Boiler room fan linkage early warning system based on combustible gas concentration threshold value

Abstract

The invention discloses a boiler room fan linkage early warning system based on a combustible gas concentration threshold value. A boiler room safety baseline model is generated based on boiler room environment parameterization configuration data; a relay control output end of the AI edge node is connected to a fan linkage control circuit; outputting a future predicted combustible gas concentration value and a leakage risk probability result; generating a bimodal linkage control instruction, wherein the bimodal linkage control instruction comprises a dynamic trend linkage control instruction and a static threshold linkage control instruction; and a relay control output end is driven according to the bimodal linkage control instruction, the start-stop state of the boiler room exhaust fan is controlled, and fan start-stop state data are generated. According to the method, the defects of response lag and single criterion failure of a traditional scheme are overcome, and the intrinsic safety of the boiler room under the complex working condition is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of boiler room fan technology, and in particular to a boiler room fan linkage early warning system based on a combustible gas concentration threshold. Background Technology

[0002] Currently, all types of natural gas boiler rooms are equipped with combustible gas alarms, but their functions are limited, only providing on-site audible and visual alarms without automatic linkage and response capabilities. When the concentration of combustible gas leaks reaches the dangerous threshold, manual activation of exhaust fans is required, which can easily lead to gas explosions, poisoning, and other safety accidents due to personnel being away from their posts or delayed responses. Furthermore, existing alarms lack remote early warning functions, meaning that staff not on-site cannot be notified of gas leaks immediately, resulting in untimely emergency response and significant blind spots in safety protection. In addition, replacing most boiler rooms with a complete set of intelligent linkage protection equipment would be costly, time-consuming, and poorly compatible with existing natural gas transmission and distribution systems, making rapid implementation and promotion difficult. Summary of the Invention

[0003] One objective of this invention is to propose a boiler room fan linkage early warning system based on a combustible gas concentration threshold. This invention avoids the shortcomings of traditional solutions, such as delayed response and failure of a single criterion, and ensures the inherent safety of boiler rooms under complex operating conditions.

[0004] A boiler room fan linkage early warning system based on a combustible gas concentration threshold, according to an embodiment of the present invention, includes: The model building module collects parameterized configuration data of the boiler room environment and generates a boiler room safety baseline model based on the parameterized configuration data. The AI ​​edge node module involves installing AI edge nodes with edge computing capabilities on-site in the boiler room, connecting the AI ​​edge nodes to the signal output terminal of the combustible gas sensor via shielded cables, and connecting the relay control output terminal of the AI ​​edge nodes to the fan linkage control circuit. The bidirectional Mamba time series prediction module deploys a bidirectional Mamba time series prediction model within the AI ​​edge node. The bidirectional Mamba time series prediction model is initialized based on the boiler room safety baseline model and outputs the predicted future combustible gas concentration value and leakage risk probability result. The dual-modal linkage control module generates dual-modal linkage control commands based on the predicted future combustible gas concentration, leakage risk probability results, and combustible gas lower explosion limit concentration threshold parameters. The dual-modal linkage control commands include dynamic trend linkage control commands and static threshold linkage control commands. The drive module drives the relay control output terminal according to the dual-modal linkage control command to control the start and stop status of the boiler room exhaust fan and generate fan start and stop status data.

[0005] The remote multi-channel alarm push module is communicatively connected to the dual-modal linkage control module and the AI ​​edge node module, and is used to push alarm information to staff through multiple channels when a warning, linkage or system failure event is triggered.

[0006] Optionally, the model building module includes: Collect a dataset of parameterized configurations for the boiler room environment; Based on gas type parameters and lower explosion limit concentration threshold parameters, a percentage lower explosion limit concentration is established and written into the boiler room environmental parameterized configuration dataset. Based on the sampling period parameter, several consecutive sampling points are collected in the initial stage of boiler room commissioning to form a boiler room baseline sampling sequence. Based on the boiler room baseline sampling sequence, a set of boiler room baseline statistical parameters is generated. Based on the parameters of the fan linkage control circuit, the set of baseline statistical parameters of the boiler room, and the static early warning threshold parameters and static linkage threshold parameters, a boiler room safety baseline model is generated.

[0007] Optionally, the AI ​​edge node module includes: Install AI edge nodes with edge computing capabilities on-site in the boiler room; Shielded cables are used to connect the AI ​​edge nodes to the signal output of the combustible gas sensor. Within the AI ​​edge node, the received combustible gas sensor output signal is periodically sampled according to the set sampling period parameters, and the combustible gas sensor output signal at each sampling moment is converted into a normalized value of combustible gas concentration based on interface sampling and dimensional mapping rules. Connect the relay control output terminal of the AI ​​edge node to the wind turbine linkage control circuit, and the AI ​​edge node outputs a relay control signal in each sampling cycle; In the fan linkage control circuit, the output state of the relay control signal corresponds one-to-one with the start and stop state of the boiler room exhaust fan. When the relay control signal is 0, the boiler room exhaust fan is in the stopped state, and when the relay control signal is 1, the boiler room exhaust fan is in the started state.

