Mine ai intelligent ventilation control system based on big data processing

Abstract

The application provides a mine AI intelligent ventilation control system based on big data processing, comprising: a heterogeneous data cross-domain semantic alignment module, which constructs an intradomain feature vector for four types of heterogeneous time series data, projects to a unified semantic space through an interdomain mapping function, and performs semantic distance inspection and correction; a key period focusing prediction module, which constructs a candidate period set through wide window coarse-grained scanning, focuses on identifying a key period at a fine-grained resolution, performs deep feature extraction, and generates a ventilation risk forward-looking prediction result; and a polarity conversion type double-target collaborative regulation module, which evaluates the compliance gap of ventilation safety margin and fan energy consumption, dynamically allocates regulation resources through a polarity conversion mechanism, and performs double-target collaborative convergence verification; the application realizes the effect of automatic correction of consistency of multi-source data at the semantic level by adopting the technical means of performing inspection on semantic distance.

Description

Technical Field

[0001] This invention relates to the field of mine safety technology, and in particular to a mine AI intelligent ventilation control system based on big data processing. Background Technology

[0002] Mine ventilation systems are core safety infrastructures that ensure the safety of underground workers, control the concentration of toxic and harmful gases, and prevent gas accumulation and coal dust explosions. With the continuous increase in coal mining depth and the increasingly complex layout of mining faces, the underground airflow network structure is becoming increasingly large, involving numerous controllable nodes such as main fans, local fans, air windows, air doors, and regulating windows. The complexity of ventilation control tasks has far exceeded the scope that can be effectively covered by human experience.

[0003] However, when it comes to achieving proactive predictive intelligent control of mine ventilation systems, existing technologies are mainly limited by the lack of systematic big data modeling and correlation analysis capabilities for multi-source heterogeneous time-series data underground (including historical gas concentration sequences, wind speed fluctuation data, mining operation data, geological structure change data, etc.). In practical applications, this makes it difficult for the system to simultaneously achieve the two core effects of "effectively ensuring ventilation safety margin" and "dynamically optimizing fan energy consumption". Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] Based on this, the present invention provides a mine AI intelligent ventilation control system based on big data processing to solve the problems mentioned in the background art, such as the lack of systematic big data modeling and correlation analysis capabilities of existing systems for multi-source heterogeneous time-series data underground.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides a mine AI intelligent ventilation control system based on big data processing, comprising:

[0008] The heterogeneous data cross-domain semantic alignment module specifically involves: constructing corresponding feature vectors for historical gas concentration sequences, wind speed fluctuation data, mining operation data, and geological structure change data; using inter-domain mapping functions to map the feature vectors of each domain to corresponding semantic points in the same semantic space; verifying the semantic distance between each pair of semantic points in the semantic space; and performing semantic point correction based on the verification results.

[0009] The key time period focusing prediction module specifically works as follows: It performs a coarse-grained scan of historical time series data through a preset wide time window, and constructs a candidate time period set based on the scan results. Based on the constructed candidate time period set, it re-scans the time series details within the candidate time periods with fine-grained resolution to determine the ordered list of key time periods for the current candidate time period. It then performs deep feature extraction on all identified key time periods. Based on the key time period deep feature vectors and combined with the current real-time multi-source unified feature set, it identifies similar key time periods between the current ventilation status features and the historical key time period deep feature vectors, and generates a baseline prediction result.

[0010] The polarity-switching dual-objective collaborative control module works as follows: Based on the generated prediction results, it assesses the current compliance gaps of the ventilation safety margin target and the fan energy consumption target respectively, and determines the primary and secondary polarities based on the calculated safety gap value and energy consumption gap value. Based on the result of the primary and secondary polarity determination, it executes the corresponding control commands. After each round of control command execution, it collects the actual response data of each controllable node, reassesses the current compliance gap of the primary polarity target, and determines whether to perform polarity switching based on the assessment results. After completing multiple control rounds, it verifies the dual-objective collaborative convergence by checking the ventilation safety margin target and the fan energy consumption target.

[0011] Specifically, the verification of the semantic distance between each pair of semantic points in the semantic space is performed as follows:

[0012] Let the two semantic points be respectively and Its semantic distance is Set the semantic alignment threshold to The set semantic alignment threshold is compared with the distance between semantic points. Based on the comparison result, the consistency status of each semantic point pair is determined, and then:

[0013] If the comparison results satisfy the formula If the two domains are determined to be semantically consistent, the corresponding feature vectors from the two domains are fused.

[0014] If the comparison results satisfy the formula If a semantic deviation is identified, the combination of the deviating semantic point pairs is recorded, and the inter-domain correction process is initiated to perform the correction.

[0015] If all semantic point pairs If so, then skip inter-domain correction and directly execute the multi-source unified feature set output.

