Mine underground carbon monoxide low-delay sensing method based on monitoring big data

Abstract

The invention discloses a mine underground carbon monoxide low-delay sensing method based on monitoring big data. A used system module comprises a multi-source sensor access module, a top-speed sensing module, a long-term sensing module and a hierarchical storage module. According to the invention, all input data streams which are received in real time and are accessed to a sensor are stored in a hot storage unit in a streaming processing mode, data which exceed one day are transmitted to a long-term sensing module, and the data are compressed through the long-term sensing module and then are stored in a cold storage unit. The top speed sensing module distributes required data one by one according to requirements set by downstream subscribers, when the subscribers request data for more than one day, the long-term sensing module recovers the compressed data and sends the data to the top speed sensing module, and then the top speed sensing module distributes the data to the downstream subscribers; according to the invention, long-term dynamic perception of the behavior of carbon monoxide in a mine well can be realized, so that rapid response and effective information transmission during catastrophe can be realized on the premise of early overrun warning in advance.

Description

Technical Field

[0001] The present invention relates to a method for sensing carbon monoxide in a mine, and in particular to a method for sensing carbon monoxide in a mine with low delay based on monitoring big data, which belongs to the field of mine safety technology. Background Art

[0002] Carbon monoxide (CO) is extremely hazardous, primarily because it is colorless, odorless, non-irritating, and poorly soluble in water. It easily binds to human hemoglobin, causing hypoxia and poisoning. It also easily accumulates in confined or poorly ventilated environments. Therefore, in relatively enclosed underground mine environments, CO poses a significant threat to the safety of underground workers and the normal operation of mines. Even low concentrations of CO can cause poisoning, and in severe cases, death. The formation of CO in mines is primarily related to incomplete combustion and high-temperature decomposition reactions of carbonaceous materials. For example, CO is generated by the incomplete combustion of carbonaceous materials such as coal, wood, and explosives in the absence of sufficient oxygen. It is also generated by excessively rich or incomplete combustion of diesel engines in underground excavators and transport vehicles during high-temperature operation. Carbon dioxide (CO2) in coal seam spontaneous combustion, fires, or high-temperature rock formations (such as geothermal areas) reacts with hot carbon to form CO through reduction reactions. And, during the explosion of ammonium nitrate explosives (such as emulsion explosives), some carbon chains decompose and react with limited oxygen to form CO. Therefore, CO2-related disasters can occur very quickly during fires and explosions, or they can develop over a long period of time during slow processes such as coal rock oxidation.

[0003] With the increasing requirements for coal mine monitoring and the advancement of mine intelligence, a large number of diverse sensors are deployed throughout mines to monitor key safety indicators such as CO, gas, oxygen, and temperature in real time. The current Coal Mine Safety Regulations stipulate that the maximum allowable concentration of CO in underground airflow is only 0.0024%, but this limit can easily be exceeded in a very short period of time during a disaster. Therefore, timely and effective information from underground monitoring big data is a key prerequisite for rapid disaster response. Collecting and storing historical data from monitoring systems is an important means of identifying long-term CO risks.

[0004] In recent years, underground mine monitoring and surveillance systems have developed significantly. However, with the increase in the number of sensors and the increase in data traffic, the system response time has gradually lengthened, becoming an increasingly restrictive factor in rapid early warning. At the same time, the massive amount of monitoring data accumulated over a long period of time not only brings huge storage burdens and organizational difficulties to mine operations, but also reduces the accuracy of early warnings to a certain extent. Summary of the Invention

[0005] In response to the problems existing in the above-mentioned existing technologies, the present invention provides a low-latency carbon monoxide perception method in mines based on monitoring big data. For mine monitoring systems with massive, multi-source, and heterogeneous data sources, it can achieve long-term dynamic perception of mine CO behavior to facilitate early warning of over-limit situations, and achieve extremely rapid response and transmission of effective information in the event of a disaster.

