature map can be calculated by the following formulas:
for the first matrix, H is the length of the input matrix, which is 4 in this embodiment; w is the width of the input matrix, 12 in this example; k is a radical of1The length of the convolution kernel is 3, k in this example2The width of the convolution kernel, which is 3 in this example; p is a fill value, 0 in this example; s is the sliding step, which is 1 in this embodiment. Thus, the output signature size is 2 × 10 × 3.
For the second matrix, H is the length of the input matrix, which is 2 in this embodiment; w is the width of the input matrix, which in this embodiment is 365; k is a radical of1The length of the convolution kernel is 1, k in this example2The width of the convolution kernel, 5 in this example; p is a fill value, 0 in this example; s is the step size of the sliding, 12 in this example. Thus, the output signature size is 2 × 31 × 1.
And performing dimensionality reduction on the characteristic diagram through a pooling layer of the convolutional neural network, and pooling the first matrix and the second matrix by using the maximum value to obtain two training matrices after dimensionality reduction. In this embodiment, the size of the pooling window of the first matrix is 2 × 2, the step size is 1, and the size of the output can be calculated to be 1 × 9 × 3 in a manner similar to convolution; the second matrix has a pooling window size of 2 x 2 with a step size of 1, resulting in an output size of 1 x 30 x 1.
And finally, expanding the two training matrixes subjected to the dimensionality reduction treatment into vectors, then merging the vectors, inputting the merged matrixes into a preset classifier, and outputting training results through the classifier. The neural network structure of the classifier is shown in fig. 2, in this embodiment, the number of neurons in the input layer of the classifier is the sum of the size 1 × 9 × 3 after the first matrix pooling and the size 1 × 30 × 1 after the second matrix pooling, which is 57 neurons in total, the hidden layer includes 8 neurons, and the output layer is one neuron. The classifier adjusts the dimension of the output result of the pooling layer through the output layer, so that the dimension can be compared with the data in the test set.
The training of the convolutional neural network based on the training result and the test set comprises:
firstly, the average value of the data in the test set is calculated, namely the average value is obtained after the electric power consumption elasticity coefficient and the kilowatt-hour power generation value are summed.
And then calculating the error of the training result of the convolutional neural network relative to the calculated average value, wherein the error is the error L between the training result f (x) of the convolutional neural network calculated based on a formula I and the calculated average value y, and the training result f (x) is the result output by the output layer of the classifier:
L＝(y-f(x))2a first formula;
wherein x is data input into the convolutional neural network, and the value ranges of x, y and f (x) are positive numbers.
If the error L does not exceed the preset threshold value, the training of the convolutional neural network is finished, and the fact that a regression model among the historical electricity selling data, the historical user quantity and the historical enterprise income of the power enterprise is established at the moment is explained, and the regression model can be used for predicting the relation among the electricity selling quantity, the historical user quantity and the enterprise income in the current year. If the error L exceeds a preset threshold value, the training of the convolutional neural network at the moment is not expected to achieve the expected effect, model parameters in the convolutional neural network need to be adjusted, and the convolutional neural network is optimized, wherein the optimization process comprises the following steps:
each parameter in the convolutional neural network is adjusted based on a gradient descent algorithm, the gradient descent is one of iterative methods and can be used for solving a least square problem, and the gradient descent is one of the most frequently adopted methods when model parameters of a machine learning algorithm, namely an unconstrained optimization problem, are solved. The gradient descent algorithm includes the formula:
wherein, thetaj' As optimized parameter, θjJ is a parameter label before optimization, alpha is a preset learning rate, and L (theta) is a preset objective function; thetaj'、θjThe value range of alpha is a real number, and the value range of j is a positive integer.
S4: and inputting the expected electricity selling data of the year into the trained convolutional neural network, determining an expected load through the convolutional neural network, and determining the electricity generating capacity of the year based on the expected load.