[0016] Specifically, the semantic point correction based on the test results is performed as follows:

[0017] Using the reference semantic point with the smallest semantic distance from the deviating semantic point pair as a benchmark, a closer correction is performed on semantic points with larger deviations. The correction direction is the line connecting the deviating semantic point to the reference semantic point, and the correction step size is based on the current deviation magnitude. Okay, after each round of correction, the semantic distance between the two points is recalculated, and the correction is repeated until... The corrected semantic point positions are updated to the final semantic representation of the corresponding data. If the semantic distance still does not converge after the preset maximum number of corrections are executed continuously, the deviated semantic points are marked as low-confidence semantic points, and the fusion weight is reduced when the multi-source unified feature set is output.

[0018] Specifically, the construction of the candidate time period set based on the scanning results is as follows:

[0019] Calculate the fluctuation amplitude of each feature component of the multi-source unified feature set within each wide time window. Set the coarse-grained detection threshold to The set coarse-grained detection threshold is compared with the fluctuation amplitude of each feature component of the multi-source unified feature set within each wide time window. Based on the comparison results, a coarse-grained scan is performed, specifically as follows:

[0020] If the comparison results satisfy the formula Mark the current wide time window as a candidate time period;

[0021] If the comparison results satisfy the formula Mark the current wide time window as a stable period;

[0022] For stable periods, only the mean summary of the feature components is retained, without further deep processing, thus forming a set of candidate periods.

[0023] Specifically, the rescanning of temporal details within the candidate time period with fine-grained resolution is as follows:

[0024] Record the start time of abnormal changes in feature components within the candidate time period. and the end time , time period The key time periods are marked as critical time periods. The boundary accuracy of the key time periods is determined by the fine-grained resolution. If multiple key time periods are identified within a candidate time period, the key time periods are sorted from largest to smallest according to the magnitude of abnormal changes, forming an ordered list of key time periods for the current candidate time period.

[0025] Specifically, the identification of similar key time periods between the current ventilation status features and the historical key time period depth feature vectors is as follows:

[0026] Set the similarity threshold to Then we have:

[0027] If the similarity between the current ventilation status characteristics and the deep feature vectors of historical key periods is... The comparison with the set similarity threshold satisfies the formula Then, the evolution sequence of ventilation status after the current historical key period is used as the reference evolution path of the current ventilation status to generate a system oriented towards the preset prediction time step. The results of the forward-looking assessment of ventilation risks include: the ventilation risk at each roadway node. Upper limit of gas concentration inside and lower limit of wind speed forecast The output is sent to the polarity-switching dual-target coordinated control module;

[0028] If the similarity between the current ventilation status characteristics and the deep feature vectors of all historical key periods satisfies the set similarity threshold, then... Then, the real-time features of the current multi-source unified feature set are used as the conservative prediction basis to generate the benchmark prediction result, which is then output to the polarity conversion dual-target collaborative control module.

[0029] Specifically, the current gap between the assessment of the ventilation safety margin target and the fan energy consumption target is as follows:

[0030] The gap between meeting the ventilation safety margin target and the target is the maximum difference between the estimated upper limit of gas concentration at each roadway node and the corresponding safe concentration threshold in the prediction results. And to determine whether the safety margin is sufficient, we have:

[0031] If the calculated gap between the ventilation safety margin target and the target satisfies the formula This indicates that there is a risk of exceeding the limit within the predicted period. If the calculated gap in achieving the ventilation safety margin target satisfies the formula... This indicates that the safety margin is sufficient;

[0032] The gap between meeting the wind turbine energy consumption target and the target energy saving value is the difference between the estimated total energy consumption under the current wind turbine speed combination. And to determine whether the energy consumption has reached the energy-saving target, we have:

[0033] If the calculated gap in meeting the wind turbine energy consumption target satisfies the formula This indicates that the current energy consumption exceeds the energy-saving target. If the calculated gap in meeting the wind turbine energy consumption target satisfies the formula... This indicates that the current energy consumption has reached the energy-saving target;

[0034] The gaps in achieving the two objectives are recorded as the safety gap value and the energy consumption gap value, respectively.

[0035] Specifically, the determination of primary and secondary polarities is as follows:

[0036] After normalizing the calculated security gap and energy consumption gap values, and determining their primary and secondary polarities based on the normalized values, we have:

[0037] If the calculated gap between the ventilation safety margin target and the target satisfies the formula Regardless of the size of the normalized safety gap value, the ventilation safety margin target is forcibly set as the primary polarity target.

[0038] If no mandatory conditions are imposed, the normalized security gap values ​​are compared. and energy deficit value Based on the magnitude of the polarity and the comparison results, the primary and secondary polarities are determined, resulting in:

[0039] If the comparison results satisfy the formula This indicates that the primary objective is ventilation safety margin, while the secondary objective is fan energy consumption.

[0040] If the comparison results satisfy the formula This indicates that the primary objective is fan energy consumption, while the secondary objective is ventilation safety margin.

