Fuel gas state sudden change intelligent decision-making method based on real-time pressure monitoring

Abstract

The invention relates to the technical field of gas pressure monitoring, and discloses a gas state sudden change intelligent decision-making method based on real-time pressure monitoring, which comprises the following steps: step 1, multi-modal data acquisition: focusing on resident scene perception requirements: a core monitoring target of a resident gas system is home-entry pipeline pressure, and pertinently configuring perception equipment; step 2, multi-dimensional data preprocessing: intensified pressure threshold interval signal reservation: the pressure fluctuation of a resident scene is mostly small and high in frequency, and the preprocessing needs to avoid loss of threshold boundary signals less than 0.1 kPa level while noise reduction is carried out; the gas state recognition precision and control flexibility are improved, the method is suitable for micro gas quantity detection, remote strategy synchronization and emergency behavior response in multiple scenes, the safety, intelligence and maintainability of equipment are enhanced, and the method is suitable for gas control in the fields of residential users, commercial kitchens and industries.

Description

Technical Field

[0001] This invention relates to the field of gas pressure monitoring technology, specifically to an intelligent decision-making method based on real-time pressure monitoring of sudden changes in gas state. Background Technology

[0002] Natural gas has become an indispensable energy source in daily life, but abnormal pressure during its use (underpressure and overpressure) is a core hidden danger that can lead to safety accidents. Underpressure can easily cause gas leaks after the gas appliances are turned off, while overpressure may cause risks such as hose rupture and loose connections. Currently, residential gas safety monitoring and control technologies are mainly divided into two categories, both of which have significant limitations and are difficult to meet the needs of precise and intelligent safety management.

[0003] The inherent defects of mechanical control technology: Traditional safety valves, overpressure protectors and other mechanical devices can only trigger passive protection under extreme pressure (such as overpressure ≥10kPa). They have no response capability to the underpressure warning range of 0.8±0.2kPa and overpressure warning range of 8±2kPa in residential scenarios, and cannot identify the potential risks brought about by small pressure fluctuations. Moreover, the mechanical structure is prone to aging and has extremely low efficiency in dealing with gradual leakage or instantaneous pressure changes.

[0004] The adaptability of simple electronic logic control is insufficient: Existing electronic monitoring equipment mostly adopts a fixed threshold judgment mode, which has three major problems: First, the misjudgment rate is high. Fixed pressure thresholds (such as underpressure set at 0.8±0.2kPa and overpressure set at 8±2kPa) cannot adapt to the dynamic fluctuations of residential gas consumption (such as the natural pressure drop caused by concentrated gas consumption during the morning peak, and the instantaneous pressure change when gas equipment is switched on and off). It is easy to misjudge normal gas consumption fluctuations as abnormalities, or react slowly to slow anomalies close to the threshold (such as pressure gradually decreasing from 1.0kPa to 0.6kPa). Second, the decision-making dimension is singular, relying only on pressure parameters and ignoring related characteristics such as flow rate, temperature, and concentration (such as if overpressure is accompanied by a change in flow rate, it may be an abnormality of gas equipment rather than a true overpressure), resulting in insufficient judgment accuracy. Third, the linkage and evolution capabilities are lacking. Most equipment adopts a local logic fixed design, which cannot realize remote alarm push and strategy update, and has poor adaptability to different community scenarios (pressure fluctuations caused by aging pipelines in old communities and stable gas supply in new communities), requiring manual calibration of thresholds for each household.

[0005] The special needs of residential scenarios have not been met: residential gas use is characterized by "small fluctuations, high frequency, and diverse scenarios". Conventional technologies are unable to accurately capture minute pressure changes at the level of 0.2 kPa, and are even less able to formulate differentiated handling strategies for the gradient risks of "underpressure warning-underpressure" and "overpressure warning-overpressure". They often fall into the dilemma of "over-control (normal fluctuations accidentally shut off valves, affecting gas use)" or "insufficient response (delayed handling of real anomalies)".

[0006] Therefore, there is an urgent need for a device that can accurately identify the default underpressure range of 0.8±0.2kPa and overpressure range of 8±2kPa in residential scenarios. This default range is configurable. Existing devices generally suffer from problems such as slow response, high misjudgment rate, or inconvenient updates when identifying small gas volumes, sudden events, or remote strategy linkages. Summary of the Invention

[0007] (a) Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides an intelligent decision-making method based on real-time pressure monitoring of gas state changes, which solves the problems of slow response, high misjudgment rate, or inconvenient updates in existing equipment when identifying small gas volumes, sudden events, or remote strategy linkages.

