Switchgear safety risk prediction method based on CEEMDAN and BiLSTM model

Abstract

The present invention relates to a switchgear safety risk prediction method based on CEEMDAN and BiLSTM models, comprising the following steps: collecting holographic data of the switchgear; decomposing the time series data of the moving contact temperatures of phases A, B, and C into multiple intrinsic mode components using a fully noise-assisted aggregated empirical mode decomposition method; calculating the correlation between each physical quantity of the switchgear and the intrinsic mode components of the moving contact temperatures of each phase; screening out the main relevant physical quantities of the moving contact temperatures of each phase; predicting the intrinsic mode components of the moving contact temperatures of each phase using a BiLSTM model; synthesizing the intrinsic mode components of the moving contact temperatures to obtain a predicted value of the moving contact temperature, and issuing an early warning of the safety risk of the switchgear based on the predicted value. The present invention achieves real-time detection and alarming of high-voltage switchgear groups, reduces the accident rate of enclosed high-voltage switchgear, improves the safe and reliable power supply capability of the power system, and reduces economic losses.

Description

Technical Field

[0001] The present invention belongs to the field of power transmission and transformation control, and specifically relates to a switch cabinet safety risk prediction method based on CEEMDAN and BiLSTM models. Background Art

[0002] Substations with voltage levels of 220kV and above are often equipped with numerous 10kV high-voltage switchgear, forming clusters of 10kV switchgear. High-voltage switchgear is constantly exposed to high voltage, high current, and strong magnetic fields. Furthermore, high-voltage switchgear is typically enclosed, making it difficult for workers to directly observe the specific conditions within the switchgear, such as the moving contacts and the three-phase busbar inlets, which are prone to heat. High-voltage switchgear is prone to internal overheating during extended operation. This is primarily due to the elevated temperatures at the moving contacts caused by prolonged mechanical movement and contact erosion. This leads to oxidation, increased contact resistance, and further temperature increases, resulting in localized welding, sparks, and even arcing, which reduces insulation and ultimately damages electrical equipment. 10kV high-voltage switchgear, essential for the stable operation of substations, is numerous and prone to heat generation. The long-term safe, stable, and reliable operation of 10kV switchgear directly impacts the safety of the substation and the power system it houses.

[0003] Therefore, the study is conducted to monitor and alarm holographic data such as temperature, current, and pressure in heat-prone parts of enclosed high-voltage switchgear. Summary of the Invention

[0004] The purpose of the present invention is to address the above-mentioned problems and provide a switchgear safety risk prediction method based on CEEMDAN and BiLSTM models. The method uses Internet of Things devices to collect physical quantities such as temperature, current and pressure in the high-voltage switchgear to obtain holographic data of the high-voltage switchgear; uses the empirical mode decomposition method to decompose the time series data of the switchgear moving contact temperature into multiple eigenmode components with different frequency domain characteristics; calculates the correlation between each physical quantity of the switchgear and the decomposed eigenmode components, and screens out the main relevant characteristic quantities affecting the moving contact temperature; when safety risks such as overheating occur, different physical quantities change at different speeds, that is, different physical quantities have different derivatives over time. The forward and backward propagation learning capabilities of the BiLSTM model are used to capture the laws of physical quantity changes, identify the changing trends of the moving contact temperature in advance, discover the safety risks of the switchgear operation as early as possible, and issue an alarm so that the operation and maintenance personnel can take timely measures to ensure power safety and reduce losses.

[0005] The technical solution of the present invention is a switch cabinet safety risk prediction method based on CEEMDAN and BiLSTM models, comprising the following steps:

[0006] Step 1: Collect the time series data of temperature, current and pressure of the switch cabinet to obtain the holographic data of the switch cabinet;

[0007] The current of the A, B, and C phase incoming lines of the switch cabinet and the temperature of the moving contacts of phases A, B, and C are collected separately; the temperature of the static contacts of phases A, B, and C are collected separately; the pressure at the contacts of phases A, B, and C in the contact box is monitored using a pressure sensor; and the temperature on the top and back of the switch cabinet is monitored using a temperature sensor.

