Isolation switch temperature prediction method and device and storage medium

Abstract

The invention discloses an isolation switch temperature prediction method and device and a storage medium, and the method is characterized in that the method comprises the following steps: obtaining a historical actual temperature value sequence of an isolation switch, employing an ARIMA model to carry out the prediction to obtain a predicted temperature value sequence of the last n historical moments, and obtaining a predicted temperature value sequence of the last n historical moments; calculating to obtain a residual error between the historical actual temperature value sequence and the predicted temperature value sequence at the last n historical moments; performing VMD decomposition on the residual error to obtain a multi-layer optimal intrinsic mode function component for training a corresponding BP model, determining an optimal initial weight threshold value between layers in the BP model through a sparrow search algorithm SSR when each BP model is trained, and accumulating predicted values obtained by prediction of all BP models to obtain a residual error predicted value; correcting a future temperature prediction value obtained by the ARIMA model through the residual prediction value; the equipment and the storage medium are used for implementing the method. The method can be applied to temperature prediction of the isolating switch.

Description

Technical Field

[0001] The present invention relates to the field of isolating switch temperature prediction methods, in particular to an isolating switch temperature prediction method, device and storage medium. Background Art

[0002] Electricity is a vital driving force for national production and life. With the rapid development of the national economy, society's demand for electricity is increasing, and the national power grid system is becoming increasingly complex. Due to the uniqueness and importance of the power system, grid security has always been an issue that cannot be ignored. Disconnectors are critical electrical equipment in the power system, and their reliability affects the normal operation of the entire power grid.

[0003] Electrical equipment is a crucial component in ensuring the proper functioning of power systems, directly impacting the normal operation of power transmission and transformation systems, the distribution and consumption systems, and even the entire power system. Therefore, real-time temperature prediction of substation primary equipment is essential. Most faults in disconnectors manifest as heat, making temperature an effective indicator of disconnector health. Research on temperature early warning for power equipment has been a hot topic in recent years, but disconnector temperature prediction suffers from low accuracy. Summary of the Invention

[0004] The present invention provides a method, a device and a storage medium for predicting the temperature of an isolating switch, so as to solve the problem of low accuracy in the prior art of predicting the temperature of an isolating switch.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for predicting the temperature of an isolating switch, the process is as follows:

[0007] Obtaining a sequence of historical actual temperature values ​​of the isolating switch;

[0008] The ARIMA model is used to predict the temperature value sequence of the disconnector at the last n historical moments based on the historical actual temperature value sequence of the disconnector before the last n historical moments. Then, the historical actual temperature value sequence X1 of the last n historical moments in the historical actual temperature value sequence of the isolating switch is combined to calculate the historical actual temperature value sequence X1 of the last n historical moments and the predicted temperature value sequence The residual E1;

[0009] Perform VMD decomposition on the residual E1 to obtain multi-layer optimal intrinsic mode function components;

[0010] A BP model is established for each layer's optimal intrinsic mode function component. The corresponding BP model is trained using the optimal intrinsic mode function component of each layer. When training each BP model, the optimal initial weight threshold between layers in the BP model is determined using the sparrow search algorithm (SSA). The predicted value of the optimal intrinsic mode function component of the corresponding layer is predicted by each trained BP model. The predicted values ​​obtained by all BP models are then accumulated to obtain the residual prediction value.

[0011] When the future temperature prediction value of the disconnector is obtained by using the ARIMA model, the future temperature prediction value is corrected by using the residual prediction value to obtain the corrected future temperature prediction value of the disconnector.

[0012] Furthermore, the ARIMA model first performs a stationarity check on the historical actual temperature value sequence of the disconnector before the last n historical moments;

[0013] When the historical actual temperature value sequence before the last n historical moments passes the stationary check, the autoregressive model AR order p and the moving average model MA order q of the ARIMA model are determined based on the historical actual temperature value sequence before the last n historical moments, and the predicted temperature value sequence of the disconnector at the last n historical moments is predicted based on the determined autoregressive model AR order p and the moving average model MA order q.

[0014] Furthermore, the autocorrelation function ACF is used to obtain the autocorrelation characteristic parameter characteristics of the historical actual temperature value sequence before the last n historical moments, which is used as the autoregressive model AR order p;

[0015] The partial autocorrelation function PACF is used to obtain the partial autocorrelation characteristic parameter characteristics of the historical actual temperature value series before the last n historical moments, which is used as the MA order q of the moving average model.

