Low-voltage switch temperature abnormity early warning system and early warning method based on time sequence prediction

Abstract

The invention discloses a low-voltage switch temperature abnormity early warning system and early warning method based on time sequence prediction. The system comprises a data acquisition module, a time sequence modeling module, a temperature prediction module, a scoring module and an early warning module. The data acquisition module is used for acquiring operating environment temperature and current information of the low-voltage switch in real time and constructing a historical input sequence; the time sequence modeling module carries out modeling on a historical sequence based on a long short-term memory neural network and extracts time sequence features; the temperature prediction module maps the hidden state of the model into a predicted temperature value at the next moment by using a decoder; the scoring module calculates a basic score and a current score based on the deviation between the predicted temperature and the actually measured temperature; and the early warning module judges whether abnormal early warning is triggered or not according to the score change trend. Accurate prediction and intelligent early warning of the temperature state of the low-voltage switch can be achieved, and the method has high real-time performance, accuracy and adaptability and is suitable for equipment state monitoring and fault early warning in scenes such as rail transit, industrial power distribution and intelligent buildings.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for power equipment, and in particular to a low-voltage switch temperature anomaly early warning system and method based on time-series prediction. Background Technology

[0002] In low-voltage electrical systems, low-voltage switches are widely used in distribution cabinets, industrial automation equipment, and building electrical systems to control circuit switching and protect the safe operation of electrical systems. However, with increased equipment operating time, changes in load conditions, and the influence of environmental factors, low-voltage switches may experience problems such as poor contact, internal overheating, and arc corrosion, leading to abnormally high operating temperatures. In severe cases, this can even cause electrical fires and other safety accidents. Therefore, real-time monitoring and early warning of temperature changes in low-voltage switches have become crucial for ensuring system safety.

[0003] In existing technologies, some solutions use temperature sensors to periodically collect the temperature of the switch body and trigger an alarm when the temperature exceeds a preset threshold. While this method is simple in structure, it has the following drawbacks: 1. Reliance on fixed thresholds and poor robustness: Traditional fixed threshold judgment methods cannot adapt to different equipment aging states, load fluctuations and other factors, and are prone to false alarms or false alarms. 2. Inability to predict trend risks in advance: This type of method only responds to "overheating" and lacks the ability to model temperature development trends, thus failing to achieve early warning; 3. Lack of intelligent scoring and judgment mechanism: Most systems have not built a quantitative scoring system and lack quantitative expression of whether the current state is abnormal, which affects the accuracy of early warning judgment; 4. Lack of adaptive learning capability: Traditional models cannot learn and optimize themselves based on the temperature distribution patterns of different devices and environments, resulting in poor versatility of early warning models.

[0004] With the development of artificial intelligence and deep learning technologies, more and more research is beginning to explore the use of time-series prediction models (such as LSTM) to predict trends and assess the health of industrial equipment status data, thereby improving the system's ability to detect potential faults. However, in the scenario of detecting abnormal temperatures in low-voltage switches, there is still a lack of a complete system structure and method that combines time-series modeling, quantitative scoring mechanisms, and early warning logic to achieve accurate identification and early response to temperature anomalies. Summary of the Invention

[0005] This invention addresses the technical problems of existing low-voltage switch temperature monitoring methods, such as poor real-time performance, inaccurate anomaly identification, and inability to provide early warnings. It proposes a low-voltage switch temperature anomaly early warning system based on time-series prediction. This system introduces a Long Short-Term Memory (LSTM) neural network model to predict temperature change trends and combines it with a scoring function and intelligent judgment logic to achieve dynamic evaluation of the operating status and early identification of abnormal trends. This significantly improves the timeliness and accuracy of equipment anomaly detection, and reduces equipment failure rates and maintenance costs.