[0008] Step 2: Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) is used to decompose the time series data of the moving contact temperatures of phases A, B, and C into multiple eigenmode components.

[0009] Step 3: Using the intrinsic mode components of the temperature of each phase moving contact of the switchgear as target variables, perform Spearman correlation analysis on the holographic data of the switchgear to calculate the correlation between each physical quantity of the switchgear and the intrinsic mode components of the temperature of each phase moving contact;

[0010] Step 4: According to the correlation with the moving contact temperature, select the main physical quantities related to the moving contact temperature of each phase;

[0011] Step 5: Establish a BiLSTM model. Use the main physical quantities related to the moving contact temperature obtained in Step 4 as the input of the BiLSTM model. Use the BiLSTM model to predict the intrinsic mode components of the moving contact temperature of each phase. Synthesize the intrinsic mode components of the moving contact temperature obtained by the BiLSTM model to obtain the predicted value of the moving contact temperature.

[0012] Step 6: Issue an early warning of the safety risk of the switch cabinet based on the predicted value of the moving contact temperature of the switch cabinet.

[0013] Preferably, wireless temperature sensors are installed at the static contacts and moving contacts of the A, B, and C three-phase inlets in the high-voltage switchgear, and at the top and back of the high-voltage switchgear respectively; wireless current sensors are installed at the moving contacts of the A, B, and C three-phase inlets in the high-voltage switchgear respectively; and wireless pressure sensors are installed at the contacts in the contact box. The above-mentioned wireless sensors transmit the holographic data of the switchgear, such as the detected temperature, current, and pressure of the switchgear, to the central server via wireless communication. The database of the central server includes a MySQL relational database and a Redis system for storing the holographic data of the switchgear. The MySQL database is used to store and query large amounts of data with typical relational characteristics, such as temperature, pressure, and current. The Redis system is a non-relational database that implements storage according to keyword-value and supports the storage and synchronization of data types such as strings, linked lists, and sets.

[0014] In step 2, the time series data of the moving contact temperatures of the A, B, and C phase incoming ports collected within the time interval are decomposed using the complete noise-assisted aggregated empirical mode decomposition method to obtain intrinsic mode components with different frequency domain characteristics. The specific process includes:

[0015] S1: Add the positive and negative white noise pairs to the original data sequence x(t), and use the empirical mode decomposition method to decompose it to obtain the first modal component IMF1:

[0016]

[0017] The residual component is:

[0018]

[0019] Where N represents the sequence length; r1(t) represents the first residual component;

[0020] S2: Add positive and negative paired Gaussian white noise to the residual component r1(t) to obtain a new sequence, and perform n times of empirical mode decomposition to obtain the second modal component IMF2:

[0021]

[0022] Where E k () represents the kth IMF component obtained through empirical mode decomposition, η j (t) is the jth Gaussian white noise sequence; e1 represents the noise coefficient;

[0023] The residual component is:

[0024]

[0025] Where r2(t) represents the second residual component;

[0026] S3: Similar to step S2, decompose to obtain the kth residual component:

[0027]

[0028] Where r k-1 (t), r k (t) represents the k-1th and kth residual components respectively;

[0029] Repeatedly use empirical mode decomposition to decompose the sequence to be processed and derive the k+1th modal component IMF k+1 :

[0030]

[0031] S4: Repeat step S3 until the residual component obtained is a monotonic function that is not suitable for further decomposition, and then end the decomposition process;

[0032] Assuming the number of modal components obtained is L, the original sequence can be expressed as:

[0033]

[0034] Where r(t) is the final residual component.

[0035] In step 5, the specific process of constructing and using the BiLSTM model to obtain the predicted value of the moving contact temperature includes:

[0036] 1) The intrinsic mode components IMF1-IMF6 and the residual component r of the time series of the moving contact temperatures of phases A, B, and C obtained in step 2 are used as the prediction variables of the BiLSTM model, and the main physical quantities related to the moving contact temperature selected in step 4 are used as the input of the BiLSTM model to construct the BiLSTM model;

[0037] Predictors for the BiLSTM model:

[0038] f(x)=ω T x+b (8)

[0039] Where ω represents the weight matrix, b represents the bias term, and x represents the input;

[0040] The collected holographic data of the switchgear is used as the training data set for the BiLSTM model.

