Sensing data fusion and abnormal dynamic weight adjustment method for transformer fire

Abstract

The invention relates to a transformer fire sensing data fusion and abnormal dynamic weight adjustment method, and belongs to the field of hydrogen production. The method comprises a data fusion module and a data preprocessing and feature extraction module. Dynamically adjusting the weight of each sensor according to the real-time change condition of the signals after data preprocessing and feature extraction of different sensors; and calculating a comprehensive fire risk score in real time by fusing data from each sensor and the dynamically adjusted weight of each sensor, and judging whether a fire occurs or not. According to the invention, various sensor data are combined, the weight of the sensor data is dynamically adjusted, accurate detection of the transformer fire is realized, and the accuracy, reliability and real-time response capability of the system are optimized.

Description

Technical Field

[0001] The invention belongs to the field of fire detection in ultra-high voltage substations, and in particular relates to a method for sensor data fusion and abnormal dynamic weight adjustment of transformer fire. Background Art

[0002] Transformers are important equipment in power systems and are widely used in important places such as substations and distribution networks. Due to the long-term operation of transformers, there are fire hazards such as overheating, overload, and electrical failure. Traditional transformer fire detection systems usually rely on single sensor monitoring, such as temperature sensors or smoke sensors. Although these sensors can provide certain fire warning functions, the data from a single sensor often cannot fully and accurately reflect the fire situation of the transformer. In addition, factors such as environmental changes and equipment aging may also affect the accuracy of the sensor, resulting in false alarms or missed alarms. Therefore, how to reduce false alarms and missed alarms of alarm events through the fusion of multiple sensor data and combined with a dynamically adjusted weight strategy has become the key to solving the current technical bottleneck. Summary of the invention

[0003] The present invention proposes a sensor data fusion and abnormal dynamic weight adjustment method for transformer fire, aiming to solve the problem of insufficient accuracy and real-time performance of traditional fire detection methods in transformer fire monitoring. By comprehensively using data from multiple sensors (such as temperature, smoke, gas concentration, sound, etc.) and combining with a dynamic weight adjustment algorithm, the present invention can accurately identify fire events in real time, thereby significantly improving the sensitivity, accuracy and response speed of fire detection. The present invention is not only highly adaptable, but also can effectively reduce false alarms and missed alarms caused by environmental changes or equipment problems.

[0004] A method for sensor data fusion and abnormal dynamic weight adjustment of transformer fire, comprising a data fusion module and a data preprocessing and feature extraction module;

[0005] The data fusion module fuses multiple sensor data to provide multi-dimensional perception information for fire detection;

[0006] The data preprocessing and feature extraction module performs data preprocessing and feature extraction on the sensor data processed by the data fusion module;

[0007] The method comprises the following steps:

[0008] Dynamically adjust the weight of each sensor according to the real-time changes of the signals after data preprocessing and feature extraction of different sensors;

[0009] By fusing the data from each sensor and the dynamically adjusted weights of each sensor, a comprehensive fire risk score is calculated in real time to determine whether a fire has occurred.

[0010] Preferably, a dynamic weight adjustment strategy is also included, specifically:

[0011] 1) Sensor priority: the priority of different sensors is dynamically adjusted according to the ability to identify fire events;

[0012] 2) Real-time data feedback: The validity and credibility of each sensor data parameter at the current moment are obtained by real-time monitoring of the changing trend, stability and historical data of sensor data;

[0013] 3) Weighted fusion: The alarm signals of each sensor are weighted and summed according to their weight values.

[0014] Preferably, dynamically adjusting the weight of each sensor includes: initial weight setting, dynamic weight adjustment, weight updating and weight normalization.

[0015] Preferably, the initial weight setting is specifically:

[0016] When the system is initialized, an initial weight is set for each sensor; the initial weight is set based on the performance of the sensor, the sensor type, and its importance in fire detection.

[0017] Preferably, the dynamic weight adjustment is specifically:

[0018] Sensor signal change rate: When the sensor signal changes rapidly, the weight of the sensor will be automatically increased;

[0019] Sensor signal stability: If a sensor outputs a stable signal for a long time, it means that the sensor is in normal working condition and is not interfered by the outside world. At this time, the weight of the sensor is increased to improve the detection accuracy;

[0020] Historical reliability of sensors: Evaluate the performance of each sensor in different environments through historical data analysis. When a sensor performs stably and accurately in a specific environment, its weight will be automatically increased. On the contrary, if a sensor frequently reports false alarms or misses alarms, its weight will be automatically reduced.