[0016] Furthermore, the ADF unit root test method is used to check the stationarity of the historical actual temperature value series before the last n historical moments.

[0017] Furthermore, when the historical actual temperature value sequence before the last n historical moments fails the stationary check, the difference order d of the ARIMA model is determined based on the historical actual temperature value sequence before the last n historical moments, and the historical actual temperature value sequence before the last n historical moments is differenced based on the determined difference order d;

[0018] Then, based on the historical actual temperature value sequence before the last n historical moments after the difference processing, the autoregressive model AR order p and the moving average model MA order q of the ARIMA model are determined, and according to the determined autoregressive model AR order p and the moving average model MA order q, the predicted temperature value sequence of the disconnector at the last n historical moments is predicted.

[0019] Furthermore, for the series that fail the stationarity check, the difference processing function is used to obtain the number of required differences, which is used as the difference order d of the ARIMA model.

[0020] Furthermore, when performing VMD decomposition on the residual E1, firstly, multiple layers of intrinsic mode function components are decomposed, and then a VMD constrained variational problem is constructed with the goal of minimizing the sum of the bandwidths of the central frequencies of the intrinsic mode function components of each layer.

[0021] Then the VMD constrained variational problem is transformed into an unconstrained problem and solved to obtain the iterative formula of the intrinsic mode function components of each layer;

[0022] The iterative formula of the eigenmode function components of each layer is iteratively solved to obtain the optimal eigenmode function components of each layer.

[0023] An electronic device includes a processor and a memory. When program instructions in the memory are read and run by the processor, the above-mentioned isolation switch temperature prediction method is executed.

[0024] A storage medium stores program instructions, and when the program instructions are read and run, the above-mentioned isolation switch temperature prediction method is executed.

[0025] In the present invention, a method for predicting the temperature of an isolating switch based on a residual correction model composed of ARIMA-VMD-SSA-BP is proposed. The ARIMA model is used to mine the linear change law in the historical actual temperature value sequence of the isolating switch, and the temperature value at the future moment is predicted. The VMD algorithm is used to decompose the residual into different IMF components to reduce the impact of different trend information on the prediction accuracy. The BP model is used to mine the characteristics of the nonlinear change law of the IMF component, and the BP model is used to predict the residual of the future temperature value, and the predicted value of the ARIMA model is corrected, thereby improving the overall prediction accuracy. In view of the problem that the BP model is prone to fall into local optimality during training, the present invention uses the SSA algorithm to optimize the initial weight threshold of the BP model to improve the defects of the BP model itself.

[0026] Finally, the prediction performance of the model was verified through actual examples. The verification results show that the disconnector temperature prediction method based on the ARIMA-VMD-SSA-BP residual correction model proposed in this invention has higher prediction accuracy and stability, and can be better applied in disconnector temperature prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Flowchart of an embodiment of the present invention.

[0028] Figure 2 This is the structure diagram of the BP neural network.

[0029] Figure 3 Flowchart of optimizing BP neural network for SSA algorithm.

[0030] Figure 4 This is the temperature data diagram of the isolation switch.

[0031] Figure 5 This is the ARIMA fitting result graph.

[0032] Figure 6 Time domain plot for VMD decomposition.

[0033] Figure 7 Frequency domain plot for VMD decomposition.

[0034] Figure 8 This is the curve of the sparrow population's fitness change.

[0035] Figure 9 Comparison chart of prediction results of different algorithms, where (a) is BP prediction result, (b) is ARIMA prediction result, (c) is ARIMA-BP prediction result, (d) is ARIMA-SSA-BP prediction result, and (e) ARIMA-VMD-SSA-BP. DETAILED DESCRIPTION

[0036] The present invention will be further described below with reference to the accompanying drawings and examples.

[0037] like Figure 1 As shown, this embodiment discloses a method for predicting the temperature of an isolating switch, comprising the following steps:

[0038] Step 1: Obtain the historical actual temperature value sequence of the disconnector, and use the ARIMA model to predict the temperature value sequence of the last n historical moments of the disconnector based on the historical actual temperature value sequence before the last n historical moments in the disconnector's historical actual temperature value sequence.

[0039] The ARIMA model used in this example is a common method for dealing with time series problems. It is good at discovering linear patterns within the sequence. Compared to the Autoregressive Moving Average (ARMA), ARIMA adds a difference operator, which can transform the original non-stationary series into a stationary series, thereby improving the prediction accuracy.