[0009] (II) Technical Solution

[0010] To achieve the above objectives, the present invention provides the following technical solution: an intelligent decision-making method based on real-time pressure monitoring of sudden changes in gas state, comprising the following steps:

[0011] Step 1: Multimodal Data Acquisition: Focusing on Residents' Perception Needs in Various Scenarios;

[0012] The core monitoring target for residential gas systems is the pressure of the inlet pipe, which requires the targeted configuration of sensing equipment;

[0013] Step 2: Multi-dimensional data preprocessing: Enhancing signal preservation within the pressure threshold range;

[0014] Pressure fluctuations in residential scenarios are mostly small and high-frequency. Preprocessing needs to reduce noise while avoiding the loss of threshold boundary signals below 0.1 kPa.

[0015] Step 3: CUSUM mutation detection: Anchored undervoltage / overvoltage threshold design logic:

[0016] Targeting the stress characteristics of residential scenarios, the baseline and threshold for mutation detection are optimized to ensure accurate capture of mutations that "approach 0.8±0.2kPa or 8±2kPa" at the default threshold, which is configurable.

[0017] Step 4: Spatiotemporal Feature Fusion and State Classification: Adding Undervoltage / Overvoltage Specific States:

[0018] Based on the original "on / off / micro-leakage / pipe burst leakage", four new residential scenario-specific statuses have been added: "underpressure warning", "underpressure", "overpressure warning" and "overpressure", with clear classification characteristics;

[0019] Step 5: Digital Twin-Driven Risk Assessment: Adapting to Risk Characteristics in Residential Scenarios

[0020] The core risks in residential scenarios are "underpressure causing gas appliances to shut down and leak" and "overpressure causing hose rupture," requiring adjustments to the risk assessment logic for underpressure / overpressure conditions.

[0021] Step Six: Layered Decision-Making and Coordinated Execution: Focusing on resident safety, gas usage convenience, and intelligent start-stop, determine the gas usage status at the downstream end of the valve based on pressure changes; close the valve when no gas is needed, and open the valve immediately when gas usage is detected.

[0022] Step 7: Evolution of Federated Learning Model: Enhancing Threshold Adaptability in Residential Scenarios;

[0023] The model training focuses on incorporating undervoltage / overvoltage samples from residential scenarios to optimize threshold recognition capabilities.

[0024] Preferably, in step one, a sensing node is deployed at the end of the gas meter pipe and the gas appliance connecting hose in the residential household. The sensing node includes a temperature and pressure sensor and a methane concentration sensor. The sensing node is connected to a 4G / 5G / NB network and uses "DC5V or 4 AA dry batteries + backup lithium battery" to ensure stable power supply in the home scenario. The collected data is uploaded to the server with a delay of ≤100ms.

[0025] Preferably, in step two, when using variational mode decomposition, the number of mode decompositions is set to 5, and only extreme noise with kurtosis values ​​< 2.5 is removed to ensure that the pressure signal at the threshold point is completely preserved;

[0026] The isolated forest algorithm only identifies invalid values ​​with pressure <0 kPa or >15 kPa, and does not judge fluctuating data in the 0.6-14 kPa range as abnormal, thus avoiding the accidental deletion of underpressure / overpressure warning signals;

[0027] Still using pressure data as a benchmark, the time deviation of flow and temperature data calculated based on differential pressure is corrected by the DTW algorithm, with a synchronization accuracy of ≤10ms, ensuring that "when the pressure reaches the threshold, the flow / temperature status can be verified synchronously".

[0028] Preferably, in step three, the "weighted average pressure of the last minute" is used as the baseline μ(t). Under normal operating conditions, the baseline is stable between 1.8-2.2 kPa. When the baseline drifts to 1.0-1.8 kPa or 2.2-8 kPa, the window is automatically shortened to 30 seconds to improve the baseline response speed.

[0029] Thresholds are set separately for "undervoltage direction" and "overvoltage direction":

[0030] Undervoltage direction: descent threshold k2 = 0.1 kPa;

[0031] Overpressure direction: rising threshold k1 = 0.5 kPa.

[0032] Preferably, in step four, the focus is on extracting the time trend of pressure approaching 0.6-1.0 kPa, and the flow characteristics are kept with "time characteristic weight 0.6, spatial characteristic weight 0.4", but for under-pressure / over-pressure states, "distance weight between pressure and threshold" is added, with the weight being higher the pressure is closer to 0.6 kPa or 10 kPa.

