CN112925825A - Multi-meteorological-factor prediction method for power transmission line - Google Patents

Multi-meteorological-factor prediction method for power transmission line Download PDF

Info

Publication number
CN112925825A
CN112925825A CN202110213155.3A CN202110213155A CN112925825A CN 112925825 A CN112925825 A CN 112925825A CN 202110213155 A CN202110213155 A CN 202110213155A CN 112925825 A CN112925825 A CN 112925825A
Authority
CN
China
Prior art keywords
information
meteorological
feature extraction
network
power transmission
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
CN202110213155.3A
Other languages
Chinese (zh)
Inventor
路通
陈俍宇
袁明磊
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.)
Nanjing University
Original Assignee
Nanjing University
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 Nanjing University filed Critical Nanjing University
Priority to CN202110213155.3A priority Critical patent/CN112925825A/en
Publication of CN112925825A publication Critical patent/CN112925825A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2465—Query processing support for facilitating data mining operations in structured databases
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2474—Sequence data queries, e.g. querying versioned data
    • 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
    • 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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs

Abstract

The invention discloses a multi-meteorological-factor prediction method for a power transmission line, which relates to the technical field of machine learning and solves the technical problem of low meteorological prediction accuracy, and the technical scheme is characterized in that a TCN (Temporal Convolutional Network) Network and a DenseNet Network are combined for feature extraction, the TCN Network can process time sequence information in parallel, has flexible receptive field and stable gradient, occupies lower memory compared with other methods, can well fit time sequence relation, and improves the accuracy of prediction of each meteorological factor; secondly, a DenseNet network is combined on the basis of a TCN network, so that overfitting can be resisted, and the accuracy of weather prediction is improved; finally, the method not only utilizes the historical microclimate information, but also combines the weather forecast information when predicting each meteorological factor, thereby improving the accuracy of meteorological prediction.