[0008] Optionally, the bidirectional Mamba time series prediction module includes: Deploy a bidirectional Mamba time series prediction model within the AI ​​edge node, and use the boiler room safety baseline model as the initialization input for the bidirectional Mamba time series prediction model to generate a set of model initialization parameters; Write the baseline mean and baseline volatility into the model initialization parameter set; An input window for the normalized value sequence of combustible gas concentration is constructed based on the sampling period parameter, and the model input sequence and the model input rate of change sequence are generated on the input window. The static early warning threshold parameters and the static linkage threshold parameters are written into the model initialization parameter set, and a threshold neighborhood distance sequence is constructed based on the static early warning threshold parameters and the static linkage threshold parameters. In the bidirectional Mamba time series prediction model, a state space update structure for boiler room linkage prior constraints is constructed. A forward state vector and a reverse state vector are defined, and the forward state vector and the reverse state vector are updated based on the model input sequence, the model input rate of change sequence and the threshold neighborhood distance sequence. Based on the forward and reverse state vectors, a sequence of context consistency test gating coefficients is constructed. The context consistency test gating coefficient sequence is applied to the model input sequence, the model input rate of change sequence, and the threshold neighborhood distance sequence, respectively, to generate the confidence-weighted input sequence, the confidence-weighted rate of change sequence, and the confidence-weighted threshold neighborhood distance sequence. Based on the confidence-weighted input sequence, the confidence-weighted rate of change sequence, and the confidence-weighted threshold neighborhood distance sequence, the predicted future combustible gas concentration and leakage risk probability results are generated.

[0009] Optionally, the dual-modal linkage control module includes: Using the normalized value of combustible gas concentration at the kth sampling time and the static early warning threshold parameter, a static early warning judgment quantity is generated; using the normalized value of combustible gas concentration at the kth sampling time and the static linkage threshold parameter, a static linkage judgment quantity is generated. A prediction linkage judgment quantity is generated using the future predicted combustible gas concentration value and the static linkage threshold parameter. When the future predicted combustible gas concentration value is greater than or equal to the static linkage threshold parameter, the prediction linkage judgment quantity is one; otherwise, it is zero. A risk trigger determination value is generated based on the leakage risk probability result and the risk trigger probability threshold. When the leakage risk probability result is greater than or equal to the risk trigger probability threshold, the risk trigger determination value is one; otherwise, it is zero. Based on the prediction linkage judgment quantity and the risk trigger judgment quantity, a dynamic trend linkage control instruction is generated. The dynamic trend linkage control instruction is one only when both the prediction linkage judgment quantity and the risk trigger judgment quantity are one; otherwise, it is zero. Based on the static linkage judgment quantity, a static threshold linkage control instruction is generated. When the static linkage judgment quantity is one, the static threshold linkage control instruction is one; otherwise, it is zero. Based on the dynamic trend linkage control command and the static threshold linkage control command, a dual-modal linkage control command is generated. If either the dynamic trend linkage control command or the static threshold linkage control command is one, the dual-modal linkage control command is one; otherwise, it is zero.

[0010] Optionally, the drive module includes: The dual-modal linkage control command is input to the relay control logic module of the AI ​​edge node. The relay control logic module performs state parsing on the dual-modal linkage control command and generates the corresponding relay control signal. The relay control output terminal is driven by the relay control signal. When the relay control signal is 1, the relay control output terminal is in a closed state. When the relay control signal is zero, the relay control output terminal is in an open state. The start / stop status of the boiler room exhaust fan is controlled by the closed or open state of the relay control output terminal. The start and stop status of the boiler room exhaust fan at each sampling time is recorded to form the fan start and stop status data.

[0011] Optionally, the start / stop status of the boiler room exhaust fan includes: When the relay control output is closed, the boiler room exhaust fan is in the start state; When the relay control output is in the off state, the boiler room exhaust fan is in the off state.

[0012] Optionally, the remote multi-channel alarm push module includes: The event triggering classification unit receives warning and linkage signals from the dual-modal linkage control module and fault signals from the AI ​​edge node module, analyzes the event type and matches the corresponding alarm level, which includes yellow warning, orange warning and red emergency. The multi-channel push unit communicates with the event triggering hierarchical unit and executes the corresponding push strategy according to the alarm level. The multi-channel push unit includes a mobile APP information push sub-unit, an automatic voice call dialing sub-unit, and a fallback SMS notification sub-unit. The status feedback log unit receives operation feedback data from staff through the interactive terminal, transmits it back to the AI ​​edge node module to update the system status, and fully records relevant information of all alarm events, realizing data retention and traceability.

[0013] The beneficial effects of this invention are: This invention deploys a bidirectional Mamba time-series prediction model within an AI edge node. By introducing dynamic features relevant to the boiler room scenario, it deeply integrates the boiler room safety baseline model with the model initialization parameter set, achieving unified modeling of the long-term cumulative trend of minor leaks and the short-term rate mutation of sudden leaks. The bidirectional Mamba model can respond in advance to the trend and speed of changes in combustible gas concentration. By predicting the gas concentration trend over several future sampling periods and combining dynamic risk probability quantification, it dynamically adjusts the linkage decision window, effectively improving the early warning capability for slow micro-leaks and the rapid response capability for fast-burst leaks.

[0014] This invention constructs a context consistency check gating coefficient sequence based on the difference between forward and reverse state vectors, the rate of change, and the threshold neighborhood in the field of industrial fan linkage early warning. It realizes dynamic weighting of signal credibility at each sampling moment and automatically suppresses sudden signal spikes caused by sensor aging drift, electromagnetic interference, or non-physical diffusion characteristics through the gating mechanism. This improves the physical consistency and discrimination robustness of combustible gas concentration data stream. After adopting the technology of this invention, the false linkage rate of boiler room exhaust fans is significantly reduced, the number of false alarms and invalid fan actions are greatly reduced, and the equipment operation and maintenance costs are effectively controlled.