[0006] To achieve the above objectives, the system modules used in this low-latency carbon monoxide perception method in mines based on monitoring big data include a multi-source sensor access module, an extreme-speed perception module, a long-term perception module, and a hierarchical storage module. The multi-source sensor access module includes hardware access channels and communication protocols for relevant sensors. The extreme-speed perception module includes a streaming processing engine, a subscription distribution component, and a computing device A. The long-term perception module includes a data verification component, a data compression component, a database, a data recovery component, and a computing device B. The hierarchical storage module includes a hot storage unit with fast read and write speeds and a cold storage unit with no read and write speed requirements.

[0007] The low-latency carbon monoxide sensing method in mines based on monitoring big data specifically includes the following steps:

[0008] Step 1: Receive input data streams from all connected sensors in real time through the multi-source sensor access module;

[0009] Step 2: The speed sensing module processes the input data in a streaming manner and stores it in a hot storage unit. The data is then distributed to each downstream subscriber one by one according to the requirements set by the subscribers, as follows:

[0010] Step 2-1: Perform fast real-time preprocessing on the event data stream input by the sensor to verify the integrity of the data format. Each event data contains at least the sensor unique number, signal category identification code, timestamp, and signal value.

[0011] Step 2-2: Store each event data in the hot storage unit;

[0012] Step 2-3: Check whether the incoming data meets the request conditions of the downstream subscribers one by one, send the qualified data to the downstream subscribers, and dynamically update the subscribers and their request rules;

[0013] Step 3: Based on the timestamp, data older than one day is transferred to the long-term perception module;

[0014] Step 4: The long-term perception module compresses the data and stores it in the cold storage unit, as follows:

[0015] Step 4-1: Check and mark missing values or outliers in the data. The outlier judgment rule is that the CO concentration value exceeds the preset threshold range or the CO concentration value is negative;

[0016] Step 4-2: Classify and compress according to the sensor's sensing content;

[0017] Step 4-3: Save the data in the database. The database is divided into several data tables according to different topics. Each sensor event has a unique primary key, which is used to associate different data tables with each other.

[0018] Step 4-4: Store the database in a cold storage unit;

[0019] Step 5: When a subscriber requests data for more than one day, the long-term perception module restores the compressed data and sends it to the fast perception module, which then distributes it to downstream subscribers.

[0020] Furthermore, Step 4-2 is as follows:

[0021] Step 4-2-1: For CO concentration, non-zero storage is implemented, and more than 10 consecutive zero-value events are removed, and only the data of the non-zero period is retained;

[0022] Step 4-2-2: For oxygen concentration and air volume, differential storage is implemented. The missing parts are first marked with metadata and filled with the previous valid value, as shown below:

[0023]

[0024] Where: x1, x2, …, x n is the sensor time series data, in which there are missing values; t is the time point recorded by the sensor;

[0025] Calculate the following difference of the filled data and save it:

[0026]

[0027] Where: v t is the difference between adjacent moments;

[0028] Step 4-2-3: Perform discrete wavelet transform on gas concentration

[0029]

[0030] Where: n is the index of the target reconstructed signal; k is the index of the wavelet coefficient of the next level; x 0,L [n] = x 0,H [n] = x[n] is the input signal; g[n] and h[n] are the discrete signals after low-pass and high-pass filters; x α,L [n] and x α,H[n] are low-frequency and high-frequency signals in the αth layer architecture respectively;

[0031] Then the difference is calculated and the difference sequence is compressed by Huffman coding: a Huffman tree is constructed, and for high-frequency numerical fields, the tree is traced from the root node to the leaf node; during the tracing process, when moving to the left branch, 0 is appended to the end of the current code, and when moving to the right branch, 1 is appended to the end of the current code.

