Mine cold load dynamic allocation and multi-stage refrigeration collaborative control method and system

Abstract

The present application relates to mine refrigeration control technical field, disclose a kind of mine cold load dynamic distribution and multistage refrigeration collaborative control method and system.The method collects multi-source heterogeneous sensor data, generates mine heat and humidity field dynamic directed graph by space-time alignment and thermodynamic normalization processing;Based on the graph combining cold load operator and equipment feasible region, through the multi-objective rolling optimizer with heat source topological perception generates cold distribution Markov decision chain;Under the driving of decision chain, through layered time scale collaborative controller generates execution strategy simplex complex;Finally under the constraint of complex, through edge-center double-ring feedback executor linkage PLC group, adjust ground cold source power, secondary network pressure difference and end evaporating temperature.The present application solves the problem that traditional method is difficult to adapt to mine dynamic heat and humidity field, and the problem of inaccurate cold distribution, realizes critical zone priority guarantee and global energy efficiency optimization, meets the demand of mine heat damage control.

Description

Technical Field

[0001] This invention relates to the field of mine refrigeration control technology, and more specifically, to a method and system for dynamic distribution of mine cooling load and multi-level refrigeration coordinated control. Background Technology

[0002] Traditional methods for allocating cooling loads and controlling refrigeration in mines often fail to accurately meet actual needs. This is because traditional methods rely on fixed formulas to estimate cooling loads, neglecting to consider the dynamic changes in the mine's thermal and humidity fields, the priority differences of heat sources in different areas, and the impact of surface climate boundary changes on the initial conditions of underground airflow. This results in a lack of targeted and dynamic adjustment capabilities in cooling load allocation. This problem leads to insufficient cooling supply in some areas, failing to meet safety regulations or operational requirements, while redundant cooling in other areas results in energy waste, and the overall refrigeration system operates with low efficiency and poor reliability. Solving this problem would not only achieve precise matching of cooling loads with the actual cooling needs of each area but also significantly improve the overall energy efficiency of the refrigeration system while ensuring safety in critical areas. This would further enhance the stability and adaptability of mine refrigeration control, providing a more reliable environmental guarantee for long-term stable operation in deep mines. Summary of the Invention

[0003] To achieve the above objectives, the present invention provides the following technical solution:

[0004] A method for dynamic distribution of cooling load and multi-stage coordinated control of cooling in mines includes:

[0005] Collect multi-source heterogeneous sensor data, and generate and output a dynamic directed graph of the mine's thermal and humid field through spatiotemporal alignment and thermodynamic normalization.

[0006] Based on the dynamic directed graph of the mine's thermal and humid field, and combined with the cooling load operator and the equipment feasible region, a Markov decision chain for cooling allocation is generated and output through a multi-objective rolling optimizer with heat source topology awareness.

[0007] Driven by the Markov decision chain for cooling capacity allocation, a simple complex of execution strategies is generated and output through a hierarchical time-scaled collaborative controller.

[0008] Under the constraint of the simple complex execution strategy, the power of the ground cooling source, the differential pressure of the secondary network and the terminal evaporation temperature are dynamically adjusted by linking the PLC group through the edge-center dual-loop feedback actuator.

[0009] Specifically, the spatiotemporal alignment is as follows:

[0010] Using the unified clock in the central control room of the mine as a reference, the acquisition time of all data in the multi-source heterogeneous sensor dataset is calibrated, and a time calibration dataset is generated after calibration.

[0011] A spatial topology network is constructed based on the heat flow path of the wind network and interpolated to generate a spatially aligned dataset.

[0012] Based on the dynamic changes of the thermal and wet field in the mine, the abnormal data marked in the time calibration dataset are repaired; after repair, a spatiotemporally consistent dataset is generated; if the deviation of the abnormal data exceeds the reasonable range, the reasonable data for the corresponding time point is calculated based on the synchronous data of adjacent areas in the spatial alignment dataset and the heat flow propagation speed.

[0013] Specifically, the structure of the directed graph of the dynamic thermal and humid field in the mine is as follows:

[0014] Nodes include location identification data, dry and wet bulb temperatures, air enthalpy, cooling load, heat source type label, and WBGT risk level; edges include air volume, airway pressure difference, air temperature drop, and humidity attenuation coefficient; the heat source priority order is pump house > working face > tunneling face, with > indicating priority over;

[0015] Specifically, the cooling load operator is:

[0016] A mathematical operator is used to quantify the urgency of cooling demand in each region by combining the normalized value of cooling load, WBGT risk level weight, and heat source priority weight. The calculation method of the cooling load operator is as follows: cooling load operator value = ω1 × normalized value of cooling load + ω2 × WBGT risk level weight + ω3 × heat source priority weight, where ω1, ω2, and ω3 are weight coefficients and the sum of the three is 1.

[0017] Specifically, the multi-objective rolling optimizer with heat source topology awareness includes:

[0018] Adaptive window segmentation for optimized scrolling;

[0019] Construct a multi-objective optimization objective function, including minimizing the duration of cooling load exceeding limits, minimizing the total system energy consumption, and maximizing safety redundancy;

[0020] Set multi-dimensional constraints, including equipment feasible domain constraints, heat flow conduction constraints, and heat source priority constraints;

[0021] An improved non-dominated sorting genetic algorithm is used to solve the problem. Each rolling optimization window outputs a set of non-dominated optimization schemes, which are then concatenated in time series to form an initial optimization scheme sequence.

[0022] Specifically, the method for constructing the Markov decision chain for cooling capacity allocation is as follows:

[0023] A Markov decision chain for cooling load allocation is constructed based on the initial optimization scheme sequence. The state space is a cooling load sorting vector, the action space is set by the source-network-end coordination, the transfer function is based on the thermal and humid field prediction data, and the reward function is a decision chain in the form of a linear combination.

[0024] Specifically, the hierarchical time-scaled collaborative controller includes:

[0025] Dynamically divide the timescale into fast, medium, and slow layers;

[0026] Extract the control tasks corresponding to each layer from the action space of the Markov decision chain for cooling capacity allocation, and verify whether the cooling capacity output of the unit in the slow time scale layer can support the pipeline transmission demand of the medium time scale layer.

[0027] Establish cross-timescale information exchange channels to achieve collaborative control of equipment.

[0028] Specifically, the execution strategy simple complex is as follows: generate a 0-simplex corresponding to the control parameters of the terminal cooling area, a 1-simplex corresponding to the pipeline network coordinated control parameters, and a 2-simplex corresponding to the combination of source-side unit operating states, construct the dew point safety boundary, and integrate them to form a complete complex.

[0029] Specifically, the edge-center dual-loop feedback actuator is:

[0030] The three core regulation targets are analyzed from the simple complex form of the execution strategy, including the target value of the ground cooling source power, the target value of the secondary network pressure difference, and the target value of the terminal evaporation temperature;

[0031] The edge ring is deployed locally in each PLC subgroup to collect the actual operating parameters of the equipment in real time, compare them with the control target parameters, calculate the deviation, and perform rapid correction.

[0032] The central ring is deployed in the mine cooling control center, and the adjustment data of each edge ring is summarized every 1-5 minutes to verify global constraints;

[0033] The central loop converts the verified adjustment parameters into an instruction format that the PLC can recognize.

[0034] A dynamic distribution and multi-stage refrigeration coordinated control system for mine cooling load, the system comprising:

[0035] Data acquisition and preprocessing module: used to acquire multi-source heterogeneous sensor data, and generate and output a dynamic directed graph of the mine's thermal and humid field through spatiotemporal alignment and thermodynamic normalization.

[0036] Markov Decision Chain Generation Module: Based on the dynamic directed graph of the mine's thermal and wet field, combined with the cold load operator and the equipment feasible region, a multi-objective rolling optimizer with heat source topology awareness is used to generate and output the Markov decision chain for cold load allocation.