[0015] This invention proposes a dual-modal linkage control strategy consisting of dynamic trend linkage control commands and static threshold linkage control commands. The static threshold linkage control commands ensure that the system always complies with national and industry compliance requirements for combustible gas safety, realizing a fallback mechanism for safety. The dynamic trend linkage control commands, based on the model's prediction of future gas concentration and risk probability, can trigger fan linkage in advance even if the static threshold is not reached. The dual-modal linkage control commands comprehensively improve the timeliness and intelligence of boiler room fan linkage, avoiding the shortcomings of traditional solutions such as response lag and failure of a single criterion, and ensuring the inherent safety of the boiler room under complex operating conditions. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a boiler room fan linkage early warning system based on a combustible gas concentration threshold proposed in this invention. Detailed Implementation

[0017] Example 1: Reference Figure 1 A boiler room fan linkage early warning system based on combustible gas concentration threshold, comprising: The model building module collects parameterized configuration data of the boiler room environment and generates a boiler room safety baseline model based on the parameterized configuration data. In this embodiment, the model building module includes: Collect a dataset of parameterized configurations for the boiler room environment; The boiler room environmental parameterized configuration dataset includes gas type parameters, lower explosive limit concentration threshold parameters, sampling period parameters, and fan linkage control circuit parameters. The gas type parameters are used to identify the types of combustible gases used or that may leak in the boiler room. The lower explosive limit concentration threshold parameters are used to represent the lower explosive limit concentration values ​​corresponding to the gas type parameters. The fan linkage control circuit parameters are used to represent the wiring and drive constraint parameters of the fan linkage control circuit.

[0018] Based on gas type parameters and lower explosion limit concentration threshold parameters, a percentage lower explosion limit concentration is established and written into the boiler room environmental parameterized configuration dataset. Based on the sampling period parameter, several consecutive sampling points are collected in the initial stage of boiler room commissioning to form a boiler room baseline sampling sequence. Based on the boiler room baseline sampling sequence, a set of boiler room baseline statistical parameters is generated. The baseline statistical parameter set for the boiler room includes the baseline mean and baseline volatility. The baseline mean is the average of the normalized values ​​of the combustible gas concentrations corresponding to all sampling points, and the baseline volatility is the square root of the average of the squares of the differences between the normalized values ​​of the combustible gas concentrations corresponding to all sampling points and the baseline mean.

[0019] Based on the parameters of the fan linkage control circuit, the set of baseline statistical parameters of the boiler room, and the static early warning threshold parameters and static linkage threshold parameters, a boiler room safety baseline model is generated.

[0020] The AI ​​edge node module involves installing AI edge nodes with edge computing capabilities on-site in the boiler room, connecting the AI ​​edge nodes to the signal output terminal of the combustible gas sensor via shielded cables, and connecting the relay control output terminal of the AI ​​edge nodes to the fan linkage control circuit. In this embodiment, the AI ​​edge node module includes: Install AI edge nodes with edge computing capabilities on-site in the boiler room; Based on the boiler room environmental parameterized configuration dataset, sampling period parameters and fan linkage control circuit parameters are written into the AI ​​edge node. The AI ​​edge node is used to perform combustible gas signal acquisition and fan linkage control on site.

[0021] Shielded cables are used to connect the AI ​​edge nodes to the signal output of the combustible gas sensor. Shielded cables are used to reduce electromagnetic coupling interference. The combustible gas sensor outputs a sensor output signal at each sampling time t, and the sensor output signal is transmitted to the AI ​​edge node via the shielded cable.

[0022] Within the AI ​​edge node, the received combustible gas sensor output signal is periodically sampled according to the set sampling period parameters, and the combustible gas sensor output signal at each sampling moment is converted into a normalized value of combustible gas concentration based on interface sampling and dimensional mapping rules. Each sampling time corresponds to a normalized value of combustible gas concentration, and the unit of all normalized values ​​of combustible gas concentration is the lower explosive limit concentration.

[0023] Connect the relay control output terminal of the AI ​​edge node to the wind turbine linkage control circuit, and the AI ​​edge node outputs a relay control signal in each sampling cycle; The output state of the relay control signal is 0 or 1, where 0 indicates that the relay control output terminal is in the open state and 1 indicates that the relay control output terminal is in the closed state.

[0024] In the fan linkage control circuit, the output state of the relay control signal corresponds one-to-one with the start and stop state of the boiler room exhaust fan. When the relay control signal is 0, the boiler room exhaust fan is in the stopped state, and when the relay control signal is 1, the boiler room exhaust fan is in the started state.

[0025] The bidirectional Mamba time series prediction module deploys a bidirectional Mamba time series prediction model within the AI ​​edge node. The bidirectional Mamba time series prediction model is initialized based on the boiler room safety baseline model and outputs the predicted future combustible gas concentration value and leakage risk probability result. In this embodiment, the bidirectional Mamba time series prediction module includes: Deploy a bidirectional Mamba time series prediction model within the AI ​​edge node, and use the boiler room safety baseline model as the initialization input for the bidirectional Mamba time series prediction model to generate a set of model initialization parameters; The model initialization parameter set is used to ensure that the bidirectional Mamba time series prediction model meets the boundary conditions of the initial gas environment and fan linkage control during boiler room commissioning.

