CN113095372A - Low-voltage transformer area line loss reasonable interval calculation method based on robust neural network - Google Patents

Low-voltage transformer area line loss reasonable interval calculation method based on robust neural network Download PDF

Info

Publication number
CN113095372A
CN113095372A CN202110302836.7A CN202110302836A CN113095372A CN 113095372 A CN113095372 A CN 113095372A CN 202110302836 A CN202110302836 A CN 202110302836A CN 113095372 A CN113095372 A CN 113095372A
Authority
CN
China
Prior art keywords
line loss
neural network
loss rate
layer
interval
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.)
Pending
Application number
CN202110302836.7A
Other languages
Chinese (zh)
Inventor
易永仙
邵雪松
周玉
崔高颖
吴伟将
张德进
仇新宇
蔡奇新
陈霄
李悦
马云龙
庞金鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Jiangsu Electric Power Co ltd Marketing Service Center
State Grid Corp of China SGCC
State Grid Jiangsu Electric Power Co Ltd
Original Assignee
State Grid Jiangsu Electric Power Co ltd Marketing Service Center
State Grid Corp of China SGCC
State Grid Jiangsu Electric Power Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by State Grid Jiangsu Electric Power Co ltd Marketing Service Center, State Grid Corp of China SGCC, State Grid Jiangsu Electric Power Co Ltd filed Critical State Grid Jiangsu Electric Power Co ltd Marketing Service Center
Priority to CN202110302836.7A priority Critical patent/CN113095372A/en
Publication of CN113095372A publication Critical patent/CN113095372A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
    • Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Abstract

A low-voltage transformer area line loss reasonable interval calculation method based on a robust neural network is characterized by comprising the following steps: step 1, extracting characteristic quantity related to line loss rate in the power utilization area based on a power utilization information acquisition system; step 2, inputting the characteristic quantity into a recurrent neural network, and calculating a line loss rate benchmark value of the power utilization area through the recurrent neural network; and 3, acquiring a line loss error and a line loss variance based on the line loss rate benchmarking value of the power utilization area and the actual line loss rate of the power utilization area, and determining a reasonable line loss rate interval of the power utilization area based on the line loss rate benchmarking value, the line loss error and the line loss variance. The method provided by the invention can solve the technical problems of high data requirement and low practicability in the conventional transformer area theoretical line loss calculation technology, thereby providing a reliable basis for reducing the line loss rate of a power grid and ensuring high-efficiency operation.