[0033] Preferably, in step five, the digital twin of the residential pipe system is simplified, focusing on modeling the "gas meter-hose-gas-use equipment" path, and the physical parameters are adapted to residential rubber hoses or stainless steel corrugated pipes.

[0034] Preferably, in step six, when the underpressure warning is level 2, a local audible and visual warning is issued, and the APP is pushed with the message "Gas supply pressure is low, it is recommended to check whether the gas-using equipment is normal";

[0035] When the underpressure level is 3, a continuous audible and visual warning will be issued, and the APP will push a message saying "Underpressure is low, gas supply may be insufficient, it is recommended to contact the gas company"; the valve will not be closed.

[0036] When a leak accompanied by a level 4 underpressure warning occurs, immediately close the valve, trigger a strong audible and visual alarm, and send an app notification saying "Underpressure and leak, valve closed, please ventilate and contact maintenance," while simultaneously reporting to the community gas management station; when an overpressure warning occurs at level 3, trigger a local audible and visual alarm, and send an app notification saying "Pressure too high, please pay attention to gas safety."

[0037] When the overpressure reaches level 4, immediately close the valve, trigger a strong audible and visual warning, and send a push notification to the app stating "Overpressure has occurred, valve has been closed, please contact the gas company for investigation." Simultaneously, report to the regional gas dispatch center. All response actions can be manually canceled by the user through the app. After the valve is closed, gas personnel must reset it on-site to ensure safety.

[0038] Preferably, in step seven, local training involves specific training for judging whether the pressure has reached a threshold.

[0039] Encrypted aggregation: Cloud-based aggregation of resident scenario data from multiple communities, with a focus on optimizing the k1 and k2 thresholds and state classification model of the CUSUM algorithm for accurate adaptation;

[0040] Model deployment: The optimized model is prioritized for distribution to cells with frequent undervoltage / overvoltage events.

[0041] (III) Beneficial Effects

[0042] Compared with existing technologies, this invention provides an intelligent decision-making method based on real-time pressure monitoring of sudden changes in gas state, which has the following beneficial effects:

[0043] 1. The intelligent decision-making method for sudden changes in gas state based on real-time pressure monitoring improves the accuracy of gas state recognition and control flexibility, adapts to the detection of small gas volumes, remote strategy synchronization, and sudden behavior response in multiple scenarios, enhances the safety, intelligence, and maintainability of equipment, and is applicable to gas control for residential users, commercial kitchens, and industrial fields. Detailed implementation manners

[0044] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0045] The present invention provides a new technical solution: an intelligent decision-making method for sudden changes in gas state based on real-time pressure monitoring, including the following steps:

[0046] Step 1: Multimodal data acquisition: Focus on the perception requirements of the residential scenario: The core monitoring target of the residential gas system is the pressure of the inlet pipeline (the normal pressure is about 2 kPa). The system detects the pressure difference through dual temperature and pressure sensors before and after the valve, and realizes flow metering by back-calculating the calibrated 'flow-pressure difference' relationship; The leakage detector can be connected to the device through a reserved interface (optional), and the perception device needs to be configured specifically.

[0047] By two temperature and pressure sensors before and after the valve, the relationship between flow and pressure difference is calibrated, and the flow metering is realized:

[0048] Specific operation:

[0049] Deploy perception nodes at the rear pipeline of the gas meter in the residential household and the front end of the connecting hose of the gas-using equipment, including:

[0050] Temperature and pressure sensor: Adopt a 'pin' shape layout (3 measuring points, spaced 3 cm), with a range of 0 - 15 kPa (covering the underpressure / overpressure threshold range of 0.6 - 14 kPa), an accuracy of ±0.05 kPa, a sampling frequency of 10 Hz (to ensure capturing minute pressure fluctuations at the 0.2 kPa level), and can also measure the pipe wall temperature (-10°C - 60°C, accuracy ±0.3°C), indirectly reflecting the usage state of the gas-using equipment;

[0051] Methane concentration sensor: Detect the concentration in the kitchen environment (0 - 1 in the kitchen environment (0 - 100% LEL, detection lower limit ≤ 1% LEL), to prevent leakage risks.