[0026] Write the baseline mean and baseline volatility into the model initialization parameter set; The baseline mean is used to represent the normalized baseline level of combustible gas concentration during the initial stage of boiler room commissioning, and the baseline volatility is used to represent the normalized volatility of combustible gas concentration during the initial stage of boiler room commissioning. The units of both the baseline mean and the baseline volatility are percentage lower explosive limit concentrations.

[0027] An input window for the normalized value sequence of combustible gas concentration is constructed based on the sampling period parameter, and the model input sequence and the model input rate of change sequence are generated on the input window. In Example 1, the input window consists of normalized values ​​of combustible gas concentration collected in multiple consecutive sampling periods, arranged in chronological order. Within the input window, the normalized values ​​of combustible gas concentration corresponding to each sampling moment are arranged sequentially to form the model input sequence. Simultaneously, the input window performs difference calculations on the normalized values ​​of combustible gas concentration corresponding to two adjacent sampling periods, and combines the sampling period parameters to convert the difference results into changes per unit time, which are then arranged sequentially to form the model input rate of change sequence. The model input rate of change sequence is used to represent the rate of increase or decrease of combustible gas concentration in the boiler room within consecutive sampling periods.

[0028] The static early warning threshold parameters and the static linkage threshold parameters are written into the model initialization parameter set, and a threshold neighborhood distance sequence is constructed based on the static early warning threshold parameters and the static linkage threshold parameters. In Example 1, the threshold neighborhood distance sequence is composed of multiple threshold neighborhood distance values ​​in the order of sampling time. Each threshold neighborhood distance value is obtained by calculating the difference between the static linkage threshold parameter and the normalized value of the combustible gas concentration corresponding to the sampling time at the corresponding sampling time. The threshold neighborhood distance sequence is used to continuously represent the dynamic proximity of the combustible gas concentration state in the boiler room to the fan linkage trigger threshold.

[0029] When the normalized value of combustible gas concentration is less than the static linkage threshold parameter, the threshold neighborhood distance value is positive, which is used to represent the remaining safety margin of the current concentration state before triggering the fan linkage.

[0030] When the normalized value of combustible gas concentration is greater than or equal to the static linkage threshold parameter, the threshold neighborhood distance value is zero or negative, which indicates that the current concentration state has reached or exceeded the fan linkage boundary.

[0031] In the bidirectional Mamba time series prediction model, a state space update structure for boiler room linkage prior constraints is constructed. A forward state vector and a reverse state vector are defined, and the forward state vector and the reverse state vector are updated based on the model input sequence, the model input rate of change sequence and the threshold neighborhood distance sequence. In Example 1, the boiler room linkage prior constraint state space update structure generates a positive state vector and a negative state vector through state recursion. The positive state vector is calculated by the model input sequence elements, model input rate of change sequence elements, threshold neighborhood distance sequence elements, and the positive state vector of the previous sampling time. The negative state vector is calculated by the model input sequence elements, model input rate of change sequence elements, threshold neighborhood distance sequence elements, and the negative state vector of the next sampling time.

[0032] During the state recursion process, the model input rate of change sequence is injected into the state update process as a state excitation term, and the threshold neighborhood distance sequence is used as a state damping term to modulate the state update process. This results in the state response intensity of the positive and negative state vectors being enhanced when the normalized value of combustible gas concentration rises rapidly and the threshold neighborhood distance continues to decrease.

[0033] Based on the forward and reverse state vectors, a sequence of context consistency test gating coefficients is constructed. In Example 1, the state consistency difference quantity is obtained by calculating the degree of difference between the positive state vector and the negative state vector in the state space; at the same time, the model input change rate corresponding to the sampling time is obtained to represent the change intensity of the current normalized value of combustible gas concentration, and the threshold neighborhood distance corresponding to the sampling time is obtained to represent the remaining safety margin of the current normalized value of combustible gas concentration from triggering the fan linkage.

[0034] The state consistency difference, model input change rate, and threshold neighborhood distance are used as joint input features and input into a pre-configured gating mapping function to generate the gating coefficients corresponding to the sampling time. By repeating the processing process for all sampling times within the input window, a sequence of context consistency test gating coefficients corresponding to the length of the input window is obtained. Each gating coefficient in the gating coefficient sequence is used to measure the consistency of the normalized value of combustible gas concentration at the corresponding sampling time in the forward and reverse time series contexts.

[0035] When the consistency between the forward and reverse state vectors is higher than the threshold, and the distance between the model input rate of change and the threshold neighborhood conforms to the continuous evolution characteristics of gas diffusion, the corresponding gating coefficient takes a large value; when the consistency between the forward and reverse state vectors is lower than the threshold, or when the distance between the model input rate of change and the threshold neighborhood exhibits isolated abrupt changes, the corresponding gating coefficient takes a small value to reduce the impact of the corresponding sampling time on the state update.

[0036] The context consistency test gating coefficient sequence is applied to the model input sequence, the model input rate of change sequence, and the threshold neighborhood distance sequence, respectively, to generate the confidence-weighted input sequence, the confidence-weighted rate of change sequence, and the confidence-weighted threshold neighborhood distance sequence. The confidence-weighted input sequence is obtained by multiplying the gating coefficients by the terms of the original model input sequence; the confidence-weighted rate of change sequence is obtained by multiplying the gating coefficients by the terms of the original model input rate of change sequence; and the confidence-weighted threshold neighborhood distance sequence is obtained by multiplying the gating coefficients by the terms of the original threshold neighborhood distance sequence.

[0037] Based on the confidence-weighted input sequence, the confidence-weighted rate of change sequence, and the confidence-weighted threshold neighborhood distance sequence, the predicted future combustible gas concentration and leakage risk probability results are generated.

