CN106570561A - System and method for predicting insulator surface non-soluble deposit density - Google Patents

Abstract

The invention discloses a system and method for predicting the density of insoluble deposits on the surface of an insulator. The system includes an original data acquisition unit, a series gray neural network prediction unit, a parallel gray neural network prediction unit, an embedded gray neural network prediction unit, NSDD predicted value output unit and NSDD early warning unit; put the insoluble sediment density data on the surface of the insulator and local meteorological data into the series gray neural network prediction unit, the parallel gray neural network prediction unit and the embedded gray neural network prediction unit, through which A single prediction unit predicts the NSDD value of the insulator respectively; then uses the test sample to judge the prediction accuracy of the three prediction units, and takes the output of the prediction unit with high prediction accuracy as the predicted value of the NSDD of the insulator; through the NSDD early warning unit according to whether An early warning is issued when the predicted values output by two or more predictive units reach the preset grading early warning threshold.

Description

A system and method for predicting the density of insoluble deposits on the surface of an insulator

technical field

The invention belongs to the technical field of external insulation of power systems, and more specifically relates to a system and method for predicting the density of insoluble deposits on the surface of an insulator.

Background technique

Due to the accumulation of dirt on the surface of insulators under normal working voltage, pollution flashover accidents are prone to occur in rainy, foggy and other bad weather, which poses a serious threat to the safe and stable operation of the power system. It is very necessary to predict the pollution degree of insulators on transmission lines in order to prevent pollution flashover accidents in time. Non-Soluble Deposit Density (NSDD) is usually used to evaluate the degree of pollution on the surface of insulators.

The gray model has been widely used in the prediction of the degree of pollution on the surface of insulators because of the advantages of less sample data required for modeling, no need to consider the distribution law and change trend, simple modeling and convenient operation. However, due to the lack of self-learning and self-organizing And self-adaptive ability, the ability to process information is weak, can not independently complete the prediction task.

Contents of the invention

Aiming at the above defects or improvement needs of the prior art, the present invention provides a system and method for predicting the density of insoluble deposits on the surface of insulators, the purpose of which is to combine gray models and neural networks to provide an Insoluble sediment density prediction method, and pollution flashover warning function.

In order to achieve the above object, according to one aspect of the present invention, a system for predicting the density of insoluble deposits on the surface of an insulator is provided, including a raw data acquisition unit, a series gray neural network prediction unit, a parallel gray neural network prediction unit, an embedded Gray neural network prediction unit and NSDD prediction value output unit;

Among them, the original data acquisition unit is used to obtain the insulator NSDD data and meteorological data on the transmission line; the series gray neural network prediction unit, parallel gray neural network prediction unit and embedded gray neural network prediction unit are respectively used for Insulator NSDD data and meteorological data predict the insulator NSDD on the transmission line; correspondingly, the series type gray neural network prediction unit outputs the first predicted value, the parallel type gray neural network prediction unit outputs the second predicted value, and the embedded gray neural network The prediction unit outputs a third predicted value;

The NSDD predicted value output unit is used to compare the above three predicted values with the sample data, select the predicted unit with the highest prediction accuracy from the above three predicted units according to the comparison results, and use the predicted value output by the predicted unit as the insulator NSDD predicted value.

Preferably, the above-mentioned insoluble deposit density prediction system on the surface of insulators also includes an NSDD early warning unit;

The NSDD early warning unit is used to generate an early warning signal according to the first predicted value, the second predicted value, the third predicted value and the preset early warning threshold; specifically, when the first predicted value, the second predicted value, and the third predicted value Two or more of the predicted values reach the early-warning threshold, and an early-warning signal is generated.