[0052] The sensing nodes use 4G / 5G / NB networks (supporting cascading of ≤50 households per building) and employ "DC5V or 4 AA dry batteries + backup lithium battery (power outage duration ≥4 hours)" to ensure stable power supply in home scenarios; collected data is uploaded to the server with a latency of ≤100ms.

[0053] Step 2: Multi-dimensional data preprocessing: Enhancing signal preservation within the pressure threshold range.

[0054] Pressure fluctuations in residential settings are mostly small and high-frequency (such as sudden pressure changes caused by switching on and off gas appliances). Preprocessing needs to reduce noise while avoiding the loss of threshold boundary signals below 0.1 kPa.

[0055] Specific operations:

[0056] Noise Removal: When using Variational Mode Decomposition (VMD), set the number of mode decompositions to 5 (to reduce signal loss near the threshold caused by over-decomposition), and remove only extreme noise with a kurtosis value <2.5 (such as >1kHz interference generated by vibration of gas equipment) to ensure that the pressure signals at threshold points such as 0.6kPa, 1.0kPa, 8kPa, and 10kPa are completely preserved.

[0057] Outlier handling: The isolated forest algorithm only identifies invalid values ​​(sensor failure) with pressure <0 kPa or >15 kPa, and does not judge fluctuating data in the 0.6-14 kPa range as abnormal, thus avoiding the accidental deletion of underpressure / overpressure warning signals;

[0058] Time synchronization: Still based on pressure data, the time deviation of flow and temperature data calculated based on differential pressure is corrected by the DTW algorithm, with a synchronization accuracy of ≤10ms, ensuring that "when the pressure reaches the threshold, the flow / temperature status can be synchronously verified".

[0059] Step 3: CUSUM mutation detection: Anchored undervoltage / overvoltage threshold design logic:

[0060] Targeting the stress characteristics of residential scenarios, the focus is on optimizing the baseline and threshold for mutation detection to ensure accurate capture of mutations "approaching 0.8±0.2kPa or 8±2kPa":

[0061] Specific operations:

[0062] Dynamic baseline construction: The baseline μ(t) is based on the "pressure weighted average of the last 1 minute" (the weight of the most recent 30 seconds of data is 0.7, and the weight of the last 30 seconds of data is 0.3). Under normal operating conditions, the baseline is stable between 1.8-2.2 kPa. When the baseline drifts to 1.0-1.8 kPa (close to the lower limit of underpressure warning) or 2.2-8 kPa (close to the lower limit of overpressure warning), the window is automatically shortened to 30 seconds to improve the baseline response speed.

[0063] Dual threshold cumulative sum calculation: Thresholds are set separately for "underpressure direction" and "overpressure direction":

[0064] Based on 14 days of gas pressure monitoring data from 100 households, the standard deviation of pressure fluctuation in the underpressure direction was calculated to be σ_under = 0.05 kPa, and the standard deviation of pressure fluctuation in the overpressure direction was calculated to be σ_over = 0.25 kPa. To balance the false positive and false negative rates, a safety factor of 2 was adopted, and the underpressure drop threshold was set to k2 = 2 × σ_under = 0.1 kPa (adapting to an underpressure fluctuation range of 0.2 kPa), and the overpressure rise threshold was set to k1 = 2 × σ_over = 0.5 kPa (adapting to an overpressure fluctuation range of 2 kPa). Experimental verification showed that this threshold setting can ensure that the false positive rate of underpressure / overpressure sudden change detection is ≤0.5%, and the false negative rate is ≤0.1%. Underpressure direction: drop threshold k2 = 0.1 kPa (adapting to an underpressure fluctuation range of 0.2 kPa).

[0065] The cumulative sum formula is S⁻(t)=max[0,S⁻(t-1)+(μ(t)-P(t)-k2)]; when S⁻(t)≥0.2kPa (corresponding to the fluctuation range of 0.6-1.0kPa in the undervoltage warning range), it is marked as "potential sudden change in the undervoltage direction";

[0066] Overpressure direction: rising threshold k1=0.5kPa (adapting to an overpressure fluctuation range of 2kPa), cumulative sum formula S⁺(t)=max[0,S⁺(t-1)+(P(t)-μ(t)-k1)];

[0067] When S⁺(t)≥1kPa (corresponding to the fluctuation range of 6.5-10kPa in the overpressure warning range), it is marked as "potential sudden change in overpressure direction";

[0068] Validity verification: Calculate the first derivative of pressure at potential mutation points |dp / dt|: Underpressure direction: |dp / dt|≥0.05kPa / s (such as pressure rebound caused by sudden shutdown of gas-using equipment, or pressure slow drop caused by insufficient gas supply), confirmed as a valid mutation; Overpressure direction: |dp / dt|≥0.2kPa / s (such as rapid overpressure caused by a sudden increase in upstream pressure), confirmed as a valid mutation.