[0041] Specifically, the determination of whether to perform a polarity switch based on the evaluation results is as follows:

[0042] Set the polarity switching threshold to The set polarity switching threshold will be compared with the current achievement gap of the reassessed primary polarity target. A comparison is performed, and a decision is made based on the comparison to determine whether to perform a polarity switch. Specifically:

[0043] If the comparison results satisfy the formula If the polarity is reversed, the current secondary polarity target will be upgraded to the new primary polarity target, and the current primary polarity target will be downgraded to the new secondary polarity target. The polarity allocation of regulatory resources will be re-executed, and the next round of regulation will begin with the new primary polarity target as the main driver.

[0044] If the comparison results satisfy the formula If the current polarity remains unchanged, the current main polarity target will continue to guide the implementation of regulation, and the actual response data of this round will be updated to the historical records of the key period focus prediction module, triggering the rolling update of the prediction results.

[0045] Specifically, the verification of dual-objective convergence by examining the ventilation safety margin target and the fan energy consumption target is as follows:

[0046] After each preset verification cycle of K control rounds, the ventilation safety margin target is verified. With wind turbine energy consumption target Whether each of their respective compliance boundary conditions is met simultaneously, specifically:

[0047] If the ventilation safety margin target and the fan energy consumption target satisfy the formula When the current combination of control parameters is recorded as a snapshot of the coordinated convergence parameters, it is stored in the big data historical database as a reference control starting point for subsequent similar ventilation conditions.

[0048] If the ventilation safety margin target and the fan energy consumption target satisfy the formula If the target is not met, the primary and secondary polarity determination will be re-executed, and a new cycle of polarity determination and resource allocation will begin until the dual targets are achieved in synergy or the preset maximum control cycle limit is reached.

[0049] (III) Beneficial Effects

[0050] As can be seen from the above technical solution, the beneficial effects of the high-speed motor hybrid winding and its assembly method proposed in this invention are as follows:

[0051] 1. This invention employs a technique of constructing intra-domain feature vectors for four types of data—historical gas concentration sequences, wind speed fluctuation data, mining operation data, and geological structure change data—and then projecting these feature vectors into a unified semantic space using an inter-domain mapping function. This technique enables the system to establish measurable feature relationships between multi-source heterogeneous time-series data that originally had different dimensions and acquisition principles within a unified semantic space. This allows the system to systematically model and perform horizontal comparative analysis of heterogeneous downhole data, thus solving the problem of existing systems lacking systematic big data modeling capabilities for multi-source heterogeneous downhole time-series data.

[0052] 2. This invention achieves automatic semantic consistency correction of multi-source data by performing semantic distance checks on each pair of semantic points in the semantic space, using the reference semantic point with the smallest semantic distance as a benchmark to perform iterative correction on deviating semantic points, marking deviating semantic points that fail to converge as low-confidence semantic points and reducing their fusion weight. This makes the credibility of the fusion result not dependent on manual weight setting but objectively driven by semantic distance, thus solving the problem of distortion of association analysis results caused by different information source nature when fusing multi-source data in existing systems.

[0053] 3. This invention employs a two-stage scanning technique: first, it performs coarse-grained scanning of historical time-series data within a preset wide time window to construct a candidate time period set; then, it performs fine-grained resolution to shrink and focus the candidate time periods and sort them according to the magnitude of abnormal changes to form an ordered list of key time periods. This technique concentrates limited in-depth analysis resources on the key time periods in historical time-series data that truly drive the evolution of ventilation status, making the input features of the prediction model targeted and highly information-dense. This solves the problem that existing systems lack the ability to systematically correlate and analyze historical time-series data and cannot identify key evolution patterns.

[0054] 4. This invention employs techniques such as performing deep feature extraction on key time periods, matching the current real-time ventilation status features with the deep feature vectors of historical key time periods, and using the evolution sequence after similar historical time periods as a reference path to generate forward-looking prediction results of ventilation risks. This enables feedforward prediction and intervention based on historical big data patterns before the formation of a ventilation risk situation, eliminating the inherent control lag window effect of traditional passive over-limit response. This solves the problem that existing systems can only respond passively and cannot guarantee ventilation safety margins due to a lack of big data correlation analysis capabilities.

[0055] 5. This invention employs a technical approach that involves normalizing the assessment of the gaps between the two objectives of ventilation safety margin and fan energy consumption, using the normalized gap value to drive the dynamic determination and control of primary and secondary polarities in resource allocation, and automatically performing polarity switching when the gap of the primary polarity objective converges to within the polarity switching threshold. This achieves the gradual and coordinated convergence of the two mutually constraining objectives of ensuring ventilation safety margin and dynamically optimizing fan energy consumption towards their respective achievement boundaries during polarity alternation, enabling both to be dynamically balanced rather than subject to fixed weight trade-offs. This solves the problem that existing systems struggle to simultaneously achieve the two core effects of effectively ensuring ventilation safety margin and dynamically optimizing fan energy consumption. Attached Figure Description

[0056] The features and advantages of the invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the invention in any way. In the drawings:

[0057] Figure 1 This is a schematic diagram of the overall composition of the mine AI intelligent ventilation control system based on big data processing according to the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] like Figure 1 As shown, the present invention provides a mine AI intelligent ventilation control system based on big data processing, including: a heterogeneous data cross-domain semantic alignment module, a key period focus prediction module, and a polarity conversion dual-objective collaborative control module;