[0038] In Example 1, the bidirectional Mamba time series prediction model performs state updates on the confidence-weighted input sequence in the forward time direction to obtain a forward state vector sequence, and performs state updates on the confidence-weighted input sequence in the reverse time direction to obtain a reverse state vector.

[0039] The bidirectional Mamba time-series prediction model fuses the forward and reverse state vectors according to a preset fusion rule at each sampling time to obtain a fused state vector. The bidirectional Mamba time-series prediction model inputs the fused state vector into the concentration prediction output head. The concentration prediction output head performs a linear mapping on the fused state vector and outputs the future predicted combustible gas concentration value corresponding to several future sampling periods. The unit of the future predicted combustible gas concentration value corresponding to several future sampling periods is the percentage lower explosive limit concentration.

[0040] The bidirectional Mamba time series prediction model inputs the fused state vector, the current term of the confidence-weighted rate of change sequence, and the current term of the confidence-weighted threshold neighborhood distance sequence into the risk probability output head. The risk probability output head first performs a linear mapping on the fused state vector to obtain the risk logarithm value. Then, it performs a weighted synthesis with the rise rate contribution term corresponding to the current term of the confidence-weighted rate of change sequence and the safety margin contribution term corresponding to the current term of the confidence-weighted threshold neighborhood distance sequence. The weighted synthesis result is then probability normalized to obtain the leakage risk probability result, which ranges from 0 to 1.

[0041] The current term of the confidence-weighted rate of change sequence is used to improve the probability of leakage risk when the rate of increase in combustible gas concentration in the boiler room increases, and the current term of the confidence-weighted threshold neighborhood distance sequence is used to improve the probability of leakage risk when the remaining safety margin in the boiler room decreases.

[0042] The dual-modal linkage control module generates dual-modal linkage control commands based on the predicted future combustible gas concentration, leakage risk probability results, and combustible gas lower explosion limit concentration threshold parameters. The dual-modal linkage control commands include dynamic trend linkage control commands and static threshold linkage control commands. In this embodiment, the dual-modal linkage control module includes: Using the normalized value of combustible gas concentration at the kth sampling time and the static early warning threshold parameter, a static early warning judgment quantity is generated; using the normalized value of combustible gas concentration at the kth sampling time and the static linkage threshold parameter, a static linkage judgment quantity is generated. The static early warning determination quantity is used to indicate whether the current sampling time of the boiler room has reached the static early warning boundary. When the normalized value of the combustible gas concentration at the kth sampling time is greater than or equal to the static early warning threshold parameter, the static early warning determination quantity is one; otherwise, it is zero. The static linkage determination quantity is used to indicate whether the current sampling time of the boiler room has reached the static linkage boundary. When the normalized value of the combustible gas concentration at the kth sampling time is greater than or equal to the static linkage threshold parameter, the static linkage determination quantity is one; otherwise, it is zero.

[0043] A prediction linkage judgment quantity is generated using the future predicted combustible gas concentration value and the static linkage threshold parameter. When the future predicted combustible gas concentration value is greater than or equal to the static linkage threshold parameter, the prediction linkage judgment quantity is one; otherwise, it is zero. The prediction linkage decision quantity is used to indicate whether the bidirectional Mamba time series prediction model predicts that the normalized value of the combustible gas concentration in the boiler room will reach the static linkage boundary within the prediction step size.

[0044] A risk trigger determination value is generated based on the leakage risk probability result and the risk trigger probability threshold. When the leakage risk probability result is greater than or equal to the risk trigger probability threshold, the risk trigger determination value is one; otherwise, it is zero. The risk trigger determination quantity serves as the probability gating input for the dynamic trend linkage control command. The risk trigger probability threshold is used to set the threshold at which the leakage risk probability result reaches the trigger condition. The value range is from zero to one. The risk trigger probability threshold is written into the model initialization parameter set.

[0045] Based on the prediction linkage judgment quantity and the risk trigger judgment quantity, a dynamic trend linkage control instruction is generated. The dynamic trend linkage control instruction is one only when both the prediction linkage judgment quantity and the risk trigger judgment quantity are one; otherwise, it is zero. The dynamic trend linkage control command is used to generate a dynamic control command for the boiler room exhaust fan in advance when the bidirectional Mamba time series prediction model predicts that the normalized value of the combustible gas concentration in the boiler room will reach the static linkage boundary and the leakage risk probability result reaches the risk trigger probability threshold.

[0046] Based on the static linkage judgment quantity, a static threshold linkage control instruction is generated. When the static linkage judgment quantity is one, the static threshold linkage control instruction is one; otherwise, it is zero. The static threshold linkage control instruction is used to indicate that when the normalized value of the combustible gas concentration at the current sampling time in the boiler room reaches the static linkage boundary, a static control instruction for linkage of the boiler room exhaust fan is generated.

[0047] Based on the dynamic trend linkage control command and the static threshold linkage control command, a dual-modal linkage control command is generated. If either the dynamic trend linkage control command or the static threshold linkage control command is one, the dual-modal linkage control command is one; otherwise, it is zero.

[0048] The dual-modal linkage control command is used to integrate dynamic trend linkage control command and static threshold linkage control command as the linkage decision for boiler room exhaust fan.

[0049] The drive module drives the relay control output terminal according to the dual-modal linkage control command to control the start and stop status of the boiler room exhaust fan and generate fan start and stop status data.