[0006] To address the aforementioned technical problems, this invention provides a low-voltage switch temperature anomaly early warning system based on time-series prediction, comprising: The data acquisition module is used to collect the operating data of the target low-voltage switch in real time at every moment. The data includes ambient temperature values. and current value And construct a historical input sequence arranged in time series. ; The time-series modeling module is used for the historical temperature sequence. Temporal modeling is performed. The temporal modeling module is based on a prediction model of a long short-term memory neural network (LSTM). The LSTM model includes an input gate, a forget gate, a candidate state generation unit, and an output gate, and outputs the hidden state and cell state at the current time. The temperature prediction module includes a decoder for mapping and decoding the hidden state into a predicted temperature value for the next moment. The scoring module is used to substitute the predicted temperature value and the measured temperature value into the basic scoring function to generate a basic score. and further Substitute into the current scoring function to calculate the current score. ; The early warning module is used to provide early warnings based on the current score. The specific logic for determining whether an abnormal state has been entered is as follows: if the current score is below 90 twice consecutively and the score shows a downward trend, it is determined to be abnormal and a warning signal is output.

[0007] Optionally, the basic scoring function is:

[0008] in, To predict temperature, For actual measured temperature, The formula introduces an error tolerance mechanism and a squared penalty mechanism to improve the sensitivity and rationality of the scoring, with 3 representing the prediction error and 3 representing the tolerance threshold.

[0009] Optionally, the current scoring function is:

[0010] in, For the current rating, The score is based on the previous moment. The base score is weighted by coefficients of 0.9 and 0.1, respectively, to smooth the trend of historical and current score changes.

[0011] Optionally, the input gate is used to calculate the weight of the current input on the cell state update based on the current input and the previous hidden state; The forget gate is used to control the degree to which the previous cell state is retained at the current moment; The candidate state generation unit is used to generate a normalized candidate state vector. The output gate is used to combine the current input and the cell state to generate the current hidden state output.

[0012] Optionally, the decoder adopts a linear fully connected structure to map the hidden state vector to the predicted temperature value, thereby realizing regression prediction of the target temperature sequence.

[0013] Optionally, the scoring module calculates a base score based on the deviation between the predicted value and the measured value, and dynamically generates the current score by combining a weighted update mechanism. The lower the current score, the more abnormal the operating status.

[0014] Optionally, the early warning module determines whether to issue an early warning based on the following logic: if the current score is below 90 for two consecutive sampling periods and the current score is less than the score at the previous moment, it is considered that there is an abnormal trend and an early warning is triggered.

[0015] Optionally, the system is deployed in a power distribution cabinet or low-voltage electrical appliance monitoring terminal, and is suitable for real-time monitoring of the temperature status of multiple low-voltage switches, and for abnormal early warning in high-power load or aging risk scenarios.

[0016] Optionally, the system is deployed in a power distribution cabinet or a low-voltage electrical appliance monitoring terminal to monitor the temperature change status of multiple low-voltage switches in real time and provide early warnings in high-risk scenarios such as high-power loads and aging equipment.

[0017] To address the aforementioned technical problems, this invention also provides a method for early warning of abnormal temperature in low-voltage switches based on time-series prediction, characterized by comprising the following steps: S1: Collect historical operating data sequences of the target low-voltage switch ; S2: Will Input the data into the trained LSTM model, and it will output the predicted temperature value for the next time step. S3: Substitute the predicted temperature and the measured temperature into the basic scoring function to generate the basic score. ; S4: Calculate the current score based on the weighted update rule. ; S5: When If the score is below 90 twice consecutively and the score shows a downward trend, an early warning mechanism will be triggered.