[0060] Specifically, the heterogeneous data cross-domain semantic alignment module is used to perform intra-domain feature representation construction, inter-domain mapping projection, and semantic distance correction on multi-source heterogeneous time-series data in the well, and output a semantically aligned multi-source unified feature set; the critical period focus prediction module performs wide-window coarse-grained scanning and narrow-window fine-grained focusing on historical time-series data based on the received multi-source unified feature set, identifies critical periods, and generates ventilation risk prospective prediction results; the polarity conversion dual-objective collaborative control module evaluates the gap between the two objectives of ventilation safety margin and fan energy consumption based on the received prospective prediction results, dynamically allocates control resources through a polarity conversion mechanism, and outputs control instructions to each controllable node.

[0061] Furthermore, the heterogeneous data cross-domain semantic alignment module is used to perform cross-domain semantic alignment processing on downhole multi-source heterogeneous time-series data, including intra-domain feature representation construction and inter-domain mapping and semantic space projection, as detailed below:

[0062] The feature representation construction within the domain is tailored to four types of data: historical gas concentration sequences, wind speed fluctuation data, mining operation data, and geological structure change data. A separate feature representation for each data type is constructed for its respective domain. Specifically:

[0063] For the historical gas concentration sequence, the average concentration, the slope of concentration change, and the time of extreme concentration of each roadway node within a preset historical time window are extracted to form a concentration domain feature vector.

[0064] For wind speed fluctuation data, the mean wind speed, standard deviation of wind speed fluctuation, and number of wind speed abrupt changes of each wind station are extracted to form a wind speed domain feature vector.

[0065] Based on the mining operation data, the change in the current mining face's advance distance, the shift intensity coefficient, and the personnel density distribution are extracted to form a feature vector of the working condition domain.

[0066] Based on the geological structure change data, the nearest distance of known faults to the current working face, the radius of influence of the goaf area, and the trigger frequency of historical gas anomaly zones are extracted to form a geological domain feature vector.

[0067] Furthermore, inter-domain mapping and semantic space projection construct four types of intra-domain feature vectors from the intra-domain feature representations. These vectors are then projected onto the same semantic space using a pre-trained inter-domain mapping function. Semantic distance checks and corrections are performed, and the processed multi-source unified feature set is sent to the critical time-segment focusing prediction module. This process includes semantic space projection, semantic distance checks, and inter-domain corrections, as specifically implemented below:

[0068] By using a pre-trained inter-domain mapping function, the feature vectors of the four domains are uniformly projected onto the same semantic space, so that the heterogeneous features with different dimensions and collection principles are formed into measurable feature points in the semantic space, as detailed below:

[0069] The dimension of the semantic space is determined offline by the system on historical operating data, with the minimum dimension that can simultaneously retain the main information of the feature vectors in the four domains as the standard;

[0070] After projection, each of the four types of data corresponds to a feature point in the semantic space, denoted as concentration semantic point, wind speed semantic point, working condition semantic point, and geological semantic point.

[0071] The semantic distance between each pair of four semantic points in the semantic space is examined, specifically as follows:

[0072] Let the two semantic points be respectively and Its semantic distance is Set the semantic alignment threshold to ;

[0073] By comparing the set semantic alignment threshold with the distance between semantic points, and determining the consistency status of each semantic point pair based on the comparison result, we have:

[0074] If the comparison results satisfy the formula If the two domains are determined to be semantically consistent, the corresponding feature vectors from the two domains are fused.

[0075] If the comparison results satisfy the formula If a semantic deviation is identified, the combination of the deviating semantic point pairs is recorded, and the inter-domain correction process is initiated to perform the correction.

[0076] If all semantic point pairs If so, then skip inter-domain correction and directly execute the multi-source unified feature set output.

[0077] Correction is performed on the semantically deviating point pairs identified in the semantic distance test, specifically as follows:

[0078] Using the reference semantic point with the smallest semantic distance from the deviating semantic point pair as a benchmark, a closer correction is performed on semantic points with larger deviations; the correction direction is the direction of the line connecting the deviating semantic point to the reference semantic point; the correction step size is based on the current deviation magnitude. Okay, after each round of correction, the semantic distance between the two points is recalculated, and the correction is repeated until... The corrected semantic point positions are then updated to the final semantic representation of this type of data.

[0079] If the semantic distance still fails to converge after performing the preset maximum number of corrections, the deviating semantic point is marked as a low-confidence semantic point, and its fusion weight is reduced when outputting the multi-source unified feature set.

[0080] It should be noted that, for the four semantic point features after inter-domain mapping and semantic space projection processing, after assigning fusion weights according to the degree of semantic consistency, a weighted fusion is performed to construct a semantically aligned multi-source unified feature set, which is then sent to the critical time-segment focusing prediction module. The feature component weights corresponding to low-confidence semantic points are reduced to a preset low-confidence weight coefficient. Semantic distance is at The feature components within the range are weighted according to the inverse of their distance from the reference semantic point.