[0041] 2) Set the error calculation interval (e - ,e + ), where e - , ε +are the lower and upper limits of the error value respectively. When the error value is greater than the error calculation interval, the error loss value is calculated;

[0042] 3) In the first LSTM layer of the BiLSTM model, errors are forward propagated to learn from historical data; in the second LSTM layer, errors are backward propagated to learn from future data. The BiLSTM model is trained in both forward and reverse directions using the switchgear holographic data, enabling the BiLSTM model to fully and effectively learn the load information in the switchgear holographic data.

[0043] 4) According to the prediction variables of the trained BiLSTM model, the intrinsic mode components IMF1-IMF6 and the residual component r of the time series of the moving contact temperatures of phases A, B, and C are obtained, and the predicted values ​​of the moving contact temperatures of phases A, B, and C are obtained by synthesis; the prediction accuracy of the BiLSTM model is evaluated using the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE).

[0044] The first and second LSTM layers of the BiLSTM model use the following gating structure to control the transmission of information to cell units through information selection method:

[0045] Forget Gate:

[0046] f t =σ(W f ·[h t-1 ,x t ]+b f ) (9)

[0047] Where f t Represents the output of the forget gate, W f represents the weight matrix of the forget gate, σ() represents the activation function; b f represents the bias term of the forget gate; h t-1 represents the hidden variable at time t-1; x t represents the input at time t;

[0048] The forget gate reads h t-1 and x t , outputs a value from 0 to 1, which determines whether the information is transmitted. The output value equals 0, which means that the information is completely discarded, and the output value equals 1, which means that the information is completely retained and transmitted;

[0049] Input Gate:

[0050] i t =σ(W i ·[h t-1 ,x t ]+b i ) (10)

[0051]

[0052] Where i t represents the output of the input gate, W i represents the weight matrix of the input gate, b i represents the bias term of the input gate, C t 、 They represent the characteristic variables of the input gated structure at time t; * represents the Hadamard product;

[0053] Output Gate:

[0054] o t =σ(W o ×[h t-1 ,x t ]+b o ) (12)

[0055] h t =o t *tanh(C t ) (13)

[0056] Where o t Represents the output of the output gate, W o represents the weight matrix of the output gate, b o Represents the bias term of the output gate, h t represents the hidden variable at time t.

[0057] In step 6, set the temperature threshold T of the moving contact threshold ,When the predicted values ​​of the moving contact temperatures of phases A, B, and C obtained by the BiLSTM model exceed the set temperature threshold, an alarm is issued.

[0058] Preferably, the temperature threshold is dynamically adjusted according to different operating scenarios of the switchgear to quickly detect safety risks and issue alarms, including:

[0059] When the ratio of the real-time collected single-phase current value to the rated current value is greater than or equal to 1.1, let T threshold =0.9T base , where T base Indicates the reference value of the moving contact temperature threshold;

[0060] When the ratio of the real-time collected contact pressure value to the rated pressure value is greater than or equal to 1.2, let T threshold =0.8T base ;

[0061] When the ratio of the real-time collected temperature value on the top or back of the switch cabinet to the standard temperature value is greater than or equal to 1.1, let T threshold =0.75Tbase .

[0062] The central server compares the predicted value of the moving contact temperature of phases A, B, and C with T in real time. threshold The size of the predicted value exceeds T threshold When an alarm occurs, an action signal is sent to the LED alarm light and buzzer of the alarm unit.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1) The present invention realizes real-time detection and alarm of high-voltage switchgear groups, reduces the accident rate of enclosed high-voltage switchgear, improves the ability of the power system to provide safe and reliable power supply, reduces economic losses, and increases social benefits;

[0065] 2) The present invention uses a fully noise-assisted aggregated empirical mode decomposition method to decompose the time series data of the switchgear moving contact temperature into multiple eigenmode components with different frequency domain characteristics. A neural network model is then used to establish a nonlinear relationship between the eigenmode components of the moving contact temperature and various physical quantities of the switchgear, thereby improving the prediction accuracy of the switchgear moving contact temperature.