[0018] The beneficial effects of the technical solution of this invention are: This invention introduces an LSTM neural network to perform time-series modeling of temperature and current data of low-voltage switches, which can capture long-term and short-term dependencies, significantly improve the accuracy of temperature prediction, and control the average error within 0.5℃. This invention uses a base score and a current scoring function to dynamically measure the deviation between predicted and measured temperatures, forming an equipment health scoring system that can quantify the degree of abnormality in real time. The scoring function of this invention introduces an error tolerance threshold and a nonlinear penalty mechanism, which effectively avoids false alarms caused by normal fluctuations, while enhancing the response sensitivity to severe anomalies. This invention combines scoring trend changes for continuous judgment, triggering an alert only when the score continuously declines and falls below a threshold, thereby improving the stability and reliability of the alert judgment. The early warning module of this invention supports local audible and visual alarms and remote platform push notifications, enabling rapid on-site response and remote centralized control; The modular design of the system structure of this invention supports multi-channel switch monitoring, adjustable sampling period and adaptive threshold adjustment, and is adaptable to various application scenarios and operation and maintenance strategies. This invention enables early warning of potential faults such as switch overheating and poor contact, significantly reducing downtime and maintenance time caused by faults, and significantly improving system stability and economic benefits. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the early warning method steps of the early warning system in an embodiment of the present invention; Figure 2 This is a diagram illustrating the overall architecture of the early warning system in an embodiment of the present invention. Figure 3 This is a schematic diagram of the LSTM encoder structure in an embodiment of the present invention; Figure 4 This is a schematic diagram of the time expansion structure of the LSTM timing modeling module in an embodiment of the present invention; Figure 5 This is a schematic diagram of the decoder structure of the temperature prediction module in an embodiment of the present invention; Figure 6 This is a trend chart of abnormal sample score changes in an embodiment of the present invention; Figure 7 This is a graph showing the trend of score changes for normal samples in an embodiment of the present invention. Detailed Implementation

[0020] To better understand the above technical solutions, the technical solutions of the embodiments of this specification will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this specification and the specific features in the embodiments are detailed descriptions of the technical solutions of the embodiments of this specification, rather than limitations on the technical solutions of this specification. In the absence of conflict, the embodiments of this specification and the technical features in the embodiments can be combined with each other.

[0021] I. Overview of Implementation Examples This embodiment discloses a low-voltage switch temperature anomaly early warning system based on time-series prediction, applicable to scenarios such as distribution cabinets and low-voltage electrical appliance monitoring terminals. It enables real-time monitoring of the temperature status of multiple low-voltage switches. This system is particularly suitable for high-risk scenarios such as high-power load operation and equipment aging. By fusing multi-dimensional time-series data of ambient temperature and current, it accurately captures temperature change trends based on a deep learning-based time-series prediction model. Combined with a dynamic scoring mechanism and intelligent early warning logic, it identifies potential faults such as poor contact, overload, or short circuit in advance, effectively reducing equipment failure risks and maintenance costs.

[0022] In this embodiment, the system is deployed in the low-voltage distribution cabinet monitoring terminal of a rail transit vehicle depot. The monitoring targets are the low-voltage switches of 12 EMU train sets, covering key temperature measurement points at the incoming and outgoing terminals of phases A, B, and C. The monitoring cycle is set to 1 minute / time, and the historical data sequence length N is selected as 60 (i.e., temperature prediction for the next moment is based on the time series data of the previous 60 minutes). The system runs on an industrial-grade Linux operating system, and the hardware configuration includes an Intel Core i7 processor, 16GB of memory, and a 1TB storage module to ensure the real-time performance of data acquisition, model calculation, and early warning response.

[0023] II. System Overall Architecture like Figure 1 As shown in the figure, the low-voltage switch temperature anomaly early warning system based on time-series prediction in this embodiment mainly includes a data acquisition module M1, a time-series modeling module M2, a temperature prediction module M3, a scoring module M4, and an early warning module M5. These modules work together to achieve fully automated processing from data acquisition to anomaly early warning. The overall system architecture is as follows: Figure 2 As shown.

[0024] (a) Data Acquisition Module The data acquisition module serves as the system's data input source. Its core function is to collect the operating data of the target low-voltage switch at each moment in real time and construct a historical input sequence arranged in time series.

[0025] In this embodiment, the data acquisition module adopts a combination of industrial-grade wireless temperature and current sensors. The temperature sensor is a PT1000 platinum resistance thermometer with a measurement range of -50℃ to 200℃ and an accuracy of ±0.1℃. It is installed at key heat-generating locations such as the input and output terminals of the low-voltage switch. The current sensor is a Hall effect sensor with a measurement range of 0 to 500A and an accuracy of ±1%, connected in series in the main circuit of the switch. The sensors communicate with the monitoring terminal via an RS485 bus at a data transmission rate of 9600bps, ensuring real-time performance and stability.