[0081] The critical period focus prediction module performs a two-stage scan and critical period depth feature extraction on historical time-series data based on the received multi-source unified feature set, generating a forward-looking prediction result for ventilation risks. This includes a wide-window coarse-grained scan and a narrow-window fine-grained focus, as detailed below:

[0082] Wide-window coarse-grained scanning performs a coarse-grained scan of historical time-series data using a preset wide time window. The length of the wide time window covers the system's preset longest ventilation state evolution cycle, as detailed below:

[0083] Calculate the fluctuation amplitude of each feature component of the multi-source unified feature set within each wide time window. Set the coarse-grained detection threshold to ;

[0084] The set coarse-grained detection threshold is compared with the fluctuation amplitude of each feature component of the multi-source unified feature set within each wide time window. Based on the comparison results, a coarse-grained scan is performed, specifically as follows:

[0085] If the comparison results satisfy the formula Mark the current wide time window as a candidate time period;

[0086] If the comparison results satisfy the formula Mark the current wide time window as a stable period;

[0087] For stable periods, only the mean summary of the feature components is retained, without further deep processing, thus forming a set of candidate periods.

[0088] Furthermore, the shrinking window fine-grained focusing involves progressively shrinking the time window from the wide time window boundary inwards to a fine-grained resolution for each candidate time period in the candidate time period set, and completing the identification and deep feature extraction of key time periods. This includes key time period identification, key time period deep feature extraction, and forward prediction generation, as detailed below:

[0089] Rescan the temporal details within the candidate time periods with fine-grained resolution, specifically:

[0090] Record the start time of abnormal changes in feature components within the candidate time period. and the end time , time period The critical time period is marked as such, and the boundary accuracy of the critical time period is determined by the fine-grained resolution.

[0091] If multiple key time periods are identified within a candidate time period, then each key time period is sorted from largest to smallest according to the magnitude of abnormal changes, forming an ordered list of key time periods for the current candidate time period.

[0092] Deep feature extraction is performed on all identified key time periods, specifically as follows:

[0093] Extract the abnormal rise rate, peak concentration, and peak duration of the gas concentration characteristic components within each key time period; extract the wind speed characteristic components and the wind speed drop amplitude before and after the start of each key time period. and recovery time Extract the work intensity change event that is temporally closest to the start time of the critical period from the work condition feature components. The time interval between its occurrence and the start time of the critical period. Extract the geological structural types and their spatial locations that contribute the most to the gas concentration anomaly during the current critical period from the geological feature components; and concatenate the four types of depth features into a depth feature vector for the critical period.

[0094] Based on the key time period deep feature vector, combined with the current real-time multi-source unified feature set, similar key time periods between the current ventilation status features and historical key time period deep feature vectors are identified, specifically:

[0095] Set the similarity threshold to Then we have:

[0096] If the similarity between the current ventilation status characteristics and the deep feature vectors of historical key periods is... The comparison with the set similarity threshold satisfies the formula Then, the evolution sequence of ventilation status after the current historical key period is used as the reference evolution path of the current ventilation status to generate a system oriented towards the preset prediction time step. The results of the forward-looking assessment of ventilation risks include: the ventilation risk at each roadway node. Upper limit of gas concentration inside and lower limit of wind speed forecast The output is sent to the polarity-switching dual-target coordinated control module;

[0097] If the similarity between the current ventilation status characteristics and the deep feature vectors of all historical key periods satisfies the set similarity threshold, then... Then, the real-time features of the current multi-source unified feature set are used as the conservative prediction basis to generate the benchmark prediction result, which is then output to the polarity conversion dual-target collaborative control module.

[0098] The polarity-switching dual-objective coordinated control module assesses the gaps in meeting the two objectives of ventilation safety margin and fan energy consumption based on forward-looking predictions. It dynamically allocates control resources through a polarity-switching mechanism, establishes a dynamic coordinated control mechanism, and adjusts the control commands of each controllable node in real time to ensure that the safety margin and energy-saving objectives gradually converge in a coordinated manner during polarity switching. The specific implementation is as follows:

[0099] This is a polarity-switching closed-loop control mechanism. The input is the forward-looking prediction result from the key period focusing prediction module, and the output is the dynamic control command for each controllable node, including the control of ventilation safety margin target, the control of fan energy consumption target, and the verification of dual-target collaborative convergence, specifically:

[0100] After receiving the forward-looking forecast results, the current gaps in meeting the ventilation safety margin target and the fan energy consumption target are assessed separately, specifically as follows:

[0101] The gap between meeting the ventilation safety margin target and the target is the maximum difference between the estimated upper limit of gas concentration at each roadway node and the corresponding safe concentration threshold in the prediction results. Therefore:

[0102]

[0103] in, Indicating the first prediction result The upper limit of the estimated gas concentration at each roadway node Indicating the first prediction result The safe concentration threshold for each roadway node This indicates the gap between the target and the actual target for ventilation safety margin, used to determine whether the safety margin is sufficient. Specifically:

[0104] If the calculated gap between the ventilation safety margin target and the target satisfies the formula This indicates that there is a risk of exceeding the limit within the predicted period;

[0105] If the calculated gap between the ventilation safety margin target and the target satisfies the formula This indicates that the safety margin is sufficient.