[0066] 3) This invention leverages the forward and backward learning capabilities of BiLSTM to learn the phenomenon of different speeds of change in the physical quantities of the switchgear when it overheats. This method captures the patterns of change in these physical quantities and proactively detects impending safety risks such as switchgear overheating. This allows for early detection of safety risks in switchgear operation and issues an alarm, enabling operators to take timely measures to ensure power safety.

[0067] 4) The present invention uses the Spearman coefficient to analyze the correlation between the physical quantities of the switchgear and the intrinsic modal components of the moving contact temperature of each phase. Compared with the Pearson coefficient and the Kendall coefficient, the Spearman coefficient is less susceptible to abnormal data interference and has lower requirements for data distribution. It is particularly suitable for correlation calculation of 10kV switchgear data.

[0068] 5) The present invention uses dynamically adjusted temperature thresholds to help quickly detect risks and issue alarms;

[0069] 6) The MySQL database used in the present invention realizes the storage and query of data with typical relational characteristics in the switch cabinet holographic data, and has the advantages of small size, low cost, fast computing speed, and safety and reliability; the Redis system realizes the efficient storage and query of non-relational data in the switch cabinet holographic data. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The present invention will be further described below with reference to the accompanying drawings and examples.

[0071] Figure 1 Schematic diagram of a switch cabinet monitoring system according to an embodiment of the present invention.

[0072] Figure 2 Schematic diagram of the distribution of stepless sensors for collecting holographic data of switch cabinets according to an embodiment of the present invention.

[0073] Figure 3 Schematic diagram of the structure of the central server of an embodiment of the present invention.

[0074] Figure 4 Schematic diagram of the eigenmode components obtained by CEEMDAN decomposition in an embodiment of the present invention.

[0075] Figure 5 Schematic diagram of a BiLSTM model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0076] The embodiment relates to safety monitoring of a 10 kV switchgear group in a substation with a voltage level of 220 kV or above.

[0077] like Figure 1-3 As shown, the switch cabinet monitoring system of the embodiment includes a wireless measuring unit and a central server. The wireless measuring unit monitors the real-time temperature value, real-time current value and real-time pressure value of the high-voltage switch cabinet. The wireless measuring unit includes a wireless temperature sensor, a wireless current sensor and a wireless pressure sensor.

[0078] The data transmission network unit receives real-time data measured by wireless sensors through the ZigBee wireless network, and then transmits the real-time data to the central server for data preprocessing and prediction. When the high-voltage switchgear is about to overheat during operation, the switchgear monitoring system will issue an alarm.

[0079] The central server unit uses a fully noise-assisted aggregated empirical mode decomposition method to decompose the moving contact time data series data into intrinsic mode components with different frequency scales, and then uses the Spearman coefficient to select appropriate historical features to establish a bidirectional long-short-term memory neural network prediction model. Combined with the high-voltage switchgear operating procedures and other physical quantities obtained by measurements related to the moving contact temperature, the risk threshold of the high-voltage switchgear in different scenarios is determined; the predicted data is compared with the alarm threshold, and if the warning value is exceeded, the alarm signal is transmitted to the LED alarm light and buzzer of the alarm module for corresponding action.

[0080] The wireless sensor group arranged on the switch cabinet of the embodiment is as shown in the attached Figure 2Eight wireless temperature sensors are installed on the moving contacts of the A, B, and C three-phase incoming lines within the high-voltage switchgear, at monitoring points on the switchgear's exterior surface, and at the stationary contacts within the switchgear. Three wireless current sensors are installed on the moving contacts of the A, B, and C three-phase incoming lines within the high-voltage switchgear. Three wireless pressure sensors are installed on the contacts within the contact box. A ZigBee wireless network module receives data from each wireless sensor; the module uses the MC13213 ZigBee platform SoC chip. Data transmitted by the wireless sensors is received by a ZigBee to IP / TCP converter, which is then transmitted to a central server via Ethernet. Staff can quickly access server data via a computer or mobile phone, enabling remote monitoring of the high-voltage switchgear's operating status. Real-time data from the wireless sensor nodes is transmitted to the ZigBee wireless network module via the RS485 bus interface. The data is then transmitted to the central server via a coordinator and an RS485 interface converter.