[0037] Simple Complex Generation Module: Used to generate and output simple complexes of execution strategies under the drive of the Markov decision chain of cooling allocation through a hierarchical time-scaled collaborative controller;

[0038] Dual-loop feedback execution and equipment adjustment module: Under the constraint of simple complex execution strategy, it is used to dynamically adjust the power of the ground cooling source, the secondary network pressure difference and the terminal evaporation temperature by linking the PLC group through the edge-center dual-loop feedback actuator.

[0039] Compared to existing technologies, the advantages of this invention are as follows: This invention effectively solves the problems of traditional mine cooling load allocation and refrigeration control methods being difficult to adapt to the dynamic thermal and humidity fields of mines and having inaccurate cooling capacity allocation. By collecting multi-source heterogeneous sensor data and processing it through spatiotemporal alignment and thermodynamic normalization, a dynamic directed graph of the mine thermal and humidity field is generated. This graph can accurately capture the changes in mine thermal and humidity, the priority of heat sources in different areas, and the influence of surface climate boundary disturbances, breaking through the limitation of traditional fixed formulas that ignore dynamic characteristics when estimating cooling load. Based on this directed graph, combined with the cooling load operator and the cooling capacity allocation Markov decision chain generated by a multi-objective rolling optimizer with heat source topology awareness, the cooling load demand of different areas and time periods can be dynamically matched, avoiding the problem of insufficient or redundant cooling capacity in some areas due to the lack of dynamic adjustment capability in traditional methods.

[0040] This invention, through a hierarchical time-scaled collaborative controller and an edge-center dual-loop feedback actuator linked PLC group, achieves both rapid response to local thermal disturbances and ensures coordinated global cooling capacity transfer and equipment operation. While ensuring adequate cooling capacity in critical areas such as pump rooms, it significantly improves the overall energy efficiency of the refrigeration system, enhances the stability and adaptability of mine refrigeration control, and provides a reliable automated solution for deep mine thermal hazard management. By constructing a dynamic thermal and humidity field model and a multi-objective collaborative optimization mechanism, it achieves precise on-demand allocation of energy and cooling capacity and efficient collaborative operation of equipment. While ensuring safety and compliance in key areas, it significantly reduces overall system energy consumption and equipment wear. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart of a method for dynamic allocation of mine cooling load and multi-stage refrigeration coordinated control in this invention;

[0043] Figure 2 This is a flowchart of the multi-objective rolling optimizer with heat source topology awareness in this invention;

[0044] Figure 3 This is a schematic diagram of the simple complex structure for executing the strategy in this invention;

[0045] Figure 4 This is a functional block diagram of a dynamic distribution and multi-level refrigeration coordinated control system for mine cooling load in this invention. Detailed Implementation

[0046] 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, and 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.

[0047] Example 1:

[0048] Please see Figure 1 As shown, this embodiment provides a method for dynamic allocation of mine cooling load and multi-stage refrigeration coordinated control, including:

[0049] Step S10: Collect multi-source heterogeneous sensor data, and generate and output a dynamic directed graph of the mine's thermal and humid field through spatiotemporal alignment and thermodynamic normalization.

[0050] This step revolves around the accurate acquisition and structured processing of mine thermal and humid environment data. The core is to collect multi-source heterogeneous sensor data by classification, and combine it with processing methods adapted to the thermal flow characteristics of the mine to transform discrete data into a dynamic directed graph of the mine thermal and humid field that retains the thermal flow topology and priority constraints, thus providing a structured data foundation for subsequent cooling load allocation optimization.

[0051] First, multi-source heterogeneous sensor data is collected, which includes four core data categories: The first category is environmental perception data, including location identification data, dry-bulb and wet-bulb temperatures, air enthalpy, and relative humidity. Location identification data refers to the unique spatial coordinates of each monitoring point within the mine, obtained through assignment using the mine's pre-set three-dimensional rectangular coordinate system, with a coordinate accuracy to 0.5 meters. Dry-bulb and wet-bulb temperatures refer to the actual dry-bulb temperature and wet-bulb evaporation cooling temperature of the air at the monitoring point, synchronously collected by dust-proof temperature and humidity sensors deployed in various areas, with a sampling period of 10 seconds. Air enthalpy refers to the total heat per unit mass of air, calculated using dry-bulb and wet-bulb temperatures combined with the air thermodynamic formula: Air enthalpy = 1.01 × dry-bulb temperature. The formula is: Dry Bulb Temperature + 0.622 × (2501 + 1.86 × Dry Bulb Temperature) × (Relative Humidity × Saturated Vapor Pressure) / (Atmospheric Pressure - Relative Humidity × Saturated Vapor Pressure). The saturated vapor pressure is calculated using the Magnus-Tetens equation. The atmospheric pressure is the average atmospheric pressure at the corresponding depth in the mine. Specifically, for every 100m increase in depth, the atmospheric pressure increases by an average of 10.1 kPa (referencing the standard atmospheric pressure gradient). The average atmospheric pressure is determined by monthly monitoring data at each depth in the mine. Relative humidity is the ratio of the actual water vapor pressure in the air to the saturated vapor pressure at the same temperature, directly collected by temperature and humidity sensors, with a range of 40%-95%. The second category is wind network operation data, including air volume, airway pressure difference, air temperature drop, and humidity attenuation coefficient. Air volume refers to the volume of air passing through the airway per unit time, collected by ultrasonic air volume sensors installed at the airway cross-section, with the unit being cubic meters per minute. Airway pressure difference refers to the pressure difference between the airway inlet and outlet, calculated by subtracting data from pressure sensors at both ends of the airway, with the unit being Pascals. Air temperature drop refers to the temperature decrease of air flowing through the airway, calculated by subtracting data from temperature sensors at the airway inlet and outlet, with the unit being degrees Celsius. The humidity attenuation coefficient is the percentage decrease in humidity after air flows through the airway, calculated by fitting humidity data from the airway inlet and outlet, with a range of 0.7-0.98; the longer the airway, the smaller the coefficient. The third category is heat source characteristic data, which includes heat source type labels, heat source power, and heat source location coordinates. Among them, the heat source type label refers to the category identifier of the heat source in the mine, such as mechanical equipment heat dissipation labeled as "equipment source", surrounding rock heat dissipation labeled as "surrounding rock source", and human body heat dissipation labeled as "personnel source", which is determined by matching the heat source ledger with the on-site identifiers. The heat source power refers to the heat released by the heat source per unit time. The heat source power of mechanical equipment is calculated by multiplying the rated power of the equipment by the heat dissipation efficiency, and the heat source power of surrounding rock is calculated by multiplying the heat flux density collected by the heat flux meter by the area of ​​influence of the heat source. The unit of value is kilowatts. The heat source location coordinates refer to the specific location of the heat source in the three-dimensional coordinate system of the mine, which corresponds to the location identifier data of the environmental perception data.The fourth category is safety monitoring data, with the core being the WBGT risk level. The WBGT risk level is an indicator for assessing the risk of heat stress in the working environment. Its Chinese name is Wet Bulb Temperature Risk Level. It is obtained by collecting data through WBGT sensors and classifying it according to the "Design Specification for Prevention and Control of Heat Hazards in Coal Mines". The values ​​are divided into four levels: low risk, medium risk, high risk, and extremely high risk.

[0052] Based on the four core data categories mentioned above, the original sensor dataset is constructed by archiving data according to the three dimensions of "acquisition time - spatial coordinates - data type". The acquisition time is accurate to the second, with values ​​covering all time points within the data acquisition period. The spatial coordinates are consistent with the location identification data, with values ​​covering all coordinate points within the mine's mining area. The data type corresponds to the specific parameters of the four core data categories, with values ​​covering the set of acquired or calculated values ​​for each parameter. Specifically, environmental sensing data is filled into the corresponding units of the original sensor dataset according to spatial coordinates, acquisition time, and parameter type; wind network operation data is filled into the corresponding units according to wind path spatial coordinates, acquisition time, and parameter type; heat source characteristic data is filled into the corresponding units according to heat source spatial coordinates, acquisition time, and parameter type; and safety monitoring data is filled into the corresponding units according to monitoring point spatial coordinates, acquisition time, and parameter type, forming a discrete dataset containing thermal and humidity information for the entire mine area.