Preferably, the insoluble deposit density prediction system on the surface of an insulator, its serial gray neural network prediction unit includes a parallel first GM (1,1) model, a second GM (1,1) model, a third GM (1 ,1) model, and neural network;

Among them, the GM(1,1) model is a kind of gray model, which is a first-order differential equation containing only a single variable; the neural network includes an input layer, a hidden layer and an output layer, and its transfer function adopts the Sigmoid function;

The input interfaces of the first GM (1,1) model, the second GM (1,1) model, and the third GM (1,1) model are all connected to the original data acquisition unit; the input terminal of the neural network is connected to the first GM ( 1,1) model, the second GM(1,1) model, and the output interface of the third GM(1,1) model are connected;

The first GM(1,1) model, the second GM(1,1) model, and the third GM(1,1) model predict the insulator NSDD according to the insulator NSDD data on the transmission line respectively, and obtain three sets of graying prediction results ; Predict the NSDD of the insulator according to the three groups of graying prediction results by the neural network, and obtain the first predicted value.

Preferably, the insoluble deposit density prediction system on the surface of an insulator, its parallel gray neural network prediction unit includes a GM (1,1) model, a neural network and a combined prediction network; wherein, the GM (1,1) model and the neural network in parallel;

The input ends of the GM (1,1) model and the neural network are connected to the original data acquisition unit; the input end of the combined prediction network is connected to the output end of the GM (1,1) model and the output end of the neural network;

The GM(1,1) model and the neural network perform insulator NSDD prediction based on the insulator NSDD data and meteorological data on the transmission line respectively; the combined prediction network weights the prediction results of the GM(1,1) model and the neural network to obtain the second prediction value.

Preferably, the above-mentioned insoluble deposit density prediction system on the surface of an insulator, its embedded gray neural network prediction unit includes a graying layer, a neural network and a whitening layer in series;

Among them, the graying layer is used for accumulative transformation and smoothing of the original insulator NSDD data and meteorological data on the transmission line; the neural network is used for insulator NSDD prediction based on the accumulated and smoothed data, and the whitening layer is used for the neural network. The output data is subjected to accumulative transformation and restoration processing to obtain a third predicted value.

According to another aspect of the present invention, based on the above-mentioned insoluble deposit density prediction system on the insulator surface, a method for predicting the insoluble deposit density on the insulator surface is provided, comprising the following steps:

(1) Establish three GM(1,1) models with different sequence lengths based on the collected original insulator NSDD data; use the predicted NSDD values of these three GM(1,1) models as input, and measure NSDD as The output is put into the neural network for training to obtain the optimal weight and threshold of the neural network; the insulator NSDD prediction is performed with the trained neural network to obtain the first predicted value;

Among them, measuring NSDD refers to the collected NSDD data;

(2) Predict the insulator NSDD through the gray model and the neural network model respectively, and obtain two initial prediction data; determine the weight coefficients of the two initial prediction data according to the test samples; carry out weighting processing on the two initial prediction data according to the weight coefficients, obtain a second predicted value;

In this step, for the time node k to predict the NSDD value, the (k-10) measured NSDD data of 1~(k-10) nodes are used as training samples, and the 10 measured NSDD data of (k-10)~k The NSDD data value is used as a test sample;

(3) Train the neural network, and ash the original insulator NSDD data; put the ashed data into the trained neural network, and predict the insulator NSDD by the trained neural network; output the neural network Performing whitening processing on the data to obtain a third predicted value; wherein, the graying processing includes cumulative change and smoothing processing; whitening processing refers to cumulative transformation;

(4) The above-mentioned first predicted value, second predicted value and third predicted value are combined with the test sample, and the predicted unit with the highest prediction accuracy is selected from the above-mentioned three predicted units according to the comparison results, and the predicted unit is used The output predicted value is used as the final insulator NSDD predicted value.

Preferably, the method for predicting the density of insoluble deposits on the surface of an insulator, its step (3) includes the following sub-steps:

(3.1) Use the meteorological data, the original insulator NSDD data before time node m, and the grayed data of 10 time nodes before time node m to train the neural network to obtain the optimal weight and threshold; according to the optimal weight and the threshold to build a trained neural network;

Among them, m is the time node for training and predicting the NSDD data value of the insulator. The input data used in the training neural network includes meteorological data, the original insulator NSDD data from time node 1 to time node (m-1), (m-10) to m The ashing data of the original insulator NSDD data at 10 time nodes after ashing processing; the meteorological data includes wind speed, precipitation, and relative humidity; the output data used for training the neural network refers to (m-9)～(m+1) The ashing data of the original insulator NSDD data at 10 time nodes after ashing processing;

(3.2) Perform graying processing on the original insulator NSDD data of the mth node; put the grayed data into the trained neural network to obtain the initial prediction value; perform whitening processing on the initial prediction value to obtain the second prediction value.