Description

Low-voltage transformer area line loss reasonable interval calculation method based on robust neural network
Technical Field
The invention relates to the field of power consumption management of a power distribution network, in particular to a low-voltage transformer area line loss reasonable interval calculation method based on a robust neural network.
Background
At present, the line loss rate plays an important role in economic operation as an important index for evaluating a power system, and therefore, line loss management is also one of important work contents of a power company. At present, according to the standard of line loss four-division management, the line loss management in China adopts the principle of voltage division, partition, branching and distribution. According to the measurement and calculation of the national power grid, the 380V low-voltage power grid is used as a heavy loss layer, and the loss amount of the 380V low-voltage power grid accounts for about 1/5 of the total loss amount of the power grid. The low-voltage power grid line loss management generally adopts a management means of the distribution area, so that the main influence factors of researching the distribution area line loss rate condition and analyzing the distribution area line loss rate have important significance for improving the economic operation level of the power distribution network.
In the prior art, the essence of the reasonable range of the line loss rate is the change range of the line loss rate when the line loss influence factors change in the reasonable value range. However, the power enterprises lack the management standard for the line loss rate index, and often use a fixed interval as the basis for the rationality evaluation of the line loss rate, and the management manner of "one switch" cannot distinguish the differences of the distribution room in structure, power supply range and load level, and neglect the influence of factors such as power consumption conditions in different areas, power consumption load characteristics of different users, topology structures of different distribution rooms and distributed power access on the distribution room line loss index. In addition, the line loss rate obtained by adopting a voltage drop method, a theoretical line loss calculation method of the platform area topology and other methods also needs detailed data information such as a terminal user voltage curve, the platform area topology, lines and the like, and the method has high requirements on data and cannot be popularized and applied in a large range.
Therefore, a novel line loss rate diagnosis method based on the power utilization area is needed.
Disclosure of Invention
In order to solve the defects in the prior art, the invention aims to provide a low-voltage transformer area line loss reasonable interval calculation method based on a robust neural network, which can calculate the line loss rate benchmarking value of a transformer area and determine the line loss rate reasonable interval of the transformer area based on the characteristic quantity related to the line loss rate in a power utilization information acquisition system.
The invention adopts the following technical scheme. A low-voltage transformer area line loss reasonable interval calculation method based on a robust neural network comprises the following steps: step 1, extracting characteristic quantity related to line loss rate in a low-voltage power station area based on a power consumption information acquisition system; step 2, inputting the characteristic quantity into a recurrent neural network, and calculating a line loss rate benchmarking value of the power utilization area through the recurrent neural network; and 3, acquiring a line loss error and a line loss variance based on the line loss rate benchmarking value of the power utilization area and the actual line loss rate of the power utilization area, and determining a reasonable line loss rate interval of the power utilization area based on the line loss rate benchmarking value, the line loss error and the line loss variance.
Preferably, step 1 further comprises: the characteristic quantity related to the line loss rate in the electricity consumption information acquisition system comprises: the type of the power utilization area, the capacity of the power utilization area, the electric quantity of the power utilization area, the type of users, the number of users, the reading of an electric meter, the power of the power utilization area, the line index and the CT transformation ratio.
Preferably, the method for calculating the line loss rate benchmarking value of the power utilization station area in the step 2 comprises the following steps: step 2.1, inputting the characteristic quantity into a recurrent neural network, and calculating to obtain the output of the recurrent neural network; step 2.2, performing regularization operation of L2 norm on the output of the recurrent neural network to eliminate overfitting of the recurrent neural network; step 2.3, inputting the regularization operation result into random inactivation layers with different probability values, and obtaining a random inactivation output result; and 2.4, averaging the random inactivation output results to obtain a line loss rate benchmarking value of the power utilization area.
Preferably, step 2.1 further comprises: the recurrent neural network comprises three main paths, the output of the recurrent neural network is the cascade combination of the output of each main path, and the calculation formula is as follows:
Figure BDA0002986940880000021
in the formula, ycIs the output of the recurrent neural network and,
c (-) is a cascade function,
fk(. cndot.) is a function of the main path,
x is the input of the kth main path,
Figure BDA0002986940880000022
for the weight of the kth main path,
Figure BDA0002986940880000023
for the offset of the kth main path,
nwas to the number of weights in the current main path,
nbis the number of deviations in the current main path.
Preferably, each main path includes a DAE branch path and a full link layer branch path;
wherein, the output calculation formula of each main path is as follows:
Figure BDA0002986940880000024
in the formula, ykFor the output of the k-th main path,
gk(. cndot.) is a function of DAE,
Figure BDA0002986940880000031
for the weight of the fully-connected layer in the kth main path,
Figure BDA0002986940880000032
is the deviation of the fully-connected layer in the kth main path.
Preferably, the DAE branch path includes five parts of an input, a noise layer, an encoding layer, a decoding layer and an output; and the input of the kth main path sequentially passes through a noise layer, an encoding layer and a decoding layer to generate output.
Preferably, the noise layer is a gaussian noise layer, and the calculation formula is as follows:
xi，gn＝xi+N(0，σ2) (3)
in the formula, xiFor the ith input in a main path,
xi，gngaussian noise floor as a main pathThe ith output of gn corresponding to the ith input,
N(0，σ2) Mean 0 and variance σ2Is normally distributed gaussian noise.
Preferably, the coding layer and the decoding layer are all connected layers, and the calculation formula is as follows:
Figure BDA0002986940880000033
