Trend analysis method suitable for coal mine underground multi-type sensors

Abstract

The invention relates to a trend analysis method suitable for multiple types of sensors in an underground coal mine, and belongs to the field of data analysis. The method comprises the following steps: S1, data preparation and normalization: acquiring original time sequence data of a selected sensor, and normalizing the original time sequence data to obtain a normalized value; s2, performing multi-scale division on a normalized value set: dividing the normalized value set by adopting a multi-time-scale sliding window method; s3, feature calculation: calculating a sample standard deviation of a long-term window and linear regression slopes in a short-term window and the long-term window; s4, state classification based on rules: distributing a state statei for each time point ti according to the feature vectors calculated in the step S3 and a classification engine; and S5, outputting and merging: merging the state classification sequences output in the step S4 into an event list. According to the method, the historical data set with the state mark is automatically generated, the industrial AI training data bottleneck is cracked, and advanced model development such as predictive maintenance is accelerated.

Description

Technical Field

[0001] This invention belongs to the field of data analysis and relates to a method for analyzing time-series data from heterogeneous sensors in an intelligent coal mine environment. Background Technology

[0002] In the development of intelligent coal mining, to ensure production safety, a massive number of sensors are deployed in mines to monitor key environmental and equipment parameters in real time, including methane concentration, carbon monoxide concentration, temperature, humidity, and equipment operating status. These sensors generate a surge in the volume of continuous time-series data. Efficiently and accurately analyzing this massive amount of time-series data is crucial for ensuring safe mine operation, enabling predictive maintenance of equipment failures, and conducting precise accident tracing.

[0003] However, current methods for monitoring and analyzing this type of time series data have the following significant limitations:

[0004] (1) Static threshold alarm method: This is currently the most common monitoring method. This method presets fixed parameter limits, and triggers an alarm when the sensor's monitored value exceeds the limit. Its main drawbacks are: First, it is a passive response mechanism and cannot provide early warning of dangerous upward trends in values ​​that are rapidly approaching the threshold. Second, this method is very sensitive to instantaneous peak values ​​or noise in the data, and is prone to false alarms due to brief, non-dangerous fluctuations. This not only increases the workload of the back-end operators but may also lead to "alarm fatigue," causing operators to become less sensitive to real alarms.

[0005] (2) Manual trend visual inspection: This method relies on experienced technicians or dispatch center operators to visually observe data curves to determine whether there are abnormal trends. The drawbacks of this method are quite obvious: First, it is highly dependent on the subjective experience of personnel, and the judgment standards are inconsistent, resulting in low consistency of interpretation; second, manual monitoring requires a large amount of manpower and is labor-intensive, especially when the number of sensors is large, making it difficult to achieve full coverage; third, it has poor scalability and cannot adapt to the ever-increasing data monitoring needs.

[0006] (3) Complex AI Prediction Models: Although prediction models based on artificial intelligence algorithms such as deep learning have shown great potential in trend analysis, they face two major obstacles in practical applications. The first is the "black box" problem, that is, the decision-making process of the model lacks interpretability. In high-risk fields such as coal mine safety, where decision-making requires high transparency, unexplainable warnings or judgments are difficult to gain the full trust of decision-makers. The second is the data dependency problem. These advanced models usually require massive amounts of precisely labeled training data to achieve ideal performance, but effective anomaly labeled data in safety-critical areas (such as coal mines) is itself very scarce and costly to obtain, and there is currently no mature automated labeling mechanism to solve this bottleneck.

[0007] In summary, existing technologies either fail to provide forward-looking insights due to overly simplistic mechanisms, are inefficient and inconsistent due to excessive reliance on manual intervention, or are difficult to implement in safety-critical fields due to complex models, lack of interpretability, and dependence on massive amounts of labeled data. Therefore, the market urgently needs an automated, scalable, and highly interpretable time-series data trend representation solution to overcome the limitations of single-threshold alarms and avoid the dependence of traditional machine learning on labeled data. Summary of the Invention