Preferably, the method for predicting the density of insoluble deposits on the surface of an insulator further includes step (5):

(5) Compare the first predicted value, the second predicted value, and the third predicted value with the preset early warning thresholds, when two or more of the first predicted value, the second predicted value, and the third predicted value reach Early warning threshold to generate early warning signals.

Generally speaking, compared with the prior art, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

(1) The system and method for predicting the density of insoluble deposits on the surface of insulators provided by the present invention organically integrates the gray model and the neural network model. Linear and inaccurate laws have the advantage of self-adaptability; through combination, different information of the system can be obtained from different angles and different models, so as to achieve the purpose of improving prediction accuracy, increasing stability and reliability of results, so that the combined model can be used for The change of data structure has stronger robustness, which effectively makes up for the inaccurate defect of a single prediction method;

(2) In the insoluble deposit density prediction system and method on the surface of insulators provided by the present invention, the series type gray neural network prediction unit that it adopts connects the neural network and the gray model in series in the system, and uses the output of the gray model as the neural network. The input of the network can be used for fault-tolerant analysis and prediction of complex systems, which greatly reduces the training time of the neural network and effectively solves the problem that the prediction of a single neural network is easy to fall into a local minimum;

(3) The system and method for predicting the density of insoluble deposits on the surface of insulators provided by the present invention combines the gray model and the neural network in a weighted manner to construct a parallel gray neural network prediction unit, which overcomes the problem that a single model is easy to lose information Defects reduce randomness and have the effect of improving prediction accuracy;

(4) In the insoluble sediment density prediction system and method on the surface of insulators provided by the present invention, its embedded gray neural network prediction unit weakens the randomness of the original data by setting the graying layer, and is easily approximated by the nonlinear excitation function of the neural network , which greatly shortens the network learning time and speeds up the convergence process while improving the prediction accuracy;

(5) The insoluble sediment density prediction system and method on the surface of insulators provided by the present invention use the gray system to assist in constructing the neural network. Since the information structure of the gray system includes deterministic information and uncertainty information, the certainty information in the gray system is used to Information is used to assist in the construction of neural networks, the structure of neural networks is guided by deterministic information, and the learning algorithm of neural networks is improved;

(6) In the insoluble sediment density prediction system and method on the surface of an insulator provided by the present invention, its neural network can effectively enhance the gray system; because the gray system has an empty set (time zone not including information) in the information time zone, it can only be established Approximate, incompletely determined gray differential equation, but it is difficult to directly use the gray differential equation in practical applications, and the gray differential equation needs to be analyzed; the present invention constructs a neural network to analyze the gray parameters of the gray differential equation, from the gray system already Extract samples from the known data to train the neural network. When the neural network converges, the parameters of the analytic gray differential equation are extracted to obtain a definite differential equation and realize accurate prediction.

Description of drawings

1 is a schematic diagram of a system for predicting the density of insoluble deposits on the surface of an insulator provided in an embodiment;

Fig. 2 is a schematic diagram of the prediction process of the GM (1,1) model in the embodiment;

Fig. 3 is the structural representation of neural network model in the embodiment;

Fig. 4 is the neural network prediction flowchart schematic diagram in the embodiment;

Fig. 5 is the structural representation of the serial gray neural network model in the embodiment;

Fig. 6 is the structural representation of the parallel type gray neural network model in the embodiment;

Fig. 7 is a schematic structural diagram of the embedded gray neural network model in the embodiment.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not constitute a conflict with each other.

