Early fire warning system and method for cable joint

Abstract

A cable joint fire early warning system and method are disclosed. In the system, a reflective parabolic focusing gas chamber collects the characteristic gas released by the overheated cable joint and provides an infrared light reflection and convergence optical path. An integrated infrared collimating light source emits collimated infrared light into the reflective parabolic focusing gas chamber. A four-channel pyroelectric detector receives the infrared radiation reflected and converged by the reflective parabolic focusing gas chamber and converts it into a measurement signal and an environmental reference signal characterizing the concentration of the characteristic gas. A pyroelectric detector signal conditioning unit is connected to the four-channel pyroelectric detector to filter and amplify the measurement signal and the environmental reference signal. The cable joint overheating early warning model module is an early warning model based on selective state-space compensation and QBEL reinforcement learning to receive and process the measurement signal, environmental reference signal, temperature value, humidity value, and pressure value collected by the cable joint overheating monitoring device, and output the cable joint fire early warning status.

Description

Technical Field

[0001] This invention relates to the field of cable condition monitoring technology, and in particular to an early warning system and method for cable joint fires. Background Technology

[0002] Cables are frequently subjected to heavy or overload conditions. Under these conditions, the cable body, especially the intermediate and terminal joints, is prone to abnormal overheating due to manufacturing defects, mechanical stress damage, or poor construction techniques. This continuous abnormal overheating can lead to pyrolytic degradation of the cable's outer sheath, potentially causing a cable fire. Cable fires are characterized by rapid spread, difficulty in extinguishing, and a high risk of causing large-scale power outages. Furthermore, the PVC outer sheath material of many operating cables produces large amounts of toxic and corrosive gases when burned intensely, posing a serious threat to the lives of rescue personnel and other equipment around the cable. Currently, the main methods for monitoring cable joint temperature include infrared imaging, fiber optic temperature sensing, and thermocouples. Infrared imaging is easily obstructed and struggles to monitor concealed cable coverage areas; fiber optic temperature sensing is laid in a zigzag pattern, limiting its applicability, making it prone to missed detections, and its wiring is complex; thermocouple temperature sensing requires direct contact between the thermocouple and the cable surface, allowing only single-point temperature monitoring and potentially causing poor local heat dissipation, posing a safety hazard.

[0003] When a cable joint overheats, it will cause the outer sheath material to produce characteristic gases. By detecting the characteristic gases of the cable joint overheating, an early warning of cable joint fires can be achieved. Currently, the commonly used methods for detecting characteristic gases of cable joint overheating are semiconductor gas sensors and laser gas sensors. Semiconductor gas sensors are easily affected by the complex environment in which the cable joint is located, have poor anti-interference capabilities, and frequently trigger false alarms. Although laser gas sensors have the advantages of high accuracy and precision, their cost is too high, making it difficult to apply them widely.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of the present invention, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] To address the shortcomings or defects of the existing technology, an early warning system and method for cable joint fires are provided. This system collects overheating characteristic gas signals, ambient temperature and humidity values, and pressure values ​​from the cable joint to achieve overheating monitoring. It features high sensitivity and strong anti-interference capabilities, enabling early fire warnings even in complex operating conditions of cable joints, thus greatly reducing the risk of greater losses from cable fires.

[0006] The objective of this invention is achieved through the following technical solutions.

[0007] An early warning system for cable joint fires includes:

[0008] A cable joint overheating monitoring device, comprising a cable joint status monitoring module, the cable joint status monitoring module including:

[0009] The reflective parabolic focusing gas chamber collects the characteristic gases released from overheated cable joints and provides an optical path for infrared light reflection and convergence.

[0010] An integrated infrared collimating light source emits collimated infrared light into the reflective parabolic focusing gas chamber.

[0011] A four-channel pyroelectric detector receives infrared radiation reflected and focused by a parabolic focusing gas cell, and converts it into measurement signals characterizing the concentration of characteristic gases and environmental reference signals.

[0012] A pyroelectric detector signal conditioning unit, connected to the four-channel pyroelectric detector, is used to filter and amplify the measurement signal and the environmental reference signal.

[0013] The ambient temperature and humidity acquisition unit measures the temperature and humidity values ​​of the environment where the cable joint is located.

[0014] An environmental pressure acquisition unit measures the pressure value of the environment where the cable joint is located.

[0015] The host computer communicates with the cable joint overheat monitoring device via a signal transmission module. The host computer includes:

[0016] The cable joint overheating early warning model module is an early warning model based on selective state space compensation and QBEL reinforcement learning. It receives and processes the measurement signals, environmental reference signals, temperature values, humidity values ​​and pressure values ​​collected by the cable joint overheating monitoring device, and outputs the early warning status of cable joint fire.

[0017] In the aforementioned early warning system for cable joint fires, the inner wall of the reflective parabolic focusing gas chamber has seven reflective mirrors and one parabolic focusing surface. The collimated infrared light emitted by the integrated infrared collimating light source undergoes multiple reflections on the reflective mirrors and converges at the final parabolic focusing surface, ultimately focusing the infrared light onto the filter of the four-channel pyroelectric detector.

[0018] In the aforementioned early warning system for cable joint fires, the infrared light center wavelengths and types of gases allowed to pass through the three detection channels of the four-channel pyroelectric detector are 9.5μm for 2-ethylhexanol, 3.3μm for benzene, and 4.6μm for carbon monoxide, respectively. The remaining 4.0μm detection channel outputs an environmental reference signal.

[0019] The cable joint fire early warning system described above includes a cable joint overheating early warning model module comprising:

[0020] The selective state-space compensation module uses a feature mapping network optimized by a selective state-space scanning mechanism to compensate for environmental temperature, humidity and pressure disturbances, and outputs a cable joint overheating characteristic gas concentration vector to eliminate environmental interference.

[0021] The QBEL reinforcement integration learning module includes a recurrent neural network (RNN), a long short-term memory network (LSTM), a gated recurrent unit (GRU), and a dynamic weight fusion unit based on the Q-learning algorithm. It is used to model the time-series evolution characteristics of multiple gases and output the fire risk index of cable joints.

[0022] In the aforementioned early warning system for cable joint fires, the input vector of the selective state-space compensation module includes three differential signals at the same timestamp: temperature, humidity, pressure, and characteristic gas of the cable joint overheating, as well as one environmental reference signal.