[0026] Operational data collected by the data acquisition module It includes two dimensions: ambient temperature value With current value The ambient temperature value is synchronously collected by temperature and humidity sensors deployed inside the distribution cabinet, eliminating the interference of ambient temperature fluctuations on switch temperature monitoring; the current value reflects the load operating status of the switch in real time, providing a key influencing factor for predicting temperature change trends. The acquisition module preprocesses the raw data, including outlier removal (using the 3σ criterion to remove extreme data caused by sensor failure) and data normalization (mapping ambient temperature and current values ​​to the [0,1] interval), and finally constructs a historical input sequence. , where n=60, meaning each input sequence contains two-dimensional running data from the first 60 time steps.

[0027] (II) Temporal Modeling Module The temporal modeling module builds a prediction model based on a long short-term memory neural network (LSTM). Its core function is to perform temporal modeling on the historical input sequence X, capture long-term dependencies in the data, and output the hidden state and cell state at the current moment.

[0028] In this embodiment, the LSTM model adopts a three-layer network structure: the input layer has a dimension of 2 (corresponding to the two features of ambient temperature and current), the hidden layer has 128 neurons, and the output layer has a dimension of 64 (corresponding to the dimension of the hidden state vector). The core structure of the model includes an input gate, a forget gate, a candidate state generation unit, and an output gate. Each unit controls the information flow through the Sigmoid activation function (value range [0,1]) and the tanh activation function (value range [-1,1]), as shown in the specific structure below. Figure 3 As shown.

[0029] Forget gate: Used to control the degree to which the previous cell state is retained in the current time step; its calculation formula is as follows: ,in Here is the forget gate weight matrix. The state was hidden in the previous moment. Input data for the current moment. For the forget gate bias term, This is the Sigmoid activation function. In this embodiment, the initial value of the forget gate weight matrix is ​​set using the Xavier initialization method, and the initial value of the bias term is set to 0.1 to ensure the effective preservation of the cell state at the previous time step in the initial state.

[0030] Input gate: Used to calculate the weight of the current input on the cell state update based on the current input and the previous hidden state. The calculation formula is as follows: ,in The input gate weight matrix, This is the input gate bias term. Input gate output. The larger the value, the greater the contribution of the current input data to the cell state update. In this embodiment, the input gate weight matrix is ​​also initialized using Xavier, and the bias term is set to 0.1.

[0031] Candidate state generation unit: used to generate normalized candidate state vectors, calculated using the following formula: in The candidate state weight matrix, For candidate state bias terms, The activation function. Candidate state vector. It reflects the feature information after the current input and the historical state are fused. Its value is normalized to the interval [-1,1] to facilitate subsequent cell state updates.

[0032] Cell state update: The cell state at the current time step is updated by combining the outputs of the forget gate, input gate, and candidate state generation unit. The calculation formula is as follows: Where ⊙ represents element-wise multiplication, This represents the cell state at the previous moment. This represents the current cell state. This update mechanism selectively retains historical information while effectively incorporating current information, thus solving the long-term dependency problem of traditional RNNs.

[0033] Output gate: Used to combine the current input and cell state to generate the current hidden state output. The calculation formula is: ,in This is the output gate weight matrix. For the output gate bias term, The hidden state is the current state. The output gate controls the output ratio of the cell state to ensure that the hidden state accurately reflects the core features of the input sequence.

[0034] In this embodiment, the LSTM model training process uses the Adam optimizer with a learning rate of 0.001, a batch size of 32, and 100 training iterations. The training dataset consists of approximately 129,600 samples (3 months × 30 days × 24 hours × 60 minutes) from the continuous operation of 12 low-voltage switches in a rail transit depot over three months. 80% of these samples are used for the training set, and 20% for the validation set. The mean squared error (MSE) is used as the loss function during training. If the validation set loss function does not decrease for five consecutive iterations, an early stopping strategy is employed to terminate training and prevent overfitting.

[0035] (III) Temperature Prediction Module The temperature prediction module includes a decoder, whose core function is to map and decode the hidden state output by the time series modeling module into the predicted temperature value for the next time step.