[0106] The gap between meeting the wind turbine energy consumption target and the target energy saving value is the difference between the estimated total energy consumption under the current wind turbine speed combination. Therefore:

[0107]

[0108] in, This indicates the estimated total energy consumption under the current wind turbine speed combination. Indicates the target energy saving value. This indicates the gap between the energy consumption target and the target, used to determine whether the energy consumption has met the energy-saving target. Specifically:

[0109] If the calculated gap in meeting the wind turbine energy consumption target satisfies the formula This indicates that the current energy consumption exceeds the energy-saving target;

[0110] If the calculated gap in meeting the wind turbine energy consumption target satisfies the formula If the value is 0, it means that the current energy consumption has reached the energy-saving target.

[0111] The gaps in achieving the two objectives are recorded as safety gap values. Energy consumption gap value .

[0112] Furthermore, based on the calculated safety gap and energy consumption gap values, the primary and secondary polarities are determined as follows:

[0113] The calculated security gap and energy gap values ​​are normalized as follows:

[0114] After normalizing the calculated security gap value, we have:

[0115]

[0116]

[0117] After normalizing the calculated energy deficit value, we have:

[0118]

[0119] in, , These represent the normalized safety gap value and energy consumption gap value, respectively.

[0120] The primary and secondary polarities are determined based on the normalized security gap and energy consumption gap values, specifically as follows:

[0121] If the calculated gap between the ventilation safety margin target and the target satisfies the formula Regardless of the size of the normalized safety gap value, the ventilation safety margin target is forcibly set as the primary polarity target.

[0122] If there are no mandatory conditions, then compare the normalized safety gap value with the energy consumption gap value, and determine the primary and secondary polarities based on the comparison results, then:

[0123] If the comparison results satisfy the formula This indicates that the primary objective is ventilation safety margin, while the secondary objective is fan energy consumption.

[0124] If the comparison results satisfy the formula This indicates that the primary objective is fan energy consumption, while the secondary objective is ventilation safety margin.

[0125] Furthermore, the control of the ventilation safety margin target involves outputting control instructions to the corresponding controllable nodes according to the polarity allocation rule when the ventilation safety margin is the primary polarity target. Specifically:

[0126] If the primary objective is to ensure ventilation safety margin, the upper limit of the gas concentration should be estimated. The increased air intake demand corresponding to the highest roadway node is prioritized, and the majority of the adjustable resources in this round (assuming a primary polarity allocation ratio) will be allocated accordingly. And 0.5 < ≤1) Allocate ventilation fans and ventilation windows along the roadway with the highest estimated upper limit of gas concentration, and sequentially send adjustment commands to increase ventilation volume to them, using the remaining control resources (proportion of 1- Assigned to the controllable nodes corresponding to the secondary polarity target;

[0127] If the ventilation safety margin is a secondary objective, then only a secondary portion of the control resources are allocated to maintain the basic safety constraint of ĉ_i≤c_i^lim at each node, while the primary control resources are controlled by the energy consumption objective.

[0128] The regulation of wind turbine energy consumption targets involves outputting regulation commands to the corresponding controllable nodes according to the polarity allocation rules when wind turbine energy consumption is the primary polarity target. Specifically:

[0129] If wind turbine energy consumption is the primary objective, the wind turbine that currently contributes the most to energy consumption should be prioritized for adjustment, while meeting the requirements of each node. Under the premise that the main part of the controllable resources in this round (the proportion is) The remaining control resources (in a ratio of 1-) are allocated to the wind turbine that currently contributes the most energy consumption, and an adjustment command to reduce its speed is sent to it. Assigned to the controllable nodes corresponding to the secondary polarity target;

[0130] If the wind turbine energy consumption is a secondary target, then only a small portion of the control resources will be allocated to perform light adjustments on wind turbines with high energy consumption, while the main control resources will be controlled by the safety margin target.

[0131] Furthermore, the polarity switching judgment involves collecting actual response data from each controllable node after each round of control command execution, and reassessing the current achievement gap of the primary polarity target. The details are as follows:

[0132] Set the polarity switching threshold to The set polarity switching threshold is compared with the current achievement gap of the reassessed primary polarity target, and a decision is made based on the comparison to determine whether to perform a polarity switch. Specifically:

[0133] If the comparison results satisfy the formula If the polarity is reversed, the current secondary polarity target will be upgraded to the new primary polarity target, and the current primary polarity target will be downgraded to the new secondary polarity target. The polarity allocation of regulatory resources will be re-executed, and the next round of regulation will begin with the new primary polarity target as the main driver.