[0008] In view of this, the purpose of this invention is to provide a trend analysis method adapted to various types of sensors in underground coal mines, addressing the shortcomings of existing technologies such as the passive lag of static threshold methods, the subjective inefficiency of manual interpretation, and the dependence on annotations and lack of interpretability of black-box models. This method breaks through the limitation of a single threshold, identifying states such as constancy, gradual trends, and sudden events through dynamic behavior analysis, providing proactive and contextualized insights for operational safety; simultaneously, it avoids the dependence of traditional machine learning on labeled data, directly extracting understandable trend features from raw time-series data, meeting the stringent requirements for decision transparency in safety-critical fields.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A trend analysis method adapted to various types of sensors in coal mines transforms raw data from heterogeneous sensors into standardized features through range-adaptive data normalization and multi-window statistical feature extraction (such as local standard deviation and linear regression slope). Based on a rule engine, deterministic state classification (such as "constant," "gradual increase," and "sudden increase") is performed on time-series segments, achieving accurate transformation from raw data to human-readable event narratives. This solution breaks through the challenges of automating and interpreting dynamic behavior representation, supporting proactive early warning, equipment health monitoring, and accident tracing. Simultaneously, it generates high-quality labels for historical data, empowering supervised machine learning development and meeting the dual needs of safety-critical fields for human-machine collaborative decision-making and downstream automated processing.

[0011] The method specifically includes the following steps:

[0012] S1: Data preparation and normalization: Obtain the raw time series data of the selected sensor and normalize it to obtain the normalized value;

[0013] S2: Multi-scale partitioning of normalized value set: The normalized value set is partitioned using a multi-time-scale sliding window method;

[0014] S3: Feature Calculation: Calculate the sample standard deviation for the long-term window, as well as the linear regression slope within the short-term and long-term windows;

[0015] S4: Rule Calculation: Based on the rule-based deterministic classification engine, the feature vector calculated in step S3 is used for each time point t. i Assign a state i ;

[0016] S5: Output Merging: Merge the state classification sequence output in step S4 into an event list.

[0017] Furthermore, in step S1, the data preparation specifically includes: Let D be the original time-series dataset of the selected sensor, represented as a timestamp-value pair sequence: D = {(t1, v1), (t2, v2), ..., (t... n ,v n )}, where t i v is the timestamp of the i-th measurement. i Let i be the corresponding sensor value, i = 1, 2, ..., n; assume the measured values ​​are sampled at a fixed frequency; key parameter c sensor_type These are the minimum and maximum values ​​that the sensor can record: c sensor_type ={R min ,R max}, where R min and R max These represent the minimum and maximum value ranges for the selected sensor type, respectively.

[0018] Furthermore, in step S1, for each value v in the dataset D... i The normalized value v′ is calculated using the Min-Max normalization formula. i :

[0019] v′ i =(v i -R min ) / (R max -R min )

[0020] Generate a new, normalized time series dataset D′: D′={(t1,v′1),(t2,v′2),…(t n,v′ n All subsequent analyses were performed on this normalized dataset D′ to ensure that the logic and thresholds are universal across all types of sensors.

[0021] Furthermore, in step S1, the key parameters are stored in repository C, which is used to store the operating parameters for each sensor type.

[0022] Furthermore, in step S2, the multi-scale partitioning of the normalized value set specifically includes: defining two window sizes: a short-term window W. s Size N s 10 data points are used to capture fast transient events; a long-term window W l Size N l N data points l >N s It is used to identify gradual, persistent trends while smoothing out small fluctuations; for each time point t in the sequence i (where i≥N) l ), calculate in t i A set of statistical characteristics of data points within a window ending at time;

[0023] Let s s(i) and s l(i) They are respectively t i The normalized value sets within the short and long windows at the endpoint:

[0024]

[0025] Where i≥N l .

[0026] Furthermore, in step S3, the long-term window W is calculated. l The sample standard deviation σ is used to measure the stability or volatility of the signal;

[0027]

[0028] Where, μ l It is set s l(i) The arithmetic mean of the medians.

[0029] Furthermore, in step S3, calculating the linear regression slope specifically includes: calculating the linear regression slope in both the short-term and long-term windows; letting x... j The time index for data points within the window;

[0030]

[0031] Where, m s(i) and m l(i)These represent the linear regression slopes within the short-term and long-term windows, respectively; sum Σ by iterating through all points j and x within each window. j It is the index of a point within the window, v′ j These are normalized sensor values.