Description

Multi-meteorological-factor prediction method for power transmission line
Technical Field
The disclosure relates to the technical field of machine learning, in particular to a multi-meteorological-factor prediction method for a power transmission line.
Background
The meteorological conditions around the power transmission line are always very important for the safety of the power grid system, the power grid system also has the phenomena of disconnection, tower collapse, flashover and the like when natural disasters such as wind disasters, ice disasters and the like of the power transmission line occur at different degrees at home and abroad, and the occurrence frequency of the disasters is on the rising trend as the global temperature gradually becomes warm and extreme weather continuously increases.
The transmission line disaster caused by high wind speed generally occurs under severe weather conditions, when strong wind or hurricane acts on the wind pressure surfaces of the wire and the insulator string, the wire deflects and displaces to a certain degree, and when the air gap between the wire and a tower component or a peripheral object is smaller than the power frequency voltage breakdown distance, air breakdown occurs to form windage yaw discharge tripping. The inter-phase discharge of the wires caused by wind deflection is caused by different swing periods and phases of the wires of each phase due to certain difference of wind receiving time of the wires of each phase under the action of gust. In addition, different swing periods of the wires can be caused by different sag degrees of the wires of each phase, different-period swing is formed, and the problem of insufficient air gaps among phases and interphase discharge is caused. In addition, due to the long-term action of wind power, hardware is abraded, and even the breakage of the line hardware can be caused in severe cases, so that the disconnection and the tripping are caused.
And the ice coating of the transmission circuit is easily caused by meteorological factors such as low temperature, high humidity and high wind speed, and the great influence is generated on the power transmission.
In the aspects of power transmission line meteorological disaster feature identification and early warning management, research of a comparison system is lacked all the time. At present, The existing power transmission line meteorological factor prediction Model is mainly used for prediction through a physical Model, such as The WRF (The Weather Research and Forecasting Model, Weather Forecasting mode), and The Model is generally difficult to fit The time series relationship of meteorological information and cannot perform accurate prediction on each meteorological.
Disclosure of Invention
The disclosure provides a multi-meteorological-factor prediction method for a power transmission line, and the technical purpose of the method is to improve the accuracy of meteorological prediction.
The technical purpose of the present disclosure is achieved by the following technical solutions:
a multi-meteorological-factor prediction method for a power transmission line comprises the following steps:
s1: acquiring microclimate information and corresponding weather forecast information near the power transmission line;
s2: grouping the microclimate information and the corresponding weather forecast information according to a time sequence, wherein each group of data comprises the microclimate information and the corresponding weather forecast information in the same time period, and dividing all data into a training set and a test set according to groups;
s3: obtaining a meteorological factor prediction model through training of the training set, wherein the meteorological factor prediction model comprises the following steps: selecting n groups of data from the training set as a first input, inputting the data into a first-layer feature extraction module of the TCN network for feature extraction, and obtaining a first output; splicing the first output and the first input through a DenseNet network to obtain a second input, and inputting the second input into a second-layer feature extraction module of the TCN network for feature extraction to obtain a second output; analogizing in sequence until the final output characteristics of the n groups of data are obtained, inputting the final output characteristics into a full-connection neural network for training to obtain an intermediate prediction result, comparing the intermediate prediction result with a real value and calculating an error, adjusting network parameters through the error, taking new n groups of data out of a training set again for training until the error is reduced to a preset order of magnitude, and obtaining the meteorological factor prediction model at the moment; wherein n represents a hyper-parameter, n ∈ [24,96 ];
s4: testing the meteorological factor prediction model through the test set, adjusting the hyper-parameters of the meteorological factor prediction model according to the test effect, and repeating the step S3 until the final meteorological factor prediction model is obtained;