[0023] In the aforementioned early warning system for cable joint fires, the QBEL reinforcement ensemble learning module models the overheating multi-gas evolution process of the cable joint as a Markov decision process. The state consists of the prediction error and characteristic gas concentration gradient of the recurrent neural network (RNN), long short-term memory network (LSTM), and gated recurrent unit (GRU) within a time window. The action is the dynamic weight vector assigned to the RNN, LSTM, and GRU. The reward function is defined based on the ensemble prediction mean square error and prediction accuracy.

[0024] The cable joint fire early warning system further includes, in the case of the cable joint overheating monitoring device:

[0025] The microcontroller control module is connected to the ambient temperature and humidity acquisition unit, the ambient pressure acquisition unit, and the signal transmission module, respectively. It is connected to the four-channel pyroelectric detector through the pyroelectric detector signal conditioning unit, and to the integrated infrared collimating light source through the infrared light source constant power drive module.

[0026] An infrared light source constant power drive module includes a buck switching regulator chip, a high-side power and current monitoring chip, and a current sampling resistor. The input terminal of the buck switching regulator chip is connected to a power supply, and its output terminal is connected to an integrated infrared collimating light source. Its feedback terminal is connected to the comparator terminal of the high-side power and current monitoring chip. The current sampling resistor is connected in series between the buck switching regulator chip and the integrated infrared collimating light source. The input terminal of the high-side power and current monitoring chip is connected across the current sampling resistor, and its output terminal is connected to the feedback terminal of the buck switching regulator chip.

[0027] In the aforementioned early warning system for cable joint fires, the microcontroller control module is used to generate a modulation signal to drive the constant power drive module of the infrared light source. The modulation signal is a 1Hz square wave signal with a 50% duty cycle. The analog-to-digital converter collects three differential signals, one environmental reference signal, temperature value, humidity value, and pressure value output by the pyroelectric detector signal conditioning unit. The signal transmission module is a LoRa module or an RS485 module connected to the microcontroller control module.

[0028] An early warning method for a cable joint fire early warning system includes,

[0029] An overheat monitoring device for cable joints is placed near the cable joints to measure the characteristic gas signals released by the outer sheath material of the overheated cable joints.

[0030] The ambient temperature and humidity acquisition unit collects the temperature and humidity values ​​of the environment where the cable joint is located, while the ambient pressure acquisition unit collects the pressure values ​​of the environment where the cable joint is located.

[0031] The characteristic gas signal, ambient temperature and humidity value, and ambient pressure value constitute the characteristic vector of the overheated state of the cable joint.

[0032] The feature vector is input into an early warning model based on selective state-space compensation and QBEL reinforcement learning, which is then processed and outputs the early warning status of cable joint fires.

[0033] The method described above includes the following steps in the early warning model processing based on selective state-space compensation and QBEL reinforcement learning:

[0034] A feature mapping network optimized by a selective state-space scanning mechanism is used to compensate for environmental temperature, humidity and pressure disturbances, and outputs a cable joint overheating characteristic gas concentration vector that eliminates environmental interference.

[0035] Three base learners—Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM), and Gated Recurrent Unit (GRU)—were constructed to model the temporal evolution characteristics of multi-gas systems.

[0036] A dynamic weight fusion mechanism based on the Q-learning algorithm is used to weight and fuse the prediction results of the three base learners to output the cable joint fire risk index.

[0037] The warning status level is determined based on the fire risk index.

[0038] Compared with existing technologies, the beneficial effects of this invention are as follows: It monitors and warns of cable joint overheating based on non-dispersive infrared technology, using a cable joint overheating early warning model based on selective state space compensation and QBEL reinforcement learning modules for data fusion, providing early warning of cable joint fires. Compared with existing detection methods based on temperature-sensing optical fibers, infrared imaging temperature measurement, or semiconductor gas sensors, it offers higher accuracy, stronger anti-interference capabilities, and significantly earlier warning time, detecting cable fires in the early stages of cable sheath material overheating. This provides more time for handling cable joint overheating faults, preventing further development of cable fires and minimizing the risk of greater losses. The cable joint overheating monitoring device has advantages such as long lifespan, small size, and ease of deployment and installation. It can flexibly adopt wireless or wired data transmission methods and is suitable for scenarios with cable joints, such as cable trenches, cable shafts, and switch cabinets.

[0039] The description provided is merely an overview of the technical solution of this invention. In order to make the technical means of this invention clearer and more understandable, so that those skilled in the art can implement it according to the contents of the specification, and to make the described and other objects, features and advantages of this invention more obvious and understandable, specific embodiments of this invention are described below. Attached Figure Description

[0040] Various other advantages and benefits of the present invention will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0041] In the attached diagram:

[0042] Figure 1 This is a schematic diagram of an early warning system for cable joint fires provided in one embodiment of the present disclosure;

[0043] Figure 2 This is a schematic diagram of the structure of a cable joint overheat monitoring device provided in one embodiment of the present disclosure;

[0044] Figure 3 This is a schematic diagram of an integrated infrared collimating light source structure provided in one embodiment of the present disclosure;

[0045] Figure 4 This is a schematic diagram of a four-channel pyroelectric detector structure provided in one embodiment of the present disclosure;

[0046] Figure 5 A schematic diagram of a reflective parabolic focusing air chamber structure provided in one embodiment of this disclosure;

[0047] Figure 6 An optical simulation result diagram of a reflective parabolic focusing gas cell provided in one embodiment of this disclosure;

[0048] Figure 7 A circuit schematic diagram of a constant power drive module for an infrared light source is provided in one embodiment of this disclosure;

[0049] Figure 8 A flowchart of an early warning method for cable joint fires provided in one embodiment of this disclosure;

[0050] Figure 9 This is a data graph obtained from a cable joint overheating simulation experiment conducted according to an embodiment of this disclosure.

[0051] The present invention will be further explained below with reference to the accompanying drawings and embodiments. Detailed Implementation

[0052] The following will refer to the appendix. Figures 1 to 9 Specific embodiments of the invention will be described in more detail below. While specific embodiments of the invention are shown in the accompanying drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0053] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that different terms may be used to refer to the same component. This specification and claims do not distinguish components based on differences in terminology, but rather on differences in function. The terms "comprising" or "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising but not limited to." The following descriptions are preferred embodiments for carrying out the invention; however, these descriptions are for the purpose of understanding the general principles of the specification and are not intended to limit the scope of the invention. The scope of protection of this invention is determined by the appended claims.

[0054] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments, and the accompanying drawings do not constitute a limitation on the embodiments of the present invention.