[0069] Step 4: Spatiotemporal Feature Fusion and State Classification: Adding Undervoltage / Overvoltage Specific States:

[0070] In addition to the existing "on / off / micro-leakage / pipe burst leakage" statuses, four new residential scenario-specific statuses have been added: "underpressure warning," "underpressure," "overpressure warning," and "overpressure," with clearly defined classification characteristics.

[0071] Specific operations:

[0072] Feature extraction:

[0073] For under-pressure related states: focus on extracting the time trend of pressure approaching 0.6-1.0 kPa (such as continuously decreasing to below 0.6 kPa), and the reverse flow characteristics (when under-pressure occurs, the reverse flow of gas-using equipment will decrease to ≤0.5 m³ / h simultaneously).

[0074] For overpressure-related conditions: focus on extracting the time trend of pressure approaching 6.5-10 kPa (such as a sudden increase or continuous rise to above 10 kPa), and temperature characteristics (if there is no back-calculated flow change and no significant fluctuation in pipe wall temperature during overpressure, false overpressure caused by the use of gas-using equipment can be ruled out).

[0075] Feature fusion: Maintain the weight of "time feature 0.6, spatial feature 0.4", but add "distance weight between pressure and threshold" for under-pressure / over-pressure states (e.g., the closer the pressure is to 0.6 kPa or 10 kPa, the higher the weight).

[0076] State output: Expanded to 8 states, the core resident scenario state definitions are as follows:

[0077] Low pressure warning: Pressure 0.6-1.0 kPa, flow rate ≤ 1 m³ / h, no abnormal concentration;

[0078] Underpressure: Pressure ≤ 0.6 kPa, Flow rate ≤ 0.5 m³ / h, Duration ≥ 30 seconds;

[0079] Overpressure warning: Pressure 6.5-10 kPa, flow rate = 0 (no gas used), pipe wall temperature stable;

[0080] Overpressure: Pressure ≥ 10 kPa, Flow rate = 0, Duration ≥ 10 seconds;

[0081] Leakage accompanied by underpressure: pressure ≤0.6kPa, concentration ≥5%LEL, flow rate with slight fluctuations (≤0.1m³ / h).

[0082] The classification accuracy has been optimized to ≥99.5% for residential scenarios, with a focus on reducing the misclassification rate of undervoltage / overvoltage and normal fluctuations.

[0083] Step 5: Digital Twin-Driven Risk Assessment: Adapting to Risk Characteristics in Residential Scenarios

[0084] The core risks in residential scenarios are "underpressure causing gas appliances to shut down and leak" and "overpressure causing hose rupture," requiring adjustments to the risk assessment logic for underpressure / overpressure conditions.

[0085] Specific operations:

[0086] Twin model construction: Simplify the digital twin of the pipeline in the residential household, focusing on modeling the path of "gas meter-hose-gas-use equipment" (pipe diameter DN15, length 2-5m), and adapting the physical parameters to residential rubber hoses or stainless steel corrugated pipes (wall thickness 2mm, design pressure 16kPa). Gas-use equipment includes, but is not limited to, stoves, water heaters and wall-mounted boilers.

[0087] Risk quantification: Based on the formula R = α × pressure deviation + β × duration + γ × concentration value, adjustment coefficients are applied for underpressure / overpressure.

[0088] Under-pressure conditions: α=0.4 (the greater the pressure is below 0.6 kPa, the higher the deviation), β=0.3 (the longer the duration, the higher the risk of flameout), γ=0.3 (if the concentration is abnormal, the risk of leakage will be compounded).

[0089] Undervoltage warning (0.6-1.0 kPa): R=0.2-0.4 (Level 2 risk);

[0090] Undervoltage (≤0.6kPa, no abnormal concentration): R=0.4-0.6 (Level 3 risk);

[0091] Leakage accompanied by undervoltage (≤0.6kPa + concentration ≥5%LEL): R=0.7-0.9 (Level 4 risk);

[0092] Overpressure conditions: α=0.5 (the higher the pressure is above 10 kPa, the higher the deviation), β=0.4 (the longer the duration, the higher the risk of hose rupture), γ=0.1 (low concentration weight when there is no leakage).