[0050] In this embodiment, the driving module includes: The dual-modal linkage control command is input to the relay control logic module of the AI ​​edge node. The relay control logic module performs state parsing on the dual-modal linkage control command and generates the corresponding relay control signal. The relay control output terminal is driven by the relay control signal. When the relay control signal is 1, the relay control output terminal is in a closed state. When the relay control signal is zero, the relay control output terminal is in an open state. The start / stop status of the boiler room exhaust fan is controlled by the closed or open state of the relay control output terminal. The start and stop status of the boiler room exhaust fan at each sampling time is recorded to form the fan start and stop status data.

[0051] In this embodiment, the start / stop states of the boiler room exhaust fan include: When the relay control output is closed, the boiler room exhaust fan is in the start state; When the relay control output is in the off state, the boiler room exhaust fan is in the off state.

[0052] In this embodiment, the remote multi-channel alarm push module, the dual-modal linkage control module, and the AI ​​edge node module establish a TCP long connection communication via the MQTT 3.1.1 protocol. The communication baud rate is set to 115200bps, and the data transmission is encrypted using AES-128 to ensure the security and real-time performance of the alarm signal transmission. The module hardware adopts an industrial-grade ARM Cortex-A72 processor, with built-in 2GB DDR4 memory and 16GB eMMC storage, supporting a wide temperature operating environment of -40℃ to 85℃, and adapting to the complex working conditions of boiler rooms.

[0053] The implementation details of the event-triggered hierarchical unit are as follows: The event-triggered hierarchical unit has a built-in event parsing rule base. The received signals include event type identifier bits (0x01=static warning, 0x02=dynamic trend warning, 0x03=static linkage, 0x04=system failure), concentration data field (precision 0.01%LEL), risk probability field (precision 0.001), and timestamp field (accurate to milliseconds). The specific implementation steps are as follows: Signal parsing: The received binary data stream is parsed according to the preset protocol format, and the event type identifier bit and core data are extracted. If the data verification fails (CRC32 check code mismatch), the module's local fault alarm is triggered, and the data retransmission is requested again. Alarm level matching: Alarm level is matched according to the event type identifier bit, where 0x01 (static warning) and 0x04 (system failure) correspond to yellow warning (priority 1), 0x02 (dynamic trend warning) corresponds to orange warning (priority 2), and 0x03 (static linkage) corresponds to red emergency (priority 3). Alarm Level Output: The matched alarm level and the parsed core data are encapsulated into an alarm command in JSON format and sent to the multi-channel push unit via the internal bus. In this embodiment, the remote multi-channel alarm push module, the dual-modal linkage control module, and the AI ​​edge node module establish a TCP long connection communication via the MQTT 3.1.1 protocol. The communication baud rate is set to 115200bps, and the data transmission is encrypted with AES-128 to ensure the security and real-time performance of the alarm signal transmission. The module hardware adopts an industrial-grade ARM Cortex-A72 processor, with built-in 2GB DDR4 memory and 16GB eMMC storage, supporting a wide temperature range of -40℃ to 85℃, and adapting to the complex working conditions of boiler rooms. Example

[0054] In the actual operating environment of the boiler room, the AI ​​edge node has completed the basic environmental parameterization configuration. The main parameters include: gas type methane, lower explosive limit concentration threshold parameter 5.0% volume fraction, corresponding linkage threshold of 20% LEL, sampling period parameter 10 seconds, and fan linkage control circuit parameter wiring type A.

[0055] Within the seven days prior to commissioning, the AI ​​edge node automatically collected a total of 6,048 baseline samples for the boiler room. Sampling sequence data showed that, after normalization, the baseline mean was 1.0% LEL and the baseline volatility was 0.18% LEL. The system used the baseline mean and baseline volatility as initialization parameters input into the bidirectional Mamba time series prediction model.

[0056] At 8:00, the AI ​​edge node took its first sample, with the sensor outputting 4.00mA, which was calibrated and converted to 0.8%LEL. Ten subsequent samples were taken (once every 10 seconds), with the normalized concentration value fluctuating between 0.9% and 1.2%LEL, and the absolute value of the rate of change being less than 0.05% / 10 seconds.

[0057] At the 361st sampling (sampling time 1 hour), the sensor output current was 5.60 mA, corresponding to a concentration of 1.9% LEL. The subsequent 6 samplings showed concentrations of 2.1%, 2.5%, 2.9%, 3.5%, 4.4%, and 5.2% LEL, respectively.

[0058] Model input sequence: [2.1,2.5,2.9,3.5,4.4,5.2]%LEL; Model input rate of change sequence: [+0.4,+0.4,+0.6,+0.9,+0.8]% / 10 seconds; Threshold neighborhood distance sequence: [17.9,17.5,17.1,16.5,15.6,14.8]%LEL; During the 367th sampling, the bidirectional Mamba prediction indicated that the concentration would reach 13.2% LEL in 30 minutes. The leakage risk probability output was 0.35, which had not yet exceeded the risk threshold. The system recorded the event and continued to track it.

[0059] During the 810th sampling, the sensor output was detected to rise from 1.5% LEL to 10.8% LEL within 10 seconds, and then continued to rise sharply, reaching 15.6% LEL at the next sampling point.

[0060] Model input sequence: [10.8, 15.6]%LEL; Model input rate of change sequence: [+9.3, +4.8]% / 10 seconds; Threshold neighborhood distance sequence: [9.2, 4.4]%LEL; Following this sampling, the bidirectional Mamba time-series prediction model, combining concentration trends, rate of change, and threshold distance, predicts that the concentration within the next two sampling points will reach 22.5% LEL, with a future risk probability outputting 0.96.