[0053] The original sensor dataset is subjected to spatiotemporal alignment processing using a heat flux topology-guided spatiotemporal interpolation alignment method. Specifically, the heat flux topology-guided spatiotemporal interpolation alignment method is as follows:

[0054] The first step is time reference calibration. Using the unified clock in the mine's central control room as the reference, the acquisition time of all data in the multi-source heterogeneous sensor dataset is calibrated to eliminate time deviations between different sensors. Data with time deviations exceeding 5 seconds are marked as abnormal data and temporarily stored. After calibration, a time calibration dataset is generated. During the calibration process, for data with different sampling periods, such as temperature sampled for 10 seconds and air volume sampled for 1 minute, linear interpolation is used to supplement missing time point data to ensure the integrity of various types of data at the same time point.

[0055] The second step is spatial topology interpolation. Based on the heat flow paths (i.e., wind path connections) of the mine ventilation network, a spatial topology network is constructed. The discrete monitoring points in the original sensor dataset are used as network nodes, and the wind paths are used as connecting edges between nodes. For wind paths or areas without monitoring points, interpolation calculations are performed based on the measured data of adjacent monitoring points and the wind volume weights of the edges (the greater the wind volume, the higher the weight) to obtain estimated data for areas without monitoring points, generating a spatially aligned dataset. This process avoids the shortcomings of traditional uniform interpolation that ignores heat flow paths, ensuring that the interpolated data conforms to the heat and moisture propagation laws in the mine. For example, if the wind volume of a certain wind path is twice that of an adjacent wind path, the interpolation result of that wind path will be more biased towards the monitoring point data with higher wind volume.

[0056] The third step is abnormal data repair. Based on the dynamic changes in the mine's thermal and wet field (e.g., heat source power fluctuations not exceeding 20% / minute), abnormal data marked in the time-calibrated dataset are repaired: if the deviation of the abnormal data is within a reasonable fluctuation range (e.g., temperature deviation ≤ 2℃), the average of the previous and next five time points is used as the replacement; if the deviation exceeds a reasonable range, reasonable data for the corresponding time point is calculated based on the contemporaneous data of adjacent areas in the spatially aligned dataset, combined with the heat source attenuation law, and a spatiotemporally consistent dataset is generated after repair.

[0057] Thermodynamic normalization was performed based on the spatiotemporally consistent dataset, using a regional thermal characteristic weighted normalization method. Specifically, this method involves first classifying regions according to heat source priority (pump station > working face > tunneling face), and assigning thermal characteristic weights to different regions. The weight for the pump station region is set to 1.0, for the working face to 0.8, and for the tunneling face to 0.5. These weights are determined based on the mine's thermal hazard risk assessment report, with higher-risk regions receiving larger weights. Then, the thermodynamic parameters (temperature, enthalpy, and cooling load) in the spatiotemporally consistent dataset are normalized using the formula: Normalized value = (Actual parameter value - Minimum regional parameter value). / (maximum value of regional parameter - minimum value of regional parameter) × regional thermal characteristic weight; where the minimum and maximum values ​​of regional parameters are the extreme values ​​in the historical monitoring data of the region over the past 3 months, determined through statistical analysis; for example, the actual temperature of a monitoring point in the pump room area is 32℃, the minimum temperature of the area is 28℃ and the maximum temperature is 36℃, then the normalized temperature value of this point = (32-28) / (36-28)×1.0=0.5; after processing by this method, a thermodynamically standardized dataset is generated. This dataset not only eliminates the difference in parameter magnitude, but also retains the difference in thermal characteristics of different regions, avoiding the problem of traditional global normalization masking the characteristics of local thermal environment.

[0058] A dynamic directed graph of the mine's thermal and wet field is generated based on a thermodynamically standardized dataset. The structure of this directed graph is as follows: nodes contain location identification data, dry-bulb and wet-bulb temperatures, air enthalpy, cooling load, heat source type labels, and WBGT risk levels; edges contain air volume, airway pressure difference, air temperature drop, and humidity attenuation coefficient; at the same time, the priority order of heat sources is marked in the directed graph, in the order of pump house > working face > tunneling face, with > indicating priority. The nodes are distinguished by color: pump house nodes are red, working face nodes are yellow, and tunneling face nodes are blue, ensuring the visualization of heat flow paths and priority constraints.

[0059] The functions and effects of this step are as follows: By collecting multi-source heterogeneous sensor data through classification, the comprehensiveness of the mine's thermal and humid environment information is ensured; the spatiotemporal interpolation alignment method guided by heat flow topology, combined with the mine's ventilation network heat flow path, performs data interpolation and repair, solving the data deviation problem caused by the traditional spatiotemporal alignment ignoring the heat propagation law, making the spatiotemporally consistent dataset more closely match the actual thermal and humid distribution of the mine; the regional thermal characteristic weighted normalization method adapts the heat hazard risk of different regions through regional weights, avoiding global normalization from masking local differences, and making the thermodynamically standardized dataset more accurately reflect the differences in heat demand in each region; the finally generated dynamic directed graph of the mine's thermal and humid field retains the core information of "heat flow path + heat source priority", completely solving the problem of traditional tables or lists losing the coupling effect of ventilation network topology and heat sources, providing a structured and highly adaptable data foundation for the subsequent step S20 cold load allocation optimization, ensuring that the optimizer can accurately perceive the heat disturbance propagation chain and improve the scientificity and rationality of cold load allocation.

[0060] Step S20: Based on the dynamic directed graph of the mine's thermal and wet field, and combining the cold load operator and the equipment feasible region, a Markov decision chain for cold load allocation is generated and output through a multi-objective rolling optimizer with heat source topology awareness.

[0061] This step revolves around the modeling of cooling capacity allocation decisions for the mine refrigeration system. The core is to rely on the dynamic directed graph of the mine's thermal and humid field output in step S10, and transform the spatial characteristics of the thermal and humid field and equipment operation constraints into a Markov decision chain for cooling capacity allocation with time-series decision-making capabilities through cooling demand quantification methods and topology-aware optimization logic, so as to provide a scientific decision-making basis for subsequent execution layer control.

[0062] First, we define the core external variable for this step: the cooling load operator. The cooling load operator is a mathematical operator used to quantify the urgency of cooling demand in different areas of the mine. It is calculated by integrating the key thermodynamic and safety characteristics of nodes in the dynamic directed graph of the mine's thermal and humidity field. Its core function is to transform discrete thermal and humidity data into a sortable cooling demand priority index. Specifically, the cooling load operator is calculated as follows: Cooling load operator value = ω1 × Normalized cooling load value + ω2 × WBGT risk level weight + ω3 × Heat source priority weight. In the formula, ω1, ω2, and ω3 are weighting coefficients, determined by the analytic hierarchy process (AHP) combined with the mine's control objective of "safety first, energy efficiency second." The sum of the three is 1. The weighting coefficients are recalibrated quarterly based on historical accident data and energy efficiency statistics, or ω2 is temporarily increased during high-risk periods such as summer. The normalized value of the cooling load converts the actual cooling load of the nodes in the dynamic directed graph of the mine's thermal and humidity field into a value in the 0-1 range. The calculation method is: Normalized value of cooling load = (Actual cooling load of the node - Minimum historical cooling load of the mine) / (Maximum historical cooling load of the mine) The cooling load (the mine's historical minimum cooling load) is calculated using data from the past three months of monitoring. The WBGT risk level weights are set at 0.2 for low risk, 0.4 for medium risk, 0.7 for high risk, and 1.0 for extremely high risk, directly derived from the WBGT risk level field of the mine's dynamic directed graph of thermal and wet field nodes. The heat source priority weights are set at 1.0 for pump rooms, 0.8 for working faces, and 0.5 for tunneling faces, with priority rules consistent with the partial order of heat source priorities in the mine's dynamic directed graph of thermal and wet field. This operator directly quantifies the urgency of cooling demand in each area, providing a foundation for defining the subsequent decision chain state.