[0055] To better understand, such as Figures 1 to 7 As shown, an early warning system for cable joint fires includes:

[0056] A cable joint overheating monitoring device, comprising a cable joint status monitoring module, the cable joint status monitoring module including:

[0057] The reflective parabolic focusing gas chamber collects the characteristic gases released from overheated cable joints and provides an optical path for infrared light reflection and convergence.

[0058] An integrated infrared collimating light source emits collimated infrared light into the reflective parabolic focusing gas chamber.

[0059] A four-channel pyroelectric detector receives infrared radiation reflected and focused by a parabolic focusing gas cell, and converts it into measurement signals characterizing the concentration of characteristic gases and environmental reference signals.

[0060] A pyroelectric detector signal conditioning unit, connected to the four-channel pyroelectric detector, is used to filter and amplify the measurement signal and the environmental reference signal.

[0061] The ambient temperature and humidity acquisition unit measures the temperature and humidity values ​​of the environment where the cable joint is located.

[0062] An environmental pressure acquisition unit measures the pressure value of the environment where the cable joint is located.

[0063] The host computer communicates with the cable joint overheat monitoring device via a signal transmission module. The host computer includes:

[0064] The cable joint overheating early warning model module is an early warning model based on selective state space compensation and QBEL reinforcement learning. It receives and processes the measurement signals, environmental reference signals, temperature values, humidity values ​​and pressure values ​​collected by the cable joint overheating monitoring device, and outputs the early warning status of cable joint fire.

[0065] In a preferred embodiment of the cable joint fire early warning system, the inner wall of the reflective parabolic focusing gas chamber has 7 reflective mirrors and 1 parabolic focusing surface. The collimated infrared light emitted by the integrated infrared collimated light source is reflected multiple times on the reflective mirrors and converges at the final parabolic focusing surface, ultimately focusing the infrared light onto the filter of the four-channel pyroelectric detector.

[0066] In a preferred embodiment of the cable joint fire early warning system, the infrared light center wavelengths and types of gases allowed to pass through the three detection channels of the four-channel pyroelectric detector are 9.5μm for 2-ethylhexanol, 3.3μm for benzene, and 4.6μm for carbon monoxide, respectively, and the remaining 4.0μm detection channel outputs an environmental reference signal.

[0067] In a preferred embodiment of the cable joint fire early warning system, the cable joint overheating early warning model module includes:

[0068] The selective state-space compensation module uses a feature mapping network optimized by a selective state-space scanning mechanism to compensate for environmental temperature, humidity and pressure disturbances, and outputs a cable joint overheating characteristic gas concentration vector to eliminate environmental interference.

[0069] The QBEL reinforcement integration learning module includes a recurrent neural network (RNN), a long short-term memory network (LSTM), a gated recurrent unit (GRU), and a dynamic weight fusion unit based on the Q-learning algorithm. It is used to model the time-series evolution characteristics of multiple gases and output the fire risk index of cable joints.

[0070] In a preferred embodiment of the cable joint fire early warning system, the input vector of the selective state space compensation module includes three differential signals of temperature, humidity, pressure, and overheating characteristic gas of the cable joint at the same timestamp, as well as one environmental reference signal.

[0071] In a preferred embodiment of the cable joint fire early warning system, the QBEL reinforcement ensemble learning module models the overheating multi-gas evolution process of the cable joint as a Markov decision process. The state is the prediction error and characteristic gas concentration gradient of the recurrent neural network (RNN), long short-term memory network (LSTM), and gated recurrent unit (GRU) within the time window. The action is the dynamic weight vector assigned to the RNN, LSTM, and GRU. The reward function is defined based on the ensemble prediction mean square error and prediction accuracy.

[0072] In a preferred embodiment of the cable joint fire early warning system, the cable joint overheat monitoring device further includes:

[0073] The microcontroller control module is connected to the ambient temperature and humidity acquisition unit, the ambient pressure acquisition unit, and the signal transmission module, respectively. It is connected to the four-channel pyroelectric detector through the pyroelectric detector signal conditioning unit, and to the integrated infrared collimating light source through the infrared light source constant power drive module.

[0074] An infrared light source constant power drive module includes a buck switching regulator chip, a high-side power and current monitoring chip, and a current sampling resistor. The input terminal of the buck switching regulator chip is connected to a power supply, and its output terminal is connected to an integrated infrared collimating light source. Its feedback terminal is connected to the comparator terminal of the high-side power and current monitoring chip. The current sampling resistor is connected in series between the buck switching regulator chip and the integrated infrared collimating light source. The input terminal of the high-side power and current monitoring chip is connected across the current sampling resistor, and its output terminal is connected to the feedback terminal of the buck switching regulator chip.

[0075] In a preferred embodiment of the cable joint fire early warning system, the microcontroller control module is used to generate a modulation signal to drive the infrared light source constant power drive module. The modulation signal is a 1Hz square wave signal with a 50% duty cycle. The analog-to-digital converter collects three differential signals, one environmental reference signal, temperature value, humidity value, and pressure value output by the pyroelectric detector signal conditioning unit. The signal transmission module is a LoRa module or an RS485 module connected to the microcontroller control module.

[0076] An early warning method for a cable joint fire early warning system includes,

[0077] An overheat monitoring device for cable joints is placed near the cable joints to measure the characteristic gas signals released by the outer sheath material of the overheated cable joints.

[0078] The ambient temperature and humidity acquisition unit collects the temperature and humidity values ​​of the environment where the cable joint is located, while the ambient pressure acquisition unit collects the pressure values ​​of the environment where the cable joint is located.

[0079] The characteristic gas signal, ambient temperature and humidity value, and ambient pressure value constitute the characteristic vector of the overheated state of the cable joint.

[0080] The feature vector is input into an early warning model based on selective state-space compensation and QBEL reinforcement learning, which is then processed and outputs the early warning status of cable joint fires.

[0081] In a preferred embodiment of the method, the early warning model processing based on selective state-space compensation and QBEL reinforcement learning includes,

[0082] A feature mapping network optimized by a selective state-space scanning mechanism is used to compensate for environmental temperature, humidity and pressure disturbances, and outputs a cable joint overheating characteristic gas concentration vector that eliminates environmental interference.

[0083] Three base learners—Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM), and Gated Recurrent Unit (GRU)—were constructed to model the temporal evolution characteristics of multi-gas systems.

[0084] A dynamic weight fusion mechanism based on the Q-learning algorithm is used to weight and fuse the prediction results of the three base learners to output the cable joint fire risk index.

[0085] The warning status level is determined based on the fire risk index.