[0021] Preferably, the weight update is specifically:

[0022] Calculate the real-time weight. The real-time calculation formula for the weight of each sensor is:

[0023]

[0024] in:

[0025] ω i (t) is the real-time weight of sensor i at time t;

[0026] ωi (t-1) is the weight of sensor i at the previous moment;

[0027] α is the adjustment factor, which is used to control the speed of weight adjustment;

[0028] is the weight change of sensor i at time t;

[0029] Calculation of weight change, Combining the sensor's real-time signal change rate with its historical stability:

[0030]

[0031] Among them: β is the rate of change factor, γ is the stability factor; i (t) is the difference change rate of the sensor data parameter signal of sensor i within T1 seconds; Stablity i (t) Calculated as the number of reliable alarms per T2 month / the total number of alarms per T2 month.

[0032] Preferably, the weight normalization is specifically as follows:

[0033] The sensor weights are normalized after each update so that the sum of all sensor weights is always 1:

[0034]

[0035] Where N is the total number of sensors.

[0036] Preferably, the calculation formula for the comprehensive fire risk score is as follows:

[0037]

[0038] in:

[0039] Risk Score(t) is the comprehensive fire risk score at time t;

[0040] S i (t) is the alarm signal of sensor i at time t.

[0041] Preferably, when the comprehensive fire risk score exceeds a set threshold, the system determines it as a fire event and activates an alarm mechanism.

[0042] Preferably, the plurality of sensors include: a temperature sensor, a smoke sensor, a CO2 concentration sensor, a noise sensor and a vibration sensor.

[0043] Beneficial technical effects:

[0044] The present invention can identify fire events in real time and accurately, thereby significantly improving the sensitivity, accuracy and response speed of fire detection, enhancing the fire protection level of UHV substations, and improving the safety of power systems; in addition, it can effectively reduce false alarms and missed alarms caused by environmental changes or equipment problems, and reduce the probability of false operation of UHV substations. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0046] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention.

[0047] Figure 2 This is a flow chart of dynamically adjusting the weight of each sensor according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0049] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0050] Embodiment 1

[0051] 1. Multi-sensor data fusion module.

[0052] In order to comprehensively and accurately monitor the risk of transformer fire, the sensor data fusion and abnormal dynamic weight adjustment method of transformer fire of the present invention includes a multi-sensor data fusion module, which combines the real-time data of multiple sensors such as temperature sensor, smoke sensor, CO2 concentration sensor, noise sensor and vibration sensor. These sensors monitor environmental variables such as temperature, gas change, smoke concentration and noise around the transformer, respectively, and can provide multi-dimensional perception information for fire detection.

[0053] Preferably, the sensor data includes: a temperature sensor provides a temperature alarm signal TS (Temperature Signal, TS for short) of the transformer, a smoke detector is used to provide a smoke alarm signal SS (Smoke Signal, SS for short) of the transformer, a CO2 concentration sensor is used to provide a CO2 concentration alarm signal CS of the transformer to determine whether the transformer releases a large amount of carbon dioxide due to overheating or failure (Concentration Signal, CS for short), a noise sensor is used to provide a noise alarm signal NS of the transformer to capture abnormal sounds of the equipment (Noise Signal, NS for short), and a vibration sensor is used to provide a vibration alarm signal VS of the transformer to capture abnormal vibration changes of the equipment (Vibration Signal, VS for short).

[0054] By fusing data from different sensors, the present invention can provide more comprehensive fire warnings and avoid the situation where a single sensor cannot cover all conditions for fire occurrence.

[0055] Optionally, the data fusion module has multiple input ports, receives multiple sensor data simultaneously, and optionally performs analog-to-digital conversion. The output port of the data fusion module transmits the multiple sensor data to the data preprocessing and feature extraction module.

[0056] 2. Sensor data preprocessing and feature extraction module.

[0057] Preferably, a sensor data preprocessing and feature extraction module is also included, which eliminates sensor errors through a series of data preprocessing steps for subsequent fire detection and judgment. The specific steps are as follows:

[0058] (1) Four statistical features, namely mean, standard deviation, maximum value and minimum value, are set for the five sensor data parameters TS, SS, CS, NS and VS to describe the overall level and fluctuation degree of the signal.

[0059] (2) Set the instantaneous rate of change (Rate of Change) for the two sets of sensor data parameters NS and VS to describe the rising or falling trend of potential risks. The instantaneous rate of change = (current value - previous value) / time difference between the current value and the previous value.