s5: and inputting the microclimate information of a certain time period near the power transmission line and the corresponding weather forecast information into the final weather factor prediction model, and predicting the microclimate information of the next time period near the power transmission line.
The beneficial effect of this disclosure lies in: according to the multi-meteorological-factor prediction method for the power transmission line, the characteristics are extracted by combining a TCN (Temporal Convolutional Network) Network and a DenseNet Network, firstly, the TCN Network can process time sequence information in parallel, flexible receptive field and stable gradient are achieved, and compared with other methods, the method occupies lower memory, can well fit time sequence relation, and improves the accuracy of prediction of each meteorological factor; secondly, a DenseNet network is combined on the basis of a TCN network, so that overfitting can be resisted, the generalization performance is improved, and the accuracy of weather prediction is improved; finally, the method not only utilizes the historical microclimate information, but also combines the weather forecast information when predicting each meteorological factor, thereby improving the accuracy of meteorological prediction.
Drawings
Fig. 1 is a flow chart of a multi-meteorological-factor prediction method for a power transmission line according to the present disclosure;
FIG. 2 is a schematic diagram of feature extraction performed by combining a TCN network and a DenseNet network;
fig. 3 is a schematic view of visualization of input and output when the TCN network and the DenseNet network are combined to perform feature extraction.
Detailed Description
The technical scheme of the disclosure will be described in detail with reference to the accompanying drawings. In the description of the present invention, it is to be understood that the terms "first" and "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of technical features indicated, but merely as distinguishing between different components.
Fig. 1 is a flowchart of a multi-meteorological-factor prediction method for a power transmission line according to the present disclosure, and as shown in fig. 1, the method specifically includes: step S1: and acquiring microclimate information and corresponding weather forecast information near the power transmission line.
Step S2: and grouping the microclimate information and the corresponding weather forecast information according to a time sequence, wherein each group of data comprises the microclimate information and the corresponding weather forecast information in the same time period, and dividing all data into a training set and a test set according to groups. In the application, the microclimate information comprises the temperature at the height of 2 meters above the ground, the relative humidity at the height of 2 meters above the ground and the wind speed at the height of 10 meters above the ground; the weather forecast information includes 3 types of weather forecast information corresponding to the microclimate information in the same time period, that is, the microclimate information includes 3 pieces of information, and the weather forecast information includes 3 pieces of information, so that each set of data has a piece of information in total, and a is 6.
Step S3: obtaining a meteorological factor prediction model through training of the training set, wherein the meteorological factor prediction model comprises the following steps: selecting n groups of data from the training set as a first input, inputting the data into a first-layer feature extraction module of the TCN network for feature extraction, and obtaining a first output; splicing the first output and the first input through a DenseNet network to obtain a second input, and inputting the second input into a second-layer feature extraction module of the TCN network for feature extraction to obtain a second output; analogizing in sequence until the final output characteristic of the n groups of data is obtained, inputting the final output characteristic into a Full Connection neural network (FC, Full Connection) for training to obtain an intermediate prediction result, comparing the intermediate prediction result with a true value (namely actual microclimate information) and calculating an error, adjusting network parameters (including weights of the TCN network, the DenseNet network and the Full Connection neural network) through the error, taking out new n groups of data from the training set again for training until the error is reduced to a preset order of magnitude, and obtaining the meteorological factor prediction model at this moment; where n represents a hyperparameter, n ∈ [24,96 ].
Specifically, step S3 further includes: step S31: one of the training sets comprises a sequence (x) of n sets of microclimate information and corresponding weather forecast information1,x2,x3,...,xt,...,xn) Inputting the input into a TCN network and a DenseNet network for feature extraction to obtain the final output features of the n groups of data, and obtaining the final output features of any t e [1, n ∈]，xtThe size of (a).