[0086] In one embodiment, a cable joint fire early warning system includes,

[0087] A cable joint overheating monitoring device, comprising a cable joint status monitoring module, wherein the cable joint status monitoring module includes,

[0088] The reflective parabolic focusing gas chamber collects characteristic gases released from the overheated outer sheath material of cable joints via diffusion and provides an optical path to an integrated infrared collimating light source. The collimated infrared light emitted by the integrated infrared collimating light source undergoes multiple reflections on the mirrors on the inner wall of the chamber and converges at the final parabolic focusing surface, ultimately focusing the infrared light onto the filter of a four-channel pyroelectric detector.

[0089] The integrated infrared collimating light source emits high-intensity infrared light through four integrated nano-amorphous carbon thin-film light sources and is equipped with a condenser cup, which is used to collimate the infrared light emitted by the thin-film light source.

[0090] The four-channel pyroelectric detector is equipped with four pyroelectric detection channels using specific passband wavelength filters. It converts infrared radiation into electrical signals related to the concentration of the gas being measured. Three of the detection channels output measurement signals for three characteristic gases of overheated cable joints: 2-ethylhexanol, benzene, and carbon monoxide, respectively. The remaining detection channel outputs an environmentally relevant reference signal.

[0091] The pyroelectric detector signal conditioning unit includes a bandpass filter circuit and an instrumentation amplifier circuit. The bandpass filter circuit is used to filter out the DC component and high-frequency interference of the four-channel pyroelectric detector output signal, and the instrumentation amplifier circuit is used to differentially amplify the weak signal output by the bandpass filter circuit.

[0092] The ambient temperature and humidity acquisition unit measures the temperature and humidity values ​​of the environment where the cable joint is located.

[0093] An environmental pressure acquisition unit measures the pressure value of the environment where the cable joint is located.

[0094] The host computer, which collects and processes data from the cable joint overheating monitoring device via a signal transmission module, includes...

[0095] The cable joint overheating early warning model based on selective state-space compensation and QBEL reinforcement learning module is input with various time series data collected by the cable joint overheating monitoring device to predict the overheating state of the cable joint and output the early warning state of cable joint fire.

[0096] The model structure and training steps are as follows:

[0097] Step S401: Collect various data from the cable joint overheat monitoring device to construct the input vector for the compensation model. Include the ambient temperature values ​​at the same time point. Ambient humidity value Environmental pressure Three differential signals of characteristic gases from overheated cable joints , , And one environmental reference signal Composition of feature vectors , as input to the compensation model:

[0098]

[0099] Step S402: Combine the eigenvectors Perform a high-dimensional projection, mapping it to a hidden layer space suitable for the evolution of the state space, to generate high-dimensional hidden layer feature vectors. :

[0100] In the formula, These are high-dimensional hidden layer feature vectors; The learnable projection matrix of the feature mapping layer; This is the bias vector of the feature mapping layer.

[0101] Step S403: Construct a selective state-space scanning mechanism to compensate for environmental temperature, humidity, and pressure disturbances. This introduces a mechanism dependent on the current input. Dynamic discretization step size This enables the compensation model to adaptively adjust the state evolution rate according to environmental parameters and compensate for environmental disturbance signals.

[0102]

[0103]

[0104]

[0105] In the formula, For dynamic discretization step size; For smoothing nonlinear activation functions; The step size prediction weight matrix; and These are the core system evolution matrix and the input projection matrix in continuous state space, respectively; For matrix exponentiation operations; It is the identity matrix; and These are the discrete state evolution matrix and the input projection matrix after dynamic discretization, respectively.

[0106] Step S404: Perform state update and feature reconstruction. The hidden states are recursively updated based on the discretized state matrix, and compensated high-dimensional features are output through residual connections. Finally, the overheating characteristic gas concentration vector of the cable joint is reconstructed by dimensionality reduction of the output layer to eliminate interference from ambient temperature, humidity, and pressure. :

[0107]

[0108]

[0109]

[0110] In the formula, and These are the internal hidden state vectors of the compensation model in the current computation step and the previous computation step, respectively; This is the mapping matrix from the hidden state to the output; To preserve the residual connection matrix of the original signal; This is the high-dimensional feature vector obtained after filtering out environmental interference signals; This is the output weight matrix used for dimensionality reduction reconstruction; This is the bias vector for the output layer; This is the compensated characteristic gas concentration vector. As input for the subsequent QBEL reinforcement ensemble learning module.

[0111] Step S405: Construct the temporal feature matrix for QBEL reinforcement ensemble learning. The overheating characteristic gas concentration of the cable joint output from step S404 is then used. , , and ambient temperature ,humidity and pressure According to time steps Perform splicing. Set the time window length to [value missing]. The data within is used as a time series sample , as input to each base learner:

[0112]

[0113] Step S406: Construct a recurrent neural network as the first base learner. The RNN processes the input sequence... Recursive processing is performed to capture the short-term characteristics of the cable joint in the initial stage of overheating. The hidden layer state update formula is as follows:

[0114]

[0115] In the formula, for The Middle Sliced ​​data of rows; The state is hidden at this moment; The hidden state from the previous moment; , These are the weight matrices; For bias terms; It is the hyperbolic tangent activation function.

[0116] Based on the updated hidden state The forward prediction output of the RNN base learner at the current time step is obtained through mapping using a fully connected layer. :

[0117]

[0118] In the formula, , These are the weight matrix and bias term of the predicted output layer of the RNN network, respectively.

[0119] Step S407: Construct a Long Short-Term Memory (LSTM) network as the second base learner. LSTM is used to extract the long-term dependencies of different feature gases during the overheating process of the cable joint.

[0120]

[0121]

[0122]

[0123]

[0124]

[0125]

[0126] In the formula, for The Middle Sliced ​​data of rows; This represents the candidate hidden state at the current moment; This is the candidate hidden state from the previous time step; These are the corresponding weight matrices; These are the corresponding bias terms; This represents element-wise multiplication; Represents the Sigmoid function; This is the output vector of the forget gate; The output vector of the input gate; The output vector of the output gate; This represents the current state of the candidate cells. , These are the cell state vectors at the current time and the previous time, respectively; It is the hyperbolic tangent activation function.

[0127] Based on the updated hidden state The forward prediction output of the LSTM base learner at the current time step is obtained through mapping of the fully connected layer. :

[0128]

[0129] In the formula, , These are the weight matrix and bias term of the predicted output layer of the LSTM network, respectively.

