CN-122020380-A - Icing thickness prediction model of digital-analog double-drive

CN122020380ACN 122020380 ACN122020380 ACN 122020380ACN-122020380-A

Abstract

The invention provides a strategy for predicting ice coating thickness by a digital-analog double-drive model, belongs to the field of power systems, and is insufficient in research on predicting the ice coating thickness of a power transmission line in the prior art. The method comprises the steps of reading operation data of a power transmission line and meteorological data of an icing area, preprocessing the data, establishing a digital-analog double-drive prediction model which takes Makkonen model as a physical model and CNN-LSTM neural network as a data driving model, and comparing the digital-analog double-drive prediction model with a pure CNN-LSTM neural network and a pure Makkonen model under the same data set and evaluation index to verify the prediction accuracy of the digital-analog double-drive prediction model. According to the invention, makkonen models are organically fused with the CNN-LSTM neural network, so that the collaborative optimization of mechanism constraint and data driving is realized. Not only improves the accuracy and the robustness of ice coating thickness prediction, but also enhances the physical interpretability and the engineering practicability of the model, and provides high-efficiency and credible technical support for the accurate early warning and the active defense of the ice coating disaster of the power grid.

Inventors

WANG ZHENGUO
HOU HUI
KE RENGUAN
LUO HUAFENG
Zhou Linfan
WANG YANJUN
ZHENG WENZHE
XUAN JIAZHUO

Assignees

国网浙江省电力有限公司电力科学研究院

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (4)

1. A digital-analog double-drive icing thickness prediction model is characterized by comprising the following steps: Step 1, reading operation data of a power transmission line and meteorological data of an ice coating area, and preprocessing the data; Step 2, establishing a digital-analog double-drive prediction model taking Makkonen model as a physical model and CNN-LSTM neural network as a data drive model; And 3, comparing the digital-analog double-drive prediction model with a pure CNN-LSTM neural network and a pure Makkonen model under the same data set and evaluation index, and verifying the prediction accuracy of the digital-analog double-drive prediction model.
2. The digital-to-analog dual-drive icing thickness prediction model according to claim 1, characterized in that: The pretreatment of the data in the step 1 is carried out according to the following steps: The method comprises the steps that 1.1, the read operation data of a power transmission line and meteorological data of an icing area comprise a reference tension value, a tension increase rate, a comprehensive suspension load, an unbalanced tension difference, light radiation intensity, rainfall measurement, temperature, maximum wind speed, standard wind speed, 10-minute average wind direction, humidity, maximum wind speed, air pressure, 10-minute average wind speed, precipitation intensity and equivalent icing thickness; And 1.2, cleaning, denoising and normalizing the read data.
3. The digital-to-analog dual-drive icing thickness prediction model according to claim 1, characterized in that: The digital-analog double-drive prediction model in the step 2 is specifically as follows: Step 2.1 basic principle of CNN-LSTM neural network: is the cell state at time t-1; outputting a hidden layer at the moment t-1; input at time t; For forgetting the door, controlling the candidate state and information reservation at the last moment; To activate a function, describing the amount of cellular information that can pass, it can define an output value between [0,1], indicating full pass when the value is 1, and no full pass when the value is 0; The input door is used for controlling the internal state input information storage at the current moment; in order to output the door, the door is provided with a door opening, controlling the output of the internal state at the current moment; is an activation function; Is that A hidden layer input at a moment; Is that Cell state at time; , Wherein, the A weight matrix for the input gate; a bias vector for the input gate; , , Wherein, the 、 The weight matrix of the forgetting gate and the memory unit respectively, As a result of the candidate memory cell, 、 Bias vectors of the forgetting gate and the memory unit respectively; , , Wherein, the To output the weight matrix of the gate, A bias vector for the output gate; ; Step 2.2 principle of Makkonen model: , In the formula, Indicating the ice coating quality of the wire per unit length, The time is represented by the time period of the day, Indicating the ambient wind speed, Indicating the radius of the wire, The length of the line is indicated and, Indicating the water content in the liquid state, The collision rate is indicated by the number of collisions, The capture rate is indicated as being indicative of the rate of capture, The freezing rate is indicated by the expression, Representing a small increment of ice coating mass, Indicating the rate of increase of the ice coating mass, Namely the ice coating quality increased on the wire in unit length in unit time; step 2.3, the fusion strategy of the data driving model and the physical model is as follows: fusing Makkonen model and CNN-LSTM neural network prediction results by weighted average method, and setting CNN-LSTM neural network prediction value as The predicted value of Makkonen model is The predicted value after fusion is The fusion formula is: , In the formula, Is the weight of the CNN-LSTM neural network, Is weight of Makkonen model, and And mixing the materials in proportion and outputting a result.
4. The digital-to-analog dual-drive icing thickness prediction model according to claim 1, characterized in that: in the step 3, the steps of comparing the digital-analog dual-drive prediction model with the CNN-LSTM neural network model and Makkonen model under the same data set and evaluation index are as follows: Step 3.1 selecting the decision coefficient Root mean square error Percentage of absolute error from average As an index of the evaluation, a color of the sample was measured, , , , In the formula, Is the first The icing thickness predictions for the individual samples, Is the first A true value of the icing thickness of each sample, Is the average value of the actual ice coating thickness, Sample size for the test set; And 3.2, outputting a prediction result by the digital-analog double-drive prediction model, and respectively evaluating the prediction result of the digital-analog double-drive prediction model, the prediction result of the pure CNN-LSTM neural network and the prediction result of the pure Makkonen model by using the three evaluation indexes to obtain a conclusion.