The insoluble sediment density prediction system on the surface of insulators provided by the embodiment of the present invention, as shown in Figure 1, includes a raw data acquisition unit, a series gray neural network prediction unit, a parallel gray neural network prediction unit, and an embedded gray neural network prediction unit. Unit, NSDD predicted value output unit and NSDD early warning unit;

Among them, the original data acquisition unit is used to obtain NSDD data and meteorological data of insulators on the transmission line; the series gray neural network prediction unit, the parallel gray neural network prediction unit and the embedded gray neural network prediction unit are respectively based on the NSDD data Predict the NSDD of the insulator with the meteorological data, correspondingly, obtain the first predicted value, the second predicted value, and the third predicted value; NSDD data of insulators on transmission lines; meteorological data are data collected around insulators to be predicted;

The NSDD prediction value output unit combines the above three prediction values with the sample data, selects the prediction unit with the highest prediction accuracy from the above three prediction units, and uses the prediction value output by the prediction unit as the insulator NSDD prediction value;

The NSDD early warning unit generates an early warning signal according to the first predicted value, the second predicted value, the third predicted value and the preset hierarchical warning threshold; specifically, when two of the first predicted value, the second predicted value, and the third predicted value If one or more predicted values reach the hierarchical early-warning threshold, an early-warning signal is generated.

The prediction process of the GM (1,1) model adopted in the present embodiment is shown in Figure 2, according to the figure as follows:

First input the original data sequence X ⁽⁰⁾ = [x ⁽⁰⁾ (1), x ⁽⁰⁾ (2), ..., x ⁽⁰⁾ (n)];

Then judge whether the above-mentioned original data sequence X ⁽⁰⁾ meets the modeling condition; If not, then carry out the smoothing processing of data, if so, then do accumulation processing to original data and obtain accumulation sequence X ⁽¹⁾ ; x ⁽¹⁾ =[x ⁽¹⁾ (1),x ⁽¹⁾ (2),…,x ⁽¹⁾ (n)]; where

Then construct a first-order linear differential equation with respect to X ⁽¹⁾ The parameters a and b are obtained by solving the least squares method;

in,

From this, the predicted value of X(1) is obtained,

The predicted value of X(0) is obtained through cumulative reduction and reduction,

The structure of the neural network model adopted in the present embodiment is as shown in Figure 3, comprises input layer, hidden layer, output layer; _{Wij is the weight value of input layer to hidden layer, and θ j is} hidden layer neuron The threshold of the element, V _ij is the connection weight from the hidden layer to the output layer, is the threshold of the output layer; the connection weights W _ij , V _ij and thresholds θ _j , acquired through training;

The input of each neuron in the hidden layer Among them, _Xi is the input amount of each input node of the neural network;

The transfer function of neural network adopts Sigmoid function f(x)=1/(1+e- ^x );

The output of the hidden layer

The input to the neurons in the output layer

The output of the neurons in the output layer, that is, the predicted value of the neural network

The process of predicting through the above neural network is shown in Figure 4, including the following steps:

First construct the basic structure of the neural network: set the number of nodes in the input layer, hidden layer, and output layer of the neural network, the initial weights and thresholds of the input hidden layer, and the initial weights and thresholds of the input and output layers;

Then select a sample from the sample library to train the neural network; during the training process, according to the error and output of the hidden layer, and the error and output of the output layer, the weights and thresholds of the hidden layer are continuously adjusted, and The weights and thresholds of the output layer;

After the neural network has been trained by all samples, the training of the neural network ends when the requirements of the learning times and error conditions are met.

In the embodiment, the structure of the series-type gray neural network prediction unit used is shown in Figure 5; it includes three GM(1,1) models, GM ₁ , GM ₂ and GM ₃ , and a 3×7×1 type Neural Networks;

The neural network includes 3 input layer nodes, 7 hidden layer nodes, and 1 output layer node; 3 GM (1,1) models GM ₁ , GM ₂ , and GM ₃ are the input quantities of the neural network; the output layer to The hidden layer and the transfer function from the hidden layer to the output layer are Sigmoid functions, and the setting of the hidden layer is based on Kolmogorov's theorem; the output of the neural network is the first predicted value.