[0130] Step S408: Construct a gated recurrent unit (GRU) as the third base learner. GRU achieves higher computational efficiency while maintaining the ability to represent long sequences. Its core state update formula is as follows:

[0131]

[0132]

[0133]

[0134]

[0135] In the formula, for The Middle Sliced ​​data of rows; This represents the candidate hidden state at the current moment; This is the candidate hidden state from the previous time step; To update the output vector of the gate; To reset the output vector of the gate; This represents the current state of the candidate cells. , , These are the weight matrices for the update gate, reset gate, and candidate hidden state, respectively. , , These are the corresponding bias terms; It is the hyperbolic tangent activation function.

[0136] Based on the updated hidden layer state The forward prediction output of the GRU base learner at the current time step is obtained through mapping via a fully connected layer. :

[0137]

[0138] In the formula, and These are the weight matrix and bias term of the predicted output layer of the GRU network, respectively.

[0139] Step S409: Introduce the Q-Learning algorithm of reinforcement learning to achieve dynamic weight fusion. Model the overheating multi-gas evolution process of the cable joint as a Markov decision process. Define the state. For the three base learners in the time window Within the prediction error and characteristic gas concentration gradient. Define the action. Let be the dynamic weight vector assigned to RNN, LSTM, and GRU, and satisfy . .

[0140] Define reward function :

[0141] In the formula, The mean square error of the integrated prediction under the current action fusion; Let be the indicator function, and let be the reward gain if the prediction accuracy improves. Otherwise, it is 0.

[0142] Step S410: Update the Q-value matrix according to the Bellman equation, and continuously optimize the weight allocation strategy of the agent in the overheating evolution state of the cable joint:

[0143]

[0144] In the formula, Indicates the state at the current time step; This indicates the action taken by the agent at the current time step; Indicates taking action The state of the environment at the next time step after the transition; The value function for state-action pairs; The learning rate controls the extent to which new information is covered. This is a discount factor that measures the weight of future rewards. The maximum expected Q value for the next state; This represents the instantaneous reward function value obtained by the agent in the next time step.

[0145] Step S411: Perform state alignment mapping using the QBEL-enhanced ensemble learning model output. The Q agent infers the state based on the current feature matrix. Query the Q matrix and select the optimal weight action. Perform weighted fusion:

[0146]

[0147] In the formula, , , These represent the forward prediction outputs of the three base learners at the current time step; , , These represent the optimal weighted actions. The optimal weighting weights are assigned to the three base learners: RNN, LSTM, and GRU. Represents any candidate action in the agent's action space; represent The state at any given moment.

[0148] Step S412: To increase the stability of the model, based on the true labels... The final fusion output with the model forward propagation Construct the mean squared error loss function :

[0149]

[0150] In the formula, This represents the total number of time-series samples in the training batch. This is the regularization penalty coefficient; It is a set of network parameters that includes the compensation model and each base learner; Let represent the squared Euclidean norm of the network parameters.

[0151] Step S413: Gradient calculation and dynamic norm pruning based on backpropagation. This involves calculating the network parameter set... Calculate the partial derivatives and then use the backpropagation algorithm to compute the cumulative gradient of the feature sequence. :

[0152]

[0153] In the formula, The total time step for backpropagation calculation; This represents the set of underlying network parameters. Perform partial derivative operations; The representative model in the first The instantaneous loss function components corresponding to each time step.

[0154] To prevent gradient explosion caused by abrupt jumps in the differential signal during backpropagation in deep networks, a dynamic pruning mechanism based on the global norm is used to prune the calculated gradients. Perform scaling:

[0155]

[0156] In the formula, The preset gradient norm safety threshold, For the current gradient Norm.

[0157] Step S414: Perform iterative parameter updates using an adaptive moment estimation algorithm with bias correction. The clipped gradient is smoothed using first and second momentum. The system calculates the first moment estimate in the current iteration step. With second-order moment estimation :

[0158]

[0159]

[0160] In the formula, This is the estimate of the first moment from the previous iteration step; This is the estimate of the second moment from the previous iteration step; The gradient calculated in the current iteration step and then clipped using the dynamic norm. and These are the exponential decay rates estimated by the first and second moments, respectively.

[0161] To eliminate initialization bias in the early stages of iteration, momentum bias correction is performed:

[0162]

[0163]

[0164] In the formula, This is the estimate of the first moment for the current iteration step; This is the estimated value of the second moment for the current iteration step; and These are the current first-order and second-order exponential decay rates, respectively.

[0165] Finally, based on the set initial learning rate For global network parameters Perform adaptive updates:

[0166]

[0167] In the formula, To prevent extremely small constant terms with a denominator of zero, this step is iterated repeatedly until the loss function converges.

[0168] Step S415: Warning Alignment and Multi-Level Output Mapping. This involves merging the weighted prediction results... Mapped to a fire risk index for cable joints ( The prediction results are divided into the following four warning states using threshold alignment: Normal ( Slight overheating ), severe overheating ( A fire broke out. ).

[0169] In one embodiment, the reflective parabolic focusing chamber is made of brass and machined using CNC technology. A hole is drilled at the top of the chamber as an air inlet, used to collect characteristic gases released from the overheated outer sheath material of the cable connector via diffusion. A light source inlet and outlet are located on the side of the chamber, housing an integrated infrared collimating light source and a four-channel pyroelectric detector, respectively. The reflective parabolic focusing chamber provides an optical path to the integrated infrared collimating light source. By machining seven reflective mirrors and a parabolic focusing surface on the inner wall of the metal chamber and polishing it, the collimated infrared light emitted by the integrated infrared collimating light source undergoes multiple reflections within the chamber and converges through the final parabolic focusing surface, ultimately focusing the infrared light onto the filter of the pyroelectric detector. This achieves a longer optical path within the chamber and focuses the infrared light, increasing the utilization efficiency of the infrared light.

[0170] The microcontroller control module generates a 1Hz square wave signal with a 50% duty cycle to modulate the constant power drive module of the infrared light source. It also acquires three differential signals, one environmental reference signal, ambient temperature and humidity values, and ambient pressure values ​​output from the pyroelectric detector signal conditioning unit via a built-in analog-to-digital converter, and communicates with the host computer. The cable joint status monitoring module collects characteristic gas signals released from the overheated outer sheath material of the cable joint, as well as ambient temperature, humidity, and pressure values. The signal transmission module sends the information collected by the cable joint status monitoring module to the host computer for analysis.