[0081] Central server such as Figure 3 As shown in the figure, since the real-time data collected by wireless sensors has non-stationary and nonlinear characteristics, the central server stores, preprocesses and performs nonlinear regression prediction on the received real-time data. The moving contact temperatures of the A, B, and C phase incoming ports collected within the time interval are decomposed using the completely noise-assisted aggregated empirical mode decomposition (CEEMDAN) to decompose the data into intrinsic mode components (IMFs) with different frequency domain characteristics.

[0082] The switchgear safety risk prediction method based on CEEMDAN and BiLSTM model includes the following steps:

[0083] Step 1: Collect the time series data of temperature, current and pressure of the switch cabinet to obtain the holographic data of the switch cabinet;

[0084] The current of the A, B, and C phase incoming lines of the switch cabinet and the temperature of the moving contacts of phases A, B, and C are collected separately; the temperature of the static contacts of phases A, B, and C are collected separately; the pressure at the contacts of phases A, B, and C in the contact box is monitored using a pressure sensor; and the temperature on the top and back of the switch cabinet is monitored using a temperature sensor.

[0085] Step 2: Use the fully noise-assisted aggregated empirical mode decomposition method to decompose the time series data of the moving contact temperature of phases A, B, and C into multiple eigenmode components;

[0086] Step 3: Using the intrinsic mode components of the temperature of each phase moving contact of the switchgear as target variables, perform Spearman correlation analysis on the holographic data of the switchgear to calculate the correlation between each physical quantity of the switchgear and the intrinsic mode components of the temperature of each phase moving contact;

[0087] Step 4: According to the correlation with the moving contact temperature, select the main physical quantities related to the moving contact temperature of each phase;

[0088] Step 5: Establish a BiLSTM model. Use the main physical quantities related to the moving contact temperature obtained in Step 4 as the input of the BiLSTM model. Use the BiLSTM model to predict the intrinsic mode components of the moving contact temperature of each phase. Synthesize the intrinsic mode components of the moving contact temperature obtained by the BiLSTM model to obtain the predicted value of the moving contact temperature.

[0089] Step 6: Issue an early warning of the safety risk of the switch cabinet based on the predicted value of the moving contact temperature of the switch cabinet.

[0090] The fully noise-assisted aggregated empirical mode decomposition method used in step 2 includes:

[0091] Step 1: Randomly add m pairs of positive and negative white noise to the original data sequence;

[0092] Step 2: Decompose the data sequence with added positive and negative white noise using the empirical mode decomposition algorithm, and perform ensemble averaging on the IMFs obtained after adding different white noise pairs to complete one iteration;

[0093] Step 3: Repeat steps 1 and 2 above, and add white noise to the residual value again during each iteration;

[0094] Step 4: After the iteration conditions are met, the complete CEEMDAN decomposition results are obtained.

[0095] A bidirectional long-short-term memory neural network prediction model was then used to perform regression prediction modeling on the individual IMF components in the CEEMDAN decomposition results. The predicted values ​​of each IMF component were then cumulatively reconstructed to obtain a complete data prediction result. In combination with high-voltage switchgear operating procedures and other relevant physical quantities, risk classification thresholds for high-voltage switchgear were determined under different scenarios. Experimental analysis showed that when the ratio of real-time current to rated current was 1.1 or above, the alarm threshold was lowered by approximately 10%; when the ratio of real-time pressure to rated pressure was 1.2 or above, the alarm threshold was lowered by approximately 20%; and when the ratio of real-time temperature other than the moving contact to rated temperature was 1.1 or above, the alarm threshold was lowered by approximately 25%.