[0040] The changing trend of the disconnector temperature exhibits a certain degree of inertia, meaning that the sampled value at the current sampling moment is correlated with the sampled value at the previous sampling moment, resulting in a time series with strong autocorrelation. The ARIMA model is a commonly used method for processing time series problems, excelling at exploiting linear patterns within the sequence. Compared to the Autoregressive Moving Average (ARMA), the ARIMA model adds a differencing operator, transforming the original non-stationary series into a stationary one, thereby improving forecast accuracy. The parameters that must be determined in the ARIMA model include the differencing order d, the autoregressive model order p, and the moving average model order q.

[0041] Assume that the historical actual temperature value sequence before the last n historical moments in the historical actual temperature value sequence of the disconnector is X = [x(1)x(2)x(3)…x(N)], and the ARIMA model first performs a stationarity check on the sequence X.

[0042] In this embodiment, the ADF unit root test method is used to perform a stationarity check on the historical actual temperature value sequence X before the last n historical moments. In the ADF unit root test method, the T statistic and the significance level P probability value of the sequence X are calculated and compared with the critical value under the significance level. When the T value is greater than the critical value of the three confidence intervals and the P value is greater than 0.05, it indicates that the sequence X is a non-stationary sequence and fails the stationarity check. When the T value is less than the critical value of the three confidence intervals and the P value is less than 0.05 and close to 0, there is no unit root and the sequence X is a stationary sequence and passes the stationarity check.

[0043] In this embodiment, when the sequence X fails the stationarity check, a difference processing function is used to obtain the number of times the historical actual temperature value sequence X before the last n historical moments needs to be differentiated, which is used as the difference order d of the ARIMA model.

[0044] Then, based on the determined differential order d, the historical actual temperature value sequence X before the last n historical moments is differentially processed so that the sequence X passes the stationarity check after differential processing. The differential processing is shown in formula (1):

[0045]

[0046] In formula (1), x(t) is the historical actual temperature value at time t before the last n historical moments, t = 1, 2, 3, ..., N; Indicates combinatorial number operation; Δ d x(t) represents the stationary series after d differences.

[0047] When the sequence X passes the stationary check, the autocorrelation function ACF is used to obtain the autocorrelation characteristic parameters of the historical actual temperature value sequence before the last n historical moments, which is used as the autoregressive model AR order p; the partial autocorrelation function PACF is used to obtain the partial autocorrelation characteristic parameters of the historical actual temperature value sequence before the last n historical moments, which is used as the moving average model MA order q.

[0048] Finally, according to the determined autoregressive model AR order p and moving average model MA order q, the predicted temperature value sequence of the disconnector at the last n historical moments is predicted. The prediction is shown in formula (2):

[0049]

[0050] In formula (2): i is the autoregressive coefficient at time lag i; μ t is the mean error term at the tth moment in the last n historical moments, indicating that there is an error between the predicted value and the actual value; μ t-i is the mean error term corresponding to the lag i moment, ε is the mean value of the sequence; x(t) is the predicted temperature value sequence of the last n historical moments The predicted temperature value at time t in θ i is the moving average coefficient at time lag i; γ i is the autoregressive coefficient at time lag i.

[0051] If the sequence X passes the stationary check directly without differential processing, the predicted temperature value sequence of the disconnector at the last n historical moments can be obtained directly through formula (2):

[0052] In this embodiment, the predicted temperature value sequence of the disconnector at the last n historical moments is obtained. Finally, the historical actual temperature value sequence X1 of the last n historical moments in the historical actual temperature value sequence of the disconnector is combined to calculate the historical actual temperature value sequence X1 of the last n historical moments and the predicted temperature value sequence The residual E1, that is

[0053] Step 2: Perform VMD decomposition on the residual E1 of step 1 to obtain multi-layer optimal intrinsic mode function components.

[0054] In this embodiment, the VMD algorithm is used to decompose the residual E1 obtained in step 1 into multiple layers of optimal intrinsic mode function (IMF) components. The VMD algorithm process of this embodiment is as follows:

[0055] (A1) First, perform VMD decomposition on the residual E1 to obtain multi-layer intrinsic mode function (IMF) components. The IMF components u decomposed by VMD are k (t) can be expressed as shown in formula (3):

[0056]

[0057] u k (t) = A k (t)cos[φ k (t)] (3.2)

[0058] In formula (3): k = 1, ..., K, K is the number of decomposition levels, A k (t) is the IMF signal amplitude; φ k (t) is the instantaneous phase; t is time.