Step S31 further includes: l feature extraction modules (k, d) through a TCN network(i)Filters) performs feature extraction on the microclimate information and the weather forecast information to obtain the final output features, and the calculation formulas of the feature extraction are shown as (1), (2) and (3):
Figure BDA0002952165840000031
Figure BDA0002952165840000032
Figure BDA0002952165840000033
wherein x represents a sequence of the input micrometeorological information and weather forecast information (x)1,x2,x3,...,xt,...,xn) The vector of the composition is then calculated,
Figure BDA0002952165840000034
representing hidden variables of i-th layer
Figure BDA0002952165840000035
Vector of components, C representing the final output characteristic, fi(.) represents a feature extraction module (k, d)(i)Filters), concat (.) means that the function is spliced before and after the time dimension, i.e. for any i e [2, l ]]And j ∈ [1, p ]i-1]，
Figure BDA0002952165840000036
l、k、d(i)Both filters represent a hyper-parameter, for any i ∈ [1, l ∈ [ ]]，pi＝n+i*filters，pl＝n+l*filters。
In the formula (1), the reaction mixture is,
Figure BDA0002952165840000037
the reason why the hidden variable of the 0 th layer is represented by the 0 th layer is that the hidden variable of the 0 th layer is actually the input x of the whole model, and strictly speaking, the hidden variable cannot be calculated, and the hidden variable is only used for the unified expression of the formula. Inputting the 0 th layer hidden variable into the 1 st layer feature extraction module, and outputting the 1 st layer hidden variable; inputting the hidden variables of the 1 st layer into a 2 nd layer feature extraction module, and outputting the hidden variables of the 2 nd layer; and so on.
As can be seen, the TCN network has l feature extraction modules (k, d)(i)Filters), the feature extraction module here refers to Residual block(k,d(i)Filters) (residual block), see in particular the paper "Bai S, Kolter J Z, Koltun V.an empirical evaluation of genetic consistent and recovery networks for sequence modification [ J]arXiv preprint arXiv:1803.01271,2018. When the meteorological factor prediction model is trained through the training set, the training set comprises a plurality of groups of data, each group of data comprises microclimate information and corresponding weather forecast information in the same time period, and according to the step S31, one sequence (x) comprising n groups of microclimate information and corresponding weather forecast information in the training set is obtained1,x2,x3,...,xt,...,xn) A first feature extraction module which is used as a first input and is input into the TCN network performs feature extraction to obtain a first output, the DenseNet network splices the first output and the first input to obtain a second input, the second input is input into a second feature extraction module of the TCN network to perform feature extraction to obtain a second output, and so on until the first input is input into the first feature extraction module of the TCN network to perform feature extraction to obtain a final output feature C of the n groups of data, as shown in fig. 2, correspondingly, fig. 3 is a visual schematic diagram of input and output when the TCN network and the DenseNet network are combined to perform feature extraction, and each rectangle in fig. 3 represents input and output data.
Step S32: using C as input of fully-connected neural network to output predicted meteorological value
Figure BDA0002952165840000041
m represents the number of sets of microclimate information to be predicted, for any given t e [1, m]，
Figure BDA0002952165840000042
Is a/2.
Step S32 specifically includes: inputting C into the fully-connected neural network to obtain a prediction vector consisting of m predicted values
Figure BDA0002952165840000043
As an output, the calculation formula is shown in (4):
Figure BDA0002952165840000044
(4) (ii) a Wherein g (.) represents a fully connected neural network.
Step S33: and calculating the mean square error (MSE (theta)) of the prediction meteorological value and the microclimate information.
Step S33 specifically includes:
Figure BDA0002952165840000045
wherein, yiThe true value of the microclimate information is represented,
Figure BDA0002952165840000046
predicted value, y, representing micrometeorological informationiAnd
Figure BDA0002952165840000047
the size of the (c) is a/2, and m represents the group number of the predicted values output by the fully-connected neural network.
Step S34: the model parameters Θ are optimized using a Stochastic Gradient Descent (SGD) method.
Step S35: reading the next sequence comprising n groups of micrometeorological information and corresponding weather forecast information, repeating the steps S31 to S34, and continuously optimizing the model parameters theta until the MSE (theta) is reduced to a preset order of magnitude.