The steps for NSDD prediction using the above series gray neural network prediction unit are as follows:

(1.1) Use the NSDD data acquired by the data acquisition module to establish three GM(1,1) models, and the sequence lengths of the three GM(1,1) models are 10, 8 and 6 respectively; (k-10), (k-18)~(k-10) and (k-16)~(k-10) time nodes data; where k refers to the time node that needs to be predicted for insulator NSDD;

(1.2) Use the three GM(1,1) models to predict respectively, and obtain three groups of NSDD prediction data, each group includes 10 NSDD prediction data values at (k-10)~k time nodes;

(1.3) The above-mentioned 3 GM(1,1) models share the NSDD prediction values (k-16)~k of these 16 NSDD data, and these 16 NSDD prediction data values are used as the input of the neural network, and the actual collection The original insulator NSDD data obtained is used as the output of the neural network to train the neural network to obtain the optimal weight and threshold of the neural network; in the embodiment, the neural network structure used in this step is 3 × 7 × 1 type;

(1.4) Use the trained neural network to predict the NSDD value of the insulator at the future moment after the k time node, and obtain the first predicted value.

In the embodiment, the structure of the adopted parallel gray neural network prediction unit is as shown in Figure 6, including GM (1,1) model, neural network and combination prediction network; GM (1,1) model is connected in parallel with neural network, The output of the two is used as the input of the combination prediction network, and the output of the combination prediction network is the second prediction value; the steps of using the parallel gray neural network prediction unit for NSDD prediction are as follows:

(2.1) Utilize the gray prediction model GM (1,1) model and the neural network model to predict respectively, obtain the initial prediction NSDD value y ₁ and y ₂ ;

(2.2) Use the test value Y ⁽⁰⁾ (t) to subtract the initial predicted NSDD values y ₁ and y ₂ respectively to obtain the prediction errors e ₁ and e ₂ ;

Calculate the weight coefficient ω 1 of the GM (1,1) model and the weight coefficient ω ₂ of the neural network model according to the prediction error _; ω ₁ +ω ₂ =1;

(2.3) Obtain the second predicted value according to y _c =ω ₁ y ₁ +ω ₂ y ₂ .

Among them, the weight coefficient ω ₁ and the weight coefficient ω ₂ are obtained according to the following method:

Obtain e ₁ , e ₂ and e _c according to the test sequence Y ⁽⁰⁾ (t) as

The variance of the second predicted value y _c of the insulator NSDD is

Find the minimum value of Var(e _c ) to get

Because, ω ₂ =1-ω ₁ therefore, weight coefficient ω ₁ and weight coefficient ω ₂ are respectively:

Wherein, Var(e ₁ )=σ ₁₁ , Var(e ₂ )=σ ₂₂ , cov(e ₁ ,e ₂ )=σ ₁₂ .

The structure of the embedded gray neural network prediction unit used in this embodiment is shown in Figure 7, the ashing layer, neural network and whitening layer are connected in series in sequence; the ashing layer performs cumulative transformation and smoothing processing on the original data, and the whitening layer The output data of the neural network is processed by cumulative transformation and restoration; the structure of the neural network is 4×9×1; the input layer of the neural network is meteorological data and NSDD data; the NSDD data is the (k-10)th node The accumulative data of 10 time nodes from the kth node; its hidden layer is set according to the Kolmogorov theorem; the output layer is the predicted NSDD data, and its value is a total of 10 time nodes from (k-9) to (k+1) Accumulated data of nodes;

The steps of NSDD prediction using the embedded gray neural network prediction unit are as follows:

(3.1) Perform cumulative transformation on the original data sequence X ⁽⁰⁾ = (x ⁽⁰⁾ (1), x ⁽⁰⁾ (2), ..., x ⁽⁰⁾ (n)) to obtain the 1-AGO sequence X ^{(1 )} = (x ⁽¹⁾ (1),x ⁽¹⁾ (2),...,x ⁽¹⁾ (n));