And determining expected electricity selling data of the year according to the income condition of the enterprise in the last year.
Inputting the predicted electricity selling data and the preset load data into the trained convolutional neural network, and outputting the income of the enterprise in the current year through the convolutional neural network;
if the prediction result does not reach the preset income threshold value, adjusting the input preset historical load until the preset income threshold value is reached;
and if the prediction result reaches the preset income limit value, the adjusted preset load data is used as the expected load, equipment for power generation is distributed according to the expected load, the predicted power generation amount of the current year is obtained, namely the power generation equipment is accessed between 6 days and 6 nights according to the input daytime load density, the power generation equipment is accessed between 6 nights and according to the input nighttime load density, and the power generation equipment outputs the predicted power generation amount of the current year.
The expected electricity sales amount and the number of users of the power enterprise in the year are input into the trained convolutional neural network, and the enterprise income which can be achieved by the power enterprise is predicted through the convolutional neural network, so that the power enterprise can plan power resources reasonably according to the prediction result, and adjust the production and operation plan in the year in time.
Example two
As shown in fig. 3, the present invention further provides a prediction apparatus 5 for power generation amount of an electric power enterprise based on a convolutional neural network, including:
training set unit 51: the method is used for acquiring historical electricity selling data and historical loads of the power enterprises through the power sensor and establishing a training set based on the acquired data. The method is specifically used for:
in this embodiment, the historical electricity selling data includes the electricity selling amount and the number of users of each type in a preset unit time of the electric power enterprise, and the historical load includes daytime load density and nighttime load density. The types of users include industrial, commercial, agricultural production, and residential class 4 users. Calculating the increase rate of the electricity sold in unit time according to the electricity sold, calculating the increase rate of the users in unit time according to the number of the users of each type, and adding the increase rate of the electricity sold and the increase rate of the users into a training set as training data. In this embodiment, the increase rate of the electricity sold per month with respect to the electricity sold in the previous month is calculated by taking a month as a unit time, that is, the electricity sold increase rate is (n-th month electricity sold/n-1-th month electricity sold) x 100%, the user increase rate is (n-th month user number/n-1-th month user number) x 100%, and the calculated electricity sold increase rate is added to the training set in the form of a row vector of 1 × 12, that is, 12-dimensional training data.
In the training set, besides the electricity sales amount and the electricity sales amount increase rate, other indexes reflecting the electricity development are specifically as follows:
(1) the total monthly power consumption of the enterprise: the method comprises the steps that the total power supply of each month of power enterprises to industrial users and commercial users is represented, index data of the monthly power consumption total of the enterprises in the current year are obtained, and 1 x 12 row vectors, namely 12-dimensional training data are formed;
(2) the total monthly electricity consumption of residents: expressing the total electricity consumption of residents in each month and the region to obtain index data of the total monthly electricity consumption of the residents in the year, and forming a 1 x 12 row vector, namely 12-dimensional training data;
(3) daytime load density: representing the ratio of the average load to the area in an area from 6 points in the day to 6 points at night, reflecting the daytime economic index in the area, and forming a 1 x 365 row vector, namely 365-dimensional training data by taking the day as a unit;
(4) load density at night: the average load-area ratio in the area from 6 nights to 6 daytime in the next day is represented, the nighttime economic index in the area is reflected, and a 1 x 365 row vector, namely 365-dimensional training data is formed by taking the day as a unit.