[0059] (A2) Taking the minimum sum of the bandwidths of the central frequencies of each layer of IMF components as the goal, the VMD constrained variational problem is constructed as shown in formulas (4) and (5):

[0060]

[0061] In formulas (4) and (5): {u k} represents the IMF component signal set of each layer, u k is the k-th layer IMF component; {ω k} represents the central frequency set of each layer’s IMF components, ω k is the center frequency of the IMF component of the kth layer; * is the convolution operation; is the gradient operation; δ(t) is the Dirac function; j represents the imaginary unit; f(t) represents the IMF signal of each layer as a function of time.

[0062] (A3) The augmented Lagrangian function is introduced to transform the VMD constrained variational problem shown in formulas (4) and (5) into an unconstrained problem. The unconstrained problem is shown in formula (6):

[0063]

[0064] In formula (6), α is the penalty factor; L() is the Lagrangian operator; <> is the inner product operation; λ is the Lagrangian multiplier; λ(t) is the Lagrangian function of time t; ω k represents the center frequency of the kth mode; δ(t) represents the pulse function; represents the time-dependent partial derivative.

[0065] (A4) Solve the unconstrained problem shown in formula (6) by the alternating direction multiplier method, and obtain the iterative formula of the IMF components of each layer in the frequency domain, where the k-th layer IMF component u in the frequency domain is k Iterative formula u of (ω) k,n+1 (ω) is shown in formula (7):

[0066]

[0067] In formula (7): ω is the frequency; -ω k,n is the frequency after the nth iteration of k modes; u k,n+1 (ω) is the modal component after the nth iteration of k modes; u i (ω) is the sum of the modal components; f(ω) is the function of the modal components with respect to frequency; λ n (ω) is the Lagrangian function with respect to frequency; ω is the center frequency.

[0068] (A5) Iterate and solve the iterative formula shown in formula (7) until the iterative termination condition is met, thereby obtaining the optimal IMF component of each layer. The iterative termination condition is shown in formula (8):

[0069]

[0070] In formula (8): ε is a threshold set manually, usually 10 -6 ;u k,n is the modal component of the kth mode after the nth iteration; u k,n+1 is the modal component of the kth mode after the n+1th iteration.

[0071] Therefore, in this embodiment, according to steps (A1)-(A5), the VMD algorithm is implemented to decompose the residual E1 into multiple layers of optimal IMF components, and the obtained signal decomposition diagram is as follows: Figures 6 and 7 As shown, from Figure 7 It can be seen that the VMD algorithm can effectively decompose the residual E1 according to the set parameters, and there is no modal aliasing phenomenon.

[0072] Step 3: Establish a BP model for each layer's optimal intrinsic mode function component, and train the corresponding BP model through the optimal intrinsic mode function component of each layer. When training each BP model, the sparrow search algorithm SSR is used to determine the optimal initial weight threshold between layers in the BP model. Each trained BP model predicts the predicted value of the optimal intrinsic mode function component of the corresponding layer, and then the predicted values ​​predicted by all BP models are accumulated to obtain the residual prediction value.

[0073] In this embodiment, a BP model is used to mine and predict the nonlinear variation patterns in the residuals of the disconnector temperature. Most time series contain not only linear components but also nonlinear components, giving the series complex characteristics. This is also true for the disconnector temperature. Although the ARIMA model can effectively mine the linear variation patterns in the series, it is not very good at capturing the nonlinear variation patterns contained within the series. The BP model is essentially a multi-layer feedforward neural network with outstanding self-learning and nonlinear mapping capabilities. Therefore, in this embodiment, the BP model is used to mine and predict the nonlinear variation patterns in the residuals of the disconnector temperature.

[0074] The network structure of BP neural network is as follows Figure 4 As shown, it consists of an input layer, a hidden layer, and an output layer. Each layer consists of multiple neurons. The input and output of each layer are coupled by connection weights, thresholds, and activation functions. Assume that the input, hidden, and output layers have a, l, and b neurons respectively, and there are a total of i (i = 1, 2, 3, ..., a) input quantities x i (1) , in the process of forward propagation, the input x of the jth (j=1,2,3,…,l) hidden layer neuron is j (2) With the output y j (2) It can be expressed as shown in formula (9):

[0075]

[0076] Where: ij (2) is the connection weight between the i-th input neuron and the j-th hidden neuron, b j (2) is the connection threshold of the jth hidden neuron, f (2) (x) is the activation function of the hidden neuron.