[0032] Furthermore, Step 5 is as follows:

[0033] Step 5-1: For CO concentration, find the time discontinuity position in the data, fill in the missing segments based on the timestamps of the beginning and end of the discontinuity, and fill the signal value with zero;

[0034] Step 5-2: For oxygen concentration and air volume, the initial value x1 = v1. At other times, the data after filling is gradually restored according to the following formula:

[0035] x t =x t-1 +v t (t≥2)

[0036] Then, all filled positions are located according to the abnormal marks in the metadata, and the filled values are restored to the missing state to obtain the original valid values;

[0037] Step 5-3: For gas concentration, first use the inverse Huffman code to return to the state after wavelet transformation, and then use the corresponding reconstruction filter and Perform upsampling and filtering, and merge to restore the original signal:

[0038]

[0039] Where: is a low-pass reconstruction filter; Reconstruct the filter for the high pass.

[0040] Furthermore, when each event data is stored in the hot storage unit in Step 2-2, storage partitions are divided in the hot storage unit, and each event data retains one master copy and at least one backup copy in the hot storage unit; when the database is stored in the cold storage unit in Step 4-4, storage partitions are divided in the cold storage unit, and the database retains one master copy and at least one backup copy in the cold storage unit.

[0041] Furthermore, when Step 2 streams the input data through the extreme speed perception module, the streaming input data uses a sliding time window to achieve atomic processing of data batches.

[0042] Furthermore, Step 2 also includes Step 2-4: randomly selecting the sensor event self-check execution time, and if the execution time from Step 2-1 to Step 2-3 exceeds 50ms, a warning is issued to prompt the use of a system device with a larger bandwidth or a distributed computing device A.

[0043] Furthermore, Apache Kafka is used as the stream processing engine in Step 2.

[0044] Compared with existing technologies, this low-latency CO sensing method for mines based on big data monitoring utilizes the collaborative operation of a multi-source sensor access module, an ultra-fast sensing module, a long-term sensing module, and a hierarchical storage module. This method processes the real-time input data streams from all connected sensors and stores them in a hot storage unit using a streaming process. Data older than one day, as determined by timestamps, is transferred to the long-term sensing module, which compresses and stores the data in a cold storage unit. The ultra-fast sensing module distributes the required data one by one according to the requirements set by downstream subscribers. When a subscriber requests data older than one day, the long-term sensing module recovers the compressed data and sends it to the ultra-fast sensing module, which then distributes it to downstream subscribers. This method enables long-term dynamic sensing of CO behavior in mines, facilitating early warning of limit violations. It also significantly reduces system response time, enabling rapid responses and effective information delivery in the event of a disaster. It also significantly alleviates the storage burden and organization difficulties of massive amounts of monitoring data. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the system modules and data flow of the present invention. DETAILED DESCRIPTION

[0046] The present invention will be further described below with reference to the embodiments and accompanying drawings.

[0047] like Figure 1As shown, the system modules used in the low-latency carbon monoxide perception method in mines based on monitoring big data include a multi-source sensor access module, an ultra-fast perception module, a long-term perception module, and a hierarchical storage module; the multi-source sensor access module is used to receive the input data streams of all sensors in real time. The multi-source sensor access module can be a data receiving method that includes hardware access channels and communication protocols for all mine-related sensors and is directly electrically connected to each sensor. The multi-source sensor access module can also be a data receiving method that is electrically connected to a switch that uniformly manages sensor signals; the ultra-fast perception module includes a streaming processing engine, a subscription distribution component, an independent computing device A, and a necessary operating system; the long-term perception module includes a data verification component, a data compression component, a database, a data recovery component, an independent computing device B, and a necessary operating system; the hierarchical storage module includes a hot storage unit with a faster read and write rate and a cold storage unit with no read and write rate requirements.

[0048] This method for low-latency carbon monoxide sensing in mines based on big data monitoring specifically includes the following steps:

[0049] Step 1: Receive the input data streams of all connected sensors in real time through the multi-source sensor access module.

[0050] Step 2: The speed sensing module processes the input data in a streaming manner and stores it in a hot storage unit. The data is then distributed one by one according to the requirements set by downstream subscribers. The details are as follows:

[0051] Step 2-1: Perform fast real-time preprocessing on the event data stream input by the sensor to verify the integrity of the data format. Each event data contains at least the sensor unique number, signal category identification code, timestamp, and signal value.