[0036] In this embodiment, the decoder adopts a linear fully connected structure, containing two fully connected layers: the first fully connected layer has an input dimension of 64 (corresponding to the dimension of the hidden state vector output by the LSTM model) and an output dimension of 32; the second fully connected layer has an input dimension of 32 and an output dimension of 1 (corresponding to the predicted temperature value at the next time step). The activation function of the fully connected layers uses the ReLU function to avoid the gradient vanishing problem, the weight matrix uses the He initialization method, and the initial value of the bias term is set to 0.

[0037] The decoder works as follows: the hidden state h_t from the LSTM model is input into the first fully connected layer. Features are integrated and dimensionality compressed through linear transformation and the ReLU activation function, outputting a 32-dimensional intermediate feature vector. This intermediate feature vector is then input into the second fully connected layer, which, through linear transformation, outputs a single-dimensional predicted temperature value x. To ensure the physical meaning of the predicted temperature value is reasonable, the decoder output is denormalized, mapping it back to the actual temperature range (-50℃ to 200℃) to obtain the final predicted temperature value for the next time step.

[0038] (iv) Scoring Module The core function of the scoring module is to substitute the predicted temperature value and the measured temperature value into the basic scoring function to generate a basic score k, and then substitute k into the current scoring function to calculate the current score Score_t. The operating status of the low-voltage switch is quantified through the scoring.

[0039] In this embodiment, the measured temperature value y is acquired in real time by the PT1000 platinum resistance sensor in the data acquisition module and is obtained synchronously with the predicted temperature value x to ensure the consistency of the data in time.

[0040] Base score calculation: The base score function uses an asymmetric scoring mechanism, focusing on penalizing the abnormal situation where "the actual temperature is higher than the predicted temperature". The formula is as follows: Where x is the predicted temperature, y is the measured temperature, and 3 is the tolerance threshold. The design logic of this formula is as follows: when the measured temperature y is lower than the predicted temperature x+3℃, it indicates that the temperature change is within the normal fluctuation range, and the base score k=100 points; when the measured temperature y is higher than the predicted temperature x+3℃, points are deducted from the base score of 100. The larger the temperature difference (yx-3), the more points are deducted. The deduction uses a square relationship to amplify the deduction, ensuring sensitivity to severe anomalies; when the score after deduction is lower than 0, the base score k is 0 points. In this embodiment, since the average temperature prediction error of the LSTM model is controlled within 0.5℃, setting a tolerance threshold of 3℃ can effectively avoid misjudgments caused by normal prediction errors.

[0041] Current score calculation: The current score function uses a weighted update mechanism, and the formula is as follows: .in For the current rating, The score is based on the previous time step, with k as the base score and weighting coefficients of 0.9 and 0.1 respectively. This weighting mechanism aims to smooth the trends between historical and current scores, reduce drastic fluctuations caused by single prediction errors, and better reflect the long-term operating status of the low-voltage switch. If the low-voltage switch has no historical health score (e.g., during the system's first operation), the base score k is directly used as the current score. .

[0042] Output of the scoring module The value ranges from 0 to 100 points. The lower the current score, the more abnormal the operating status of the low-voltage switch is, providing a quantitative basis for subsequent early warning modules.

[0043] (v) Early warning module The core function of the early warning module is to determine the current score. It determines whether the low-voltage switch has entered an abnormal state. When the preset warning conditions are met, it outputs a warning signal to assist maintenance personnel in responding quickly.

[0044] In this embodiment, the judgment logic of the early warning module is as follows: the current score is below 90 points for two consecutive sampling periods (i.e., two consecutive minutes), and the current score... Less than the score of the previous moment When a score is below 90 for two consecutive sampling periods, it is considered an abnormal trend and an early warning is triggered. This logic is designed based on the following considerations: a score below 90 for two consecutive sampling periods avoids false warnings caused by a single score fluctuation; a downward trend in the score indicates that the abnormal state is continuously deteriorating and requires timely intervention.