[0077] Then the output input x of the jth (j=1,2,3,…,b) output neuron is j (3) y j (3) It can be expressed as shown in formula (10):

[0078]

[0079] Where: ij (3) is the connection weight between hidden neurons and output neurons, b j (3) is the connection threshold of the jth hidden neuron, f(3) (x) is the activation function of the output neuron. The model structure of the BP neural network is as Figure 2 shown.

[0080] In this embodiment, when the optimal intrinsic mode function components of each layer are respectively input into their corresponding BP models for training, after each BP model completes the forward propagation of the optimal intrinsic mode function components of each layer of the input, the error between the output and the actual value is calculated, and the error is reversely transmitted in the direction of the output - hidden - input layer. During the transmission process, the weight thresholds of each neuron of the BP model are updated by the method of gradient descent to make the error value reach the minimum.

[0081] ​​​​​​​​​​​​​​​​​​​​​​​​is the overall worst position at the tth iteration; n is the total number of sparrows, and i>n / 2 means that the i-th sparrow is in a very hungry state (because its energy is extremely low, that is, its fitness value is very poor) and needs to go somewhere else to find food; is the optimal position occupied by the discoverer at the t+1th iteration; A is a single-row d-dimensional matrix with random elements of 1 or -1, A + =A T (AA T ) -1 .

[0088] Considering their own safety and the ability to successfully obtain food, sparrows will select 10% to 20% of individuals from the population for reconnaissance, and their positions are updated as shown in formula (13):

[0089]

[0090] Where: is the overall optimal position at the tth iteration. β is the step size correction coefficient, which obeys the standard normal distribution. i This is the sparrow's fitness at this time, f w and f g Each represents the overall worst fitness and optimal fitness at this time; when f i >f g When , it means that the sparrows are located at the edge of the population and are easy to become prey; when f i＝ f g When , it means that the sparrows in the middle of the flock sense the threat of their predators and should immediately move towards the other sparrows to escape danger. k∈(0,1) is a random number. ε is the smallest constant, Planck's constant, to avoid division by zero errors.

[0091] When the SSA algorithm reaches the maximum number of iterations, the output iteration result is the initial weight threshold of each BP model.

[0092] In this embodiment, the corresponding BP model is trained by combining the optimal intrinsic mode function component of each layer with the SSA algorithm, and finally each trained BP model predicts the predicted value of the optimal intrinsic mode function component of the corresponding layer, and then the predicted values ​​predicted by all BP models are accumulated to obtain the residual prediction value The process of the SSA-BP algorithm is as follows Figure 3 shown.

[0093] Step 4: When the ARIMA model is used to predict the future temperature of the disconnector, When the residual value is predicted Correct the future temperature prediction value to obtain the corrected future temperature prediction value of the disconnector

[0094] This embodiment further discloses an electronic device, including a processor and a memory. When program instructions in the memory are read and executed by the processor, the above-mentioned isolation switch temperature prediction method is executed.

[0095] This embodiment further discloses a storage medium storing program instructions. When the program instructions are read and executed, the above-mentioned isolating switch temperature prediction method is executed.

[0096] The effectiveness of the method in this embodiment is verified by actual examples below:

[0097] The temperature data of the disconnector of a 220kV substation in a certain place is monitored. The time span is 10 days from the 1st to the 10th day. The monitoring data is collected every hour. A total of 240 sets of temperature data are collected, such as Figure 4 As shown. Figure 1 The process shown first uses the ARIMA model to fit the 48 sets of temperature values ​​for the last two days in a single step using the sliding window method. The steps are as follows:

[0098] 1) Construct a historical temperature sequence X0 with a length of 48, and perform an ADF test on the sequence. If it fails the test, perform differential processing according to formula (1) until it passes the test.

[0099] 2) Determine the upper limit of p and q in the ARIMA model, use the ergodic method to determine the order, and determine the optimal order of p and q based on the minimum sum of Akaike information quantifier (AIC) and Bayesian information quantifier (BIC). At the same time, solve the parameter values ​​in formula (2) by the least squares method.