[0093] Overpressure warning (6.5-10 kPa): R=0.3-0.5 (Level 3 risk);

[0094] Overpressure (≥10kPa): R=0.6-0.8 (Level 4 risk).

[0095] Step Six: Layered Decision-Making and Coordinated Execution: Focusing on Resident Safety, Gas Usage Convenience, and Intelligent Start-up and Shutdown:

[0096] The core functions of the intelligent control valve are: real-time judgment of the gas consumption status at the downstream end of the valve based on pressure changes; automatic closure of the built-in valve when no gas demand is detected; and immediate opening of the valve when a gas pressure fluctuation signal is detected, so as to achieve safety protection and energy-saving control in unattended operation, with a response delay of ≤100ms.

[0097] When the low pressure warning (0.6-1.0kPa) is level 2, a local audible and visual warning will be issued (green light flashing + buzzer once every 10 seconds), and a push notification will be sent to the APP to indicate "Gas supply pressure is low, it is recommended to check whether the gas-using equipment is normal";

[0098] When the underpressure (≤0.6kPa, no leakage) is level 3, a continuous audible and visual warning will be issued (yellow light flashing + buzzer once every 5 seconds), and the APP will push a message "Underpressure has occurred, gas supply may be insufficient, it is recommended to contact the gas company"; do not close the valve (to avoid affecting the subsequent restoration of gas supply);

[0099] When a leak accompanied by a level 4 underpressure warning occurs, immediately close the valve, triggering a strong audible and visual alarm (flashing red light + continuous buzzer), and the app will send a notification saying "Underpressure and leak, valve closed, please ventilate and contact maintenance," while simultaneously reporting to the community gas management station; when an overpressure warning (6.5-10kPa) is level 3, a local audible and visual alarm will be triggered (flashing yellow light + buzzer once every 5 seconds), and the app will send a notification saying "Pressure is too high, please pay attention to gas safety";

[0100] When the overpressure (≥10kPa) reaches level 4, the valve will be immediately closed, a strong audible and visual warning will be issued, and the APP will push a message saying "Overpressure has occurred, valve has been closed, please contact the gas company for investigation," and the regional gas dispatch center will be notified simultaneously. All response actions can be manually canceled by the user through the APP (if it is confirmed that the underpressure is a temporary gas supply fluctuation). After the valve is closed, gas personnel must reset it on-site to ensure safety. This function can determine the gas usage at the downstream end of the valve based on pressure changes. If no gas is needed, the valve will be closed; if gas usage is detected, the valve will be opened immediately.

[0101] Step 7: Federated Learning Model Evolution: Enhancing Threshold Adaptability in Resident Scenarios

[0102] The model training focuses on incorporating undervoltage / overvoltage samples from residential scenarios to optimize threshold recognition capabilities.

[0103] Specific operations:

[0104] Local training: Each building's server collects ≥500 labeled data points daily (including real-world scenario samples such as "underpressure warning - insufficient actual gas supply" and "overpressure warning - upstream fluctuations"), and conducts specialized training for judging whether the pressure has reached the 0.6kPa / 10kPa threshold. The training is conducted on the server side via 4G / NB network, and the device side can run small models, simplified versions / core function small models.

[0105] Encrypted aggregation: Cloud-based aggregation of residential scenario data from multiple communities, with a focus on optimizing the k1 and k2 thresholds of the CUSUM algorithm and the "underpressure / overpressure feature weights" of the state classification model, to ensure accurate adaptation to different communities (such as pressure fluctuations caused by aging pipelines in old communities and stable gas supply in newly built communities).

[0106] Model Deployment: The optimized model is prioritized for distribution to communities with frequent undervoltage / overvoltage issues. It is quickly adapted through "small sample fine-tuning" (50 local data points iterating 10 times). After 50 rounds of federated learning, the accuracy rate of undervoltage / overvoltage identification in residential scenarios is ≥99.3%.

[0107] The adjusted method is fully adapted to the 0.8±0.2kPa underpressure and 8±2kPa overpressure threshold requirements for residential scenarios. Through a closed loop of "sensing-processing-detection-classification-evaluation-execution-optimization", it achieves precise control over underpressure, overpressure and related risks in residential gas systems, taking into account both safety and convenience of gas use.