[0063] Next, we define another external variable—the equipment feasible region. The equipment feasible region refers to the allowable range of operating parameters of each core piece of equipment in the mine refrigeration system. It is used to constrain the action boundaries during the optimization process and ensure that the decision is physically executable. The specific range includes: the number of ground-based cooling units to be started and stopped is 1-N, where N is the total number of units, determined by calculation based on the maximum design cooling load of the mine and the rated cooling capacity of a single unit, typically 3-5 units; the frequency adjustment range of the secondary network circulation pump is 30-50 Hz, with the lower limit based on the minimum speed at which the pump unit avoids cavitation (determined by the performance curve provided by the pump manufacturer), and the upper limit based on the rated speed of the motor and the maximum allowable pressure of the pipeline network (determined by hydraulic calculations of the pipeline network); the opening range of the cooling capacity distribution valve is 5%-100%, with a minimum opening of 5% to prevent jamming caused by long-term closure of the valve, and the maximum opening corresponding to the valve's design flow capacity; the adjustment range of the terminal evaporation temperature is 4-8 degrees Celsius, with the lower limit based on the safe temperature at which the evaporator avoids frosting, determined by the refrigerant property manual, and the upper limit based on the minimum cooling temperature to meet the regional cooling load requirements, determined by heat balance calculations of the terminal heat exchanger. The boundary parameters of the feasible domain of the equipment need to be updated quarterly. The update basis includes the degree of equipment aging and the changes in the mine's cooling load: The degree of equipment aging is assessed by the equipment's operating current and vibration data. The assessment standard is that when the operating current exceeds the rated value by 5% for 30 consecutive minutes, or the vibration acceleration exceeds the equipment's factory threshold (such as centrifugal pump ≤0.8mm / s, compressor ≤1.0mm / s), it is determined to be an aging-affected feasible domain. The changes in the mine's cooling load are statistically analyzed by the nodes of the dynamic directed graph of the mine's thermal and humid field. The statistical judgment standard is that when the daily average cooling load fluctuation exceeds 10% in the past 3 months, it is determined to be a cooling load change and the feasible domain needs to be updated.

[0064] Based on the dynamic directed graph of the mine's thermal and humid field, the cooling load operator, and the feasible region of the equipment, a multi-objective rolling optimizer with heat source topology awareness is used to generate an initial optimization scheme sequence. For details, please refer to [link to relevant documentation]. Figure 2 As shown. Specifically, the multi-objective rolling optimizer with heat source topology awareness includes:

[0065] The first step is to dynamically divide the rolling optimization window. The window duration is adaptively adjusted based on the rate of change of the cooling load at the nodes in the dynamic directed graph of the mine's thermal and wet field: when the rate of change of the cooling load exceeds 10% / minute (i.e., the thermal and wet field fluctuates drastically), the window duration is set to 5 minutes to ensure rapid response to sudden thermal disturbances; when the rate of change is less than 3% / minute, the window duration is set to 15 minutes to reduce unnecessary optimization calculations; when the rate of change is between the two, the window duration is set to 10 minutes. The core basis for window division is the heat flow conduction relationship between "edges and nodes" in the dynamic directed graph of the mine's thermal and wet field—the nodes on the heat flow path have high synchronicity in cooling load changes, and the window duration needs to match the heat flow propagation time (calculated based on air volume, usually 1-3 minutes) to avoid optimization lagging behind the propagation of thermal disturbances.

[0066] The second step is to construct a multi-objective optimization objective function. This objective function comprises three core dimensions: First, minimizing the duration of cold load exceeding the limit, i.e., the cumulative time during which the actual cold load supply in each area is lower than the cold load demand. This is calculated by comparing the node cold load in the dynamic directed graph of the mine's thermal and wet field with the predicted cold load supply value, and statistically analyzing the duration when the difference is greater than 0. Second, minimizing the total system energy consumption, including the energy consumption of ground units (calculated based on the unit power curve), secondary pump energy consumption (calculated based on the pump frequency-power fitting formula), and terminal equipment energy consumption (calculated based on the evaporation temperature-energy consumption relationship). Third, maximizing safety redundancy, i.e., the degree to which the WBGT risk level in each area is lower than the safety threshold. This is calculated as (safety threshold - actual WBGT value) / safety threshold, where the safety threshold is set as the upper limit for medium risk according to the "Coal Mine Safety Regulations".

[0067] The third step is to set multi-dimensional constraints. In addition to the equipment feasible domain constraints, two new mine-specific constraints are added: one is the heat flow conduction constraint, which, based on the relationship between side air volume and temperature drop in the dynamic directed graph of the mine's thermal and humid field, requires that the cold energy transfer volume be ≥ side air volume × air density × air specific heat × node temperature drop requirement, to ensure that the cold energy supply matches the heat flow transfer; the other is the heat source priority constraint, which requires that the cold energy satisfaction rate in the pump room area be ≥95%, the working face ≥90%, and the tunneling face ≥85%, and the satisfaction rate is calculated as actual cold energy supply / node cold load × 100%.

[0068] The fourth step is to improve the algorithm for solving the problem. An improved Non-Dominated Sorting Genetic Algorithm (NSGA-II) is used for optimization. The core of this method is to introduce a heat source topology sensing factor into the algorithm's selection operator. This factor uses the edge airflow of the dynamic directed graph of the mine's thermal and humid field as a weight, ensuring that the optimization process prioritizes cooling allocation schemes that cover critical heat flow paths (nodes connected by high airflow edges), avoiding wasted cooling on non-critical heat flow paths. After solving, each rolling optimization window outputs a set of non-dominated optimization schemes, which are concatenated over time to form an initial optimization scheme sequence. This sequence includes the source-grid-end coordination settings and corresponding cooling demand matching results at each time point. The source-grid-end coordination settings include the number of ground units, secondary pump frequency, valve opening degree, and terminal evaporation temperature.

[0069] Based on the initial optimization scheme sequence, a Markov decision chain for cooling load allocation is constructed: its state space is a cooling load sorting vector, with vector elements sorted from largest to smallest cooling load operator value. Each element contains a region ID and an operator value, and the data is directly taken from the cooling demand priority results at each time step in the initial optimization scheme sequence; the action space is set by source-network-end coordination, with each action corresponding to a set of equipment parameter combinations in the initial optimization scheme sequence, and must satisfy the equipment feasible region constraint; the transfer function is constructed based on the next-time prediction data of the dynamic directed graph of the mine's thermal and wet field, and the nodes are analyzed using an ARIMA model. Short-term predictions of cooling load and side air volume are made, and the cooling load ranking vector for the next moment is calculated by combining the current action. The parameters are determined by training with historical monitoring data from the past month. The model is recalibrated once a month to adapt to the load change pattern. The reward function adopts a linear combination form: reward value = -α × over-limit duration - β × energy consumption + γ × redundancy safety degree. Among them, α is obtained by weighting the production loss caused by over-limit cooling and equipment maintenance costs, β is obtained by fitting unit electricity price, carbon emission cost and energy consumption data, and γ is obtained by correlation analysis between safety redundancy degree and accident-free operation time.