Step S35 is to repeat the process from step S31 to step S34, first executing step S31, inputting the next sequence including n sets of microclimate information and corresponding weather forecast information in the training set as the first input again to the first feature extraction module of the TCN network for feature extraction until the final output feature C' of the sequence of n sets of microclimate information and corresponding weather forecast information is obtained; then, steps S32 to S34 are performed. The steps S31 to S34 are repeated until the model parameter Θ in step S34 falls to a certain value.
Step S4: and testing the meteorological factor prediction model through the test set, adjusting the hyper-parameters of the meteorological factor prediction model according to the test effect, and repeating the step S3 until the final meteorological factor prediction model is obtained.
Step S5: and inputting the microclimate information of a certain time period near the power transmission line and the corresponding weather forecast information into the final weather factor prediction model, and predicting the microclimate information of the next time period near the power transmission line.
For the public understanding, the technical scheme of the invention is explained in detail by a preferred embodiment and the accompanying drawings.
In step S1, microclimate information near the power transmission line is collected, and weather forecast information of the same time period as the microclimate information near the power transmission line is recorded. The method comprises the following steps that a microclimate information acquisition device is installed near a power transmission line in a high-rise area of a circuit disaster (such as line galloping, line breakage, tower falling, ice coating and the like), the microclimate information acquisition device is required to be capable of acquiring microclimate information of the power transmission line in real time, and the period and the content of acquired data are as follows:
a data acquisition cycle: one hour;
data acquisition content: at least comprises 6 information of t2m, rh2m and w10m (respectively representing the temperature at the height of 2 meters above the ground, the relative humidity at the height of 2 meters above the ground and the wind speed at the height of 10 meters above the ground).
In step S2, the collected microclimate information and the corresponding weather forecast information are divided into a training set and a test set.
Firstly, establishing a one-to-one correspondence relationship between the collected microclimate information and the weather forecast information according to time. And secondly, sequencing and grouping the sorted microclimate information and weather forecast information according to time, wherein each group of data comprises microclimate information and corresponding weather forecast information in the same time period. And finally, counting the number of the groups, and dividing the data sets of all the groups into a training set and a test set according to the ratio of 7: 3.
Step S3: and training through a training set to obtain a meteorological factor prediction model.
And fitting the current n groups of micro-meteorological information and the corresponding meteorological forecast information by using a combination of a DenseNet network, a TCN network and a fully-connected neural network to obtain m groups of predicted values of future meteorological factors (micro-meteorological information) as output meteorological factor prediction models of the power transmission line. The specific training steps of the model are as follows:
and step S31, performing feature extraction on the microclimate information and the data corresponding to the weather forecast information sequence by using a DenseNet network and a TCN network.
Inputting: the serialized microclimate information and the corresponding weather forecast information are spliced into a sequence: (x)1,x2,x3,...,xt,...,xn)；
And (3) treatment: a feature extraction module of the TCN network extracts features of an input sequence to obtain output features, and the DenseNet network splices the input sequence and the output features to be used as new input features to be input into a next feature extraction module of the TCN network, and so on until a last feature extraction module of the TCN network outputs final output features C;
and (3) outputting: and C, feature extraction result.
The relation between the input and the output of each feature extraction module is shown in formulas (1), (2) and (3):
Figure BDA0002952165840000051
Figure BDA0002952165840000052
Figure BDA0002952165840000053
wherein
Figure BDA0002952165840000054
Represents the input of the feature extraction module of the i +1 th layer, and is also the output of the feature extraction module of the i-th layer
Figure BDA0002952165840000055
Results stitched according to time dimensionAnd x represents a sequence of the input micrometeorological information and weather forecast information (x)1,x2,x3,...,xt,...,xn) The vector of the composition, C represents the final output feature, l represents the number of feature extraction modules, k, d(i)And the filters are parameters of the ith layer of feature extraction module. In the formula (2), fi(.) represents the i-th feature extraction module (k, d)(i)Filters), concat (.) indicates that will be
Figure BDA0002952165840000056
And
Figure BDA0002952165840000057