[0096] In this embodiment, after collecting the holographic data of the switchgear, the temperatures of the moving contacts of the A, B, and C phase inlets collected within the time interval are decomposed using a completely noise-assisted aggregated empirical mode decomposition method to decompose the data into eigenmode components with different frequency domain characteristics. The specific process is as follows:

[0097] S1: Add the positive and negative white noise pairs to the original data sequence and use the EMD method to decompose the first modal component IMF1:

[0098]

[0099] The residual component is:

[0100]

[0101] S2: Add positive and negative paired Gaussian white noise to the residual component r1(t) to obtain a new sequence and perform multiple EMD decompositions to obtain IMF2:

[0102]

[0103] Where E k (·) represents the kth IMF component obtained by the EMD algorithm, η j (t) is the jth Gaussian white noise sequence.

[0104] The residual component is:

[0105]

[0106] S3: Repeat the above steps to decompose and obtain the kth residual component:

[0107]

[0108] S4: Repeatedly use EMD to decompose the sequence to be processed and derive the k+1th modal component IMF k+1 :

[0109]

[0110] Repeat the above steps until the residual component is a monotonic function that is not suitable for further decomposition. The decomposition process ends.

[0111] S5: Assuming that L modal components are obtained, the original sequence can be expressed as:

[0112]

[0113] Where r(t) is the final residual component.

[0114] The decomposed eigenmode components are as follows: Figure 4 As shown in Figure 2, Original represents the original measured temperature series, IMF1-IMF6 are the main modal components obtained by decomposition, and r is the residual component.

[0115] For the data sequence of the decomposed intrinsic mode components, the Spearman coefficient is used to select the appropriate historical characteristics of the switchgear physical quantity to construct a bidirectional long short-term memory neural network prediction model to realize the moving contact temperature prediction. The specific process is as follows:

[0116] 1) Using the data samples IMF1-IMF7 obtained by CEEMDAN decomposition, where IMF7 is the residual component, and the Spearman coefficient, appropriate historical features are selected to construct a BiLSTM neural network model and perform learning and training to make f(x) as close as possible to the original sample, that is, the error is closer to zero:

[0117] f(x)=ω T x+b (8)

[0118] where d i is the order difference after variable sorting, ω and b are the parameters of the training model.

[0119] 2) Set the error calculation interval (ε - ,ε + ), when the error value is greater than the error calculation interval, the error loss value is calculated.

[0120] The proposed BiLSTM prediction model is expressed as follows:

[0121] Based on RNN, LSTM improves on problems such as gradient disappearance. It uses unique gating structures such as input gate, forget gate and output gate to control the transmission of information to cell units through information selection method.

[0122] Forget Gate:

[0123] f t =σ(W f ×[h t-1 ,x t ]+b f ) (9)

[0124] The forget gate reads h t-1 and x t Output a value between 0 and 1 to determine whether the information is transmitted, where 0 means the information is completely discarded, and 1 means the information is completely retained and transmitted.

[0125] Input Gate:

[0126] i t =σ(W i ·[h t-1 ,x t ]+b i ) (10)

[0127]

[0128] Output Gate:

[0129] o t =σ(W o ·[h t-1 ,x t ]+b o ) (12)

[0130] h t =o t *tanh(C t ) (13)

[0131] Where, f t 、i t 、 C t 、o t 、h t represents the characteristic variable of each gating structure at time t; W represents the weight matrix of each gating structure; b represents the bias term; * represents the Hadamard product; h t-1 、h t Represents the input at time t-1 and time t.

[0132] 3) BiLSTM adopts a two-layer LSTM structure. The first layer learns historical information, and the second layer transmits information in the reverse direction to consider future data. By training the data in both the positive and negative directions, the model network can fully and effectively learn the information of the load data. Its network structure is as follows: Figure 5 shown.

[0133] 4) The predicted values ​​of each modal component IMF1-IMF7 are accumulated and reconstructed to achieve the final data prediction, and the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are used to evaluate the prediction model.

[0134] The calculation method of the Spearman coefficient of the embodiment refers to the calculation method of the Spearman coefficient disclosed in the paper "Electric Vehicle Charging Load Interval Prediction Based on Multi-Correlated Daily Scenarios Generation" by Huang Nantian et al. published in the 41st issue of "Proceedings of the Chinese Society of Electrical Engineering" in 2021.