[0052] Step 2-2: Each event data is stored in a hot storage unit. The hot storage unit can be divided into storage partitions. Each event data is retained in the hot storage unit with one master copy and at least one backup copy.

[0053] Step 2-3: Check whether the incoming data meets the request conditions of the downstream subscribers one by one, send the qualified data to the downstream subscribers, and dynamically update the subscribers and their request rules.

[0054] Streaming input data can use sliding time windows to achieve atomic processing of data batches.

[0055] To further reduce the time delay, the sensor event self-test execution time can be randomly selected. If the response time for executing the above steps exceeds 50ms, a warning will be issued to prompt the use of system equipment with larger bandwidth or distributed independent computing device A to reduce the time delay in each step.

[0056] Step 3: Based on the timestamp, data older than one day is transferred to the long-term perception module.

[0057] Step 4: The data is compressed and stored in the cold storage unit through the long-term perception module. The details are as follows:

[0058] Step 4-1: Check and mark missing values or outliers in the data. The outlier judgment rule is that the CO concentration value exceeds the preset threshold range or the CO concentration value is negative.

[0059] Step 4-2: Classify and compress the data based on the sensor's sensing content, as follows:

[0060] Step 4-2-1: For CO concentration, non-zero storage is implemented, and more than 10 consecutive zero-value events are removed, and only the data of the non-zero period is retained.

[0061] Step 4-2-2: For oxygen concentration and air volume, differential storage is implemented. The missing parts are first marked with metadata and filled with the previous valid value, as shown below:

[0062]

[0063] Where: x1, x2, …, x n is the sensor time series data, in which there are missing values; t is the time point recorded by the sensor.

[0064] Calculate the following difference of the filled data and save it:

[0065]

[0066] Where: v t is the difference between adjacent moments.

[0067] Step 4-2-3: Perform discrete wavelet transform on gas concentration

[0068]

[0069]

[0070] Where: n is the index of the target reconstructed signal (i.e. the number of data points to be calculated); k is the index of the wavelet coefficient of the next level; x 0,L [n] = x 0,H [n] = x[n] is the input signal; g[n] and h[n] are the discrete signals after low-pass and high-pass filters; x α,L [n] and x α,H [n] are the low-frequency signal and high-frequency signal in the α-th layer architecture respectively.

[0071] Then the difference is calculated and the difference sequence is compressed by Huffman coding: a Huffman tree is constructed, and for numerical fields that appear frequently, trace from the root node to the leaf node; in the process of tracing downward, when moving to the left branch, 0 is appended to the end of the current code, and when moving to the right branch, 1 is appended to the end of the current code; the more frequently appearing fields will be in the upper layer and the shorter the code, and the less frequently appearing fields will be in the lower layer and the longer the code.

[0072] Step 4-3: Save the data in a database. The database is divided into several data tables according to different themes such as sensor category and main location. The data represented by each sensor event may be stored in multiple data tables at the same time, but each sensor event has a unique primary key. Each sensor event can be linked between different data tables through the primary key.

[0073] Step 4-4: Store the database in a cold storage unit. The cold storage unit can be divided into storage partitions. The database retains one master copy and at least one backup copy in the cold storage unit.

[0074] Step 5: When a subscriber requests data for more than one day, the long-term perception module recovers the compressed data and sends it to the fast perception module, which then distributes it to downstream subscribers. The details are as follows:

[0075] Step 5-1: For CO concentration, find the time discontinuity position in the data, fill in the missing segments based on the timestamps of the start and end of the discontinuity, and fill the signal value with zero.

[0076] Step 5-2: For oxygen concentration and air volume, the initial value x1 = v1. At other times, the data after filling is gradually restored according to the following formula:

[0077] x t =x t-1 +v t (t≥2)

[0078] Then, all filled positions are located according to the abnormal marks in the metadata, and the filled values are restored to the missing state to obtain the original valid values.