[0100] 3) Obtain the ARIMA model fitting value through formula (2)

[0101] 4) The window slides 1 unit to the right and updates the data in X0.

[0102] Repeat steps 1) to 4) until the fitting is completed. The single-step fitting values ​​of the ARIMA model for each set of historical temperature series and the corresponding actual values ​​are shown in Table 1. The fitting results of the ARIMA model are shown in Figure 5 shown.

[0103] Table 1 Historical temperature, single-step fitting value and actual value

[0104]

[0105] The nonlinear residual E1 = [e(1)e(2)e(3)…e(48)] is extracted and decomposed using VMD. The IMFs obtained after decomposition are Figure 6 The corresponding modal spectra are shown in Figure 7As shown. Each IMF subsequence is normalized and input into SSA-BP as a training set for training prediction. SSA-BP training also uses the sliding window method, and the window length is set to 3. The matrix composed of the input vector is:

[0106]

[0107] The corresponding output matrix is:

[0108] T trian =[e(4)e(5)e(6)…e(48)] (15)

[0109] The number of neurons in the hidden layer of the network l is calculated according to the following empirical formula:

[0110]

[0111] According to the structures of Equations (14) and (15), the values ​​of a and b are 3 and 1, respectively. The number of neurons in the hidden layer is determined by trial and error. After multiple simulations, it is found that the optimal effect is achieved when the value of l is 8. Finally, a 3-8-1 BP topology is determined, with a total of 41 weight thresholds.

[0112] In the SSA algorithm, 20 randomly generated sets of BP initial weight threshold vectors are reconstructed to form a 20×41 sparrow population location information matrix. The number of iterations is set to 30, and the proportions of discoverers and trackers are 80% and 20% respectively. Scouts are randomly generated in the population in each iteration, and the fitness function is the sum of squares of BP training errors. During the iteration process, the fitness of the SSA algorithm population changes as follows: Figure 8 As shown. Figure 8 It can be seen that starting from the 10th generation, the fitness of the population has basically not changed, and the optimal weight threshold has been obtained at this time.

[0113] Finally, the data from the 9th to the 10th day is used as the test set. The 48 temperature values ​​and residuals predicted by ARIMA and SSA-BP are summed up, and the final prediction value of the residual correction model is output.

[0114] The control groups selected include single BP model, single ARIMA model, combined ARIMA-BP model, combined ARIMA-SSA-BP model and combined ARIMA-VMD-SSA-BP model, which are compared with the residual correction model from the perspective of image and numerical value. The combination of the combined ARIMA-BP model is a simple weighted method. The prediction results of the five models are as follows: Figure 9As shown in the figure, it can be seen intuitively that the prediction errors of the two single models are large; the prediction accuracy of the weighted combination of the two models using the simple averaging method is improved to a certain extent compared with the prediction accuracy of any single model; and the residual correction model with the VMD decomposition process has a significant improvement in accuracy compared with the previous four models, and is highly consistent with the actual temperature change trend.

[0115] From a numerical perspective, we selected the root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) as model evaluation criteria. The results are shown in Table 2. The closer the RMSE, MAE, and MAPE values ​​are to 0, the higher the model's prediction accuracy. Calculations show that the RMSE of the proposed residual correction model is reduced by 49.66%, 50.67%, 40.78%, and 17.86%, respectively, compared to the other four models; the MAE is reduced by 59.40%, 47.26%, 45.48%, and 13.47%, respectively; and the MAPE is reduced by 83.09%, 81.82%, 77.15%, and 42.41%, respectively, compared to the other four models. This indicates that the prediction bias of the ARIMA model is significantly reduced after VMD-SSA-BP correction.

[0116] Table 2 Evaluation indicators of five models

[0117]

[0118] The preferred embodiments of the present invention are described in detail above with reference to the accompanying drawings. The embodiments described in the present invention are merely descriptions of the preferred embodiments of the present invention and do not limit the concept and scope of the present invention. The various specific technical features described in the above specific embodiments can be combined in any suitable manner unless there is any contradiction. Such combinations should also be regarded as the contents disclosed in this disclosure as long as they do not violate the concept of the present invention. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0119] The present invention is not limited to the specific details of the above-mentioned embodiments. Within the scope of the technical concept of the present invention and without departing from the design concept of the present invention, various modifications and improvements made to the technical solution of the present invention by those skilled in the art should fall within the scope of protection of the present invention. The technical contents for which protection is sought in the present invention have been fully recorded in the claims.