[0135] A bidirectional long-short-term memory neural network prediction model provides high-quality data support for effective prediction and alarm generation. Based on the operational requirements of 10kV high-voltage switchgear and the operating procedures for high-voltage equipment in power systems, risk grading thresholds for enclosed high-voltage switchgear in different scenarios are determined. Based on previously acquired physical quantities such as current and pressure, experimental analysis indicates that the alarm threshold is lowered by approximately 10% when the ratio of real-time current to rated current is 1.1 or above; by approximately 20% when the ratio of real-time pressure to rated pressure is 1.2 or above; and by approximately 25% when the ratio of real-time temperature other than the moving contact to rated temperature is 1.1 or above. When relevant data from the high-voltage switchgear exceeds the alarm threshold, the relevant alarm devices are activated: the LED alarm light flashes continuously red, a buzzer sounds, and the display terminal indicates a fault. A central server displays real-time temperature, current, and pressure values ​​at each monitoring point in the high-voltage switchgear, along with fault information, and sends corresponding alarm messages, allowing operations and maintenance personnel to promptly monitor the switchgear's operating status, address any faults, and eliminate potential risks as soon as possible to prevent further losses.

[0136] The above description specifically describes and introduces the preferred implementation cases of the present invention, but does not impose any form of limitation on the present invention. Those skilled in the art should understand that, based on the technical essence of the present invention, within the spirit and principles of the present invention, any alternative solutions such as no creative work is made in the above implementation cases, as well as any simple modifications, changes and improvements, all fall within the scope of protection of the technical solution of the present invention and the claims.

Claims

1. A switchgear safety risk prediction method based on CEEMDAN and BiLSTM models is characterized by: The following steps are involved: Step 1: Collect the time series data of temperature, current and pressure of the switch cabinet to obtain the holographic data of the switch cabinet; The current of the A, B, and C phase incoming lines of the switch cabinet and the temperature of the moving contacts of phases A, B, and C are collected separately; the temperature of the static contacts of phases A, B, and C are collected separately; the pressure at the contacts of phases A, B, and C in the contact box is monitored using a pressure sensor; and the temperature on the top and back of the switch cabinet is monitored using a temperature sensor. Step 2: Use the fully noise-assisted aggregated empirical mode decomposition method to decompose the time series data of the moving contact temperature of phases A, B, and C into multiple eigenmode components; Step 3: Using the intrinsic mode components of the temperature of each phase moving contact of the switchgear as target variables, perform Spearman correlation analysis on the holographic data of the switchgear to calculate the correlation between each physical quantity of the switchgear and the intrinsic mode components of the temperature of each phase moving contact; Step 4: According to the correlation with the moving contact temperature, select the main physical quantities related to the moving contact temperature of each phase; Step 5: Establish a BiLSTM model, use the main physical quantities related to the moving contact temperature obtained in step 4 as the input of the BiLSTM model, and use the BiLSTM model to predict the intrinsic mode components of the moving contact temperature of each phase; The intrinsic mode components of the moving contact temperature obtained by the BiLSTM model are synthesized to obtain the predicted value of the moving contact temperature; Step 6: Issue an early warning of the safety risk of the switch cabinet based on the predicted value of the moving contact temperature of the switch cabinet.

2. The switch cabinet safety risk prediction method according to claim 1, characterized in that: In step 2, the time series data of the moving contact temperatures of the A, B, and C phase incoming ports collected within the time interval are decomposed using the complete noise-assisted aggregated empirical mode decomposition method to obtain intrinsic mode components with different frequency domain characteristics. The specific process includes: S1: Add the positive and negative white noise pairs to the original data sequence x(t), and use the empirical mode decomposition method to decompose it to obtain the first modal component IMF1: The residual component is: Where N represents the sequence length; r1(t) represents the first residual component; S2: Add positive and negative paired Gaussian white noise to the residual component r1(t) to obtain a new sequence, and perform n times of empirical mode decomposition to obtain the second modal component IMF2: Where E k (·) represents the kth IMF component obtained through empirical mode decomposition, h j (t) is the jth Gaussian white noise sequence; ε1 represents the noise coefficient; The residual component is: Where r2(t) represents the second residual component; S3: Similar to step S2, decompose to obtain the kth residual component: Where r k-1 (t), r k (t) represents the k-1th and kth residual components respectively; Repeatedly use empirical mode decomposition to decompose the sequence to be processed and derive the k+1th modal component IMF k+1 : S4: Repeat step S3 until the residual component obtained is a monotonic function that is not suitable for further decomposition, and then end the decomposition process; Assuming the number of modal components obtained is L, the original sequence can be expressed as: Where r(t) is the final residual component.