Test set unit 52: the method is used for establishing a test set based on historical enterprise revenue data of the power enterprise. The method is specifically used for:
the test set of the convolutional neural network comprises two parts of data of power consumption elasticity coefficient and kilowatt-hour power generation value. The method comprises the steps of firstly, obtaining pre-stored historical enterprise income data of the electric power enterprise, wherein the historical enterprise income data comprises the electric power charge increase rate and the generated energy of the electric power enterprise in a preset unit time. A Gross Domestic Product (GDP) and a rate of increase in the Gross Domestic Product (GDP) are obtained in a preset unit time, and in this embodiment, the GDP and the rate of increase in the Gross Domestic Product are both known in advance. And calculating the ratio of the electricity charge growth rate to the domestic total production value growth rate in a preset time period to obtain the electric power consumption elasticity coefficient of the electric power enterprise. In the present embodiment, the ratio of the electricity rate increase in a certain year to the total domestic production rate increase in the same year is calculated in units of years, and the electricity consumption elasticity coefficient reflects the relationship between electricity consumption and national economic development. And calculating the ratio of the generated energy to the total domestic production value to obtain the kilowatt-hour power generation value of the power enterprise, wherein the kilowatt-hour power generation value reflects the GDP generated by each kilowatt-hour. In this embodiment, the annual power consumption elasticity coefficient and the kilowatt-hour power generation value are added to the test set as test data on an annual basis.
The data in the training set reflect the condition of power development, and the data in the testing set reflect the condition of economic benefit created by power enterprises.
The training unit 53: and the test set is used for inputting the training set into a preset convolutional neural network to obtain a training result of the convolutional neural network, and training the convolutional neural network based on the training result and the test set. The method is specifically used for:
in the embodiment, the convolutional neural network comprises four parts, namely an input layer, a convolutional layer, a pooling layer and a classifier.
The training data of 12 dimensions in the training set are spliced into a first matrix of 4 x 12, the training data of 365 dimensions are spliced into a second matrix of 2 x 365, and the first matrix and the second matrix are input into the convolutional neural network through an input layer, namely 4 types of training data including the electricity selling growth rate of 12 months in a year, the number of newly added users in each month, the total electricity consumption of an enterprise and the total electricity consumption of a resident month obtained in S1, and 2 types of training data including the night economic index and the daytime economic index of 365 days in the year are respectively input into the convolutional neural network.
And performing feature extraction on the input training data through a convolutional layer of the convolutional neural network, and outputting a feature map. In this embodiment, the size of the convolution kernel used by the first matrix composed of 12-dimensional vectors is set to 3 × 3, the number is 3, and the sliding step s is 1; the convolution kernel used by the second matrix composed of 365-dimensional vectors is set to 1 × 5, the number of the convolution kernels is 1, and the sliding step s is 12. The length H 'and the width w' of the output feature map can be calculated by the following formulas:
for the first matrix, H is the length of the input matrix, which is 4 in this embodiment; w is the width of the input matrix, 12 in this example; k is a radical of1The length of the convolution kernel is 3, k in this example2The width of the convolution kernel, which is 3 in this example; p is a fill value, 0 in this example; s is the sliding step, which is 1 in this embodiment. Thus, the output signature size is 2 × 10 × 3.
For the second matrix, H is the length of the input matrix, which is 2 in this embodiment; w is the width of the input matrix, which in this embodiment is 365; k is a radical of1The length of the convolution kernel is 1, k in this example2The width of the convolution kernel, 5 in this example; p is a fillerRecharge, which is 0 in this embodiment; s is the step size of the sliding, 12 in this example. Thus, the output signature size is 2 × 31 × 1.
And performing dimensionality reduction on the characteristic diagram through a pooling layer of the convolutional neural network, and pooling the first matrix and the second matrix by using the maximum value to obtain two training matrices after dimensionality reduction. In this embodiment, the size of the pooling window of the first matrix is 2 × 2, the step size is 1, and the size of the output can be calculated to be 1 × 9 × 3 in a manner similar to convolution; the second matrix has a pooling window size of 2 x 2 with a step size of 1, resulting in an output size of 1 x 30 x 1.