[0171] The microcontroller control module is connected to the environmental temperature and humidity acquisition unit, the environmental pressure acquisition unit, and the signal transmission module. It is connected to the four-channel pyroelectric detector via a pyroelectric detector signal conditioning unit and to the integrated infrared collimating light source via an infrared light source constant power drive module. The microcontroller control module includes a microcontroller and a crystal oscillator circuit. The microcontroller integrates a high-precision analog-to-digital converter for acquiring various data. The cable connector status monitoring module includes a reflective parabolic focusing chamber, an integrated infrared collimating light source, a four-channel pyroelectric detector, a pyroelectric detector signal conditioning unit, an environmental temperature and humidity acquisition unit, and an environmental pressure acquisition unit. The integrated infrared collimating light source and the four-channel pyroelectric detector are fixed at the light source inlet and outlet of the reflective parabolic focusing chamber, respectively. The environmental temperature and humidity acquisition unit includes a temperature and humidity sensor connected to the microcontroller control module. The environmental pressure acquisition unit includes a pressure-sensitive sensor connected to the microcontroller control module. The signal transmission module is a LoRa module or an RS485 module, connected to the microcontroller control module, and communicates wirelessly or wired with the host computer.

[0172] In one embodiment, the integrated infrared collimating light source includes a substrate integrating four nano-amorphous carbon thin-film light sources and a collimating concentrator. The collimating concentrator is connected to the substrate, and the substrate pins are connected to an infrared light source constant power drive module and a microcontroller control module. The nano-amorphous carbon thin-film light source can emit 2-15 ppm of light. Non-dispersive infrared light.

[0173] Because four-channel pyroelectric detectors require high stability of infrared signals, if the infrared light source is driven by constant current or constant voltage, the output power of the infrared light source is prone to drift with changes in device temperature, causing fluctuations in the pyroelectric detector signal. To solve this problem, a constant power driving method is used to drive the infrared light source, such as... Figure 7 As shown in the diagram, the main components include the LT1933 buck switching regulator chip and the MAX4210EEUA+ high-side power and current monitoring chip. For pins 1-6 of the buck switching regulator chip, pin 1 is the BOOST boost pin, connected to a capacitor and diode, used to provide the drive voltage for the internal switching transistor; pin 2 is the GND pin, connected to analog ground; pin 3 is the FB feedback pin, receiving external feedback signals for voltage regulation; pin 4 is the SHDN# enable control pin, connected to an external modulation signal terminal through a protection resistor, receiving modulation signals from the microcontroller control module to control the start and stop of the power supply chip; pin 5 is the VIN power input pin, connected to the power supply; and pin 6 is the SW switch pin, connected to the energy storage inductor and freewheeling diode, and outputting the drive voltage after capacitor filtering. For pins 1-8 of the power and current monitoring chip, pin 1 is the GND pin, connected to analog ground; pin 2 is the IN voltage monitoring input pin, connected between the voltage divider resistors to collect the output voltage at the load end; pins 3 and 6 are NC pins, floating and not connected to internal circuitry; pin 4 is the VCC power supply pin, connected to a 3.3V power supply and connected to a filter capacitor; pin 5 is the RS+ pin, and pin 7 is the RS- pin, both connected across the current sampling resistors to monitor the actual current flowing through the load; pin 8 is the POUT power output pin, outputting an analog voltage signal representing the actual power consumed. This pin is directly connected to feedback pin FB (pin 3) of the voltage regulator chip. During operation, the buck switching regulator chip converts the input power supply into the voltage required by the load. The high-side power monitoring chip collects the current flowing through the sampling resistor and the voltage divider signal in real time. Internally, a hardware multiplier multiplies the current and voltage signals and outputs an analog voltage signal representing the output power of the infrared light source on the POUT pin. The power signal is directly injected into the FB pin of the voltage regulator chip, forming a hardware analog closed-loop feedback loop, which dynamically stabilizes the output power at the set value, thereby achieving constant power drive of the infrared light source.

[0174] In one embodiment, the pyroelectric detector signal conditioning unit includes a bandpass filter circuit and an instrumentation amplifier circuit. The bandpass filter circuit is used to filter out the DC component and high-frequency interference of the output signal of the four-channel pyroelectric detector, and its passband is 0.5Hz to 1.5Hz. The instrumentation amplifier circuit is used to differentially amplify the weak signal output by the bandpass filter circuit.

[0175] In one embodiment, to verify the actual effect of the cable joint overheating monitoring device and the early warning model based on selective state-space compensation and QBEL reinforcement learning module for early warning of cable joint fires, the following simulation experiment was conducted, and the experimental results are as follows: Figure 9 As shown. The experimental object was a 60cm long 10kV cross-linked polyethylene insulated, PVC sheathed cable joint, placed inside a constructed cable joint overheating test chamber. A thermocouple was installed inside the cable joint to monitor its internal temperature. The cable joint overheating monitoring device of this invention was positioned approximately 30cm above the cable joint for real-time monitoring, and a smoke detector was placed next to the overheating monitoring device for comparison. During the experiment, an overload current was applied to the cable using an external high-current generator to simulate continuous overheating of the cable joint. Turn on the high-current generator and start timing. Every 10 minutes, extract the gas from the test chamber. Use GC-MS and a carbon monoxide detector to measure and record the concentrations of 2-ethylhexanol, benzene, and carbon monoxide in the test chamber. When an overload current is applied to the experimental cable joint for 38 minutes, the host computer warning model of the cable joint overheating monitoring device starts to send a "slight overheating" warning signal. At this time, the thermocouple shows a temperature of 105℃. At 63 minutes, the host computer warning model of the cable joint overheating monitoring device starts to send a "severe overheating" warning signal. At this time, the thermocouple shows a temperature of 146℃. As the temperature continues to rise, at 129 minutes, the host computer warning model of the cable joint overheating monitoring device starts to send a "fire" warning signal. At this time, the thermocouple shows a temperature of 219℃. At 141 minutes, the surface of the outer sheath of the cable joint begins to melt and produce smoke. The smoke detector sounds an alarm. 103 minutes have passed since the system first issued the warning signal.