in the formula (I), the compound is shown in the specification,
Figure BDA0002986940880000034
the ith output for the ith intermediate layer of a primary path in the neural network,
Figure BDA0002986940880000035
is the jth input of the ith layer in a main path, which is the jth output from the ith-1 layer,
wijis the weight of the jth input and ith output of the ith layer in a main path,
bjis the weight of the jth input and ith output of the ith layer in a main path.
Preferably, step 2.2 further comprises:
the formula for the regularization operation of the L2 norm is:
Figure BDA0002986940880000036
wherein L is a loss function,
λ is a hyper-parameter of the penalty term,
huber is a function of the Huber loss,
nsthe number of training samples is determined by the number of the platform areas and the training days,
delta is a self-defined hyper-parameter.
Preferably, step 2.3 further comprises:
respectively inputting the regularization operation results into 10 random inactivation layers with different probabilities to obtain 10 different line loss rate outputs;
the calculation formula of the line loss rate output is as follows:
Figure BDA0002986940880000041
wherein p is the probability of a random deactivation layer.
Preferably, step 3 further comprises:
the line loss error is calculated by the formula:
Figure BDA0002986940880000042
in the formula, e1The absolute error between the line loss rate benchmarking value and the actual line loss rate in the power utilization area is adopted,
nsin order to train the number of samples,
Figure BDA0002986940880000043
for the line loss rate benchmarking value of the xlyb training sample in the power utilization area,
Figure BDA0002986940880000044
for the actual line loss rate of the xlyb training sample in the utilization area,
denotes conjugation;
the line loss variance is calculated by the following formula:
Figure BDA0002986940880000045
in the formula, e2The variance between the line loss rate benchmarking value and the actual line loss rate in the power utilization area is shown,
Figure BDA0002986940880000046
and the glth output is the xlyb line loss rate, wherein the value of the gl is an integer between 0 and 10, and the glth output represents the probability of 10 different random inactivation layers respectively.
Preferably, the reasonable line loss rate interval of the power utilization area is as follows:
Figure BDA0002986940880000047
in the formula IxlybIs the lowest limit of the rational interval, uxlybIs the highest limit value of a reasonable interval.
Preferably, the method further comprises step 4: adjusting parameters of each layer in the neural network; and the adjustment is determined based on the parameter values of each layer when the mean square error between the line loss rate benchmarking values of all the platform area samples and the actual line loss rate is minimum.
Compared with the prior art, the low-voltage transformer area line loss reasonable interval calculation method based on the robust neural network has the advantages that the method can obtain implicit useful information by utilizing various data mining technologies on the basis of the data of the power utilization information acquisition system, calculate the transformer area line loss rate benchmarking value and the reasonable interval, and achieve the purpose of diagnosing the line loss rate abnormity. By adopting the method, the technical problems of high data requirement and low practicability in the conventional transformer area theoretical line loss calculation technology can be solved, so that the line loss rate of a power grid is reduced, and a reliable basis is provided for high-efficiency operation.
Drawings
FIG. 1 is a schematic flow chart of a method in a low-voltage transformer area line loss reasonable interval calculation method based on a robust neural network of the present invention;
FIG. 2 is a schematic diagram of a recurrent neural network in the low-voltage transformer area line loss reasonable interval calculation method based on the robust neural network of the present invention;
FIG. 3 is a schematic structural diagram of a denoising automatic encoder in the low-voltage transformer area line loss reasonable interval calculation method based on the robust neural network;
FIG. 4 is a schematic diagram of a principle of a random deactivation layer in the low-voltage transformer area line loss reasonable interval calculation method based on the robust neural network;
fig. 5 is a schematic diagram of a reasonable interval and a line loss rate benchmarking value of a power consumption station area obtained in an embodiment of the low-voltage station area line loss reasonable interval calculation method based on the robust neural network.
Detailed Description
The present application is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present application is not limited thereby.
As shown in fig. 1, a low-voltage transformer area line loss reasonable interval calculation method based on a robust neural network includes steps 1 to 3.
Step 1, extracting characteristic quantity related to line loss rate in a low-voltage power station area based on a power utilization information acquisition system.
Specifically, the invention can be used for the electricity utilization data related to the line loss rate in the power grid power utilization area as the input quantity of the neural network. In one embodiment, the characteristic field may be obtained from a power consumption information collection system.
In one embodiment of the invention, the total number of the characteristic fields of the low-voltage power utilization station area extracted from the power utilization information acquisition system is 18. Table 1 is a characteristic field table of the power consumption station area in the power consumption information acquisition system of the present invention. As shown in table 1, the 18 feature fields can be roughly classified as follows:
preferably, step 1 further comprises: the characteristic quantity related to the line loss rate in the electricity consumption information acquisition system comprises: the type of the power utilization area, the capacity of the power utilization area, the electric quantity of the power utilization area, the type of users, the number of users, the reading of an electric meter, the power of the power utilization area, the line index and the CT transformation ratio.
Table 1 characteristic field table of power consumption area in power consumption information collecting system of the present invention
Serial number Characteristic field
1 Number of residents
2 Number of non-resident
3 Capacity of residents
4 Non-residential capacity
5 Capacity of cell
6 CT transformation ratio
7 Comprehensive multiplying power
8 Three-phase electric energy meter
9 Single phase electric energy meter
10 Rate of daily load
11 Daily power factor
12 Daily active power
13 Amount of power supply
14 Category of urban rural network
15 Type of cell
16 Electric quantity for accessing internet