[0060] (3) Correlation coefficients are set for TS and SS in pairs; correlation coefficients are set for NS and VS in pairs to assist in identifying the multi-sensor synergy effect when a fire occurs.

[0061] Correlation coefficient = Cov(TS,SS) / ((D(TS)) 0.5 (D(SS)) 0.5 ), Cov(TS,SS) is the covariance of TS and SS, D(TS) and D(SS) are the variances of TS and SS respectively.

[0062] 3. Dynamic weight adjustment algorithm

[0063] The present invention introduces a dynamic weight adjustment algorithm to automatically adjust the weight of each sensor according to the real-time changes in the input signals of different sensors. Traditional fire detection methods usually fix the weight of each sensor, but the present invention uses dynamic weight adjustment to enable the system to more intelligently deal with false alarms and missed alarms caused by environmental changes, sensor failures or other factors.

[0064] 3.1 Dynamic Weight Adjustment Strategy

[0065] The dynamic weight adjustment strategy is:

[0066] Sensor priority: The priority of different sensors is dynamically adjusted according to the ability to identify fire events. Optionally, TS and SS are set to have higher weights for alarm identification in the early stage of a fire. For example, the weights of TS and SS are set to 0.9 and 0.95 respectively, the weight of CS is set to 0.8 in the early stage of a fire, and the weights of NS and VS are both set to 0.7 in the early stage of a fire; the weights of TS and SS are set to 0.8 and 0.8 respectively in the middle stage of a fire, the weight of CS is set to 0.9 in the middle stage of a fire, and the weights of NS and VS are both set to 0.85 in the middle stage of a fire. Preferably, NS and VS have higher weights in equipment failure identification, for example, the weights of NS and VS are set to 0.8 and 0.85 respectively in equipment failures caused by fire events.

[0067] Real-time data feedback: By real-time monitoring of the changing trends, stability and historical data of sensor data, the validity and credibility of each sensor data parameter at the current moment are calculated, and its weight is adjusted dynamically.

[0068] Weighted fusion: The alarm signals of each sensor are weighted and summed according to their weight values ​​to obtain a comprehensive fire risk score (Risk Score). The comprehensive fire risk score is used to judge transformer fires.

[0069] 3.2 Weight calculation and update mechanism

[0070] The key to dynamic weight adjustment is how to calculate the weight of each sensor in real time and update these weights according to real-time data. The specific steps are as follows:

[0071] 3.2.1 Initial weight setting

[0072] When the system is initialized, an initial weight is set for each sensor. The initial weight is set based on the performance of the sensor, the sensor type, and its importance in fire detection. Optionally, the sensor includes a temperature sensor, a smoke sensor, a CO2 concentration sensor, a noise sensor, and a vibration sensor.

[0073] Preferably, the initial weights of TS, SS, CS, NS and VS decrease in sequence.

[0074] Preferably, the initial weights of TS and SS are greater than the initial weights of CS, NS and VS. For example, the initial weights of TS, SS, CS, NS and VS are 0.4, 0.3, 0.2, 0.15 and 0.25 respectively.

[0075] Preferably, the initial weights of TS and SS are greater than the sum of the initial weights of CS, NS and VS. For example, the initial weights of TS and SS are 0.4 and 0.5 respectively, and the sum of the initial weights of CS, NS and VS is 0.15+0.1+0.1=0.35.

[0076] Optionally, temperature sensor (TS): the initial weight is set to 0.3;

[0077] Smoke sensor (SS): The initial weight is set to 0.25;

[0078] CO2 concentration sensor (CS): the initial weight is set to 0.2;

[0079] Noise sensor (NS): the initial weight is set to 0.15;

[0080] Vibration sensor (VS): The initial weight is set to 0.10.

[0081] 3.2.2 Dynamic Weight Adjustment

[0082] The steps for dynamic adjustment of weights are as follows:

[0083] Sensor signal change rate: When the sensor signal changes rapidly, the weight of the sensor will be automatically increased. This is because sudden changes in the signal usually indicate the occurrence of fire or fault.

[0084] Signal change rate = (current value - previous value) / time difference between current value and previous value.

[0085] The rapid change of the sensor signal may be: the sensor signal change rate exceeds a set threshold.

[0086] The weight of the sensor will be automatically increased as follows: the weight of the sensor will be automatically increased according to the set step size, for example, the set step size is 0.05 or 0.1; the weight of the sensor will be automatically increased by percentage, for example, if the previous weight of the sensor is a, then increasing by percentage means increasing by (1+b%) a, where b is a non-zero positive number.