[0070] The functions and effects of this step are as follows: The cooling load operator, by integrating the multi-dimensional features of the dynamic directed graph of the mine's thermal and wet field, achieves accurate quantification of the urgency of cooling demand, avoiding the shortcomings of traditional single cooling load indicators that ignore safety priorities; the multi-objective rolling optimizer with heat source topology awareness relies on the heat flow topology relationship of the directed graph, enabling the optimization scheme to actively adapt to the mine's ventilation network structure, solving the problem of traditional optimizers ignoring heat disturbance propagation paths; the Markov decision chain for cooling allocation, through "state-action-reward" closed-loop modeling, possesses the foresight to predict sudden heat source fluctuations and the game-like nature of cross-regional cooling demand competition, completely overcoming the pain points of ordinary PID or static threshold scheduling being unable to handle time-varying boundaries and dynamic competition. Simultaneously, this step closely follows the dynamic directed graph of the mine's thermal and wet field in S10, driving the state definition of the decision chain through the spatial heat conduction structure of the directed graph, ensuring that variables are progressively added, laying the decision foundation for the subsequent generation and execution strategy in S30.

[0071] For example, when the directed graph of the mine's thermal and wet field shows a cooling load of 800 kW and a high risk of WBGT at the pump house node, and a cooling load of 600 kW and a medium risk of WBGT at the working face node, the cooling load operator calculates 0.92 for the pump house and 0.75 for the working face. The multi-objective rolling optimizer with heat source topology awareness solves the problem within a 10-minute window and outputs an optimization scheme of "2 units, 42 Hz pump frequency, pump house valve opening of 90%, and terminal evaporation temperature of 4.5℃". After the multi-window schemes are connected in series to form an initial optimization scheme sequence, in the constructed Markov decision chain for cooling load allocation, the state is [pump house (0.92), working face (0.75)], the action is the above parameter combination, the transfer function predicts that the cooling load ranking will remain unchanged at the next moment, and the reward value is calculated as -α×0-β×280 kW·h+γ×0.85, forming a complete decision unit.

[0072] Step S30: Driven by the Markov decision chain of the cooling capacity allocation, the execution strategy simple complex is generated and output through the hierarchical time-scaled collaborative controller.

[0073] This step revolves around the topological construction of the mine cooling execution strategy. The core is to rely on the Markov decision chain for cooling capacity allocation output in step S20, and through a hierarchical time-scaled collaborative controller, transform the time-sequential "state-action" decision into a simple complex execution strategy with spatial coordination and safety constraints. This solves the pain points of "asynchronous actions" and "fuzzy safety boundaries" in traditional control, and provides a precise basis for equipment execution in step S40.

[0074] First, we define the core control component for this step: the hierarchical time-scaled collaborative controller. This controller is a key module connecting the Markov decision chain for cooling capacity allocation with the simple complex form of the execution strategy. Its design logic is based on the dynamic response differences of mine refrigeration equipment, achieving collaborative control of "high-frequency adjustment for fast-response equipment and advance planning for slow-response equipment" by dividing the time-scale hierarchy. Specifically, the working mechanism of the hierarchical time-scaled collaborative controller is as follows:

[0075] The first step is to dynamically divide the timescale levels. Based on the state transition frequency of the Markov decision chain for cooling capacity allocation, i.e., the intensity of thermal and humidity field fluctuations, the control process is divided into fast timescale, medium timescale, and slow timescale levels: The fast timescale level corresponds to terminal temperature and humidity regulation, with a response period of 10-30 seconds. When the state transition frequency exceeds 0.5 times / minute (intense fluctuations in the thermal and humidity field), the response period is shortened to 10 seconds; otherwise, it is extended to 30 seconds. This is because terminal temperature and humidity fluctuate the fastest due to the influence of ventilation airflow, requiring high-frequency correction. The medium timescale level corresponds to the linkage between pipeline valve positions and secondary pump frequencies, with a response period of 1-5 minutes. The length of the period is matched with the hydraulic inertia of the pipeline network—the larger the pipe diameter and the longer the pipe length, the longer the period, to avoid pressure oscillations caused by excessively frequent regulation. The slow timescale level corresponds to the start-up and shutdown combination of the source-side units, with a response period of 10-30 minutes. The period needs to cover the complete start-up and shutdown process of the units (such as compressor loading and lubricating oil circulation) to avoid frequent start-ups and shutdowns that shorten equipment life.

[0076] The second step is to decompose the decision chain actions to the time scale layer. The corresponding control tasks for each layer are extracted from the action space of the cooling capacity allocation Markov decision chain: the "terminal evaporation temperature" parameter is allocated to the fast time scale layer as the calculation benchmark for terminal temperature and humidity settings; the "secondary pump frequency" and "distribution valve opening" parameters are allocated to the medium time scale layer, transforming them into core parameters for pipeline valve position linkage; and the "number of ground units" parameter is allocated to the slow time scale layer to determine the quantity benchmark for source-side unit start-up and shutdown combinations. After decomposition, an "action compatibility check" needs to be performed, i.e., checking whether the cooling capacity output of the units in the slow time scale layer can support the pipeline transmission requirements of the medium time scale layer. The calculation formula is: total cooling capacity output of the units ≥ upper limit of pipeline cooling capacity transmission corresponding to the secondary pump frequency. If this is not met, feedback is sent to the cooling capacity allocation Markov decision chain to request an update of the action space until the actions of each layer are compatible.

[0077] The third step is cross-timescale information collaboration. Information exchange channels are established at each level: The start-up and shutdown plans of units at the slow timescale level need to be synchronized to the medium timescale level one time-scale cycle in advance, allowing for advance adjustments to pipeline valve positions and pump frequencies to match changes in cooling supply; pipeline pressure data at the medium timescale level needs to be synchronized to the fast timescale level every one fast time-scale cycle, enabling fine-tuning of terminal temperature and humidity settings based on actual cooling capacity. During the collaboration process, the information exchange frequency for high-priority areas such as pump rooms is increased by 50%, ensuring priority response to critical zone control actions.

[0078] After processing by the hierarchical time-scaled co-controller, the three main components of the execution strategy simplex complex are generated respectively:

[0079] First, there is the 0-simplex, with terminal temperature and humidity settings. As the basic unit of the complex, each 0-simplex corresponds to the control parameters of a terminal cooling zone, with the structure {zone ID, target temperature, target relative humidity}. The region ID is consistent with the node position ID of the directed graph of the dynamic thermal and humid field of the mine; the target temperature is based on the "terminal evaporation temperature" parameter of the fast timescale layer, combined with the regional WBGT risk level correction, and the formula is target temperature = reference temperature - k × risk level coefficient, where the reference temperature is determined by the terminal evaporation temperature + heat exchange temperature difference (usually 15-18℃), and k is the correction coefficient, which is 2 for high risk, 1 for medium risk, and 0.5 for low risk. The risk level is taken from the WBGT risk level field of the node of the directed graph of the dynamic thermal and humid field of the mine; the target relative humidity range is 50%-70%, constrained by the "temperature-humidity" coupling relationship, and the formula is target relative humidity = max[50%, min(70%, 70%-(target temperature-22℃)×2%)], that is, when the calculated result is lower than 50%, it is forced to be 50%, and when it is higher than 70%, it is forced to be 70%, to avoid the high temperature and high humidity environment triggering the dew point risk. The value must ensure that the subsequent dew point safety constraints are met. max(·) means to take the maximum value, and min(·) means to take the minimum value.

[0080] Second, there is the 1-simplex, with valve position linkage in the pipeline network. As an edge element of the complex, each 1-simplex corresponds to the coordinated control parameters of a pipeline network, with the structure {pipeline ID, distribution valve opening, secondary pump frequency, linkage coefficient}. The pipeline ID corresponds to the edge ID of the directed graph of the mine's thermal and humid field; the distribution valve opening and secondary pump frequency are directly taken from the decomposition parameters of the mid-time scale layer, and the opening range needs to be between 5% and 100% (5% is the minimum safe opening); the linkage coefficient is used to quantify the coordinated relationship between valve position and pump frequency, and the calculation formula is linkage coefficient = distribution valve opening change rate / secondary pump frequency change rate, with a value range of 0.8-1.2, obtained through pipeline network hydraulic characteristic experiments. When the linkage coefficient is close to 1, the influence of valve position and pump frequency adjustment on flow rate is linearly matched, which can avoid hydraulic imbalance.