Among them, x ⁽⁰⁾ (k)≥0, k=1,2,...,n;

(3.2) Carry out three-point smoothing processing to above-mentioned 1-AGO sequence;

For k=2,3,...,n-1 nodes:

The smoothing formula is

For both endpoints k=1 and k=n:

The smoothing formula is

(3.3) Predict according to the result of graying processing by neural network;

(3.4) Perform a cumulative transformation on the output sequence X ⁽³⁾ = (x ⁽³⁾ (1), x ⁽³⁾ (2), ..., x ⁽³⁾ (n)) of the neural network,

Obtain the third predicted value X ⁽⁴⁾ = (x ⁽⁴⁾ (1), x ⁽⁴⁾ (2), ..., x ⁽⁴⁾ (n)), where x ⁽³⁾ (k) ≥ 0, k =1,2,...,n;

x ⁽³⁾ (k)=x ⁽²⁾ (k)-x ⁽²⁾ (k-1).

In this embodiment, the NSDD early warning unit can be used to perform early warning prediction according to the predicted values of the above units; the early warning unit is set with four early warning levels of A, B, C, and D.

Among them, when the NSDD prediction value output by two of the three prediction units in the system reaches 95% of the NSDD value ρ _F of the insulator when pollution flashover may occur, that is, 95% ρ _F , the system issues an A-level early warning; when When the NSDD prediction value of two of the three prediction models in this system reaches 90% of the NSDD value ρ _F of the insulator when pollution flashover may occur, that is, 90% ρ _F , the system issues a B-level early warning; when the three of the system When the NSDD prediction value of two of the prediction models reaches 85% of the NSDD value ρ _F of the insulator when pollution flashover may occur, that is, 85% ρ _F , the system issues a C-level early warning; when the three prediction models of the system When the NSDD prediction value of the two prediction models reaches 80% of the NSDD value ρ _F of the insulator when pollution flashover may occur, that is, 80% ρ _F , the system issues a D-level warning.

Through the NSDD early warning unit, the predicted NSDD value of the insulator is compared with the NSDD value of the insulator when pollution flashover occurs to generate early warning information for the operator to process, which can play a role in timely and effective prevention of pollution flashover accidents on the transmission line.

Those skilled in the art can easily understand that the above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, All should be included within the protection scope of the present invention.

Claims (8)