[0134] If the comparison results satisfy the formula If the current polarity remains unchanged, the current main polarity target will continue to guide the implementation of regulation, and the actual response data of this round will be updated to the historical records of the key period focus prediction module, triggering the rolling update of the prediction results.

[0135] Dual-objective collaborative convergence verification involves verifying the ventilation safety margin target after each preset verification cycle of K control rounds. With wind turbine energy consumption target Whether each of their respective compliance boundary conditions is met simultaneously, specifically:

[0136] If the ventilation safety margin target and the fan energy consumption target satisfy the formula When the current combination of control parameters is recorded as a snapshot of the coordinated convergence parameters, it is stored in the big data historical database as a reference control starting point for subsequent similar ventilation conditions.

[0137] If the ventilation safety margin target and the fan energy consumption target satisfy the formula If the target is not met, the primary and secondary polarity determination will be re-executed, and a new cycle of polarity determination and resource allocation will begin until the dual targets are achieved in synergy or the preset maximum control cycle limit is reached.

[0138] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A mine AI intelligent ventilation control system based on big data processing, characterized in that, include: The heterogeneous data cross-domain semantic alignment module specifically involves: constructing corresponding feature vectors for historical gas concentration sequences, wind speed fluctuation data, mining operation data, and geological structure change data; using inter-domain mapping functions to map the feature vectors of each domain to corresponding semantic points in the same semantic space; verifying the semantic distance between each pair of semantic points in the semantic space; and performing semantic point correction based on the verification results. The key time period focusing prediction module specifically works as follows: It performs a coarse-grained scan of historical time series data through a preset wide time window, and constructs a candidate time period set based on the scan results. Based on the constructed candidate time period set, it re-scans the time series details within the candidate time periods with fine-grained resolution to determine the ordered list of key time periods for the current candidate time period. It then performs deep feature extraction on all identified key time periods. Based on the key time period deep feature vectors and combined with the current real-time multi-source unified feature set, it identifies similar key time periods between the current ventilation status features and the historical key time period deep feature vectors, and generates a baseline prediction result. The polarity-switching dual-objective collaborative control module works as follows: Based on the generated prediction results, it assesses the current compliance gaps of the ventilation safety margin target and the fan energy consumption target respectively, and determines the primary and secondary polarities based on the calculated safety gap value and energy consumption gap value. Based on the result of the primary and secondary polarity determination, it executes the corresponding control commands. After each round of control command execution, it collects the actual response data of each controllable node, reassesses the current compliance gap of the primary polarity target, and determines whether to perform polarity switching based on the assessment results. After completing multiple control rounds, it verifies the dual-objective collaborative convergence by checking the ventilation safety margin target and the fan energy consumption target.

2. The method according to claim 1, characterized in that, The specific steps for verifying the semantic distance between each pair of semantic points in the semantic space are as follows: Let the two semantic points be respectively and Its semantic distance is Set the semantic alignment threshold to The set semantic alignment threshold is compared with the distance between semantic points. Based on the comparison result, the consistency status of each semantic point pair is determined, and then: If the comparison results satisfy the formula If the two domains are determined to be semantically consistent, the corresponding feature vectors from the two domains are fused. If the comparison results satisfy the formula If a semantic deviation is identified, the combination of the deviating semantic point pairs is recorded, and the inter-domain correction process is initiated to perform the correction. If all semantic point pairs If so, then skip inter-domain correction and directly execute the multi-source unified feature set output.

3. The method according to claim 1, characterized in that, The semantic point correction based on the test results is performed as follows: Using the reference semantic point with the smallest semantic distance from the deviating semantic point pair as a benchmark, a closer correction is performed on semantic points with larger deviations. The correction direction is the line connecting the deviating semantic point to the reference semantic point, and the correction step size is based on the current deviation magnitude. Okay, after each round of correction, the semantic distance between the two points is recalculated, and the correction is repeated until... The corrected semantic point positions are updated to the final semantic representation of the corresponding data. If the semantic distance still does not converge after the preset maximum number of corrections are executed continuously, the deviated semantic points are marked as low-confidence semantic points, and the fusion weight is reduced when the multi-source unified feature set is output.

4. The method according to claim 1, characterized in that, The process of constructing a candidate time period set based on the scan results is as follows: Calculate the fluctuation amplitude of each feature component of the multi-source unified feature set within each wide time window. Set the coarse-grained detection threshold to The set coarse-grained detection threshold is compared with the fluctuation amplitude of each feature component of the multi-source unified feature set within each wide time window. Based on the comparison results, a coarse-grained scan is performed, specifically as follows: If the comparison results satisfy the formula Mark the current wide time window as a candidate time period; If the comparison results satisfy the formula Mark the current wide time window as a stable period; For stable periods, only the mean summary of the feature components is retained, without further deep processing, thus forming a set of candidate periods.