[0032] Furthermore, in step S4, the classification engine is a set of predefined adjustable thresholds: the slope threshold T of the short window. slope_sudden The slope threshold T of the long window slope_slow The standard deviation threshold T for long windows std_const ;

[0033] The rules are as follows:

[0034] If the linear regression slope m within the short-term window s(i) Greater than threshold T slope_sudden Then the state i It was marked as "sudden rise";

[0035] Otherwise, if m s(i) Less than negative T slope_sudden (i.e. -T) slope_sudden If the state is 0, then the state is 0. i It was marked as "sudden drop";

[0036] Otherwise, if the standard deviation σ within the long-term window l(i) Less than threshold T std_const Then the state i Marked as "constant";

[0037] Otherwise, if the linear regression slope m within the long window l(i) Greater than threshold T slope_slow Then the state i It was marked as "slow and steady rise";

[0038] Otherwise, if m l(i) Less than negative T slope_slow (i.e. -T) slope_slow If the state is 0, then the state is 0. i It was marked as "slow and continuous decline";

[0039] Otherwise, the state is the default state. i It is marked as "fluctuating up and down".

[0040] Furthermore, step S5 specifically includes: iterating over the state sequence state. i It groups consecutive identical states into time periods; finally, it outputs a list of events, where each event is defined by a start time, an end time, and the characteristic states of that time period.

[0041] The beneficial effects of this invention are as follows:

[0042] (1) By employing multi-scale feature extraction and rule-based classification analysis, this invention breaks through the limitations of traditional static threshold methods, realizes dynamic monitoring and real-time analysis of complex underground coal mine environments, and effectively improves early warning efficiency.

[0043] (2) This invention does not rely on a large amount of manually labeled data, but directly extracts understandable trend features from the original time series data, which overcomes the data bottleneck problem of traditional machine learning methods and meets the strict requirements of decision transparency in safety-critical fields.

[0044] (3) This invention proposes a unified analysis framework that can adapt to different types of sensors, realize the effective integration and analysis of heterogeneous data, and solve the problem of lack of unified analysis standards in the prior art.

[0045] (4) By introducing a multi-scale analysis mechanism of short-term and long-term windows, this invention can flexibly process data at different time scales, improve the comprehensiveness and adaptability of trend analysis, and overcome the shortcomings of existing technologies in not considering multi-scale windows in the feature extraction process.

[0046] (5) This invention achieves refined analysis of complex trends by deeply exploring multi-dimensional features such as standard deviation and linear regression slope, effectively supporting the decision-making needs of underground coal mine safety operation and providing comprehensive and accurate safety management support for coal mining enterprises.

[0047] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0048] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0049] Figure 1 This invention provides a trend analysis method adapted to various types of sensors in underground coal mines. Detailed Implementation

[0050] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0051] Please see Figure 1 This invention provides a trend analysis method adapted to various types of sensors in underground coal mines. The method transforms raw numerical data into a series of discrete, interpretable states, which characterize the dynamic behavior of the monitored parameters over time. This process comprises several consecutive stages: sensor data normalization, multi-scale feature extraction, and rule-based state classification.

[0052] The method specifically includes the following steps:

[0053] S1: Data Preparation and Normalization

[0054] (1) Data preparation: Let D be the raw time series dataset of the selected sensor, represented as a timestamp-value pair sequence: D = {(t1, v1), (t2, v2), ... (t n ,v n )}, where t i v is the timestamp of the i-th measurement. i Let i = 1, 2, ..., n be the corresponding sensor values; assuming the measurements are sampled at a fixed frequency. Additionally, a configuration repository C is maintained to store the operating parameters for each sensor type. In this method, the key parameters are the minimum and maximum values ​​that the sensor can record: c sensor_type ={R min ,R max}, where R min R represents the minimum range of values ​​for this specific sensor type. max This represents the maximum value range.

[0055] (2) Data normalization:

[0056] To achieve a unified analytical framework applicable to any sensor, regardless of its original unit of measurement or scale (e.g., Celsius, percentage, etc.), each raw data value v i All values ​​should be normalized to a dimensionless scale, preferably within the range [0,1]. For each value v in dataset D... i The normalized value v′ is calculated using the Min-Max normalization formula. i :

[0057] v′ i =(v i -R min ) / (R max -R min )

[0058] Generate a new, normalized time series dataset D′: D′={(t1,v′1),(t2,v′2),…(t n ,v′ n All subsequent analyses were performed on this normalized dataset D′ to ensure that the logic and thresholds are universal across all types of sensors.