And finally, expanding the two training matrixes subjected to the dimensionality reduction treatment into vectors, then merging the vectors, inputting the merged matrixes into a preset classifier, and outputting training results through the classifier. The neural network structure of the classifier is shown in fig. 2, in this embodiment, the number of neurons in the input layer of the classifier is the sum of the size 1 × 9 × 3 after the first matrix pooling and the size 1 × 30 × 1 after the second matrix pooling, which is 57 neurons in total, the hidden layer includes 8 neurons, and the output layer is one neuron. The classifier adjusts the dimension of the output result of the pooling layer through the output layer, so that the dimension can be compared with the data in the test set.
The training unit 53 is further configured to:
firstly, the average value of the data in the test set is calculated, namely the average value is obtained after the electric power consumption elasticity coefficient and the kilowatt-hour power generation value are summed.
And then calculating the error of the training result of the convolutional neural network relative to the calculated average value, wherein the error is the error L between the training result f (x) of the convolutional neural network calculated based on a formula I and the calculated average value y, and the training result f (x) is the result output by the output layer of the classifier:
L＝(y-f(x))2a first formula;
wherein x is data input into the convolutional neural network, and the value ranges of x, y and f (x) are positive numbers.
If the error L does not exceed the preset threshold value, the training of the convolutional neural network is finished, and the fact that a regression model among the historical electricity selling data, the historical user quantity and the historical enterprise income of the power enterprise is established at the moment is explained, and the regression model can be used for predicting the relation among the electricity selling quantity, the historical user quantity and the enterprise income in the current year. If the error L exceeds a preset threshold value, the training of the convolutional neural network at the moment is not expected to achieve the expected effect, model parameters in the convolutional neural network need to be adjusted, and the convolutional neural network is optimized, wherein the optimization process comprises the following steps:
each parameter in the convolutional neural network is adjusted based on a gradient descent algorithm, the gradient descent is one of iterative methods and can be used for solving a least square problem, and the gradient descent is one of the most frequently adopted methods when model parameters of a machine learning algorithm, namely an unconstrained optimization problem, are solved. The gradient descent algorithm includes the formula:
wherein, thetaj' As optimized parameter, θjJ is a parameter label before optimization, alpha is a preset learning rate, and L (theta) is a preset objective function; thetaj'、θjThe value range of alpha is a real number, and the value range of j is a positive integer.
The prediction unit 54: and the expected power selling data of the year is input into the trained convolutional neural network, the expected load is determined through the convolutional neural network, and the power generation amount of the year is determined based on the expected load. The method is specifically used for:
and determining expected electricity selling data of the year according to the income condition of the enterprise in the last year.
Inputting the predicted electricity selling data and the preset load data into the trained convolutional neural network, and outputting the income of the enterprise in the current year through the convolutional neural network;
if the prediction result does not reach the preset income threshold value, adjusting the input preset historical load until the preset income threshold value is reached;
and if the prediction result reaches the preset income limit value, the adjusted preset load data is used as the expected load, equipment for power generation is distributed according to the expected load, the predicted power generation amount of the current year is obtained, namely the power generation equipment is accessed between 6 days and 6 nights according to the input daytime load density, the power generation equipment is accessed between 6 nights and according to the input nighttime load density, and the power generation equipment outputs the predicted power generation amount of the current year.
The expected electricity sales amount and the number of users of the power enterprise in the year are input into the trained convolutional neural network, and the enterprise income which can be achieved by the power enterprise is predicted through the convolutional neural network, so that the power enterprise can plan power resources reasonably according to the prediction result, and adjust the production and operation plan in the year in time.
The sequence numbers in the above embodiments are merely for description, and do not represent the sequence of the assembly or the use of the components.
The above description is only exemplary of the present invention and should not be taken as limiting the invention, as any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. The prediction method of the power generation capacity of the power enterprise based on the convolutional neural network is characterized by comprising the following steps:
historical electricity selling data and historical loads of an electric power enterprise are obtained through an electric power sensor, a training set is established based on the obtained data,
establishing a test set based on historical enterprise income data of the power enterprise;
inputting the training set into a preset convolutional neural network to obtain a training result of the convolutional neural network, and training the convolutional neural network based on the training result and the test set;
and inputting the expected electricity selling data of the year into the trained convolutional neural network, determining an expected load through the convolutional neural network, and determining the electricity generating capacity of the year based on the expected load.