Specifically, the station zone category refers to: and the city rural power grid category in the electricity utilization information acquisition system is a service data item. Based on the data item, whether the current power utilization transformer area is an urban network transformer area or a rural network transformer area can be known. Since different station types have different power user types and user distributions, different line loss calculation modes can be provided based on the data item.
The cell capacity refers to the load capacity of the current cell, and the cell capacity may also include a residential capacity and a non-residential capacity, i.e. the load capacity is allocated according to different user types.
The electric quantity of the transformer area comprises the generated energy, the power supply quantity and the internet surfing electric quantity of the transformer area. The power generation amount refers to the electric quantity output by the generator end of the transformer area, the power supply amount refers to the electric quantity obtained by subtracting the power consumption of the power plant from the electric quantity output by the generator, the on-grid electric quantity refers to the electric quantity transmitted to the power grid through the transformer, and the loss of the transformer is deducted.
The user types in the transformer area can be classified according to the use of the electric energy and are divided into resident types and non-resident types. Thus, the number of users may include the number of residential and non-residential users.
The electric meter reading comprises three-phase electric energy meter reading and single-phase electric energy meter reading in the power grid.
The power indexes of the transformer area respectively comprise daily power factors which can represent functional rates and daily reactive power quantities which represent reactive power.
The line indexes also comprise the maximum zero line current for ensuring the safety performance of the power grid, and the theoretical line loss rate and the comprehensive multiplying power for representing the line loss rate characteristic of the current transformer area.
In addition, the characteristic field may further include a detection value of a current transformer for detecting the magnitude of the current in the station area line, such as a CT transformation ratio.
And taking the characteristic field as the input of the neural network, calculating and obtaining the line loss rate data of the current station area based on the recurrent neural network.
And 2, inputting the characteristic quantity into a recurrent neural network, and calculating the line loss rate benchmarking value of the power utilization area through the recurrent neural network.
FIG. 2 is a schematic diagram of a recurrent neural network in the low-voltage transformer area line loss reasonable interval calculation method based on the robust neural network of the present invention; FIG. 3 is a schematic structural diagram of a denoising automatic encoder in the low-voltage transformer area line loss reasonable interval calculation method based on the robust neural network; fig. 4 is a schematic diagram of the principle of a random inactivation layer in the low-voltage transformer area line loss reasonable interval calculation method based on the robust neural network. Preferably, as shown in fig. 2-4, the method for calculating the line loss rate benchmarking value of the power utilization area in step 2 further includes four steps of step 2.1 to step 2.4.
And 2.1, inputting the characteristic quantity into the recurrent neural network, and calculating to obtain the output of the recurrent neural network.
Preferably, step 2.1 further comprises: the recurrent neural network comprises three main paths, the output of the recurrent neural network is the cascade combination of the output of each main path, and the calculation formula is as follows:
Figure BDA0002986940880000071
in the formula, ycIs to circulateThe output of the recurrent neural network, C (-) is the cascade function, fk(. h) is the main path function, x is the input to the kth main path,
Figure BDA0002986940880000081
for the weight of the kth main path,
Figure BDA0002986940880000082
is the deviation of the kth main path, nwIs the number of weights, n, in the current main pathbIs the number of deviations in the current main path.
Preferably, each main path includes a DAE branch path and a full link layer branch path;
wherein, the output calculation formula of each main path is as follows:
Figure BDA0002986940880000083
in the formula, ykFor the output of the kth main path, gk(. cndot.) is a function of DAE,
Figure BDA0002986940880000084
for the weight of the fully-connected layer in the kth main path,
Figure BDA0002986940880000085
is the deviation of the fully-connected layer in the kth main path.
Preferably, the DAE branch path includes five parts of an input, a noise layer, an encoding layer, a decoding layer and an output; and the input of the kth main path sequentially passes through a noise layer, an encoding layer and a decoding layer to generate output.
Wherein, the noise layer can be a gaussian noise layer, and the calculation formula is:
xi，gn＝xi+N(0，σ2) (3)
in the formula, xiFor the ith input, x, in the kth main pathi，gnThe output corresponding to the ith input in the gaussian noise layer gn for the kth main path, N (0,σ2) Mean 0 and variance σ2Is normally distributed gaussian noise.
Preferably, the coding layer and the decoding layer are all fully connected layers, and the calculation formulas are respectively:
Figure BDA0002986940880000086
in the formula (I), the compound is shown in the specification,
Figure BDA0002986940880000087
for the ith output of the ith intermediate layer of a primary path in the neural network,
Figure BDA0002986940880000088
is the jth input of the ith layer in a main path, which is the jth output from the ith-1 layer, wijIs the weight of the jth input and ith output of the ith layer in a main path, bjThe weight of the jth input and ith output of the ith layer in the kth main path.
It is understood that the intermediate layer mentioned above refers to an intermediate layer of the neural network, and in the present invention, specifically, may be an encoding layer, a decoding layer, a regularization layer, and a random deactivation layer. The intermediate layers represent different layers by using different values of the parameter l.
According to the content of the formula, the output of the recurrent neural network can be calculated.
Step 2.2, the output of the recurrent neural network is subjected to regularization operation of L2 norm to eliminate overfitting of the recurrent neural network.
In addition, to prevent loop fitting of the output results in step 2.1, the output of the recurrent neural network can also be regularized. In an embodiment of the invention, the L2 norm may be selected to implement a regularization operation.
Preferably, the regularization operation of the L2 norm is calculated by the formula:
Figure BDA0002986940880000091
where L is the loss function, λ is the hyperparameter of the penalty term, Huber is the Huber loss function, nsTo train the number of samples, δ is a custom hyper-parameter.
In the neural network of the present invention, the power consumption data of one station area per day can be trained as one sample. Therefore, the number of training samples can be determined according to the number of cells with better line loss indexes in the system and the number of training days.