[0059] S2: Multi-scale partitioned normalized value set:

[0060] To analyze the dynamic behavior of the data, this method employs a sliding window approach with multiple time scales.

[0061] Define two window sizes: a short-lived window W s Size N s A number of data points are used to capture fast transient events. A long-term window W l Size N l N data points l >N s This is used to identify gradual, persistent trends while smoothing out small fluctuations. For each time point t in the sequence... i (where i≥N) l ), calculate in t i A set of statistical characteristics of data points within a window ending at a given time.

[0062] Let s s(i) and s l(i) They are respectively t i The normalized value sets within the short and long windows at the endpoint:

[0063]

[0064] S3: Feature Calculation

[0065] (1) Calculation of standard deviation characteristics:

[0066] Calculate the long window W l The sample standard deviation σ is used to measure the stability or volatility of a signal. A lower standard deviation indicates a constant or near-constant state.

[0067]

[0068] Where, μ l It is set s l(i)The arithmetic mean of the medians.

[0069] (2) Calculation of linear regression slope:

[0070] To quantify the rate and direction of change, a simple linear regression is performed on the data points within each window. The slope m of the regression line is the most critical feature in trend detection.

[0071] The linear regression slope is calculated in both the short-term and long-term windows; let x j This is the time index for the data points within the window.

[0072]

[0073] Where, m s(i) and m l(i) These represent the linear regression slopes within the short-term and long-term windows, respectively; sum Σ by iterating through all points j and x within each window. j It is the index of a point within the window, v′ j These are normalized sensor values.

[0074] S4: Rule Calculation

[0075] Rule-based deterministic classification engines will use the feature vector calculated in the previous step to classify each time point t. i Assign a state i The classification engine uses a set of predefined, adjustable thresholds:

[0076] T slope_sudden A larger slope value in a short window indicates a sudden change.

[0077] T slope_slow A moderate slope value for a long window indicates a continuous and slow rise.

[0078] T std_const A very low standard deviation for a long window indicates a stable signal.

[0079] The rules are as follows:

[0080] If the linear regression slope m within the short-term window s(i) Greater than threshold T slope_sudden Then the state i It was marked as "sudden rise";

[0081] Otherwise, if m s(i) Less than negative T slope_sudden (i.e. -T) slope_sudden If the state is 0, then the state is 0. i It was marked as "sudden drop";

[0082] Otherwise, if the standard deviation σ within the long-term window l(i) Less than threshold T std_const Then the state i Marked as "constant";

[0083] Otherwise, if the linear regression slope m within the long window l(i) Greater than threshold T slope_slow Then the state i It was marked as "slow and steady rise";

[0084] Otherwise, if m l(i) Less than negative T slope_slow (i.e. -T) slope_slow If the state is 0, then the state is 0. i It was marked as "slow and continuous decline";

[0085] Otherwise, the state is the default state. i It is marked as "fluctuating up and down".

[0086] S5: Output Merging:

[0087] The output of step 5 is a series of state classifications, one for each time point. The final step is to merge this sequence into a more meaningful and compact representation. The algorithm iterates through the state sequence. i The system groups consecutive identical states into time intervals. The final output is a list of events, where each event is defined by a start time, an end time, and the characteristic states of that time interval. For example: [{start time: t a End time: t b The state is "slowly and continuously rising" (...).

[0088] This invention's method can identify dangerous trends such as "slow and continuous rise" at an early stage, triggering intervention before exceeding safety thresholds, transforming passive monitoring into proactive early warning; it replaces subjective manual interpretation with automated algorithms, ensuring consistency in interpretation across operators, and its sensor-independent characteristics support unified analysis of heterogeneous equipment, significantly reducing implementation complexity and expansion costs; it outputs interpretable state labels such as "sudden drop" and "constant," providing real-time contextual awareness to improve decision-making quality; and it automatically generates historical datasets with state labels, breaking through the bottleneck of industrial AI training data and accelerating the development of advanced models such as predictive maintenance.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A trend analysis method adapted to multiple types of sensors in underground coal mines, characterized in that, The method specifically includes the following steps: S1: Data preparation and normalization: Obtain the raw time series data of the selected sensor and normalize it to obtain the normalized value; S2: Multi-scale partitioning of normalized value set: The normalized value set is partitioned using a multi-time-scale sliding window method; S3: Feature Calculation: Calculate the sample standard deviation for the long-term window, as well as the linear regression slope within the short-term and long-term windows; S4: Rule-based state classification: Based on the feature vector calculated in step S3 and the classification engine, for each time point t... i Assign a state i ; S5: Output Merging: Merge the state classification sequence output in step S4 into an event list.