Claims

1. A method for predicting the temperature of an isolating switch, characterized in that: The process is as follows: Obtaining a sequence of historical actual temperature values ​​of the isolating switch; The ARIMA model is used to predict the temperature value sequence of the disconnector at the last n historical moments based on the historical actual temperature value sequence of the disconnector before the last n historical moments. Then, the historical actual temperature value sequence X1 of the last n historical moments in the historical actual temperature value sequence of the isolating switch is combined to calculate the historical actual temperature value sequence X1 of the last n historical moments and the predicted temperature value sequence The residual E1; Perform VMD decomposition on the residual E1 to obtain multi-layer optimal intrinsic mode function components; A BP model is established for each layer's optimal intrinsic mode function component. The corresponding BP model is trained using the optimal intrinsic mode function component of each layer. When training each BP model, the optimal initial weight threshold between layers in the BP model is determined using the sparrow search algorithm (SSA). The predicted value of the optimal intrinsic mode function component of the corresponding layer is predicted by each trained BP model. The predicted values ​​obtained by all BP models are then accumulated to obtain the residual prediction value. When the future temperature prediction value of the disconnector is obtained by using the ARIMA model, the future temperature prediction value is corrected by using the residual prediction value to obtain the corrected future temperature prediction value of the disconnector.

2. The method for predicting the temperature of an isolating switch according to claim 1, wherein: The ARIMA model first performs a stationarity check on the historical actual temperature value sequence before the last n historical moments of the disconnector; When the historical actual temperature value sequence before the last n historical moments passes the stationary check, the autoregressive model AR order p and the moving average model MA order q of the ARIMA model are determined based on the historical actual temperature value sequence before the last n historical moments, and the predicted temperature value sequence of the disconnector at the last n historical moments is predicted based on the determined autoregressive model AR order p and the moving average model MA order q.

3. The method for predicting the temperature of an isolating switch according to claim 2, wherein: The autocorrelation function ACF is used to obtain the autocorrelation characteristic parameter characteristics of the historical actual temperature value sequence before the last n historical moments, which is used as the autoregressive model AR order p; The partial autocorrelation function PACF is used to obtain the partial autocorrelation characteristic parameter characteristics of the historical actual temperature value series before the last n historical moments, which is used as the MA order q of the moving average model.

4. A method for predicting temperature of an isolating switch according to claim 2 or 3, characterized in that: The ADF unit root test method is used to check the stationarity of the historical actual temperature value series before the last n historical moments.

5. The method for predicting the temperature of an isolating switch according to claim 4, characterized in that: When the historical actual temperature value sequence before the last n historical moments fails the stationary test, the difference order d of the ARIMA model is determined based on the historical actual temperature value sequence before the last n historical moments, and the historical actual temperature value sequence before the last n historical moments is differenced based on the determined difference order d; Then, based on the historical actual temperature value sequence before the last n historical moments after the difference processing, the autoregressive model AR order p and the moving average model MA order q of the ARIMA model are determined, and according to the determined autoregressive model AR order p and the moving average model MA order q, the predicted temperature value sequence of the disconnector at the last n historical moments is predicted.

6. The method for predicting the temperature of an isolating switch according to claim 5, characterized in that: For the series that fail the stationary test, the difference processing function is used to obtain the number of required differences, which is used as the difference order d of the ARIMA model.

7. The method for predicting the temperature of an isolating switch according to claim 1, characterized in that: When performing VMD decomposition on the residual E1, first decomposition is performed to obtain multiple layers of intrinsic mode function components, and then a VMD constrained variational problem is constructed with the goal of minimizing the sum of the bandwidths of the central frequencies of the intrinsic mode function components of each layer. Then the VMD constrained variational problem is transformed into an unconstrained problem and solved to obtain the iterative formula of the intrinsic mode function components of each layer; The iterative formula of the eigenmode function components of each layer is iteratively solved to obtain the optimal eigenmode function components of each layer.

8. An electronic device comprising a processor and a memory, characterized in that: When the program instructions in the memory are read and executed by the processor, the isolating switch temperature prediction method according to any one of claims 1 to 7 is executed.

9. A storage medium storing program instructions, characterized in that: When the program instructions are read and executed, the disconnector temperature prediction method according to any one of claims 1 to 7 is executed.

Images