[0081] Thirdly, the 2-simplex represents the start-stop combination of source-side units. As a complex surface unit, each 2-simplex corresponds to a combination of the operating states of all source-side units, with the structure {unit group ID, start-stop status of each unit, total cooling output}. The unit group ID is the set identifier of the source-side units; the start-stop status of each unit is represented by "1" (start) or "0" (stop), selected based on the unit COP (coefficient of performance) ranking, prioritizing the start of the unit with the highest COP. The COP data is calculated through the fitting relationship of "cooling output - power consumption" in historical operation; the total cooling output is calculated based on the "number of ground units" in the slow timescale layer and the rated cooling capacity of a single unit, with the formula: total cooling output = number of started units × rated cooling capacity of a single unit × load rate. The load rate is determined based on the cooling demand of the pipe network in the medium timescale layer, with a value range of 70%-100%, to avoid energy efficiency degradation caused by full-load operation of the units.

[0082] Finally, surface constraints (dew point safety boundaries) are constructed and integrated into the complex. The dew point safety boundary is expressed by the function f(φ,T)≥0, where φ is the target relative humidity and T is the target temperature. The core is to ensure that the actual temperature is not lower than the dew point temperature at the corresponding humidity to avoid condensation. Specifically, the dew point temperature Td is calculated using the Magnus-Tetens equation, with the formula ln(φ / 100)=a×(Td-T0) / (Td-b), where T0 is the reference temperature (273.15K), and a and b are coefficients. These coefficients are fitted and corrected using temperature and humidity data from the mine over the past month. During the rainy season, b can be slightly increased to adapt to the high humidity environment. Furthermore, f(φ,T)=T-Td is derived, and the safety constraint is satisfied when f(φ,T)≤0. During the integration process, each element needs to be verified individually: For all 0-simplexes and 1-simplexes associated with each 2-simplex, substitute them into f(φ,T) for calculation. If f(φ,T)<0 exists, increase the target temperature of the corresponding 0-simplex by 0.5-1℃ or decrease the target relative humidity by 3%-5%, until all elements satisfy the constraints, ultimately forming a complete execution strategy simplex complex. For the specific structure, please refer to [link / reference]. Figure 3 As shown.

[0083] The functions and effects of this step are as follows: The hierarchical time-scaled collaborative controller solves the problem of cooling supply and demand mismatch caused by asynchronous responses from the terminal, pipeline, and unit in traditional control through dynamic time-scale division. The fast time-scale layer can respond to local thermal disturbances within 10-30 seconds, while the slow time-scale layer can plan unit start-up and shutdown in advance to avoid temporary adjustments. The execution strategy simplex complex naturally expresses the "action coordination dimension" through the topological structure, clearly defining the cooling transmission of 0-simplex dependent on 1-simplex and the cooling supply of 1-simplex dependent on 2-simplex, avoiding action dependencies that cannot be reflected by traditional parameter vectors, and fundamentally preventing hydraulic imbalance in the pipeline network. The dew point safety boundary is calculated through an improved mature formula to ensure that the constraints are scientific and reliable, effectively preventing equipment condensation and corrosion. At the same time, this step closely follows the S20 cooling distribution Markov decision chain, driving the action coordination dimension of the simplex complex through the temporal decision dependencies of the decision chain, ensuring that variables are progressively layered, and providing a safe and implementable control strategy for the S40 PLC execution.

[0084] For example, when the action of the cooling capacity allocation Markov decision chain is "2 ground units, secondary pump frequency 42 Hz, distribution valve opening (pump room 90%, working surface 75%), terminal evaporation temperature 5℃", the hierarchical timescale collaborative controller allocates "terminal evaporation temperature 5℃" to the fast timescale layer, calculates the target temperature of the pump room = (5+17)-2×1=20℃ (WBGT high risk, k=2), and the target relative humidity = 70%-(20-22)×2%=74%; and allocates "42 Hz, 90% / 75% opening" to the medium timescale layer. At the time-scaled layer, the linkage coefficient is calculated as 0.9 / 1.05≈0.86. The "2 units" are allocated to the slow time-scaled layer. The units that start COP 2 are determined, and the total cooling output is 2×600kW×0.9=1080kW. Substituting into the dew point formula, f(74%,20℃)=20-19.8=0.2>0. The humidity of the pump room is reduced to 72%, and finally f(72%,20℃)=20-19.6=0.4≤0, forming a simple complex form of execution strategy containing the corresponding 0-simplex, 1-simplex, and 2-simplex.

[0085] Step S40: Under the simple complex constraint of the execution strategy, the PLC group is linked by the edge-center dual-loop feedback actuator to dynamically adjust the power of the ground cold source, the secondary network pressure difference and the terminal evaporation temperature, so as to achieve priority protection of the critical zone and optimal global energy efficiency.

[0086] This step revolves around the physical execution and closed-loop control of the mine refrigeration system. The core is to rely on the simple complex execution strategy output in step S30, and through the edge-center dual-loop feedback actuator, to transform the topological control strategy into equipment actions that can be executed by the PLC group, simultaneously achieving "local rapid response" and "global safety constraints", and ultimately achieving precise adjustment of ground cold source power, secondary network pressure difference, and terminal evaporation temperature, ensuring the cooling demand in the critical zone and optimizing global energy efficiency.

[0087] First, we define the core execution component of this step: the PLC group. The PLC group is a collection of control units that directly drive the mine refrigeration equipment. It is divided into three subgroups according to equipment type, each corresponding one-to-one with a component in the simplex execution strategy: The first subgroup is the ground-based cold source PLC subgroup, responsible for controlling the source-side units. Each unit is equipped with one PLC, and the control commands received originate from the 2-simplex (source-side unit start / stop combination) of the simplex execution strategy. It collects unit current and condensing temperature data to provide feedback on the operating status, ensuring that the total cooling output conforms to the 2-simplex setting. The second subgroup is the secondary pipeline PLC subgroup, responsible for controlling the secondary... The network circulation pumps and distribution valves are each equipped with one PLC. The control commands received originate from the 1-simplex (network valve position linkage). Hydraulic status is fed back by collecting inlet and outlet pressure differentials and flow data to ensure valve position and pump frequency match the 1-simplex linkage requirements. The third category is the terminal equipment PLC subgroup, responsible for controlling the terminal evaporators. Each terminal area is equipped with one PLC, receiving control commands from the 0-simplex (terminal temperature and humidity setting). Cooling status is fed back by collecting terminal dry-bulb and wet-bulb temperatures and evaporation pressure data to ensure temperature and humidity meet 0-simplex constraints. These three subgroups are linked via an industrial Ethernet bus with a communication cycle of 100-500 milliseconds. The communication cycle for high-priority areas such as pump rooms is shortened to 100 milliseconds to prioritize responses to critical zone regulation needs. The adjustment of the communication cycle is determined based on the partial order of heat source priority in the execution strategy simplex (pump room > working face > tunneling face).