[0079] Step 5-3: For gas concentration, first use the inverse Huffman code to return to the state after wavelet transformation, and then use the corresponding reconstruction filter and Perform upsampling and filtering, and merge to restore the original signal:

[0080]

[0081] Where: is a low-pass reconstruction filter; Reconstruct the filter for the high pass.

[0082] The following is an example of applying the low-latency carbon monoxide sensing method in a mine based on monitoring big data.

[0083] The data streams generated by a CO sensor and a nearby methane sensor on January 22, 2024, are shown in Tables 1 and 2 below, respectively. The time column in the table indicates the moment when the sensor data is transmitted to the system module of the present invention.

[0084] Table 1 Monitoring values of a CO sensor over a period of time

[0085]

[0086] Table 2 Monitoring values of a certain exhaust gas sensor over a period of time

[0087]

[0088]

[0089] This example uses Apache Kafka as the stream processing engine. A downstream subscriber request is created for "CO and gas sensor data when the CO concentration is greater than or equal to 5 ppm." The data received by the downstream subscriber is shown in bold in Tables 1 and 2, with both a false negative rate and a false positive rate of 0%. The system module in this example receives data from over 70 sensors, with an average data throughput of 2.4 data items per second. The average latency for subscriber-requested data within the system module is 18 milliseconds. Using conventional batch processing methods under the same test conditions, the average latency is over 10 seconds.

[0090] This low-latency carbon dioxide (CO) sensing method for mines based on big data monitoring utilizes the collaborative operation of a multi-source sensor access module, an ultra-fast sensing module, a long-term sensing module, and a hierarchical storage module. It streams the real-time input data streams from all connected sensors and stores them in a hot storage unit. Data older than one day, as determined by timestamps, is transferred to the long-term sensing module, which compresses and stores the data in a cold storage unit. The ultra-fast sensing module distributes the required data one by one according to the requirements set by downstream subscribers. When a subscriber requests data older than one day, the long-term sensing module recovers the compressed data and sends it to the ultra-fast sensing module, which then distributes it to downstream subscribers. This method enables long-term dynamic sensing of CO2 behavior in mines, facilitating early warning of limit violations. It also significantly reduces system response time, enabling rapid responses and effective information delivery in the event of a disaster. It also significantly alleviates the storage burden and organization difficulties of massive monitoring data.

Claims

1. A low-latency carbon monoxide sensing method in mines based on monitoring big data, characterized in that: The system modules used include a multi-source sensor access module, an extreme-speed perception module, a long-term perception module, and a hierarchical storage module; the multi-source sensor access module includes the hardware access channel and communication protocol for the relevant sensors; the extreme-speed perception module includes a streaming processing engine, a subscription distribution component, and a computing device A; the long-term perception module includes a data verification component, a data compression component, a database, a data recovery component, and a computing device B; the hierarchical storage module includes a hot storage unit with a fast read / write speed and a cold storage unit with no read / write speed requirements; The low-latency carbon monoxide sensing method in mines based on monitoring big data specifically includes the following steps: Step 1: Receive input data streams from all connected sensors in real time through the multi-source sensor access module; Step 2: The speed sensing module processes the input data in a streaming manner and stores it in a hot storage unit. The data is then distributed to each downstream subscriber one by one according to the requirements set by the subscribers, as follows: Step 2-1: Perform fast real-time preprocessing on the event data stream input by the sensor to verify the integrity of the data format. Each event data contains at least the sensor unique number, signal category identification code, timestamp, and signal value. Step 2-2: Store each event data in the hot storage unit; Step 2-3: Check whether the incoming data meets the request conditions of the downstream subscribers one by one, send the qualified data to the downstream subscribers, and dynamically update the subscribers and their request rules; Step 3: Based on the timestamp, data older than one day is transferred to the long-term perception module; Step 4: The long-term perception module compresses the data and stores it in the cold storage unit, as follows: Step 4-1: Check and mark missing values or outliers in the data. The outlier judgment rule is that the CO concentration value exceeds the preset threshold range or the CO concentration value is negative; Step 4-2: Classify and compress according to the sensor's sensing content; Step 4-3: Save the data in the database. The database is divided into several data tables according to different topics. Each sensor event has a unique primary key, which is used to associate different data tables with each other. Step 4-4: Store the database in a cold storage unit; Step 5: When a subscriber requests data for more than one day, the long-term perception module restores the compressed data and sends it to the fast perception module, which then distributes it to downstream subscribers.