and the splicing modules are spliced together according to the time dimension. n, l, k, d(i)The filters all represent hyper-parameters, n is suggested to be between 24 and 96, k is 2, d(i)Get 2i-1Taking 8 from filters, and taking l
Figure BDA0002952165840000058
For example, if data from 1 month, 1 day, 0 to 1 month, 2 day, 7 are used to predict data from 1 month, 2 day, 8 to 1 month, 2 day, 13, and if the data is input in units of hours, n is 32, and the output m is 6. When filters takes 8, then the length of the input of the first layer feature extraction module is 32, and the length of the output is 8; the input of the second layer of feature extraction module is the concatenation of the input and the output of the first layer of feature extraction module, the length is 40, the output length of the second layer of feature extraction module is 8, the recursion is carried out in sequence, and the input length of the first layer of feature extraction module is 32+8 x (l-1).
And step S32, outputting the predicted microclimate information by using the fully-connected neural network.
Inputting: a feature extraction result C;
and (3) treatment: the fully connected neural network processes the feature extraction result;
and (3) outputting: predicting weather values
Figure BDA0002952165840000061
The calculation formula is as shown in (4)The following steps:
Figure BDA0002952165840000062
where C denotes the feature extraction result output in step S31,
Figure BDA0002952165840000063
representing a predictor vector consisting of m predictors
Figure BDA0002952165840000064
And g (.) represents a fully-connected neural network, m is a hyper-parameter, and the value of the set number of the future micrometeorological information is predicted according to the requirement.
Step S33, calculating the mean square error MSE (theta) of the prediction meteorological value and the actual microclimate information, wherein the calculation process is shown as formula (5):
Figure BDA0002952165840000065
wherein, yiThe true value of the microclimate information is represented,
Figure BDA0002952165840000066
predicted value, y, representing micrometeorological informationiAnd
Figure BDA0002952165840000067
is a/2, i.e., a vector of size 3.
And step S34, performing back propagation, and optimizing the model parameter theta by using a random gradient descent method.
And S35, selecting the next sequence comprising n groups of micrometeorological information and weather forecast information, repeating the steps S31 to S34, and continuously optimizing the model parameters theta until the MSE (theta) is reduced to a preset order of magnitude, wherein the MSE (theta) is generally recommended to be reduced to be less than 1.
Step S4: and testing the meteorological factor prediction model through the test set, adjusting the hyper-parameters of the meteorological factor prediction model according to the test effect, and repeating the step S3 until the final meteorological factor prediction model is obtained.
Deploying the test set on a meteorological factor prediction model for testing, detecting the effect of the model, and adjusting the hyper-parameter information according to the effect of the model, such as: learning rate lr, sequence length n of input microclimate information and corresponding weather forecast information, sequence length m of microclimate information to be predicted, l, k, d in step S31(i)Filters, etc. And then jumping to step S3 to retrain until the model can obtain satisfactory effect on the test set, and obtaining the final power transmission line multi-meteorological-factor prediction model through the step.
Step S5: and (3) taking the micro-meteorological information sequence and the weather forecast information sequence in the last (such as within three days) as the input sequence of the model, and inputting the sequence into the multi-meteorological-factor prediction model of the power transmission line to realize the prediction of the multiple meteorological factor values of the power transmission line in a short term.
The specific prediction results are in the form:
the temperature is 12.8 ℃ at 8 o' clock of 1 month, 2 days, the relative humidity is 68, the wind speed is 5.6m/s,
the temperature is 12.3 ℃ at 9 o' clock in 1 month and 2 days, the relative humidity is 66, the wind speed is 8.8m/s,
the temperature is 12.9 ℃ at 10 o' clock of 1 month and 2 days, the relative humidity is 63, the wind speed is 7.4m/s,
the temperature of 11 o' clock in 1 month and 2 days is 13.2 ℃, the relative humidity is 60, the wind speed is 7.3m/s,
the temperature is 13.8 ℃ at 12 o' clock in 1 month and 2 days, the relative humidity is 57, the wind speed is 8.2m/s,
13 o' clock of 1 month and 2 days, the temperature is 13.9 ℃, the relative humidity is 56, and the wind speed is 6.8 m/s.
After the prediction result is obtained, the engineering maintenance unit can judge whether the risks of galloping, broken lines, tower falling or ice coating and the like exist according to various meteorological factors.
The foregoing is an exemplary embodiment of the present disclosure, and the scope of the present disclosure is defined by the claims and their equivalents.