[0176] Furthermore, according to the Beer-Lambert Law, gas absorption intensity is positively correlated with optical path length. The inner wall of the reflective parabolic focusing gas chamber is fabricated with seven reflective mirrors and one parabolic focusing surface. Through multiple reflections, a significant improvement in detection sensitivity is achieved without increasing the size of the gas chamber. Experiments show that the monitoring device can detect characteristic gas concentration changes at the ppm level. The parabolic focusing surface converges the infrared light, after multiple reflections, onto the filter of the four-channel pyroelectric detector, improving light energy utilization, effectively enhancing the signal-to-noise ratio, and lowering the detection limit. Achieving a long optical path design within a limited space facilitates deployment and installation in cable trenches, cable shafts, and other spaces, solving the problems of large size and difficult installation of traditional gas detection equipment. The top of the gas chamber adopts a diffuser-type air inlet design, avoiding increased power consumption and mechanical failure risks associated with air extraction.

[0177] This invention employs an integrated design with four nano-amorphous carbon thin-film light sources and a collimating cup. The nano-amorphous carbon thin-film light source emits a broad-spectrum infrared light of 2-15 μm, covering the characteristic absorption peaks of the main characteristic gases generated by overheating of the cable outer sheath material (2-ethylhexanol 9.5 μm, benzene 3.3 μm, carbon monoxide 4.6 μm), ensuring simultaneous detection of multiple gases. The collimating cup design collimates the emitted infrared light, ensuring multiple reflections on the reflective mirror surface of the gas chamber wall, reducing light energy loss.

[0178] This invention employs a four-channel pyroelectric detector. Three channels correspond to the characteristic absorption wavelengths of 2-ethylhexanol (9.5 μm), benzene (3.3 μm), and carbon monoxide (4.6 μm), respectively. The fourth channel serves as a reference channel at 4.0 μm. Through selective transmission using a specific passband wavelength filter, the detection of three characteristic gases associated with overheating in cable joints is achieved. 2-Ethylhexanol is an early characteristic product of overheating in PVC outer sheath material, benzene is a mid-stage overheating product, and carbon monoxide is a late-stage pyrolysis product. By analyzing the concentration trends of these three gases, the overheating stage of the cable joint can be accurately determined. The fourth reference channel does not correspond to any characteristic gas absorption peak and is only used to detect ambient background radiation and light source fluctuations. By differentially processing the signals from the three measurement channels and the reference channel signal, interference from ambient temperature changes and light source power fluctuations is significantly reduced.

[0179] Experimental data shows that when the cable joint temperature reaches 105℃, the concentration of 2-ethylhexanol has increased significantly, and the system can issue a "slight overheating" warning, buying valuable time for fault handling. The development stage of cable joint overheating can be determined by observing the concentration evolution relationship of the three characteristic gases. For example, when the concentration of 2-ethylhexanol begins to decrease while the concentrations of benzene and carbon monoxide rise rapidly, it indicates that the overheating has entered the intermediate stage; when the carbon monoxide concentration rises sharply, it indicates that a fire is imminent.

[0180] This invention employs a constant power drive circuit based on a buck switching regulator chip and a high-side power monitoring chip. In traditional constant current or constant voltage drive methods, the output power of the infrared light source drifts with changes in device temperature. This invention's constant power drive method dynamically adjusts the output voltage by real-time monitoring of load current and voltage, performing hardware multiplication within the chip, thus controlling the output power fluctuation of the infrared light source within ±5%, fundamentally solving the signal drift problem of pyroelectric detectors. This drive circuit uses pure hardware analog closed-loop feedback with a response time of <50μs, far faster than software-controlled digital feedback systems (typically >1ms), enabling real-time compensation for power changes caused by ambient temperature variations and power supply voltage fluctuations, ensuring long-term stability of the infrared light source's output power. A 1Hz square wave modulation signal generated by the microcontroller control module controls the start and stop of the power supply chip, achieving modulation of the infrared light source. This modulation method, combined with a bandpass filter circuit, effectively filters out environmental interference and detector noise, enabling the system to operate stably in complex cable trench environments.

[0181] The signal conditioning unit of this invention includes a bandpass filter circuit and an instrumentation amplifier circuit. The output signal of the pyroelectric detector contains a DC component and a weak AC component. The bandpass filter circuit (passband 0.5-1.5Hz) effectively filters out the DC component and high-frequency noise, retaining only the useful signal with the same frequency as the modulation signal. The instrumentation amplifier circuit has a high common-mode rejection ratio, which can effectively suppress common-mode signals caused by power supply noise, ground interference, etc., ensuring accurate amplification of weak differential signals.

[0182] This invention employs a feature mapping network optimized by a selective state-space scanning mechanism to compensate for temperature, humidity, and pressure interference, as the temperature, humidity, and pressure of the cable joint's environment interfere with the detection signal. This invention uses the QBEL reinforcement ensemble learning method to fuse three base learners: RNN, LSTM, and GRU. The RNN base learner excels at capturing short-term features in the initial stage of cable joint overheating and has a rapid response capability to sudden overheating faults. The LSTM base learner effectively alleviates the gradient vanishing problem of long sequences through a gating mechanism, enabling the extraction of long-term dependencies between different characteristic gases during the cable joint overheating process. The GRU base learner offers higher computational efficiency while maintaining long-sequence representation capabilities. This invention models the multi-gas evolution process of cable joint overheating as a Markov decision process and uses the Q-learning algorithm to achieve dynamic weight allocation. The state definition includes the prediction error and characteristic gas concentration gradient of the three base learners within a time window. The action definition includes the dynamic weight vectors assigned to the RNN, LSTM, and GRU. The reward function is defined based on the ensemble prediction mean square error and prediction accuracy. The strategy optimization involves continuously updating the Q-value matrix through the Bellman equation to optimize the weight allocation strategy. In the simulation experiment, the system issued a "slight overheating" warning when the cable joint temperature reached 105℃ and a "severe overheating" warning when the temperature reached 146℃, providing a fire warning 103 minutes earlier than the smoke alarm and giving more time for fault handling. Through the integrated learning of three base learners, the system effectively suppressed overfitting and abnormal predictions that might occur with a single model.

[0183] A reflective parabolic focusing gas chamber and a constant-power drive module ensure high-quality raw detection signals. A selective state-space compensation module compensates for environmental temperature, humidity, and pressure interference, while a QBEL-enhanced integrated learning module enables accurate prediction of cable joint overheating conditions. By simultaneously detecting three characteristic gases and employing a multi-base learner for fusion, the system can accurately identify the development stage of overheating, providing maintenance personnel with rich decision-making information. The system can issue an early warning when the cable joint temperature reaches 105℃, 103 minutes earlier than traditional smoke detectors, and can determine the severity of overheating by analyzing the type and concentration ratio of the characteristic gases. The hardware design considers adaptability to harsh environments such as cable trenches, and the algorithm design considers long-term operational stability, making the system highly reliable and easy to maintain in practical engineering applications, suitable for widespread application.