[0108] This invention improves the accuracy of gas status identification and control flexibility, adapts to small gas volume detection, remote strategy synchronization and emergency behavior response in multiple scenarios, enhances the safety, intelligence and maintainability of the equipment, and is suitable for gas control in residential users, commercial kitchens and industrial fields.

[0109] Based on a risk assessment model driven by digital twins, underpressure, overpressure, and related conditions are divided into 2-4 levels of risk, with corresponding differentiated handling strategies. For level 2 underpressure warnings, only a prompt is triggered without shutting off the valve, while for level 4 leaks accompanied by underpressure, the valve is shut off immediately and a coordinated report is filed, avoiding the "one-size-fits-all" control mode of traditional technology. At the same time, users can manually review abnormal conditions through the APP, which not only ensures the bottom line of safety but also reduces unnecessary gas interruptions and improves the user experience for residents.

[0110] Example: Resident scene parameters, dynamic baseline and threshold calculation example of the CUSUM algorithm:

[0111] Taking a residential user as an example, their 14-day gas consumption data shows:

[0112] Normal standby pressure: 1.8-2.2 kPa (average μ=2.0 kPa);

[0113] Pressure fluctuation range during morning rush hour (7:00-9:00): 1.6-2.0 kPa (standard deviation σ1=0.15 kPa);

[0114] Pressure fluctuation range during nighttime standby (22:00-6:00): 1.9-2.1 kPa (standard deviation σ2=0.05 kPa);

[0115] Multi-parameter fusion weights and calculation examples:

[0116] Fusion formula: Mutation confidence = Pressure contribution value × 0.5 + Flow rate contribution value × 0.3 + Concentration contribution value × 0.2;

[0117] Pressure contribution value: If the pressure deviates from the baseline by 0.3 kPa, assign a value of 0.8; if it deviates by 0.1-0.3 kPa, assign a value of 0.4; if it deviates by <0.1 kPa, assign a value of 0.1.

[0118] Flow contribution value: When there is no gas usage, if the flow rate is >0.05 m³ / h, assign a value of 0.8; when there is gas usage, if the flow rate matches the pressure change, assign a value of 0.2.

[0119] Concentration contribution value: ≥5%LEL, assign 1.0; 1%-5%LEL, assign 0.5; <1%LEL, assign 0.1;

[0120] Example: During the morning rush hour, the pressure deviates from the baseline by 0.2 kPa (contribution 0.4), the flow rate is 0.08 m³ / h (contribution 0.8), and the concentration is 0.5% LEL (contribution 0.1). Then the confidence level of the abrupt change is 0.4 × 0.5 + 0.8 × 0.3 + 0.1 × 0.2 = 0.46 (the alarm threshold of 0.6 is not reached, so no alarm is triggered).

[0121] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent decision making based on real-time pressure monitoring of gas state change, characterized in that: Includes the following steps: Step 1: Multimodal Data Acquisition: Focusing on Residents' Perception Needs in Various Scenarios; The core monitoring target for residential gas systems is the pressure of the inlet pipe, which requires the targeted configuration of sensing equipment; Step 2: Multi-dimensional data preprocessing: Enhancing signal preservation within the pressure threshold range; Pressure fluctuations in residential scenarios are mostly small and high-frequency. Preprocessing needs to reduce noise while avoiding the loss of threshold boundary signals below 0.1 kPa. Step 3: CUSUM mutation detection: anchoring undervoltage / overvoltage threshold design logic; In response to the stress characteristics of residential scenarios, the baseline and threshold of mutation detection were optimized to ensure accurate capture of mutations that "approach 0.8±0.2kPa or 8±2kPa"; Step 4: Spatiotemporal Feature Fusion and State Classification: Add specific states for undervoltage / overvoltage; Based on the existing "on / off / micro-leakage / pipe burst leakage", four new residential scenario-specific statuses have been added: "underpressure warning", "underpressure", "overpressure warning" and "overpressure", with clear classification characteristics; Step 5: Digital Twin-Driven Risk Assessment: Adapting to Risk Characteristics in Residential Scenarios; The core risks in residential scenarios are "underpressure causing gas appliances to shut down and leak" and "overpressure causing hose rupture," requiring adjustments to the risk assessment logic for underpressure / overpressure conditions. Step Six: Layered Decision-Making and Coordinated Execution: Focusing on Residents' Safety and Gas Convenience; Step 7: Evolution of Federated Learning Model: Enhancing Threshold Adaptability in Residential Scenarios; The model training focuses on incorporating undervoltage / overvoltage samples from residential scenarios to optimize threshold recognition capabilities.