Claims (6)

1. A multi-meteorological-factor prediction method for a power transmission line is characterized by comprising the following steps:
s1: acquiring microclimate information and corresponding weather forecast information near the power transmission line;
s2: grouping the microclimate information and the corresponding weather forecast information according to a time sequence, wherein each group of data comprises the microclimate information and the corresponding weather forecast information in the same time period, and dividing all data into a training set and a test set according to groups;
s3: obtaining a meteorological factor prediction model through training of the training set, wherein the meteorological factor prediction model comprises the following steps: selecting n groups of data from the training set as a first input, inputting the data into a first-layer feature extraction module of the TCN network for feature extraction, and obtaining a first output; splicing the first output and the first input through a DenseNet network to obtain a second input, and inputting the second input into a second-layer feature extraction module of the TCN network for feature extraction to obtain a second output; analogizing in sequence until the final output characteristics of the n groups of data are obtained, inputting the final output characteristics into a full-connection neural network for training to obtain an intermediate prediction result, comparing the intermediate prediction result with a real value and calculating an error, adjusting network parameters through the error, taking new n groups of data out of a training set again for training until the error is reduced to a preset order of magnitude, and obtaining the meteorological factor prediction model at the moment; wherein n represents a hyper-parameter, n ∈ [24,96 ];
s4: testing the meteorological factor prediction model through the test set, adjusting the hyper-parameters of the meteorological factor prediction model according to the test effect, and repeating the step S3 until the final meteorological factor prediction model is obtained;
s5: and inputting the microclimate information of a certain time period near the power transmission line and the corresponding weather forecast information into the final weather factor prediction model, and predicting the microclimate information of the next time period near the power transmission line.
2. The multi-meteorological-factor prediction method for the power transmission line according to claim 1, wherein the microclimate information comprises a temperature at a height of 2 meters above the ground, a relative humidity at a height of 2 meters above the ground, and a wind speed at a height of 10 meters above the ground; the weather forecast information includes 3 types of weather forecast information corresponding to the microclimate information in the same time period, and each group of data has a total of information, where a is 6.
3. The multi-meteorological-factor prediction method for the power transmission line according to claim 2, wherein the step S3 includes:
step S31: one of the training sets comprises a sequence (x) of n sets of microclimate information and corresponding weather forecast information1,x2,x3,...,xt,...,xn) Inputting the input into a TCN network and a DenseNet network for feature extraction to obtain the final output features of the n groups of data, and obtaining the final output features of any t e [1, n ∈]，xtThe size of (a);
step S32: taking the final output characteristics as the input of a fully-connected neural network, and outputting a predicted meteorological value
Figure FDA0002952165830000011
m represents the number of sets of microclimate information to be predicted, for any given t e [1, m]，
Figure FDA0002952165830000012
The size of (a) is a/2;
step S33: calculating the mean square error (MSE (theta)) of the prediction meteorological value and the microclimate information;
step S34: optimizing the model parameter theta by using a random gradient descent method;
step S35: reading the next sequence comprising n groups of micrometeorological information and corresponding weather forecast information, repeating the steps S31 to S34, and continuously optimizing the model parameters theta until the MSE (theta) is reduced to a preset order of magnitude.
4. The multi-meteorological-factor prediction method for the power transmission line according to claim 3, wherein the step S31 includes:
l feature extraction modules (k, d) through a TCN network(i)Filters) extracting the characteristics of the microclimate information and the weather forecast information to obtain the final output characteristics,the calculation formulas of feature extraction are shown as (1), (2) and (3):
Figure FDA0002952165830000021
Figure FDA0002952165830000022
Figure FDA0002952165830000023
wherein x represents a sequence of the input micrometeorological information and weather forecast information (x)1,x2,x3,...,xt,...,xn) The vector of the composition is then calculated,
Figure FDA0002952165830000024
representing hidden variables of i-th layer
Figure FDA0002952165830000025
The vector of the components, that is,
Figure FDA0002952165830000026
represents the input of the (i +1) th layer feature extraction module; c represents the final output characteristic, fi(.) represents a feature extraction module (k, d)(i)Filters), concat (.) indicates that it is going to be
Figure FDA0002952165830000027
And
Figure FDA0002952165830000028
splicing according to time dimension, namely splicing any i e to [2, l ∈]And j ∈ [1, p ]i-1]，
Figure FDA0002952165830000029
l、k、d(i)Each of the filters represents a hyper-parameter, k is 2, d(i)＝2i-1，filters＝8，l＝2[log2 n]For any i e [1, l ∈ ]]Input p to the i-th layer feature extraction ModuleiN + (i-1) filters, input p to the l-th level feature extraction modulel＝n+(l-1)*filters。
5. The multi-meteorological-factor prediction method for the power transmission line according to claim 4, wherein the step S32 includes:
inputting the final output characteristics into a full-connection neural network to obtain a prediction vector consisting of m predicted values
Figure FDA00029521658300000210
As an output, the calculation formula is shown in (4):
Figure FDA00029521658300000211
wherein g (.) represents a fully connected neural network.
6. The multi-meteorological-factor prediction method for the power transmission line according to claim 5, wherein the step S33 includes:
Figure FDA00029521658300000212
wherein, yiThe true value of the microclimate information is represented,
Figure FDA00029521658300000213
predicted value, y, representing micrometeorological informationiAnd
Figure FDA00029521658300000214
the size of the (c) is a/2, and m represents the group number of the predicted values output by the fully-connected neural network.
CN202110213155.3A 2021-02-25 2021-02-25 Multi-meteorological-factor prediction method for power transmission line Pending CN112925825A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110213155.3A CN112925825A (en) 2021-02-25 2021-02-25 Multi-meteorological-factor prediction method for power transmission line