2. The method for predicting the power generation amount of the power enterprise based on the convolutional neural network as claimed in claim 1, wherein the step of acquiring historical power selling data and historical load of the power enterprise through a power sensor and establishing a training set based on the acquired data comprises the following steps:
acquiring historical electricity selling data and historical loads of an electric power enterprise through an electric power sensor, wherein the historical electricity selling data comprises historical electricity selling quantity of the electric power enterprise in a preset unit time, and the historical loads comprise daytime load density and nighttime load density;
calculating the increase rate of the electricity sold in unit time according to the historical electricity sold;
and adding the electricity selling quantity increase rate, historical electricity selling data and historical load into a training set as training data.
3. The prediction method for power generation capacity of power enterprises based on convolutional neural network of claim 1, wherein the test set is established based on historical enterprise revenue data of the power enterprises, and comprises the following steps:
acquiring pre-stored historical enterprise income data of the power enterprise, wherein the historical enterprise income data comprises the electricity charge increase rate and the power generation amount of the power enterprise in a preset unit time;
acquiring a domestic production total value and a domestic production total value increase rate in a preset unit time;
calculating the ratio of the electric charge growth rate to the domestic total production value growth rate to obtain the electric power consumption elasticity coefficient of the electric power enterprise;
calculating the ratio of the generated energy to the total domestic production value to obtain the kilowatt-hour power generation value of the power enterprise;
and adding the power consumption elasticity coefficient and the kilowatt-hour power generation value into a test set as test data.
4. The method for predicting the power generation amount of the power enterprise based on the convolutional neural network as claimed in claim 1, wherein the step of inputting the training set into a preset convolutional neural network to obtain the training result of the convolutional neural network comprises the following steps:
inputting training data in the training set into a convolutional neural network;
performing feature extraction on input training data through a convolutional layer of a convolutional neural network, and outputting a feature map;
carrying out dimensionality reduction on the characteristic graph through a pooling layer of the convolutional neural network to obtain a training matrix subjected to dimensionality reduction;
and inputting the training matrix into a preset classifier, and outputting a training result through the classifier.
5. The method for predicting the power generation amount of the power enterprise based on the convolutional neural network as claimed in claim 1, wherein the training of the convolutional neural network based on the training result and the test set comprises:
calculating the average value of the data in the test set;
and calculating the error of the training result of the convolutional neural network relative to the calculated average value, and finishing the training of the convolutional neural network if the error does not exceed a preset threshold value.
6. The method for predicting the power generation amount of the power enterprise based on the convolutional neural network as claimed in claim 5, wherein the calculating the error of the training result of the convolutional neural network relative to the calculated average value comprises:
calculating an error L between a training result f (x) of the convolutional neural network and the calculated average value y based on a formula I:
L＝(y-f(x))2a first formula;
wherein x is data input into the convolutional neural network, and the value ranges of x, y and f (x) are positive numbers.
7. The convolutional neural network based power generation enterprise prediction method of claim 1, wherein the training of the convolutional neural network further comprises a process of optimizing the convolutional neural network, and the process of optimizing comprises:
adjusting various parameters in a convolutional neural network based on a gradient descent algorithm, the gradient descent algorithm comprising the formula:
wherein, thetaj' As optimized parameter, θjFor the parameters before optimization, j is the index of the parameter, alpha is the preset learning rate, L (theta)j) For a predetermined parameter thetajThe objective function of (1); thetaj'、θjThe value range of alpha is a real number, and the value range of j is a positive integer.