2. The method for low-latency sensing of carbon monoxide in mines based on monitoring big data according to claim 1 is characterized in that: Step 4-2 is as follows: Step 4-2-1: For CO concentration, non-zero storage is implemented, and more than 10 consecutive zero-value events are removed, and only the data of the non-zero period is retained; Step 4-2-2: For oxygen concentration and air volume, differential storage is implemented. The missing parts are first marked with metadata and filled with the previous valid value, as shown below: Where: x1, x2, …, x n is the sensor time series data, in which there are missing values; t is the time point recorded by the sensor; Calculate the following difference of the filled data and save it: Where: v t is the difference between adjacent moments; Step 4-2-3: Perform discrete wavelet transform on gas concentration Where: n is the index of the target reconstructed signal; k is the index of the wavelet coefficient of the next level; x 0,L [n] = x 0,H [n] = x[n] is the input signal; g[n] and h[n] are the discrete signals after passing through the low-pass and high-pass filters; x α,L [n] and x α,H [n] are low-frequency and high-frequency signals in the αth layer architecture respectively; Then the difference is calculated and the difference sequence is compressed by Huffman coding: a Huffman tree is constructed, and for high-frequency numerical fields, the tree is traced from the root node to the leaf node; during the tracing process, when moving to the left branch, 0 is appended to the end of the current code, and when moving to the right branch, 1 is appended to the end of the current code.

3. The method for low-latency sensing of carbon monoxide in mines based on monitoring big data according to claim 2 is characterized in that: Step 5 is as follows: Step 5-1: For CO concentration, find the time discontinuity position in the data, fill in the missing segments based on the timestamps of the beginning and end of the discontinuity, and fill the signal value with zero; Step 5-2: For oxygen concentration and air volume, the initial value x1 = v1. At other times, the data after filling is gradually restored according to the following formula: x t =x t-1 +v t (t≥2) Then, all filled positions are located according to the abnormal marks in the metadata, and the filled values are restored to the missing state to obtain the original valid values; Step 5-3: For gas concentration, first use the inverse Huffman code to return to the state after wavelet transformation, and then use the corresponding reconstruction filter and Perform upsampling and filtering, and merge to restore the original signal: Where: is a low-pass reconstruction filter; Reconstruct the filter for the high pass.

4. The method for low-latency sensing of carbon monoxide in mines based on monitoring big data according to claim 1, 2 or 3, characterized in that: When storing each event data in the hot storage unit in Step 2-2, storage partitions are divided in the hot storage unit, and each event data retains one master copy and at least one backup in the hot storage unit; when storing the database in the cold storage unit in Step 4-4, storage partitions are divided in the cold storage unit, and the database retains one master copy and at least one backup in the cold storage unit.

5. The method for low-latency sensing of carbon monoxide in mines based on monitoring big data according to claim 1, 2 or 3, characterized in that: Step 2: When streaming input data through the extreme speed perception module, the streaming input data uses a sliding time window to achieve atomic processing of data batches.

6. The method for low-latency sensing of carbon monoxide in mines based on monitoring big data according to claim 1, 2 or 3, characterized in that: Step 2 also includes Step 2-4: randomly selecting the sensor event self-check execution time. If the response time from Step 2-1 to Step 2-3 exceeds 50ms, a warning is issued to prompt the use of a system device with a larger bandwidth or a distributed computing device A.

7. The method for low-latency sensing of carbon monoxide in mines based on monitoring big data according to claim 1, 2 or 3, characterized in that: In Step 2, Apache Kafka is used as the stream processing engine.

Images