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110213155.3A CN112925825A (en) 2021-02-25 2021-02-25 Multi-meteorological-factor prediction method for power transmission line

Publications (1)

Publication Number Publication Date
CN112925825A true CN112925825A (en) 2021-06-08

Family

ID=76171957

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110213155.3A Pending CN112925825A (en) 2021-02-25 2021-02-25 Multi-meteorological-factor prediction method for power transmission line

Country Status (1)

Country Link
CN (1) CN112925825A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113642234A (en) * 2021-08-09 2021-11-12 贵州电网有限责任公司 Power grid icing prediction method based on multi-source characteristic time convolution deep learning
CN114792174A (en) * 2022-06-23 2022-07-26 国网天津市电力公司电力科学研究院 Distribution network line meteorological prediction method based on wavelet neural network
KR20230038839A (en) * 2021-09-13 2023-03-21 금오공과대학교 산학협력단 3D printer monitoring system applying machine learning algorithm based on Temporal Convolution Neural Network with Two-Stage Sliding Window Strategy

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110442860A (en) * 2019-07-05 2019-11-12 大连大学 Name entity recognition method based on time convolutional network
CN111311027A (en) * 2020-03-23 2020-06-19 南京大学 Wind speed prediction method for power transmission line

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110442860A (en) * 2019-07-05 2019-11-12 大连大学 Name entity recognition method based on time convolutional network
CN111311027A (en) * 2020-03-23 2020-06-19 南京大学 Wind speed prediction method for power transmission line

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113642234A (en) * 2021-08-09 2021-11-12 贵州电网有限责任公司 Power grid icing prediction method based on multi-source characteristic time convolution deep learning
KR20230038839A (en) * 2021-09-13 2023-03-21 금오공과대학교 산학협력단 3D printer monitoring system applying machine learning algorithm based on Temporal Convolution Neural Network with Two-Stage Sliding Window Strategy
KR102581949B1 (en) * 2021-09-13 2023-09-22 금오공과대학교 산학협력단 3D printer monitoring system applying machine learning algorithm based on Temporal Convolution Neural Network with Two-Stage Sliding Window Strategy
CN114792174A (en) * 2022-06-23 2022-07-26 国网天津市电力公司电力科学研究院 Distribution network line meteorological prediction method based on wavelet neural network

Similar Documents

Publication Publication Date Title
CN112925825A (en) Multi-meteorological-factor prediction method for power transmission line
CN109492823B (en) Method for predicting icing thickness of power transmission line
CN111311027A (en) Wind speed prediction method for power transmission line
CN111458769B (en) Method and system for predicting environmental meteorological data of power transmission line
CN116256602B (en) Method and system for identifying state abnormality of low-voltage power distribution network
CN102478584B (en) Wind power station wind speed prediction method based on wavelet analysis and system thereof
CN104252649A (en) Regional wind power output prediction method based on correlation between multiple wind power plants
CN103440541A (en) Joint probability density prediction method of short-term output power of plurality of wind power plants
CN111753893A (en) Wind turbine generator power cluster prediction method based on clustering and deep learning
CN110059845B (en) Metering device clock error trend prediction method based on time sequence evolution gene model
CN107038453A (en) The fault early warning system and determination methods of a kind of Wind turbines
CN106651007A (en) Method and device for GRU-based medium and long-term prediction of irradiance of photovoltaic power station
CN103996079B (en) Wind power weighting predication method based on conditional probability
CN103914740B (en) A kind of powerline ice-covering prediction and automatic correcting method based on data-driven
CN111178585A (en) Fault reporting amount prediction method based on multi-algorithm model fusion
CN109657835A (en) A kind of power distribution network area fault number prediction technique and system
CN114202129A (en) Wind power output prediction method, electronic device, storage medium and system
CN106570790A (en) Wind farm output power data restoration method considering segmental characteristics of wind speed data
CN106570582A (en) Method and system for building transmission line dancing tripping risk prediction network model
CN114912355A (en) Method and device for predicting short-term icing of power transmission line and storage medium
CN114548601A (en) Power distribution network power failure prediction method and system under extreme disasters based on BP neural network
CN114357670A (en) Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder
CN105279582A (en) An ultra-short-term wind electricity power prediction method based on dynamic correlation characteristics
CN112684294A (en) Power distribution network fault first-aid repair positioning method based on environment dynamic influence
CN113779892B (en) Method for predicting wind speed and wind direction

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