And 2.3, inputting the regularization operation result into random inactivation layers with different probability values, and obtaining a random inactivation output result.
In order to accurately obtain the benchmarking value of the line loss rate, a plurality of random inactivation layers with different probabilities can be selected for calculating the benchmarking value.
Preferably, the regularization operation result is input into 10 random deactivation layers with different probabilities to obtain 10 different line loss rate outputs; the calculation formula of the line loss rate output is as follows:
Figure BDA0002986940880000092
wherein p is the probability of a random deactivation layer.
In an embodiment of the present invention, 10 random inactivation layers with different probabilities may be selected, and the probability of the 10 random inactivation layers may be evaluated within a range of [0, 0.5 ]. Specifically, the probability p takes on values of 0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, and 0.5.
And 2.4, averaging the random inactivation output results to obtain a line loss rate benchmarking value of the power utilization area.
The calculation formula of the line loss rate benchmarking value is as follows:
Figure BDA0002986940880000093
wherein the content of the first and second substances,
Figure BDA0002986940880000094
the benchmark value of the line loss rate of the xlyb training sample in the power utilization area,
Figure BDA0002986940880000095
and the glth output of the line loss rate of the xlyb training sample, wherein the value of gl is an integer between 0 and 10, and the glth output represents the probability of 10 different random inactivation layers respectively.
And 3, acquiring a line loss error and a line loss variance based on the line loss rate benchmarking value of the power utilization area and the actual line loss rate of the power utilization area, and determining a reasonable line loss rate interval of the power utilization area based on the line loss rate benchmarking value, the line loss error and the line loss variance.
Preferably, step 3 further comprises: the line loss error is calculated by the formula:
Figure BDA0002986940880000101
in the formula, e1Is the absolute error between the line loss rate benchmarking value and the actual line loss rate in the power station area, nsIn order to train the number of samples,
Figure BDA0002986940880000102
the line loss rate benchmark value, y, of the xlyb training sample in the power utilization areaxlybThe actual line loss rate of the xlyb training sample in the power utilization area is obtained; denotes conjugation.
The line loss variance is calculated by the following formula:
Figure BDA0002986940880000103
in the formula, e2The variance between the line loss rate benchmarking value and the actual line loss rate in the power utilization area is shown,
Figure BDA0002986940880000104
is the xlyb numberThe gl th output of the line loss rate of the training sample.
Preferably, the reasonable line loss rate interval of the power utilization area is as follows:
Figure BDA0002986940880000105
in the formula IxlybIs the lowest limit of the rational interval, uxlybIs the highest limit value of a reasonable interval.
Step 4, adjusting parameters of each layer in the neural network; and the adjustment is determined based on the parameter values of each layer when the mean square error between the line loss rate benchmarking values of all the platform area samples and the actual line loss rate is minimum.
In an embodiment of the invention, a station area with better line loss management conditions of twenty thousand in Jiangsu province is selected as a test station area. And selecting data of 30 days in a certain month of the transformer area as sample data to train a model of the neural network so as to obtain parameters such as weight, deviation and the like of each layer in the neural network, thereby constructing the recurrent neural network.
After the construction of the recurrent neural network is completed, one of the areas is selected, and in the embodiment of the invention, the area with the ID of 170001414803 is selected for calculating the line loss rate benchmarking value and the reasonable interval. Table 2 is a table of data related to the line loss rate calculated based on the method of the present invention. As shown in table 2, the benchmarking values and the reasonable intervals of the line loss rate were calculated every 4 months.
TABLE 2 line loss rate-related data table calculated based on the method of the present invention
Figure BDA0002986940880000111
Figure BDA0002986940880000121
To more clearly view the data, the line loss rate data can be plotted using table 2. Fig. 5 is a schematic diagram of a reasonable interval and a line loss rate benchmarking value of a power consumption platform area obtained in an embodiment of the power consumption platform area line loss rate diagnosis method based on the neural network. As shown in fig. 5, the difference between the actual line loss rate value and the benchmark value in 30 days is not large, and most of the actual line loss rate value data is in the benchmark interval. Only the data difference between the actual line loss rate data at day 12 and the benchmark value is large and falls outside the boundary of the benchmark interval. This indicates that the data at day 12 is anomalous. The power grid maintenance manager can accurately diagnose abnormal power consumption data in the power grid according to the related data in the transformer area and the diagnosis method, and adopts a reasonable adjustment mode.
Compared with the prior art, the method for diagnosing the line loss rate of the power utilization area based on the neural network has the advantages that the method for diagnosing the line loss rate of the power utilization area based on the neural network can obtain implicit useful information by utilizing various data mining technologies on the basis of data of a power utilization information acquisition system, and achieves the purpose of diagnosing the abnormal line loss rate by calculating the benchmarking value and the reasonable interval of the line loss rate of the power utilization area. By adopting the method, the technical problems of high data requirement and low practicability in the conventional transformer area theoretical line loss calculation technology can be solved, so that the line loss rate of a power grid is reduced, and a reliable basis is provided for high-efficiency operation.
The present applicant has described and illustrated embodiments of the present invention in detail with reference to the accompanying drawings, but it should be understood by those skilled in the art that the above embodiments are merely preferred embodiments of the present invention, and the detailed description is only for the purpose of helping the reader to better understand the spirit of the present invention, and not for limiting the scope of the present invention, and on the contrary, any improvement or modification made based on the spirit of the present invention should fall within the scope of the present invention.