3. The switch cabinet safety risk prediction method according to claim 2, characterized in that: In step 5, the specific process of constructing and using the BiLSTM model to obtain the predicted value of the moving contact temperature includes: 1) The intrinsic mode components IMF1-IMF6 and the residual component r of the time series of the moving contact temperatures of phases A, B, and C obtained in step 2 are used as the prediction variables of the BiLSTM model, and the main physical quantities related to the moving contact temperature selected in step 4 are used as the input of the BiLSTM model to construct the BiLSTM model; Predictors for the BiLSTM model: f(x)=ω T x+b (8) Where ω represents the weight matrix, b represents the bias term, and x represents the input; The collected holographic data of the switchgear is used as the training data set for the BiLSTM model. 2) Set the error calculation interval (ε - ,ε + ), when the error value is greater than the error calculation interval, the error loss value is calculated; 3) In the first LSTM layer of the BiLSTM model, errors are forward propagated to learn from historical data; in the second LSTM layer, errors are backward propagated to learn from future data. The BiLSTM model is trained in both forward and reverse directions using the switchgear holographic data, enabling the BiLSTM model to fully and effectively learn the load information in the switchgear holographic data. 4) According to the prediction variables of the trained BiLSTM model, the intrinsic mode components IMF1-IMF6 and the residual component r of the time series of the moving contact temperatures of phases A, B, and C are obtained, and the predicted values ​​of the moving contact temperatures of phases A, B, and C are obtained by synthesis; the prediction accuracy of the BiLSTM model is evaluated using the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE).

4. The switch cabinet safety risk prediction method according to claim 3, characterized in that: The first and second LSTM layers of the BiLSTM model use the following gating structure to control the transmission of information to cell units through information selection method: Forget Gate: f t =σ(W f ·[h t-1 ,x t ]+b f ) (9) Where f t Represents the output of the forget gate, W f represents the weight matrix of the forget gate, σ(·) represents the activation function; b f represents the bias term of the forget gate; h t-1 represents the hidden variable at time t-1; x t represents the input at time t; The forget gate reads h t-1 and x t , outputs a value from 0 to 1, which determines whether the information is transmitted. The output value equals 0, which means that the information is completely discarded, and the output value equals 1, which means that the information is completely retained and transmitted; Input Gate: i t =σ(W i ·[h t-1 ,x t ]+b i ) (10) Where i t represents the output of the input gate, W i represents the weight matrix of the input gate, b i represents the bias term of the input gate, C t 、 Represent the state at time t and the current input state respectively; * represents the Hadamard product; Output Gate: the t =σ(W o ·[h t-1 ,x t ]+b o ) (12) h t =o t *fishy(C) t ) (13) Where o t Represents the output of the output gate, W o represents the weight matrix of the output gate, b o Represents the bias term of the output gate, h t represents the hidden variable at time t.

5. The switch cabinet safety risk prediction method according to claim 4, characterized in that: In step 6, set the temperature threshold T of the moving contact threshold ,When the predicted values ​​of the moving contact temperatures of phases A, B, and C obtained by the BiLSTM model exceed the set temperature threshold, an alarm is issued; Dynamically adjust temperature thresholds based on different switchgear operation scenarios to quickly identify safety risks and generate alarms, including: When the ratio of the real-time collected single-phase current value to the rated current value is greater than or equal to 1.1, let T threshold =0.9T base , where T base Indicates the reference value of the moving contact temperature threshold; When the ratio of the real-time collected contact pressure value to the rated pressure value is greater than or equal to 1.2, let T threshold =0.8T base ; When the ratio of the real-time collected temperature value on the top or back of the switch cabinet to the standard temperature value is greater than or equal to 1.1, let T threshold =0.75T base .

Images