[0087] Signal stability of the sensor: If a sensor outputs a stable signal for a long time, it means that the sensor is in normal working condition and is not disturbed by the outside world. At this time, the weight of the sensor can be increased to improve the detection accuracy. Optionally, the weight of the sensor will be automatically increased by a set step size, for example, the step size is set to 0.05 or 0.1; the weight of the sensor will be automatically increased by a percentage. For example, if the weight of the sensor is a before, the percentage increase means increasing by (1+b%) a, where b is a non-zero positive number.

[0088] Historical reliability of sensors: Through historical data analysis, the performance of each sensor in different environments is evaluated. When a sensor performs stably and accurately in a specific environment, its weight will be automatically increased; conversely, if a sensor frequently falsely reports or misses a signal, its weight will be automatically reduced. Specific environments can be fire, electrical failure, set time period, etc.

[0089] 3.2.3 Weight Update

[0090] 1. Calculate real-time weight: Design the real-time calculation formula of the weight of each sensor:

[0091]

[0092] in:

[0093] ω i (t) is the real-time weight of sensor i at time t;

[0094] ω i (t-1) is the weight of sensor i at the previous moment;

[0095] α is the adjustment factor, which is used to control the speed of weight adjustment, and the adjustment range is 0.05 to 0.1;

[0096] is the weight change of sensor i at time t.

[0097] 2. Calculation of weight change: Combining the sensor's real-time signal change rate with its historical stability:

[0098]

[0099] Among them: β is the rate of change factor, γ is the stability factor; i (t) is the rate of change of the difference of the sensor data parameter signal of sensor i within T1 seconds, that is, the difference of the change / T1, optionally, T1 is 3, 4, 5, 6 or 3 to 10 seconds; Stablity i (t) Stability of the sensor data parameter signal of sensor i in the past T2 months, optionally, T2 is 1, 2, 3, 4, 5, 6 or 1 to 6, optionally, Stablity i (t) can be calculated as the number of reliable alarms per T2 month / the total number of alarms per T2 month.

[0100] Preferably, for the temperature sensor (TS): temperature change is an important indication signal in fire detection, so a higher rate of change response and a higher stability factor are required;

[0101] The β of TS is taken as 0.15 (rate of change factor, moderate value, sensitive to sharp temperature rise);

[0102] The γ of TS is taken as 0.25 (stability factor, sensors with higher stability can take higher weights).

[0103] Preferably, for smoke sensors (SS): smoke sensors are crucial for early detection of fires, and both the rate of change and stability should be high;

[0104] The β of SS is taken as 0.2 (rate of change factor, smoke sensor reacts quickly when a fire occurs);

[0105] The γ of SS is taken as 0.2 (stability factor, moderate stability value).

[0106] Preferably, for the CO2 sensor (CS): the CO2 sensor is more sensitive to environmental changes after a fire, so its rate of change should be relatively low and its stability should be high;

[0107] The β of CS is taken as 0.1 (the rate of change factor, a lower rate of change factor adapts to the characteristics of the steady growth of CO2 concentration);

[0108] The γ of CS is taken as 0.3 (stability factor, a higher stability factor avoids weight fluctuations due to environmental changes).

[0109] Preferably, a noise sensor (NS): the noise sensor is greatly affected by the environment, so a lower rate of change and stability factor are required;

[0110] The β of NS is taken as 0.05 (rate of change factor, a lower rate of change factor to reduce excessive response to environmental noise);

[0111] The γ of NS is taken as 0.1 (stability factor, the less stable the noisy sensor is, the lower the stability factor is).

[0112] Preferably, a vibration sensor (VS): a vibration sensor has a low responsiveness and is usually used to detect equipment failure or vibration signals;

[0113] The β of VS is taken as 0.05 (rate of change factor, the vibration signal changes less and responds slower);

[0114] The γ of VS is taken as 0.1 (stability factor, a lower stability factor is suitable for the equipment operating environment).

[0115] 3.2.4 Weight Normalization

[0116] To avoid a sensor having too high or too low weight, the weights of all sensors are normalized after each update so that the sum of all sensor weights is always 1:

[0117]

[0118] Where N is the total number of sensors.

[0119] 3.3 Comprehensive Fire Risk Score (Risk Score)

[0120] By integrating the data from various sensors and their dynamically adjusted weights, the present invention can calculate a comprehensive fire risk score in real time to determine whether a fire has occurred. The calculation formula for the comprehensive fire risk score is as follows:

[0121]

[0122] in:

[0123] Risk Score(t) is the comprehensive fire risk score at time t;

[0124] S i (t) is the alarm signal of sensor i at time t. Optionally, when sensor i sends an alarm signal at time t, S i(t) = 1, when sensor i does not send out an alarm signal at time t, S i (t)=0.