Claims (13)

1. A low-voltage transformer area line loss reasonable interval calculation method based on a robust neural network is characterized by comprising the following steps:
step 1, extracting characteristic quantity related to line loss rate in a low-voltage power station area based on a power consumption information acquisition system;
step 2, inputting the characteristic quantity into a recurrent neural network, and calculating a line loss rate benchmark value of the power utilization area through the recurrent neural network;
and 3, acquiring a line loss error and a line loss variance based on the line loss rate benchmarking value of the power utilization area and the actual line loss rate of the power utilization area, and determining a reasonable line loss rate interval of the power utilization area based on the line loss rate benchmarking value, the line loss error and the line loss variance.
2. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 1, wherein the reasonable line loss interval is calculated by the following steps:
the step 1 further comprises:
the characteristic quantity related to the line loss rate in the electricity utilization information acquisition system comprises: the power utilization area comprises the area type, the area capacity, the area electric quantity, the user type, the user number, the electric meter reading, the area power, the line index and the CT transformation ratio.
3. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 1, wherein the reasonable line loss interval is calculated by the following steps:
the method for calculating the line loss rate benchmarking value of the power utilization area in the step 2 comprises the following steps:
step 2.1, inputting the characteristic quantity into a recurrent neural network, and calculating to obtain the output of the recurrent neural network;
step 2.2, carrying out regularization operation of L2 norm on the output of the recurrent neural network so as to eliminate overfitting of the recurrent neural network;
step 2.3, inputting the regularization operation result into random inactivation layers with different probability values, and obtaining a random inactivation output result;
and 2.4, averaging the random inactivation output results to obtain a line loss rate benchmarking value of the power utilization area.
4. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 3, wherein the reasonable line loss interval is calculated by the following steps:
the step 2.1 further comprises:
the recurrent neural network comprises three main paths, the output of the recurrent neural network is the cascade combination of the output of each main path, and the calculation formula is as follows:
Figure FDA0002986940870000021
in the formula, ycIs the output of the recurrent neural network and,
c (-) is a cascade function,
fk(. h) is the main path function, x is the input to the kth main path,
Figure FDA0002986940870000022
for the weight of the kth main path,
Figure FDA0002986940870000023
for the offset of the kth main path,
nwas to the number of weights in the current main path,
nbis the number of deviations in the current main path.
5. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 4, wherein the reasonable line loss interval is calculated by the following steps:
each main path comprises a DAE branch path and a full connection layer branch path;
wherein, the output calculation formula of each main path is as follows:
Figure FDA0002986940870000024
in the formula, ykFor the output of the k-th main path,
gk(. cndot.) is a function of DAE,
Figure FDA0002986940870000025
for the weight of the fully-connected layer in the kth main path,
Figure FDA0002986940870000026
is the deviation of the fully-connected layer in the kth main path.
6. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 5, wherein the reasonable line loss interval is calculated by the following steps:
the DAE branch path comprises five parts of an input part, a noise layer, an encoding layer, a decoding layer and an output part;
and the input of the kth main path sequentially passes through a noise layer, an encoding layer and a decoding layer to generate output.
7. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 6, wherein the reasonable line loss interval is calculated by the following steps:
the noise layer is a Gaussian noise layer, and the calculation formula is as follows:
xi，gn＝xi+N(0，σ2) (3)
in the formula, xiFor the ith input in a main path,
xi，gnthe ith output of the gaussian noise layer gn for a main path corresponding to the ith input,
N(0，σ2) Mean 0 and variance σ2Is normally distributed gaussian noise.
8. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 7, wherein:
the coding layer and the decoding layer are all connected layers, and the calculation formulas are as follows:
Figure FDA0002986940870000031
in the formula (I), the compound is shown in the specification,
Figure FDA0002986940870000032
for the ith output of the ith intermediate layer of a primary path in the neural network,
Figure FDA0002986940870000033
is the jth input of the ith layer in a main path, which is the jth output from the ith-1 layer,
wijis the weight of the jth input and ith output of the ith layer in a main path,
bjis the weight of the jth input and ith output of the ith layer in a main path.
9. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 8, wherein:
the step 2.2 further comprises:
the calculation formula of the regularization operation of the L2 norm is as follows:
Figure FDA0002986940870000034
wherein L is a loss function,
λ is a hyper-parameter of the penalty term,
huber is a function of the Huber loss,
nsthe number of training samples is determined by the number of the platform areas and the training days,
delta is a self-defined hyper-parameter.
10. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 9, wherein:
the step 2.3 further comprises:
respectively inputting the regularization operation results into 10 random inactivation layers with different probabilities to obtain 10 different line loss rate outputs;
wherein, the calculation formula of the line loss rate output is as follows:
Figure FDA0002986940870000041
wherein p is the probability of the random deactivation layer.
11. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 1, wherein the reasonable line loss interval is calculated by the following steps:
the step 3 further comprises:
the calculation formula of the line loss error is as follows:
Figure FDA0002986940870000042
in the formula, e1The absolute error between the line loss rate benchmarking value and the actual line loss rate in the power utilization area is obtained,
nsin order to train the number of samples,
Figure FDA0002986940870000046
the line loss rate benchmarking value of the xlyb training sample in the power utilization area,
yxlybthe actual line loss rate of the xlyb training sample in the power utilization area is obtained;
denotes conjugation;
the line loss variance is calculated according to the formula:
Figure FDA0002986940870000043
in the formula, e2The variance between the line loss rate benchmarking value and the actual line loss rate in the power utilization area is obtained,
Figure FDA0002986940870000044
and the glth output of the line loss rate of the xlyb training sample, wherein the value of gl is an integer between 0 and 10, and the glth output represents the probability of 10 different random inactivation layers respectively.
12. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 11, wherein:
the reasonable line loss rate interval of the power utilization area is as follows:
Figure FDA0002986940870000045
in the formula IxlybIs the lowest limit of the rational interval, uxlybIs the highest limit value of a reasonable interval.
13. The method for calculating the reasonable line loss interval of the low-voltage transformer area based on the robust neural network as claimed in claim 1, wherein the reasonable line loss interval is calculated by the following steps:
the method further comprises the step 4:
adjusting parameters of each layer in the neural network;
and the adjustment is determined based on the parameter values of each layer when the mean square error between the line loss rate benchmarking values of all the platform area samples and the actual line loss rate is minimum.
CN202110302836.7A 2021-03-22 2021-03-22 Low-voltage transformer area line loss reasonable interval calculation method based on robust neural network Pending CN113095372A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110302836.7A CN113095372A (en) 2021-03-22 2021-03-22 Low-voltage transformer area line loss reasonable interval calculation method based on robust neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110302836.7A CN113095372A (en) 2021-03-22 2021-03-22 Low-voltage transformer area line loss reasonable interval calculation method based on robust neural network