[0045] The early warning module offers two methods: local and remote. Local early warning is achieved through an audible and visual alarm on the distribution cabinet, which flashes a red warning light and sounds a buzzer to alert on-site personnel. Remote early warning sends the warning signal to the operation and maintenance management platform via industrial Ethernet. The platform then notifies maintenance personnel via SMS and APP push notifications. The warning information includes key information such as the abnormal switch number, monitoring point location, current score, predicted temperature value, and measured temperature value, facilitating accurate fault location and the development of handling solutions by maintenance personnel.

[0046] In addition, the early warning module also has an adaptive adjustment function for the early warning threshold, which can dynamically optimize the early warning judgment logic based on factors such as the operating time of the low-voltage switch and load characteristics. For example, for aging equipment that has been in operation for more than 5 years, the early warning score threshold can be adjusted to 95 points to improve the early warning sensitivity; for high-power load scenarios, the continuous sampling period judgment standard can be shortened to 1 sampling period to speed up the early warning response.

[0047] III. System Workflow The workflow of the low-voltage switch temperature anomaly early warning system based on time-series prediction in this embodiment is as follows: Figure 1 As shown, the specific steps include the following: S1: Acquire historical operating data sequence X of the target low-voltage switch. The data acquisition module uses a PT1000 platinum resistance sensor and a Hall effect sensor to collect the ambient temperature value at each moment in real time. With current value After outlier removal and data normalization preprocessing, a historical input sequence arranged in time series is constructed. The data dimension at each time point is 2, and the sequence length is 60 (corresponding to the running data of the first 60 minutes).

[0048] S2: Input X into the trained LSTM model and output the predicted temperature value for the next time step. The time series modeling module inputs the historical input sequence X into the trained LSTM model. Through the collaborative work of the forget gate, input gate, candidate state generation unit, and output gate, it captures the long-term dependencies in the sequence data and outputs the hidden state at the current time step. The fully connected decoder in the temperature prediction module will hide its state. The mapping is decoded into the predicted temperature value x for the next moment, and then denormalized to obtain the predicted temperature value in the actual physical sense.

[0049] S3: Substitute the predicted and measured temperatures into the basic scoring function to generate the basic score k. The scoring module simultaneously obtains the predicted temperature value x and the measured temperature value y, and substitutes them into the basic scoring function. The base score k is calculated to quantify the degree of deviation between the predicted and measured temperature values ​​at the current moment.

[0050] S4: Calculate the current score based on the weighted update rule. The scoring module uses different calculation methods depending on whether a historical score exists: if a score exists from a previous moment... Then, through the current scoring function The current score is calculated; if no historical score exists, the base score k is directly used as the current score. .

[0051] S5: When Score_t falls below 90 twice consecutively and the score shows a downward trend, an early warning mechanism is triggered. The early warning module monitors the current score in real time. The changes in the situation, if they satisfy "two consecutive sampling periods" <90” and “ < Upon receiving the warning conditions, local audible and visual warnings and remote push warnings will be activated immediately, outputting warning signals containing information such as the abnormal switch number and monitoring point location. Maintenance personnel will then promptly carry out equipment maintenance based on the warning information.

[0052] IV. System Performance Verification To verify the effectiveness and reliability of the system in this embodiment, a one-month field test was conducted on 12 low-voltage switches in a rail transit depot. The test results are as follows: (I) Verification of Temperature Prediction Accuracy The A-phase outgoing terminal of circuit breaker D and the B-phase incoming terminal of circuit breaker E were selected as the test objects. The predicted and measured temperature values ​​were recorded continuously for 24 hours. The visualization results are shown below. Figure 4 , Figure 5 As shown in the figure. The test results show that the average temperature prediction error of the system is controlled within 0.5℃, and the predicted temperature curve almost completely overlaps with the measured temperature curve, which can accurately capture the dynamic trend of temperature change and verify the prediction accuracy of the LSTM model and the fully connected decoder.