[0125] When the comprehensive risk score exceeds the set threshold, the system determines it as a fire incident and activates the alarm mechanism.

[0126] 4. Alarm and emergency response mechanism

[0127] When the system determines that a fire event has occurred, the fire detection system of the present invention can quickly send out an alarm signal and start the emergency response mechanism. The alarm system transmits the fire alarm to the on-duty personnel and the emergency response team through buzzers, flashing lights, text messages, etc. At the same time, the system can also automatically control the transformer equipment to shut down or open the switch according to the nature of the fire, reducing the risk of fire spreading.

[0128] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0129] It should be noted that the serial numbers of the embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0130] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, and any combination of embodiments or schemes are also included in the patent protection scope of the present invention.

Claims

1. A method for sensor data fusion and abnormal dynamic weight adjustment of transformer fire, characterized in that: It includes data fusion module and data preprocessing and feature extraction module; The data fusion module fuses multiple sensor data to provide multi-dimensional perception information for fire detection; The data preprocessing and feature extraction module performs data preprocessing and feature extraction on the sensor data processed by the data fusion module; The method comprises the following steps: Dynamically adjust the weight of each sensor according to the real-time changes of the signals after data preprocessing and feature extraction of different sensors; By fusing the data from each sensor and the dynamically adjusted weights of each sensor, a comprehensive fire risk score is calculated in real time to determine whether a fire has occurred.

2. The method according to claim 1, characterized in that It also includes dynamic weight adjustment strategies, specifically: 1) Sensor priority: the priority of different sensors is dynamically adjusted according to the ability to identify fire events; 2) Real-time data feedback: The validity and credibility of each sensor data parameter at the current moment are obtained by real-time monitoring of the changing trend, stability and historical data of sensor data; 3) Weighted fusion: The alarm signals of each sensor are weighted and summed according to their weight values.

3. The method according to claim 2, characterized in that Dynamically adjusting the weight of each sensor includes: initial weight setting, dynamic weight adjustment, weight update and weight normalization.

4. The method according to claim 3, characterized in that The initial weight setting is specifically: When the system is initialized, an initial weight is set for each sensor; the initial weight is set based on the performance of the sensor, the sensor type, and its importance in fire detection.

5. The method according to claim 3, characterized in that The dynamic weight adjustment is specifically as follows: Sensor signal change rate: When the sensor signal changes rapidly, the weight of the sensor will be automatically increased; Sensor signal stability: If a sensor outputs a stable signal for a long time, it means that the sensor is in normal working condition and is not interfered by the outside world. At this time, the weight of the sensor is increased to improve the detection accuracy; Historical reliability of sensors: Evaluate the performance of each sensor in different environments through historical data analysis. When a sensor performs stably and accurately in a specific environment, its weight will be automatically increased. On the contrary, if a sensor frequently reports false alarms or misses alarms, its weight will be automatically reduced.

6. The method according to claim 3, characterized in that The weight update is specifically as follows: Calculate the real-time weight. The real-time calculation formula for the weight of each sensor is: in: ω i (t) is the real-time weight of sensor i at time t; ω i (t-1) is the weight of sensor i at the previous moment; α is the adjustment factor, which is used to control the speed of weight adjustment; is the weight change of sensor i at time t; Calculation of weight change, Combining the sensor's real-time signal change rate with its historical stability: Among them: β is the rate of change factor, γ is the stability factor; i (t) is the difference change rate of the sensor data parameter signal of sensor i within T1 seconds; Stablity i (t) Calculated as the number of reliable alarms per T2 month / the total number of alarms per T2 month.

7. The method according to claim 3, characterized in that The weight normalization is specifically as follows: The sensor weights are normalized after each update so that the sum of all sensor weights is always 1: Where N is the total number of sensors.

8. The method according to claim 1, characterized in that The calculation formula for the comprehensive fire risk score is as follows: in: Risk Score(t) is the comprehensive fire risk score at time t; S i (t) is the alarm signal of sensor i at time t.

9. The method according to claim 1, characterized in that When the comprehensive fire risk score exceeds the set threshold, the system determines it as a fire incident and activates the alarm mechanism.

10. The method according to claim 1, characterized in that The multiple sensors include: temperature sensor, smoke sensor, CO2 concentration sensor, noise sensor and vibration sensor.

Images