Publications (1)

Publication Number Publication Date
CN113095372A true CN113095372A (en) 2021-07-09

Family

ID=76669173

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110302836.7A Pending CN113095372A (en) 2021-03-22 2021-03-22 Low-voltage transformer area line loss reasonable interval calculation method based on robust neural network

Country Status (1)

Country Link
CN (1) CN113095372A (en)

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110110887A (en) * 2019-03-22 2019-08-09 国网浙江省电力有限公司信息通信分公司 To the prediction technique of low-voltage platform area line loss per unit
CN110489783A (en) * 2019-07-05 2019-11-22 江苏方天电力技术有限公司 A kind of low-voltage platform area line loss per unit zone of reasonableness estimation method based on QRNN
CN110782181A (en) * 2019-11-05 2020-02-11 国网重庆市电力公司电力科学研究院 Low-voltage transformer area line loss rate calculation method and readable storage medium
US20200134439A1 (en) * 2018-10-24 2020-04-30 Equifax Inc. Machine-learning techniques for monotonic neural networks
CN111123039A (en) * 2019-12-31 2020-05-08 国网北京市电力公司 Power distribution network abnormal line loss diagnosis method based on synchronization characteristics and improved K-means clustering
WO2020221278A1 (en) * 2019-04-29 2020-11-05 北京金山云网络技术有限公司 Video classification method and model training method and apparatus thereof, and electronic device
CN112054507A (en) * 2020-08-07 2020-12-08 国网辽宁省电力有限公司沈阳供电公司 Power distribution low-voltage distribution area theoretical line loss interval calculation method based on convolutional neural network
CN112149873A (en) * 2020-08-25 2020-12-29 北京合众伟奇科技有限公司 Low-voltage transformer area line loss reasonable interval prediction method based on deep learning
CN112288303A (en) * 2020-11-05 2021-01-29 国家电网有限公司 Method and device for determining line loss rate
CN112288172A (en) * 2020-10-30 2021-01-29 国家电网有限公司 Prediction method and device for line loss rate of transformer area
CN112308348A (en) * 2020-12-02 2021-02-02 国网江苏省电力有限公司徐州供电分公司 Intelligent analysis method for medium-voltage line loss abnormity