[0088] Secondly, the core control module, the edge-center dual-loop feedback actuator, is defined. This edge-center dual-loop feedback actuator is crucial for connecting the simple complex execution strategy with the PLC group. Through a dual-loop structure of rapid local adjustment in the edge loop and global constraint verification in the central loop, it addresses the pain points of traditional control, namely "response lag" and "global imbalance." The edge loop is a feedback adjustment module deployed locally in each PLC subgroup, responsible for quickly handling local device parameter deviations. The central loop is a feedback control module deployed in the control center, responsible for global constraint verification and cross-regional coordination. Specifically, the working mechanism of the edge-center dual-loop feedback actuator is as follows:

[0089] The first step is to extract the control target parameters. Three core adjustment targets are analyzed from the simplex simplex of the execution strategy: Ground cooling source power target value, calculated based on the total cooling output of the 2-simplex simplex, with the formula: Ground cooling source power target value = Total cooling output / Average unit COP. The average unit COP is obtained by fitting historical data of "cooling output - power consumption" over the past month using the least squares method to ensure calculation accuracy. Secondary network pressure difference target value, directly taken from the network design pressure difference of the 1-simplex simplex. If the actual flow rate deviates from the design flow rate by more than 5%, it is corrected by "Secondary network pressure difference target value = Design pressure difference × (Actual flow rate / Design flow rate)". The actual flow rate data is collected in real-time by the secondary network PLC subgroup. Terminal evaporation temperature target value, calculated based on the terminal target temperature of the 0-simplex simplex, with the formula: Terminal evaporation temperature target value = Terminal target temperature - Heat exchange temperature difference. The heat exchange temperature difference is determined through experiments on the heat exchange efficiency of the terminal evaporator, typically 15-18℃; the lower the efficiency, the larger the temperature difference.

[0090] The second step is local feedback adjustment via the edge ring. The edge ring is deployed locally in each PLC subgroup to collect real-time data on the actual operating parameters of the equipment and compare them with the target control parameters. The deviation is calculated and quickly corrected. For example, the edge ring of the terminal equipment PLC subgroup collects the actual terminal temperature and compares it with the target temperature set by the 0-simplex algorithm. If the deviation exceeds ±0.5℃, the target value is corrected using the formula "Terminal Evaporation Temperature Adjustment Amount = a2 × Temperature Deviation". The coefficient a2 ranges from 0.1 to 0.3, determined through experiments on the terminal temperature response speed. An exemplary correspondence is: a 0.3℃ change in terminal temperature within 10 seconds (fast response) yields a2 = 0.3; a 0.2℃ change within 10 seconds (medium response) yields a2 = 0.2; and a 0.1℃ change within 10 seconds (slow response) yields a2 = 0.1. The response speed is calculated using data from the terminal temperature sensor at 10-second sampling intervals, ensuring that the temperature deviation is controlled within ±0.2℃ within 30 seconds. Edge ring adjustments must strictly adhere to the local constraints of the simplex execution strategy. For example, when correcting valve positions in the edge ring of a secondary pipeline PLC subgroup, it is necessary to ensure that the opening is always within the range of 5%-100% set by the simplex to avoid exceeding the hydraulic safety boundary.

[0091] The third step is to verify the global constraints of the central loop. The central loop is deployed in the mine's refrigeration control center, and the adjustment data from each edge loop is summarized every 1-5 minutes. Two key global constraints are verified: first, the surface constraint of the simplex of the execution strategy, i.e., the dew point safety boundary. The central loop collects the actual temperature and humidity data of all terminal areas and substitutes them into the dew point safety boundary function f(φ,T). If f(φ,T)<0, i.e., the temperature is below the dew point, a correction command is immediately sent to the corresponding terminal device PLC subgroup, increasing the evaporation temperature by 0.5-1℃ or decreasing the relative humidity by 3%-5%, until f(φ,T)≤0; second, the heat source priority constraint. The central loop statistically analyzes the data of each area... The cooling capacity supply satisfaction rate (actual cooling capacity / required cooling capacity × 100%) ensures that the pump room satisfaction rate is ≥95%, the working face is ≥90%, and the tunneling face is ≥85%. If the pump room satisfaction rate is lower than the threshold, the power allocation is adjusted by "ground cooling source power allocation coefficient = pump room cooling demand ratio × b2", where the coefficient b2 ≥ 1.1. The value of b2 is determined by the WBGT risk level of the pump room node in the dynamic directed graph of the mine thermal and wet field. The higher the risk, the larger the value of b2. The value range of b2 is 1.1–2.0. If the pump room satisfaction rate is still lower than 95% after adjustment, an "insufficient cooling capacity" alarm is triggered, prompting the additional unit to be started.

[0092] The fourth step is to generate standardized PLC instructions. The central loop converts the verified control target parameters into instruction formats that the PLC can recognize: the instruction for the ground cooling source PLC subgroup is "unit number - start / stop status - power setting value", the instruction for the secondary piping network PLC subgroup is "piping network number - pump frequency setting value - valve position setting value", and the instruction for the terminal equipment PLC subgroup is "area number - evaporation temperature setting value". The instructions are synchronously sent to each subgroup through the industrial bus to ensure coordinated equipment operation.

[0093] After receiving commands from the edge-center dual-loop feedback actuator, the PLC groups execute adjustment actions respectively: the ground cooling source PLC subgroup controls the compressor frequency (adjustment range 25%-100% of rated frequency) or the number of units starting and stopping to match the actual power with the target power value of the ground cooling source; the secondary piping network PLC subgroup adjusts the frequency of the circulating pump inverter (30-50 Hz) to change the pump speed, combined with the stepper motor action of the electric regulating valve (accuracy 0.1% opening), to stabilize the secondary network pressure difference to the target value; the terminal equipment PLC subgroup controls the refrigerant flow by adjusting the opening of the evaporator electronic expansion valve (0%-100%) to maintain the terminal evaporation temperature at the corrected target value. During the adjustment process, each PLC subgroup transmits equipment operation data back to the edge-center dual-loop feedback actuator in real time, forming a closed-loop control of "target-execution-feedback-correction" to ensure that the three core parameters continuously meet the constraints of the simplex of the execution strategy.

[0094] The function and effect of this step are as follows: The edge-center dual-loop feedback actuator achieves rapid local adjustment through the edge loop, solving the problem of 1-2 minute response lag in traditional centralized control. It can handle local thermal disturbances (such as sudden increases in equipment heat dissipation at the working face) within 30 seconds. The central loop performs global constraint verification, avoiding the shortcomings of traditional distributed control that ignore dew point risks and priorities, ensuring that the entire system does not exceed safety boundaries. PLC groups are linked by equipment type, combined with high-speed communication via the industrial bus, ensuring the synchronization of actions of the ground cold source, secondary piping network, and terminal equipment. This avoids cold energy delivery losses due to asynchronous actions, such as pressure waste caused by pump frequency adjustment but valve position not being moved. Furthermore, this step closely follows the S30 execution strategy simplex, driving the physical feasibility of PLC execution through simplex geometric safety constraints. It also reuses the heat source priority information from the dynamic directed graph of the mine's thermal and humidity field, ensuring priority protection in the critical zone. Compared to traditional open-loop control, this step improves the cold energy satisfaction rate in the critical zone, enhances system energy efficiency (COP), and completely solves the core pain points of slow local response and weak global constraints in mine cooling systems.

[0095] For example, when the execution strategy simplex is set as follows: 2-simplex setting "2 units, cooling output 1000kW" (corresponding to a ground cooling source power target of 400kW), 1-simplex setting "pump station network pressure difference 0.6MPa, valve position 90%, pump frequency 42Hz", and 0-simplex setting "pump station terminal temperature 20℃ (corresponding to evaporation temperature 5℃)", the edge loop collects the actual pump station temperature of 20.5℃ (deviation +0.5℃), network pressure difference of 0.58MPa (deviation -0.02MPa), and unit power of 380kW (deviation -20kW). It then quickly corrects the evaporation temperature to 5.1℃, the pump frequency to 43Hz, and the unit power to 390kW. Because the unit power of 390kW is still different from the target value of 400kW... There was a 10kW deviation in W. The edge loop collected power data again at 10-second intervals. The correction was calculated as "correction amount = 0.3 × (400-390) = 3kW", and the second correction was made to 393kW. If the deviation still exceeded 5kW after three consecutive corrections, the central loop was triggered to intervene. The unit load rate was adjusted (from 0.9 to 0.92) to further correct the deviation and ensure that the power met the standard. The central loop verification found that the humidity in the pump room was 65% and the temperature was 20.5℃. The calculation f(65%, 20.5℃) = 20.5 - 19.9 = 0.6 > 0 was performed. The terminal PLC was further instructed to reduce the humidity to 63%, so that the pump room parameters met the constraints. At the same time, the parameters of the working face and the tunneling face did not exceed the threshold, achieving the unity of priority in the critical zone and global stability.