[0053] (II) Verification of anomaly detection capability The A-phase outgoing terminal of circuit breaker F and the A-phase incoming terminal of circuit breaker G were selected as test objects. Both test objects had previously experienced over-temperature alarms during operation. During the test, the system calculates the current score in real time through the scoring module, and the visualization results are shown below. Figure 6 , Figure 7 As shown. The results indicate that when the measured temperature at the A-phase output terminal of circuit breaker F is more than 3°C higher than the predicted temperature, the basic score k drops rapidly, and the current score... When two consecutive sampling periods showed scores below 90 and a downward trend, the system promptly triggered an early warning. Similarly, when a similar anomaly occurred at the A-phase input terminal of circuit breaker G, the system accurately identified and issued an early warning. According to subsequent maintenance reports, both circuit breakers had potential contact problems. The system's early warning provided precise guidance for troubleshooting and verified its anomaly identification capabilities.

[0054] (III) Real-time performance and reliability verification During testing, the system's average data acquisition latency was 50ms, model prediction latency was 80ms, and score calculation and early warning response latency was 20ms, with the total latency controlled within 150ms, meeting real-time monitoring requirements. In 30 consecutive days of operation, no hardware failures or software crashes occurred, with an average false alarm rate of 0.8% and a false negative rate of 0%, verifying the system's high reliability.

[0055] (iv) Verification of Operation and Maintenance Cost Optimization Compared with the traditional manual inspection mode, after the deployment of this system, the inspection cycle of low-voltage switches has been extended from 2 hours / time to 24 hours / time, and the manpower input has been reduced by 80%; the average handling time of equipment failure has been shortened from 30 minutes to 10 minutes, and the downtime loss has been reduced by 67%; the equipment failure rate has been reduced from 3.2% to 0.5%, effectively reducing operation and maintenance costs and equipment failure risks.

[0056] V. Implementation Examples Extension and Adaptation The low-voltage switch temperature anomaly early warning system based on time-series prediction in this embodiment can be flexibly expanded and adapted according to actual application scenarios: Monitoring Channel Expansion: The system supports simultaneous monitoring of multiple low-voltage switches. By increasing the number of sensors and expanding the data acquisition interface, the number of monitoring channels can be expanded to more than 100, adapting to the centralized monitoring needs of large power distribution rooms.

[0057] Sampling period adjustment: The sampling period can be flexibly adjusted according to the monitoring accuracy requirements. The shortest sampling period can be set to 10 seconds / time to meet the high-sensitivity monitoring scenario; the longest sampling period can be set to 5 minutes / time to reduce the system resource consumption in low-load scenarios.

[0058] Early warning threshold adaptation: The early warning scoring threshold can be adaptively adjusted according to factors such as equipment type, years of operation, and load characteristics. For example, the early warning threshold for new equipment can be set to 85 points, and the early warning threshold for aging equipment can be set to 95 points, thereby improving the targeting of early warnings.

[0059] Deployment scenario adaptation: The system can be deployed on various hardware platforms such as power distribution cabinets, low-voltage electrical appliance monitoring terminals, and industrial IoT gateways, adapting to the low-voltage switch temperature monitoring needs of different fields such as rail transit, industrial production, and commercial buildings.

[0060] In summary, the low-voltage switch temperature anomaly early warning system based on time-series prediction in this embodiment achieves accurate temperature prediction by integrating multi-dimensional time-series data and using an LSTM model and a fully connected decoder. Combined with an asymmetric scoring mechanism and intelligent early warning logic, it effectively improves the real-time performance, accuracy, and reliability of temperature monitoring, reduces operation and maintenance costs and equipment failure risks, and has broad application prospects.

[0061] This embodiment introduces an LSTM neural network to perform time-series modeling of the temperature and current data of the low-voltage switch, which can capture the long-term and short-term dependencies, significantly improve the accuracy of temperature prediction, and control the average error within 0.5℃. This embodiment uses a base score and the current scoring function to dynamically measure the deviation between the predicted and measured temperatures, forming an equipment health scoring system that can quantify the degree of abnormality in real time. The scoring function in this embodiment introduces an error tolerance threshold and a nonlinear penalty mechanism to effectively avoid false alarms caused by normal fluctuations, while enhancing the response sensitivity to severe anomalies. This embodiment combines scoring trend changes for continuous judgment, and triggers an alert only when the score continuously declines and falls below a threshold, thereby improving the stability and reliability of the alert judgment. The early warning module in this embodiment supports local audible and visual alarms and remote platform push notifications, enabling rapid on-site response and remote centralized control. The modular design of the system structure in this embodiment supports multi-switch monitoring, adjustable sampling period and adaptive threshold adjustment, and adapts to a variety of application scenarios and operation and maintenance strategies. This embodiment enables early warning of potential faults such as switch overheating and poor contact, significantly reducing downtime and maintenance time caused by faults, and significantly improving system stability and economic benefits.