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200134439A1 (en) * 2018-10-24 2020-04-30 Equifax Inc. Machine-learning techniques for monotonic neural networks
CN110110887A (en) * 2019-03-22 2019-08-09 国网浙江省电力有限公司信息通信分公司 To the prediction technique of low-voltage platform area line loss per unit
WO2020221278A1 (en) * 2019-04-29 2020-11-05 北京金山云网络技术有限公司 Video classification method and model training method and apparatus thereof, and electronic device
CN110489783A (en) * 2019-07-05 2019-11-22 江苏方天电力技术有限公司 A kind of low-voltage platform area line loss per unit zone of reasonableness estimation method based on QRNN
CN110782181A (en) * 2019-11-05 2020-02-11 国网重庆市电力公司电力科学研究院 Low-voltage transformer area line loss rate calculation method and readable storage medium
CN111123039A (en) * 2019-12-31 2020-05-08 国网北京市电力公司 Power distribution network abnormal line loss diagnosis method based on synchronization characteristics and improved K-means clustering
CN112054507A (en) * 2020-08-07 2020-12-08 国网辽宁省电力有限公司沈阳供电公司 Power distribution low-voltage distribution area theoretical line loss interval calculation method based on convolutional neural network
CN112149873A (en) * 2020-08-25 2020-12-29 北京合众伟奇科技有限公司 Low-voltage transformer area line loss reasonable interval prediction method based on deep learning
CN112288172A (en) * 2020-10-30 2021-01-29 国家电网有限公司 Prediction method and device for line loss rate of transformer area
CN112288303A (en) * 2020-11-05 2021-01-29 国家电网有限公司 Method and device for determining line loss rate
CN112308348A (en) * 2020-12-02 2021-02-02 国网江苏省电力有限公司徐州供电分公司 Intelligent analysis method for medium-voltage line loss abnormity

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
何伟民等: "基于RNN的低压变压器区域日线损率基准测试", 计算机测量与控制, no. 9, pages 1 - 5 *
李清涛;任宇驰;王远;李林松;迟振烨;: "基于人工神经网络全连接层优化的线损异常诊断方法研究", 电气应用, no. 04 *

Similar Documents

Publication Publication Date Title
CN102866321B (en) Self-adaptive stealing-leakage prevention diagnosis method
CN110516912B (en) Method for identifying household transformer relation of distribution station
CN103617568B (en) Setting method for abnormal data determination threshold in steady-state power quality early-warning mechanism
CN110288208B (en) Comprehensive evaluation method for reliability and economy of radiation type power distribution network
CN102868160B (en) Macrozone load modeling method in intelligent power system
CN115018139A (en) Current transformer error state online identification method and system based on interphase characteristics
CN112130109B (en) Detection method for metering performance abnormality of intelligent electric energy meter
CN103886518A (en) Early warning method for voltage sag based on electric energy quality data mining at monitoring point
CN111275295B (en) Distributed photovoltaic fault diagnosis method based on inverse distance weight interpolation
CN111817299B (en) Fuzzy reasoning-based intelligent identification method for line loss rate abnormal cause of power distribution station
CN107860987A (en) A kind of low-voltage platform area drop damage aid decision-making system
CN114862139A (en) Data-driven transformer area line loss rate abnormity diagnosis method
CN111612326A (en) Comprehensive evaluation method for power supply reliability of distribution transformer
CN108062616B (en) Electricity consumption anomaly detection system with constraint conditions
CN112485514A (en) Method for evaluating uncertainty of power consumption of electric balance
CN113095372A (en) Low-voltage transformer area line loss reasonable interval calculation method based on robust neural network
CN114156865B (en) Low-voltage distribution network topology generation and fault prediction method considering state perception
CN115760400A (en) Mining behavior detection method based on electric power data and storage medium
CN114580938A (en) Comprehensive power distribution effect evaluation method with multiple data dimensions
Wu et al. Survey on negative line loss rate of transformer region: Rectification measures and challenges
Yang et al. AI-Based Detection of Power Consumption Behavior of People in a Smart City
Lingang et al. Research on integrated calculation method of theoretical line loss of MV and LV distribution Network based on Adaboost integrated learning
Ma et al. Electric Energy Meter State Management System Based on Index Analysis System
CN110873857A (en) Intelligent electric energy meter running state evaluation method and system based on multi-source data fusion
Dong et al. A deep learning-based approach for identifying bad data in power systems

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