1. A prediction system for insoluble deposit density on the surface of an insulator, characterized in that it comprises an original data acquisition unit, a series gray neural network prediction unit, a parallel gray neural network prediction unit, an embedded gray neural network prediction unit and NSDD prediction value output unit; The original data acquisition unit is used to obtain the insulator NSDD data and meteorological data on the transmission line; the series gray neural network prediction unit, the parallel gray neural network prediction unit and the embedded gray neural network prediction unit are used for respectively according to the power transmission line Online insulator NSDD data and meteorological data are used to predict the insulator NSDD on the transmission line, and three predicted values are obtained; The NSDD predictive value output unit is used to combine the three predictive values with the sample data, select the predictive unit with the highest predictive accuracy from the three predictive units according to the comparison results, and use the predictive value output by the predictive unit As an insulator NSDD prediction value. 2. The insoluble deposit density prediction system on the surface of an insulator according to claim 1, further comprising an NSDD early warning unit; The NSDD early warning unit is used to generate an early warning signal according to the three predicted values and a preset early warning threshold; specifically, when two or three of the three predicted values reach the early warning threshold, an early warning signal is generated . 3. the insoluble deposit density prediction system on the surface of insulators as claimed in claim 1 or 2, is characterized in that, described series type gray neural network prediction unit comprises the first GM (1,1) model that is juxtaposed, the second GM (1,1) model, the third GM (1,1) model, and neural network; The input interfaces of the first GM (1,1) model, the second GM (1,1) model, and the third GM (1,1) model are all connected to the original data acquisition unit; the input of the neural network The end is connected to the output interface of the first GM (1,1) model, the second GM (1,1) model, and the third GM (1,1) model; The first GM(1,1) model, the second GM(1,1) model, and the third GM(1,1) model respectively predict the insulator NSDD according to the insulator NSDD data on the transmission line, and obtain three groups of graying Prediction result: the neural network performs NSDD prediction of the insulator according to the three groups of ashing prediction results, and obtains the first prediction value. 4. the insoluble deposit density prediction system on the surface of insulators as claimed in claim 1 or 2, is characterized in that, described parallel type gray neural network prediction unit comprises GM (1,1) model, neural network and combined prediction network; The GM (1,1) model is connected in parallel with the neural network; the input ends of the GM (1,1) model and the neural network are connected with the original data acquisition unit; the input end of the combined prediction network is connected with the GM (1, 1) The output end of the model is connected to the output end of the neural network; The GM (1,1) model and the neural network are used to carry out insulator NSDD prediction according to the insulator NSDD data and meteorological data on the transmission line respectively; the combined prediction network is used for the GM (1,1) model and the neural network The prediction results are weighted to obtain the second prediction value. 5. The insoluble deposit density prediction system on the surface of an insulator as claimed in claim 1 or 2, wherein said embedded gray neural network prediction unit comprises a graying layer, a neural network and a whitening layer connected in succession; The ashing layer is used for accumulative transformation and smoothing of the insulator NSDD data and meteorological data on the transmission line; The output data of the network is processed by cumulative subtraction transformation and restoration to obtain the third predicted value. 6. A method for predicting the density of insoluble deposits on the surface of insulators based on the prediction system for the density of insoluble deposits on the surface of insulators according to any one of claims 1 to 5, characterized in that it comprises the following steps: (1) Establish three GM(1,1) models with different sequence lengths based on the collected original insulator NSDD data; use the predicted NSDD values output by these three GM(1,1) models as input to measure NSDD The data is used as the output to carry out neural network training; use the trained neural network to predict the NSDD of the insulator, and obtain the first predicted value; Among them, measuring NSDD refers to the collected NSDD data; (2) Carry out insulator NSDD prediction by gray model and neural network model respectively, obtain two initial prediction data; Determine the weight coefficient of described two initial prediction data according to test sample; Carry out two initial prediction data according to described weight coefficient Weighting processing to obtain a second predicted value; (3) Train the neural network, and perform graying processing on the measured NSDD data; use the grayed data as input data, and use the trained neural network to predict the NSDD of the insulator; perform whitening processing on the output data of the neural network , to obtain a third predicted value; the graying process includes cumulative change and smoothing; the whitening process refers to cumulative transformation; (4) Combine the first predicted value, the second predicted value and the third predicted value with the test sample, and select the one with the highest prediction accuracy from the three predicted values as the predicted value of the insulator NSDD according to the comparison results. 7. The insoluble deposit density prediction system on the surface of an insulator as claimed in claim 6, wherein said step (3) comprises the following sub-steps: (3.1) Use the meteorological data, the original insulator NSDD data before time node m, and the grayed data of 10 time nodes before time node m to train the neural network to obtain the optimal weight and threshold; according to the optimal weight and the threshold to build a trained neural network; Among them, m is the time node for training and predicting the NSDD data value of the insulator. The input data used in the training neural network includes meteorological data, the original insulator NSDD data from time node 1 to time node (m-1), (m-10) to m The ashing data of the original insulator NSDD data at 10 time nodes after ashing processing; the meteorological data includes wind speed, precipitation, and relative humidity; the output data used for training the neural network refers to (m-9)～(m+1) The ashing data of the original insulator NSDD data at 10 time nodes after ashing processing; (3.2) Perform graying processing on the original insulator NSDD data of the mth node; put the grayed data into the trained neural network to obtain the initial prediction value; perform whitening processing on the initial prediction value to obtain the second prediction value. 8. The insoluble deposit density prediction system on the surface of an insulator according to claim 6, further comprising step (5): (5) Comparing the first predicted value, the second predicted value, and the third predicted value with preset warning thresholds respectively, when two of the first predicted value, the second predicted value, and the third predicted value or Three reach the warning threshold, generating a warning signal.