[0184] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0185] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. An early warning system for cable joint fires, characterized in that, It includes, A cable joint overheating monitoring device, comprising a cable joint status monitoring module, the cable joint status monitoring module including: The reflective parabolic focusing gas chamber collects the characteristic gases released from overheated cable joints and provides an optical path for infrared light reflection and convergence. An integrated infrared collimating light source emits collimated infrared light into the reflective parabolic focusing gas chamber. A four-channel pyroelectric detector receives infrared radiation reflected and focused by a parabolic focusing gas cell, and converts it into measurement signals characterizing the concentration of characteristic gases and environmental reference signals. A pyroelectric detector signal conditioning unit, connected to the four-channel pyroelectric detector, is used to filter and amplify the measurement signal and the environmental reference signal. The ambient temperature and humidity acquisition unit measures the temperature and humidity values ​​of the environment where the cable joint is located. An environmental pressure acquisition unit measures the pressure value of the environment where the cable joint is located. The host computer communicates with the cable joint overheat monitoring device via a signal transmission module. The host computer includes: The cable joint overheating early warning model module is an early warning model based on selective state space compensation and QBEL reinforcement learning. It receives and processes the measurement signals, environmental reference signals, temperature values, humidity values ​​and pressure values ​​collected by the cable joint overheating monitoring device, and outputs the early warning status of cable joint fire.

2. The early warning system for cable joint fires as described in claim 1, characterized in that, Preferably, the inner wall of the reflective parabolic focusing gas chamber has 7 reflective mirrors and 1 parabolic focusing surface. The collimated infrared light emitted by the integrated infrared collimating light source is reflected multiple times on the reflective mirrors and converges at the final parabolic focusing surface, ultimately focusing the infrared light onto the filter of the four-channel pyroelectric detector.

3. The early warning system for cable joint fires as described in claim 1, characterized in that, The infrared light center wavelengths and detector gas types allowed by the three detection channels of the four-channel pyroelectric detector are 9.5 μm for 2-ethylhexanol, 3.3 μm for benzene, and 4.6 μm for carbon monoxide, respectively. The remaining 4.0 μm detection channel outputs an environmental reference signal.

4. The early warning system for cable joint fires as described in claim 1, characterized in that, The cable joint overheating early warning model module includes: The selective state-space compensation module uses a feature mapping network optimized by a selective state-space scanning mechanism to compensate for environmental temperature, humidity and pressure disturbances, and outputs a cable joint overheating characteristic gas concentration vector to eliminate environmental interference. The QBEL reinforcement integration learning module includes a recurrent neural network (RNN), a long short-term memory network (LSTM), a gated recurrent unit (GRU), and a dynamic weight fusion unit based on the Q-learning algorithm. It is used to model the time-series evolution characteristics of multiple gases and output the fire risk index of cable joints.

5. The early warning system for cable joint fires as described in claim 4, characterized in that, The input vector of the selective state-space compensation module includes three differential signals at the same timestamp: temperature, humidity, pressure, and overheating characteristic gas of the cable joint, as well as one environmental reference signal.

6. The early warning system for cable joint fires as described in claim 4, characterized in that, The QBEL reinforcement ensemble learning module models the overheating multi-gas evolution process of the cable joint as a Markov decision process. The state is the prediction error and characteristic gas concentration gradient of the recurrent neural network (RNN), long short-term memory network (LSTM), and gated recurrent unit (GRU) within the time window. The action is the dynamic weight vector assigned to the RNN, LSTM, and GRU. The reward function is defined based on the ensemble prediction mean square error and prediction accuracy.

7. The early warning system for cable joint fires as described in claim 1, characterized in that, The cable joint overheat monitoring device also includes: The microcontroller control module is connected to the ambient temperature and humidity acquisition unit, the ambient pressure acquisition unit, and the signal transmission module, respectively. It is connected to the four-channel pyroelectric detector through the pyroelectric detector signal conditioning unit, and to the integrated infrared collimating light source through the infrared light source constant power drive module. An infrared light source constant power drive module includes a buck switching regulator chip, a high-side power and current monitoring chip, and a current sampling resistor. The input terminal of the buck switching regulator chip is connected to a power supply, and its output terminal is connected to an integrated infrared collimating light source. Its feedback terminal is connected to the comparator terminal of the high-side power and current monitoring chip. The current sampling resistor is connected in series between the buck switching regulator chip and the integrated infrared collimating light source. The input terminal of the high-side power and current monitoring chip is connected across the current sampling resistor, and its output terminal is connected to the feedback terminal of the buck switching regulator chip.

8. The early warning system for cable joint fires as described in claim 7, characterized in that, The microcontroller control module is used to generate a modulation signal to drive the constant power drive module of the infrared light source. The modulation signal is a 1Hz square wave signal with a 50% duty cycle. The analog-to-digital converter collects three differential signals, one environmental reference signal, temperature value, humidity value, and pressure value output by the pyroelectric detector signal conditioning unit. The signal transmission module is a LoRa module or an RS485 module connected to the microcontroller control module.

9. A method for early warning of cable joint fires as described in any one of claims 1-8, characterized in that, It includes, An overheat monitoring device for cable joints is placed near the cable joints to measure the characteristic gas signals released by the outer sheath material of the overheated cable joints. The ambient temperature and humidity acquisition unit collects the temperature and humidity values ​​of the environment where the cable joint is located, while the ambient pressure acquisition unit collects the pressure values ​​of the environment where the cable joint is located. The characteristic gas signal, ambient temperature and humidity value, and ambient pressure value constitute the characteristic vector of the overheated state of the cable joint. The feature vector is input into an early warning model based on selective state-space compensation and QBEL reinforcement learning, which is then processed and outputs the early warning status of cable joint fires.

10. The method as described in claim 9, characterized in that, Early warning model processing based on selective state-space compensation and QBEL reinforcement learning includes, A feature mapping network optimized by a selective state-space scanning mechanism is used to compensate for environmental temperature, humidity and pressure disturbances, and outputs a cable joint overheating characteristic gas concentration vector that eliminates environmental interference. Three base learners—Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM), and Gated Recurrent Unit (GRU)—were constructed to model the temporal evolution characteristics of multi-gas systems. A dynamic weight fusion mechanism based on the Q-learning algorithm is used to weight and fuse the prediction results of the three base learners to output the cable joint fire risk index. The warning status level is determined based on the fire risk index.

Images