[0096] Example 2:

[0097] This embodiment, based on Embodiment 1, provides a dynamic distribution and multi-level refrigeration coordinated control system for mine cooling load, such as... Figure 4 As shown, it includes:

[0098] Data acquisition and preprocessing module: used to acquire multi-source heterogeneous sensor data, and generate and output a dynamic directed graph of the mine's thermal and humid field through spatiotemporal alignment and thermodynamic normalization.

[0099] Markov Decision Chain Generation Module: Based on the dynamic directed graph of the mine's thermal and wet field, combined with the cold load operator and the equipment feasible region, a multi-objective rolling optimizer with heat source topology awareness is used to generate and output the Markov decision chain for cold load allocation.

[0100] Simple Complex Generation Module: Used to generate and output simple complexes with execution strategies under the drive of the Markov decision chain for cooling allocation through a hierarchical time-scaled collaborative controller;

[0101] Dual-loop feedback execution and equipment adjustment module: used to dynamically adjust the power of the ground cooling source, the secondary network pressure difference and the terminal evaporation temperature by linking the PLC group through the edge-center dual-loop feedback actuator under the simple complex constraint of the execution strategy.

Claims

1. A method for dynamic distribution of cooling load and multi-stage coordinated control of cooling systems in mines, characterized in that, The method includes: Collect multi-source heterogeneous sensor data, and generate and output a dynamic directed graph of the mine's thermal and humid field through spatiotemporal alignment and thermodynamic normalization. Based on the dynamic directed graph of the mine's thermal and humid field, and combined with the cooling load operator and the equipment feasible region, a Markov decision chain for cooling allocation is generated and output through a multi-objective rolling optimizer with heat source topology awareness. Driven by the Markov decision chain for cooling capacity allocation, a simple complex of execution strategies is generated and output through a hierarchical time-scaled collaborative controller. Under the constraint of the simple complex execution strategy, the power of the ground cooling source, the secondary network pressure difference and the terminal evaporation temperature are dynamically adjusted by linking the PLC group through the edge-center dual-loop feedback actuator; The structure of the dynamic directed graph of the mine thermal and humid field is as follows: nodes include location identification data, dry and wet bulb temperatures, air enthalpy, cooling load, heat source type label and WBGT risk level; edges include air volume, airway pressure difference, air temperature drop and humidity attenuation coefficient; the heat source priority partial order relationship is set in the order of pump room > working face > tunneling face, where > indicates priority over; The spatiotemporal alignment is as follows: using the unified clock in the mine's central control room as a reference, the acquisition time of all data in the multi-source heterogeneous sensor dataset is calibrated, and a time calibration dataset is generated after calibration; a spatial topology network is constructed based on the wind network heat flow path and interpolation is performed to generate a spatial alignment dataset; combined with the dynamic change law of the mine's thermal and humid field, the marked abnormal data in the time calibration dataset is repaired; after repair, a spatiotemporally consistent dataset is generated; if the deviation of the abnormal data exceeds the reasonable range, the reasonable data for the corresponding time point is calculated based on the synchronous data of adjacent areas in the spatial alignment dataset and the heat flow propagation speed. The cooling load operator is a mathematical operator that quantifies the urgency of cooling demand in each region by combining the normalized cooling load value, WBGT risk level weight, and heat source priority weight. The cooling load operator is calculated as follows: Cooling load operator value = ω1 × normalized cooling load value + ω2 × WBGT risk level weight + ω3 × heat source priority weight, where ω1, ω2, and ω3 are weight coefficients, and their sum is 1. The multi-objective rolling optimizer with heat source topology awareness includes: adaptively dividing the rolling optimization window; constructing a multi-objective optimization objective function, including minimizing the cooling load over-limit time, minimizing the total system energy consumption, and maximizing safety redundancy; setting multi-dimensional constraints, including equipment feasible region constraints, heat flow conduction constraints, and heat source priority constraints; and using an improved non-dominated sorting genetic algorithm to solve the problem. Each rolling optimization window outputs a set of non-dominated optimization schemes, which are concatenated in a time series to form an initial optimization scheme sequence.

2. The method for dynamic allocation of mine cooling load and multi-stage refrigeration coordinated control according to claim 1, characterized in that, The method for constructing the Markov decision chain for cold energy allocation is as follows: A Markov decision chain for cooling load allocation is constructed based on the initial optimization scheme sequence. The state space is a cooling load sorting vector, the action space is set by source-network-end collaboration, the transfer function is based on the next moment prediction data of the dynamic directed graph of the mine's thermal and wet field, and the reward function is a linear combination form.

3. The method for dynamic allocation of mine cooling load and multi-stage refrigeration coordinated control according to claim 1, characterized in that, The hierarchical time-scaled collaborative controller includes: Dynamically divide the timescale into fast, medium, and slow layers; Extract the control tasks corresponding to each layer from the action space of the Markov decision chain for cooling capacity allocation, and verify whether the cooling capacity output of the unit in the slow time scale layer can support the pipeline transmission demand of the medium time scale layer. Establish cross-timescale information exchange channels to achieve collaborative control of equipment.

4. The method for dynamic allocation of mine cooling load and multi-stage refrigeration coordinated control according to claim 1, characterized in that, The execution strategy simple complex is as follows: generate a 0-simplex corresponding to the control parameters of the terminal cooling area, a 1-simplex corresponding to the pipeline network coordinated control parameters, and a 2-simplex corresponding to the combination of source-side unit operating states, construct the dew point safety boundary, and integrate them to form a complete complex.

5. The method for dynamic allocation of mine cooling load and multi-stage refrigeration coordinated control according to claim 1, characterized in that, The edge-center dual-loop feedback actuator is: The three core regulation targets are analyzed from the simple complex form of the execution strategy, including the target value of the ground cooling source power, the target value of the secondary network pressure difference, and the target value of the terminal evaporation temperature; The edge ring is deployed locally in each PLC subgroup to collect the actual operating parameters of the equipment in real time, compare them with the control target parameters, calculate the deviation, and perform rapid correction. The central ring is deployed in the mine cooling control center, and the adjustment data of each edge ring is summarized every 1-5 minutes to verify global constraints; The central loop converts the verified control target parameters into an instruction format that the PLC can recognize.

6. A dynamic distribution and multi-stage refrigeration coordinated control system for mine cooling load, used to implement the dynamic distribution and multi-stage refrigeration coordinated control method for mine cooling load as described in any one of claims 1-5, characterized in that, The system includes: Data acquisition and preprocessing module: used to acquire multi-source heterogeneous sensor data, and generate and output a dynamic directed graph of the mine's thermal and humid field through spatiotemporal alignment and thermodynamic normalization. Markov Decision Chain Generation Module: Based on the dynamic directed graph of the mine's thermal and wet field, combined with the cold load operator and the equipment feasible region, a multi-objective rolling optimizer with heat source topology awareness is used to generate and output the Markov decision chain for cold load allocation. Simple Complex Generation Module: Used to generate and output simple complexes of execution strategies under the drive of the Markov decision chain of cooling allocation through a hierarchical time-scaled collaborative controller; Dual-loop feedback execution and equipment adjustment module: Under the constraint of simple complex execution strategy, it is used to dynamically adjust the power of the ground cooling source, the secondary network pressure difference and the terminal evaporation temperature by linking the PLC group through the edge-center dual-loop feedback actuator.

Images