Description

Icing thickness prediction model of digital-analog double-drive Technical Field The invention belongs to the field of power systems, and particularly relates to a digital-analog double-drive icing thickness prediction model. Background When the air temperature is low, the phenomenon of icing of the power transmission line is very easy to occur. Especially when serious ice and snow disasters occur, serious accidents such as wire breakage, tower collapse and the like can be caused, so that personal safety is endangered and normal work and life of residents are influenced. Therefore, the ice coating thickness of the power transmission line is accurately predicted, effective precautionary measures are adopted in advance, and the method has important significance for guaranteeing safe and reliable operation of a power system. Therefore, the influence of meteorological factors during the icing process is researched, the process of icing under different micro-terrains is discussed, and further, support is provided in the aspects of reducing economic loss, improving the anti-icing and disaster-reducing capabilities of the electric power departments and the like, so that the method has very important scientific significance and practical application value. Disclosure of Invention Aiming at the defect of insufficient research on predicting the icing thickness of the power transmission line in the prior art, the invention provides a digital-analog double-drive icing thickness prediction model. In order to achieve the above purpose, the digital-analog double-drive icing thickness prediction model of the invention comprises the following steps: Step 1, reading operation data of a power transmission line and meteorological data of an ice coating area, and preprocessing the data; Step 2, establishing a digital-analog double-drive prediction model taking Makkonen model as a physical model and CNN-LSTM neural network as a data drive model; And 3, comparing the digital-analog double-drive prediction model with a pure CNN-LSTM neural network and a pure Makkonen model under the same data set and evaluation index, and verifying the prediction accuracy of the digital-analog double-drive prediction model. Preferably, the preprocessing of the data in step 1 is performed according to the following steps: The method comprises the steps that 1.1, the read operation data of a power transmission line and meteorological data of an icing area comprise a reference tension value, a tension increase rate, a comprehensive suspension load, an unbalanced tension difference, light radiation intensity, rainfall measurement, temperature, maximum wind speed, standard wind speed, 10-minute average wind direction, humidity, maximum wind speed, air pressure, 10-minute average wind speed, precipitation intensity and equivalent icing thickness; And 1.2, cleaning, denoising and normalizing the read data. Preferably, the digital-analog dual-drive prediction model in the step 2 is specifically as follows: step 2.1 basic principle of CNN-LSTM neural network model: is the cell state at time t-1; outputting a hidden layer at the moment t-1; input at time t; For forgetting the door, controlling the candidate state and information reservation at the last moment; To activate a function, describing the amount of cellular information that can pass, it can define an output value between [0,1], indicating full pass when the value is 1, and no full pass when the value is 0; The input door is used for controlling the internal state input information storage at the current moment; in order to output the door, the door is provided with a door opening, controlling the output of the internal state at the current moment; is an activation function; Is that A hidden layer input at a moment; Is that Cell state at time; , Wherein, the A weight matrix for the input gate; a bias vector for the input gate; , , Wherein, the 、The weight matrix of the forgetting gate and the memory unit respectively,As a result of the candidate memory cell,、Bias vectors of the forgetting gate and the memory unit respectively; , , Wherein, the To output the weight matrix of the gate,A bias vector for the output gate; ; step 2.2 Makkonen principle: , In the formula, Indicating the ice coating quality of the wire per unit length,The time is represented by the time period of the day,Indicating the ambient wind speed,Indicating the radius of the wire,The length of the line is indicated and,Indicating the water content in the liquid state,The collision rate is indicated by the number of collisions,The capture rate is indicated as being indicative of the rate of capture,The freezing rate is indicated by the expression,Representing a small increment of ice coating mass,Indicating the rate of increase of the ice coating mass,Namely the ice coating quality increased on the wire in unit length in unit time; step 2.3, the fusion strategy of the data driving model and the physical model is as follows