8. The method for predicting the power generation amount of the power enterprise based on the convolutional neural network as claimed in claim 1, wherein the step of inputting expected power selling data of the current year into the trained convolutional neural network, determining an expected load through the convolutional neural network, and determining the power generation amount of the current year based on the expected load comprises the following steps:
determining expected electricity selling data of the year according to the income condition of the enterprise in the previous year;
inputting the predicted electricity selling data and the preset load data into the trained convolutional neural network, and outputting the income of the enterprise in the current year through the convolutional neural network;
if the prediction result does not reach the preset income threshold value, adjusting the input preset load data until the preset income threshold value is reached;
and if the prediction result reaches the preset income limit value, the adjusted preset load data is used as the expected load, and the equipment for power generation is distributed according to the expected load to obtain the predicted power generation amount of the year.
9. The prediction device of electric power enterprise generated energy based on convolutional neural network, the prediction device includes:
training set unit: the system comprises a power sensor, a data acquisition module and a data processing module, wherein the power sensor is used for acquiring historical electricity selling data and historical loads of an electric power enterprise and establishing a training set based on the acquired data;
a test set unit: the method comprises the steps of establishing a test set based on historical enterprise income data of the power enterprise;
a training unit: the test set is used for inputting the training set into a preset convolutional neural network to obtain a training result of the convolutional neural network, and the convolutional neural network is trained based on the training result and the test set;
a prediction unit: and the expected power selling data of the year is input into the trained convolutional neural network, the expected load is determined through the convolutional neural network, and the power generation amount of the year is determined based on the expected load.
10. The convolutional neural network-based power generation amount prediction device for an electric power enterprise as claimed in claim 9, wherein the prediction unit is configured to:
determining expected electricity selling data of the year according to the income condition of the enterprise in the previous year;
inputting the predicted electricity selling data and the preset load data into the trained convolutional neural network, and outputting the income of the enterprise in the current year through the convolutional neural network;
if the prediction result does not reach the preset income threshold value, adjusting the input preset load data until the preset income threshold value is reached;
and if the prediction result reaches the preset income limit value, the adjusted preset load data is used as the expected load, and the equipment for power generation is distributed according to the expected load to obtain the predicted power generation amount of the year.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011363422.7A CN112508734B (en) | 2020-11-27 | Method and device for predicting power generation capacity of power enterprise based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011363422.7A CN112508734B (en) | 2020-11-27 | Method and device for predicting power generation capacity of power enterprise based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112508734A true CN112508734A (en) | 2021-03-16 |
| CN112508734B CN112508734B (en) | 2024-04-26 |
Family
ID=
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115169991A (en) * | 2022-09-02 | 2022-10-11 | 南京华盾电力信息安全测评有限公司 | Multi-type user energy supply and energy supply intelligent analysis system based on data mining |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040215551A1 (en) * | 2001-11-28 | 2004-10-28 | Eder Jeff S. | Value and risk management system for multi-enterprise organization |
| CN107578124A (en) * | 2017-08-28 | 2018-01-12 | 国网山东省电力公司电力科学研究院 | The Short-Term Load Forecasting Method of GRU neutral nets is improved based on multilayer |
| CN108090563A (en) * | 2017-12-15 | 2018-05-29 | 烟台港股份有限公司 | A kind of electric flux Forecasting Methodology based on BP neural network |
| CN109543901A (en) * | 2018-11-20 | 2019-03-29 | 国网辽宁省电力有限公司经济技术研究院 | Short-Term Load Forecasting Method based on information fusion convolutional neural networks model |