[0062] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A low voltage switch temperature anomaly early warning system based on time series prediction, characterized in that, include: The data acquisition module is used to collect the operating data of the target low-voltage switch in real time at every moment. The data includes ambient temperature values. and current value And construct a historical input sequence arranged in time series. ; a time series modeling module for performing time series modeling on the historical temperature sequence based on a long short-term memory neural network (LSTM) prediction model, the LSTM model comprising an input gate, a forget gate, a candidate state generation unit, and an output gate, and outputting a hidden state and a cell state of a current time point; The temperature prediction module includes a decoder for mapping and decoding the hidden state into a predicted temperature value for the next moment. a scoring module configured to generate a base score by substituting the predicted temperature value and a measured temperature value into a base score function and further substituting into a current score function to obtain a current score ; The early warning module is used to provide early warnings based on the current score. The specific logic for determining whether an abnormal state has been entered is as follows: if the current score is below 90 twice consecutively and the score shows a downward trend, it is determined to be abnormal and a warning signal is output.

2. The system of claim 1, wherein, The basic scoring function is: ; in, To predict temperature, For actual measured temperature, The formula introduces an error tolerance mechanism and a squared penalty mechanism to improve the sensitivity and rationality of the scoring, with 3 representing the prediction error and 3 representing the tolerance threshold.

3. The system according to claim 2, characterized in that, The current scoring function is: ; in, For the current rating, The score is based on the previous moment. The base score is weighted by coefficients of 0.9 and 0.1, respectively, to smooth the trend of historical and current score changes.

4. The system according to claim 3, characterized in that, The input gate is used to calculate the weight of the current input on the cell state update based on the current input and the previous hidden state; The forget gate is used to control the degree to which the previous cell state is retained at the current moment; The candidate state generation unit is used to generate a normalized candidate state vector. The output gate is used to combine the current input and the cell state to generate the current hidden state output.

5. The system of claim 4, wherein, The decoder adopts a linear fully connected structure, which maps the hidden state vector to the predicted temperature value, thereby realizing regression prediction of the target temperature sequence.

6. The system of claim 5, wherein, The scoring module calculates a base score based on the deviation between the predicted and measured values, and dynamically generates the current score using a weighted update mechanism. The lower the current score, the more abnormal the operating status.

7. The system of claim 6, wherein, The early warning module determines whether to issue an early warning based on the following logic: if the current score is below 90 for two consecutive sampling periods and the current score is less than the score at the previous moment, it is considered to have an abnormal trend and an early warning is triggered.

8. The system of claim 7, wherein, The system is deployed in a power distribution cabinet or low-voltage electrical appliance monitoring terminal and is suitable for real-time monitoring of the temperature status of multiple low-voltage switches, and for providing early warning of abnormalities in high-power loads or aging risk scenarios.

9. The system of claim 8, wherein, The system is deployed in a distribution cabinet or low-voltage electrical monitoring terminal to monitor the temperature changes of multiple low-voltage switches in real time and provide early warnings in high-risk scenarios such as high-power loads and aging equipment.

10. A low-voltage switch temperature anomaly early warning method based on time series prediction, characterized in that, Includes the following steps: S1: collect a historical operation data sequence of a target low-voltage switch ; S2: Will Input the data into the trained LSTM model, and it will output the predicted temperature value for the next time step. S3: Substitute the predicted temperature and the measured temperature into the basic scoring function to generate the basic score. ; S4: Calculate the current score based on the weighted update rule ; S5: When Trigger an early warning mechanism when the score is below 90 for two consecutive times and the score is in a downward trend.

Images