2. The trend analysis method for adapting to multiple types of sensors in underground coal mines according to claim 1, characterized in that, In step S1, the data preparation specifically includes: Let D be the original time-series dataset of the selected sensor, represented as a timestamp-value pair sequence: D = {(t1, v1), (t2, v2), ..., (t... n ,v n )}, where t i v is the timestamp of the i-th measurement. i Let i be the corresponding sensor value, i = 1, 2, ..., n; assume the measured values ​​are sampled at a fixed frequency; key parameter c sensor_type These are the minimum and maximum values ​​that the sensor can record: c sensor_type ={R min ,R max }, where R min and R max These represent the minimum and maximum value ranges for the selected sensor type, respectively.

3. The trend analysis method for adapting to multiple types of sensors in underground coal mines according to claim 2, characterized in that, In step S1, for each value v in dataset D i The normalized value v′ is calculated using the Min-Max normalization formula. i : v′ i =(v i -R min ) / (R max -R min ) Generate a new, normalized time series dataset D′: D′={(t1,v′1),(t2,v′2),…(t n ,v′ n )}.

4. The trend analysis method for adapting to multiple types of sensors in underground coal mines according to claim 2, characterized in that, In step S1, the key parameters are stored in repository C, which is used to store the operating parameters for each sensor type.

5. The trend analysis method for adapting to multiple types of sensors in underground coal mines according to claim 3, characterized in that, In step S2, the multi-scale partitioning of the normalized value set specifically includes: defining two window sizes: a short-term window W. s Size N s 10 data points are used to capture fast transient events; a long-term window W l Size N l N data points l >N s It is used to identify gradual, persistent trends while smoothing out small fluctuations; for each time point t in the sequence i Calculate in t i A set of statistical characteristics of data points within a window ending at time; Let s s(i) and s l(i) They are respectively t i The normalized value sets within the short and long windows at the endpoint: Where i≥N l .

6. The trend analysis method for adapting to multiple types of sensors in underground coal mines according to claim 5, characterized in that, In step S3, the long-term window W is calculated. l The sample standard deviation σ is used to measure the stability or volatility of the signal; Where, μ l It is set s l(i) The arithmetic mean of the medians.

7. The trend analysis method for adapting to multiple types of sensors in underground coal mines according to claim 5, characterized in that, In step S3, calculating the linear regression slope specifically includes: calculating the linear regression slope in both the short-term and long-term windows; letting x... j The time index for data points within the window; Where, m s(i) and m l(i) These represent the linear regression slopes within the short-term and long-term windows, respectively; sum Σ by iterating through all points j and x within each window. j v is the index of a point within the window, and v′ is the normalized sensor value.

8. The trend analysis method for adapting to multiple types of sensors in underground coal mines according to claim 1, characterized in that, In step S4, the classification engine is a set of predefined adjustable thresholds: the slope threshold T of the short window. slope_sudden The slope threshold T of the long window slope_slow The standard deviation threshold T for long windows std_const ; The rules are as follows: If the linear regression slope m within the short-term window s(i) Greater than threshold T slope_sudden Then the state i It was marked as "sudden rise"; Otherwise, if m s(i) Less than negative T slope_sudden Then the state i It was marked as "sudden drop"; Otherwise, if the standard deviation σ within the long-term window l(i) Less than threshold T std_const Then the state i Marked as "constant"; Otherwise, if the linear regression slope m within the long window l(i) Greater than threshold T slope_slow Then the state i It was marked as "slow and steady rise"; Otherwise, if m l(i) Less than negative T slope_slow Then the state i It was marked as "slow and continuous decline"; Otherwise, the state is the default state. i It is marked as "fluctuating up and down".

9. The trend analysis method for adapting to multiple types of sensors in underground coal mines according to claim 1, characterized in that, Step S5 specifically includes: iterating over the state sequence state. i It groups consecutive identical states into time periods; finally, it outputs a list of events, where each event is defined by a start time, an end time, and the characteristic states of that time period.

Images