[0061] The dynamic trend linkage control command is set to 1, while the static linkage judgment quantity has not yet been activated (the current actual sampled value does not exceed 20% LEL). However, the predicted linkage judgment quantity and the risk trigger judgment quantity have both reached the triggering conditions, and the wind turbine relay operates ahead of schedule.

[0062] During the 1023rd sampling, the large motor in the boiler room started, and the sensor output abnormally jumped to 27.3% LEL. The next sampling restored it to 1.1% LEL.

[0063] Model input sequence: [1.0, 1.1, 27.3, 1.1]%LEL; Model input rate of change sequence: [-0.1, +26.2, -26.2]% / 10 seconds; Gating coefficients: [0.98, 0.97, 0.06, 0.97]; The context consistency gating coefficient reaches a minimum value at this point. The AI ​​edge node suppresses the contribution of this mutation data to the linkage judgment, the wind turbine status remains unchanged, and no malfunction occurs.

[0064] At the 1580th sampling, the sampled value first reached 20.2% LEL, the static linkage judgment value was set to 1, the fan relay closed, and the fan started. In the same event, the traditional system only started linkage at this point, while the AI ​​intelligent system had already acted in advance at the 1577th sampling (sampled value of 16.8% LEL, model prediction of 21.3% LEL, risk probability of 0.78), with a lead time of 30 seconds (3 sampling periods). Refer to Table 1 below for the full-process linkage data tracking table: Table 1. Full-Process Linkage Data Tracking Table (Partial Display) Sampling number Current concentration (%LEL) Rate of change (% / 10 seconds) Risk probability Static determination Prediction and Judgment Dynamic instructions Fan operation Remark 810 10.8 +9.3 0.89 0 0 0 0 Sudden rise 811 15.6 +4.8 0.96 0 1 1 1 Fans started early 812 19.5 +3.9 0.99 0 1 1 1 The fan runs continuously 813 22.3 +2.8 1.00 1 1 1 1 Both static and dynamic conditions are satisfied 1023 27.3 +26.2 0.13 0 0 0 0 Electromagnetic interference not linked 1580 20.2 +0.8 0.58 1 1 1 1 Static threshold linkage In the training samples, the false alarm rate of the traditional method was 1.85%, while that of the method of this invention was 0.12%. Example of a static threshold false alarm sample: Sample No. 2197, with a sampled value of 19.8% LEL, momentarily jumped to 21.1% LEL, and then recovered to 16.5% LEL. In the traditional system linkage, the AI ​​model, with its low gating coefficient, did not take any action.

[0065] Example of a missed detection in a dynamic trend (traditional method): Sample numbers 7215-7223. The sampled value steadily increased from 6.2% LEL to 16.4% LEL, with a short-term increase rate of 0.9% / 10 seconds, and a cumulative increase over 21 periods. The traditional method did not report it because it did not exceed the threshold. However, the AI ​​model issued an early warning at a risk probability of 0.61 in the 15th period.

[0066] Real leak detection rate: The method of this invention has no missed detections in all real leak events, with an average linkage lead time of 40 seconds.

[0067] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A boiler room fan linkage early warning system based on combustible gas concentration threshold, characterized in that, include: The model building module collects parameterized configuration data of the boiler room environment and generates a boiler room safety baseline model based on the parameterized configuration data. The AI ​​edge node module involves installing AI edge nodes with edge computing capabilities on-site in the boiler room, connecting the AI ​​edge nodes to the signal output terminal of the combustible gas sensor via shielded cables, and connecting the relay control output terminal of the AI ​​edge nodes to the fan linkage control circuit. The bidirectional Mamba time series prediction module deploys a bidirectional Mamba time series prediction model within the AI ​​edge node. The bidirectional Mamba time series prediction model is initialized based on the boiler room safety baseline model and outputs the predicted future combustible gas concentration value and leakage risk probability result. The dual-modal linkage control module generates dual-modal linkage control commands based on the predicted future combustible gas concentration, leakage risk probability results, and combustible gas lower explosion limit concentration threshold parameters. The dual-modal linkage control commands include dynamic trend linkage control commands and static threshold linkage control commands. The drive module, based on the dual-modal linkage control commands, drives the relay control output to control the start and stop status of the boiler room exhaust fan and generates fan start and stop status data. The remote multi-channel alarm module, together with the dual-modal linkage control module and the AI ​​edge node module, works in real time to remotely push alarm information when an alarm, linkage, or system failure is triggered.

2. The boiler room fan linkage early warning system based on combustible gas concentration threshold according to claim 1, characterized in that, The model building module includes: Collect a dataset of boiler room environmental parameter configurations; Based on gas type parameters and lower explosion limit concentration threshold parameters, a percentage lower explosion limit concentration is established and written into the boiler room environmental parameterized configuration dataset. Based on the sampling period parameter, several consecutive sampling points are collected in the initial stage of boiler room commissioning to form a boiler room baseline sampling sequence. Based on the boiler room baseline sampling sequence, a set of boiler room baseline statistical parameters is generated. Based on the parameters of the fan linkage control circuit, the set of baseline statistical parameters of the boiler room, and the static early warning threshold parameters and static linkage threshold parameters, a boiler room safety baseline model is generated.