5. The method according to claim 1, characterized in that, The process of rescanning the temporal details within the candidate time period with fine-grained resolution is as follows: Record the start time of abnormal changes in feature components within the candidate time period. and the end time , time period The key time periods are marked as critical time periods. The boundary accuracy of the key time periods is determined by the fine-grained resolution. If multiple key time periods are identified within a candidate time period, the key time periods are sorted from largest to smallest according to the magnitude of abnormal changes, forming an ordered list of key time periods for the current candidate time period.

6. The method according to claim 1, characterized in that, The identification of similar key time periods between the current ventilation status features and the historical key time period deep feature vectors is as follows: Set the similarity threshold to Then we have: If the similarity between the current ventilation status characteristics and the deep feature vectors of historical key periods is... The comparison with the set similarity threshold satisfies the formula Then, the evolution sequence of ventilation status after the current historical key period is used as the reference evolution path of the current ventilation status to generate a system oriented towards the preset prediction time step. The results of the forward-looking assessment of ventilation risks include: the ventilation risk at each roadway node. Upper limit of gas concentration inside and lower limit of wind speed forecast The output is sent to the polarity-switching dual-target coordinated control module; If the similarity between the current ventilation status characteristics and the deep feature vectors of all historical key periods satisfies the set similarity threshold, then... Then, the real-time features of the current multi-source unified feature set are used as the conservative prediction basis to generate the benchmark prediction result, which is then output to the polarity conversion dual-target collaborative control module.

7. The method according to claim 1, characterized in that, The current gap between the assessment of the ventilation safety margin target and the fan energy consumption target is as follows: The gap between meeting the ventilation safety margin target and the target is the maximum difference between the estimated upper limit of gas concentration at each roadway node and the corresponding safe concentration threshold in the prediction results. And to determine whether the safety margin is sufficient, we have: If the calculated gap between the ventilation safety margin target and the target satisfies the formula This indicates that there is a risk of exceeding the limit within the predicted period. If the calculated gap in achieving the ventilation safety margin target satisfies the formula... This indicates that the safety margin is sufficient; The gap between meeting the wind turbine energy consumption target and the target energy saving value is the difference between the estimated total energy consumption under the current wind turbine speed combination. And to determine whether the energy consumption has reached the energy-saving target, we have: If the calculated gap in meeting the wind turbine energy consumption target satisfies the formula This indicates that the current energy consumption exceeds the energy-saving target. If the calculated gap in meeting the wind turbine energy consumption target satisfies the formula... This indicates that the current energy consumption has reached the energy-saving target; The gaps in achieving the two objectives are recorded as the safety gap value and the energy consumption gap value, respectively.

8. The method according to claim 1, characterized in that, The determination of primary and secondary polarities is as follows: After normalizing the calculated security gap and energy consumption gap values, and determining their primary and secondary polarities based on the normalized values, we have: If the calculated gap between the ventilation safety margin target and the target satisfies the formula Regardless of the size of the normalized safety gap value, the ventilation safety margin target is forcibly set as the primary polarity target. If no mandatory conditions are imposed, the normalized security gap values ​​are compared. and energy deficit value Based on the magnitude of the polarity and the comparison results, the primary and secondary polarities are determined, resulting in: If the comparison results satisfy the formula This indicates that the primary objective is ventilation safety margin, while the secondary objective is fan energy consumption. If the comparison results satisfy the formula This indicates that the primary objective is fan energy consumption, while the secondary objective is ventilation safety margin.

9. The method according to claim 1, characterized in that, The process of determining whether to perform a polarity switch based on the evaluation results is as follows: Set the polarity switching threshold to The set polarity switching threshold will be compared with the current achievement gap of the reassessed primary polarity target. A comparison is performed, and a decision is made based on the comparison to determine whether to perform a polarity switch. Specifically: If the comparison results satisfy the formula If the polarity is reversed, the current secondary polarity target will be upgraded to the new primary polarity target, and the current primary polarity target will be downgraded to the new secondary polarity target. The polarity allocation of regulatory resources will be re-executed, and the next round of regulation will begin with the new primary polarity target as the main driver. If the comparison results satisfy the formula If the current polarity remains unchanged, the current main polarity target will continue to guide the implementation of regulation, and the actual response data of this round will be updated to the historical records of the key period focus prediction module, triggering the rolling update of the prediction results.

10. The method according to claim 1, characterized in that, The verification of dual-objective convergence is achieved by examining the ventilation safety margin target and the fan energy consumption target, as detailed below: After each preset verification cycle of K control rounds, the ventilation safety margin target is verified. With wind turbine energy consumption target Whether each of their respective compliance boundary conditions is met simultaneously, specifically: If the ventilation safety margin target and the fan energy consumption target satisfy the formula When the current combination of control parameters is recorded as a snapshot of the coordinated convergence parameters, it is stored in the big data historical database as a reference control starting point for subsequent similar ventilation conditions. If the ventilation safety margin target and the fan energy consumption target satisfy the formula If the target is not met, the primary and secondary polarity determination will be re-executed, and a new cycle of polarity determination and resource allocation will begin until the dual targets are achieved in synergy or the preset maximum control cycle limit is reached.

Images