2. The method as claimed in claim 1, wherein the method is based on real-time pressure monitoring for intelligent decision making for gas state sudden change. In step one, a sensing node is deployed at the end of the gas meter pipe and the front end of the gas appliance connecting hose in the residential household. The sensing node includes a temperature and pressure sensor and a methane concentration sensor. The sensing node is connected to a 4G / 5G / NB network and uses "DC5V or 4 AA dry batteries + backup lithium battery" to ensure stable power supply in the home scenario. The collected data is uploaded to the server with a delay of ≤100ms.

3. The method as claimed in claim 1, wherein the method is based on real time pressure monitoring gas condition abrupt change intelligent decision making. In step two, when using variational mode decomposition, the number of mode decompositions is set to 5, and only extreme noise with kurtosis values ​​< 2.5 is removed to ensure that the pressure signal at the threshold point is completely preserved. The isolated forest algorithm only identifies invalid values ​​with pressure <0 kPa or >15 kPa, and does not judge fluctuating data in the 0.6-14 kPa range as abnormal, thus avoiding the accidental deletion of underpressure / overpressure warning signals; Still using pressure data as a benchmark, the time deviation of flow and temperature data calculated based on differential pressure is corrected by the DTW algorithm, with a synchronization accuracy of ≤10ms, ensuring that "when the pressure reaches the threshold, the flow / temperature status can be verified synchronously".

4. The method as claimed in claim 1, wherein the method is based on real time pressure monitoring gas condition abrupt change intelligent decision making. In step three, the "weighted average pressure of the last minute" is used as the baseline μ(t). Under normal operating conditions, the baseline is stable between 1.8-2.2 kPa. When the baseline drifts to 1.0-1.8 kPa or 2.2-8 kPa, the window is automatically shortened to 30 seconds to improve the baseline response speed. Thresholds are set separately for "underpressure direction" and "overpressure direction".

5. The real-time pressure based intelligent decision making method for gas state change according to claim 1, wherein: In step four, the focus is on extracting the time trend of pressure approaching 0.6-1.0 kPa, and the flow characteristics are kept with "time characteristic weight 0.6, spatial characteristic 0.4", but for under-pressure / over-pressure states, "distance weight between pressure and threshold" is added. The closer the pressure is to 0.6 kPa or 10 kPa, the higher the weight.

6. The real-time pressure based intelligent decision making method for gas state sudden change as claimed in claim 1, wherein: In step five, the digital twin of the residential pipe system is simplified, focusing on modeling the "gas meter-hose-gas-use equipment" path, and the physical parameters are adapted to residential rubber pipes or stainless steel corrugated pipes.

7. The intelligent decision-making method based on real-time pressure monitoring and sudden changes in gas state as described in claim 1, characterized in that: In step six, when the underpressure warning is level 2, a local audible and visual warning is issued, and the APP is pushed with the message "Gas supply pressure is low, it is recommended to check whether the gas-using equipment is normal"; When the underpressure level is 3, there will be a continuous audible and visual warning, and the APP will push a message saying "Underpressure is low, gas supply may be insufficient, it is recommended to contact the gas company"; the valve will not be closed. When a leak accompanied by a level 4 underpressure warning occurs, immediately close the valve, trigger a strong audible and visual alarm, and send an app notification saying "Underpressure and leak, valve closed, please ventilate and contact maintenance," while simultaneously reporting to the community gas management station; when an overpressure warning occurs at level 3, trigger a local audible and visual alarm, and send an app notification saying "Pressure too high, please pay attention to gas safety." When the overpressure reaches level 4, immediately close the valve, trigger a strong audible and visual warning, and send a push notification to the app saying "Overpressure has occurred, valve has been closed, please contact the gas company for investigation." Simultaneously, report to the regional gas dispatch center. All response actions can be manually canceled by the user through the app. After the valve is closed, gas personnel must reset it on-site to ensure safety.

8. The real-time pressure based gas condition abrupt change intelligent decision method according to claim 1, characterized in that: In step seven, local training involves specific training for judging whether the pressure has reached the threshold. Encrypted aggregation: Cloud-based aggregation of resident scenario data from multiple communities, with a focus on optimizing the k1 and k2 thresholds and state classification model of the CUSUM algorithm for accurate adaptation; Model deployment: The optimized model is prioritized for distribution to cells with frequent undervoltage / overvoltage events.