3. The boiler room fan linkage early warning system based on combustible gas concentration threshold according to claim 1, characterized in that, The AI ​​edge node module includes: Install AI edge nodes with edge computing capabilities on-site in the boiler room; Shielded cables are used to connect the AI ​​edge nodes to the signal output of the combustible gas sensor. Within the AI ​​edge node, the received combustible gas sensor output signal is periodically sampled according to the set sampling period parameters, and the combustible gas sensor output signal at each sampling moment is converted into a normalized value of combustible gas concentration based on interface sampling and dimensional mapping rules. Connect the relay control output terminal of the AI ​​edge node to the wind turbine linkage control circuit, and the AI ​​edge node outputs a relay control signal in each sampling cycle; In the fan linkage control circuit, the output state of the relay control signal corresponds one-to-one with the start and stop state of the boiler room exhaust fan. When the relay control signal is 0, the boiler room exhaust fan is in the stopped state, and when the relay control signal is 1, the boiler room exhaust fan is in the started state. When the boiler room exhaust fan is off, information is pushed to the user via mobile app, automatic voice calls, and remote SMS notifications.

4. The boiler room fan linkage early warning system based on combustible gas concentration threshold according to claim 1, characterized in that, The bidirectional Mamba time series prediction module includes: Deploy a bidirectional Mamba time series prediction model within the AI ​​edge node, and use the boiler room safety baseline model as the initialization input for the bidirectional Mamba time series prediction model to generate a set of model initialization parameters; Write the baseline mean and baseline volatility into the model initialization parameter set; An input window for the normalized value sequence of combustible gas concentration is constructed based on the sampling period parameter, and the model input sequence and the model input rate of change sequence are generated on the input window. The static early warning threshold parameters and the static linkage threshold parameters are written into the model initialization parameter set, and a threshold neighborhood distance sequence is constructed based on the static early warning threshold parameters and the static linkage threshold parameters. In the bidirectional Mamba time series prediction model, a state space update structure for boiler room linkage prior constraints is constructed. A forward state vector and a reverse state vector are defined, and the forward state vector and the reverse state vector are updated based on the model input sequence, the model input rate of change sequence and the threshold neighborhood distance sequence. Based on the forward and reverse state vectors, a sequence of context consistency test gating coefficients is constructed. The context consistency test gating coefficient sequence is applied to the model input sequence, the model input rate of change sequence, and the threshold neighborhood distance sequence, respectively, to generate the confidence-weighted input sequence, the confidence-weighted rate of change sequence, and the confidence-weighted threshold neighborhood distance sequence. Based on the confidence-weighted input sequence, the confidence-weighted rate of change sequence, and the confidence-weighted threshold neighborhood distance sequence, the predicted future combustible gas concentration and leakage risk probability results are generated.

5. A boiler room fan linkage early warning system based on a combustible gas concentration threshold as described in claim 1, characterized in that, The dual-modal linkage control module includes: Using the normalized value of combustible gas concentration at the kth sampling time and the static early warning threshold parameter, a static early warning judgment quantity is generated; using the normalized value of combustible gas concentration at the kth sampling time and the static linkage threshold parameter, a static linkage judgment quantity is generated. A prediction linkage judgment quantity is generated using the future predicted combustible gas concentration value and the static linkage threshold parameter. When the future predicted combustible gas concentration value is greater than or equal to the static linkage threshold parameter, the prediction linkage judgment quantity is one; otherwise, it is zero. A risk trigger determination value is generated based on the leakage risk probability result and the risk trigger probability threshold. When the leakage risk probability result is greater than or equal to the risk trigger probability threshold, the risk trigger determination value is one; otherwise, it is zero. Based on the prediction linkage judgment quantity and the risk trigger judgment quantity, a dynamic trend linkage control instruction is generated. The dynamic trend linkage control instruction is one only when both the prediction linkage judgment quantity and the risk trigger judgment quantity are one; otherwise, it is zero. Based on the static linkage judgment quantity, a static threshold linkage control instruction is generated. When the static linkage judgment quantity is one, the static threshold linkage control instruction is one; otherwise, it is zero. Based on the dynamic trend linkage control command and the static threshold linkage control command, a dual-modal linkage control command is generated. If either the dynamic trend linkage control command or the static threshold linkage control command is one, the dual-modal linkage control command is one; otherwise, it is zero.

6. A boiler room fan linkage early warning system based on a combustible gas concentration threshold as described in claim 1, characterized in that, The driving module includes: The dual-modal linkage control command is input to the relay control logic module of the AI ​​edge node. The relay control logic module performs state parsing on the dual-modal linkage control command and generates the corresponding relay control signal. The relay control output terminal is driven by the relay control signal. When the relay control signal is 1, the relay control output terminal is in a closed state. When the relay control signal is zero, the relay control output terminal is in an open state. The start / stop status of the boiler room exhaust fan is controlled by the closed or open state of the relay control output terminal. The start and stop status of the boiler room exhaust fan at each sampling time is recorded to form the fan start and stop status data.

7. A boiler room fan linkage early warning system based on a combustible gas concentration threshold as described in claim 1, characterized in that, The start / stop status of the boiler room exhaust fan includes: When the relay control output is closed, the boiler room exhaust fan is in the start state; When the relay control output is in the off state, the boiler room exhaust fan is in the off state.

8. A boiler room fan linkage early warning system based on a combustible gas concentration threshold as described in claim 1, characterized in that... The remote multi-channel alarm push module also includes a log retention unit, in which the trigger time, push channel, receiving status, and personnel confirmation time of all alarm events are fully recorded and stored on a cloud server.

Images