| CN110651273A (en) * | 2017-11-17 | 2020-01-03 | 华为技术有限公司 | Data processing method and equipment |
| CN111080477A (en) * | 2019-12-06 | 2020-04-28 | 北京国电通网络技术有限公司 | Household power load prediction method and system |
| CN111695731A (en) * | 2020-06-09 | 2020-09-22 | 中国电力科学研究院有限公司 | Load prediction method, system and equipment based on multi-source data and hybrid neural network |
| CN111985701A (en) * | 2020-07-31 | 2020-11-24 | 国网上海市电力公司 | Power utilization prediction method based on power supply enterprise big data model base |
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040215551A1 (en) * | 2001-11-28 | 2004-10-28 | Eder Jeff S. | Value and risk management system for multi-enterprise organization |
| CN107578124A (en) * | 2017-08-28 | 2018-01-12 | 国网山东省电力公司电力科学研究院 | The Short-Term Load Forecasting Method of GRU neutral nets is improved based on multilayer |
| CN110651273A (en) * | 2017-11-17 | 2020-01-03 | 华为技术有限公司 | Data processing method and equipment |
| CN108090563A (en) * | 2017-12-15 | 2018-05-29 | 烟台港股份有限公司 | A kind of electric flux Forecasting Methodology based on BP neural network |
| CN109543901A (en) * | 2018-11-20 | 2019-03-29 | 国网辽宁省电力有限公司经济技术研究院 | Short-Term Load Forecasting Method based on information fusion convolutional neural networks model |
| CN111080477A (en) * | 2019-12-06 | 2020-04-28 | 北京国电通网络技术有限公司 | Household power load prediction method and system |
| CN111695731A (en) * | 2020-06-09 | 2020-09-22 | 中国电力科学研究院有限公司 | Load prediction method, system and equipment based on multi-source data and hybrid neural network |
| CN111985701A (en) * | 2020-07-31 | 2020-11-24 | 国网上海市电力公司 | Power utilization prediction method based on power supply enterprise big data model base |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115169991A (en) * | 2022-09-02 | 2022-10-11 | 南京华盾电力信息安全测评有限公司 | Multi-type user energy supply and energy supply intelligent analysis system based on data mining |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Han et al. | Enhanced deep networks for short-term and medium-term load forecasting | |
| CN108596242A (en) | Power grid meteorology load forecasting method based on wavelet neural network and support vector machines | |
| Cheng et al. | Enhanced state estimation and bad data identification in active power distribution networks using photovoltaic power forecasting | |
| CN111460001B (en) | Power distribution network theoretical line loss rate evaluation method and system | |
| CN111368259B (en) | Distribution network voltage data missing filling method and related device | |
| CN112149873A (en) | Low-voltage transformer area line loss reasonable interval prediction method based on deep learning | |
| CN115147012B (en) | Neural network model-based carbon emission accounting method and device | |
| CN114330934A (en) | Model parameter self-adaptive GRU new energy short-term power generation power prediction method | |
| CN114492191A (en) | Heat station equipment residual life evaluation method based on DBN-SVR | |
| CN115660725A (en) | Method for depicting multi-dimensional energy user portrait | |
| CN114202174A (en) | Electricity price risk grade early warning method and device and storage medium | |
| CN111697560B (en) | Method and system for predicting load of power system based on LSTM | |
| CN117543537A (en) | Agent electricity purchasing user electric quantity prediction method, device and storage medium | |
| CN112598181A (en) | Load prediction method, device, equipment and storage medium | |
| CN116341728A (en) | Ultra-short-term photovoltaic output power prediction method based on data driving | |
| CN112508734A (en) | Method and device for predicting power generation capacity of power enterprise based on convolutional neural network | |
| CN116090635A (en) | Meteorological-driven new energy generation power prediction method | |
| CN112508734B (en) | Method and device for predicting power generation capacity of power enterprise based on convolutional neural network | |
| CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
| CN115759336A (en) | Prediction method and storage medium for short-term power load prediction | |
| CN112581311B (en) | Method and system for predicting long-term output fluctuation characteristics of aggregated multiple wind power plants | |
| CN114925931A (en) | Platform area load prediction method and system | |
| CN114638421A (en) | Method for predicting requirement of generator set spare parts | |
| CN108615091A (en) | Electric power meteorology load data prediction technique based on cluster screening and neural network | |
| CN112288140A (en) | Keras-based short-term power load prediction method, storage medium and equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |