A large-scale constrained dispatching method and system for a coal mine integrated energy system

Abstract

The application provides a coal mine comprehensive energy system large-scale constraint scheduling method and system, and belongs to the technical field of intelligent mine and energy system optimization scheduling. The method comprises the following steps: S1, collecting various types of data required for energy system scheduling in real time through multiple types of sensors combined with an information system, and completing key parameter prediction for the next 24 hours based on a machine learning method; S2, based on the prediction result of step S1, constructing a 24-hour scheduling optimization model of the coal mine comprehensive energy system, and solving the model by using a search direction learning evolutionary optimization algorithm; S3, converting and checking the optimization scheme obtained in step S2 to form executable instructions and issuing the instructions to each energy device, and monitoring the execution effect. The coal mine comprehensive energy system large-scale constraint scheduling method and system can realize dynamic balance of energy supply and demand and economic operation of the system.

Description

Technical Field

[0001] This invention relates to the field of smart mine and energy system optimization scheduling technology, and in particular to a large-scale constrained scheduling method and system for a coal mine integrated energy system. Background Technology

[0002] The coal mine integrated energy system is guided by the energy demand of coal mine production. It integrates external energy sources such as external power grids and external gas grids, associated energy sources such as coalbed methane, mine water waste heat, and exhaust wind waste heat, as well as renewable energy sources such as photovoltaic and wind power. Through energy conversion equipment, energy storage devices, and load regulation methods, it constructs a coupled system of source-grid-load-storage, aiming to achieve the goals of energy cascade utilization, associated resource recycling, controllable carbon emissions, and safe and reliable supply.

[0003] The optimization of integrated energy systems in coal mines faces severe technical challenges. At the system structure level, the deep coupling of various energy flows, including electricity, heat, and cooling, and the dynamic interaction among the source-grid-load-storage links result in complex mathematical models characterized by high dimensionality, nonlinearity, and strong constraints. At the operating environment level, wind and solar power output is intermittent and random, while core production loads such as mining and drainage fluctuate significantly; these uncertainties increase the difficulty of operational optimization. At the market mechanism level, time-of-use pricing policies and carbon emission trading mechanisms necessitate stronger market responsiveness in operational decisions.

[0004] The aforementioned factors collectively determine that the day-ahead scheduling optimization problem of a coal mine integrated energy system belongs to a large-scale constrained multi-objective optimization problem. The decision variables in this type of problem typically reach 300 to 400 dimensions, covering the operating power and on / off status of each piece of equipment at different time periods, while simultaneously needing to consider multiple optimization objectives such as economic cost, carbon emissions, and external dependence. As the number of decision variables and optimization objectives increases, the search space expands exponentially, triggering the curse of dimensionality, making it difficult for traditional multi-objective evolutionary optimization algorithms to effectively explore the vast search space, resulting in a sharp decline in search efficiency. Furthermore, the evolutionary population is prone to prematurely getting trapped in local optima or becoming overly divergent during the search process, making it difficult to find feasible or optimal solutions that meet the requirements of safety, economy, environmental protection, and reliability. Summary of the Invention

[0005] The purpose of this invention is to provide a large-scale constrained scheduling method and system for integrated energy systems in coal mines, in order to solve the technical problems in the prior art that lead to low search efficiency and difficulty in balancing multiple optimization objectives such as economic costs and carbon emissions due to the high dimensionality of decision variables and complex constraints, so as to achieve dynamic balance of energy supply and demand and efficient and economical operation of the system.

[0006] To achieve the above objectives, the present invention provides a large-scale constrained scheduling method for a coal mine integrated energy system, comprising the following steps: Step S1: Collect various types of data required for energy system scheduling in real time through multiple types of sensors and information systems, and predict key parameters for the next 24 hours based on machine learning methods to provide input for optimized scheduling; Step S2: Based on the prediction results of step S1, construct a 24-hour scheduling optimization model for the coal mine integrated energy system, and solve the model using a search direction learning evolution optimization algorithm; Step S3: Convert and verify the optimization scheme obtained in step S2, form executable instructions and send them to each energy device, and monitor the execution effect.

[0007] Preferably, in step S1, the multiple types of sensors include meteorological sensors, production data acquisition units, equipment status acquisition units, energy network acquisition units, and video image acquisition units; Among them, the video image acquisition unit is equipped with a machine vision-based physical state recognition system to identify the coal bunker level and water level. Key parameter forecasts include electricity load forecast, heat load forecast, renewable energy output forecast, associated energy availability forecast, and grid interaction parameter forecast. Among them, the electrical load prediction and heat load prediction adopt the spatiotemporal graph convolutional network model enhanced by attention mechanism; Photovoltaic power generation output prediction employs a method that integrates physical models with long short-term memory neural networks; Wind power output prediction adopts a method combining numerical weather prediction downscaling and extreme gradient enhancement model; The prediction of the available associated energy resources adopts a method combining a hydro-thermal coupling model and a graph neural network. The prediction of power grid interaction parameters adopts a hybrid model that combines the Prophet model and the limit gradient boosting model.

[0008] Preferably, in step S1, the prediction system adopts a three-stage architecture of offline training-online prediction-rolling update, and the model update triggering condition includes any of the following: Time-triggered, meaning an update is automatically triggered every 4 hours; Error trigger: The average absolute percentage error of the predictions over the most recent 24 hours is greater than 10%. Event triggering occurs when equipment maintenance plan information, policy adjustment instructions, or extreme weather warning information are received.

[0009] Preferably, in step S2, the search direction learning evolutionary optimization algorithm includes the following steps: Initialize the population and introduce a search direction learning module between crossover and mutation operations; Principal component analysis is used to extract the mainstream search direction from the population distribution. The search direction of the branch is learned based on the degree of constraint violation, which is used to guide infeasible solutions to move towards the feasible region; The decision variables are selectively activated by a variable importance identification method, and masks are maintained independently for the optimization objective and constraints to guide evolution.

[0010] The preferred learning process for mainstream search directions includes: Calculate the population center and centralize each individual; Calculate the covariance matrix and perform eigenvalue decomposition, retaining the first half of the principal components; The population is divided into a first group and a second group based on the alignment score of each individual with the principal components, with the alignment score of the first group being higher than that of the second group; The mainstream direction vector is defined as the vector from an individual randomly selected from the second group to an individual randomly selected from the first group, and after introducing Gaussian perturbation, it is stored in the search direction library.

[0011] Preferably, the learning process for the tributary search direction includes: Calculate the degree of violation of each constraint for each individual, and calculate the learning rate based on the degree of violation; The population is divided into a third group and a fourth group based on the constraint violation value, with the constraint violation value of the third group being lower than that of the fourth group. The tributary direction vector is defined as the vector from an individual randomly selected from the fourth group to an individual randomly selected from the third group. After multiplying by the learning rate and introducing random perturbation, it is stored in the search direction library.

[0012] Preferred methods for identifying variable importance include: Estimate the gradient of each decision variable with respect to each constraint, and calculate the combined impact of each decision variable on feasibility; Analysis of variance was used to identify decision variables that significantly affect the optimization objective. Calculate the original importance score for each decision variable and normalize it; The decision variables are activated through a probabilistic selection mechanism, and the masks are maintained independently for the optimization objective and constraints respectively.

[0013] Preferably, in step S2, the 24-hour scheduling optimization model for the coal mine integrated energy system includes the following constraints: Equipment output upper and lower limits constraints, external power purchase constraints, electric power balance constraints, thermal power balance constraints, electric thermal energy storage equipment operation constraints, total coal gangue power generation constraints, and total water inrush heat extraction constraints. The optimization objectives of the model include minimizing economic costs, minimizing carbon emissions, minimizing dependence on external power grids, and minimizing the rate of renewable energy abandonment.

[0014] Preferably, the process of converting the optimized scheme into executable instructions in step S3 includes: Convert the scheduling scheme selected from the Pareto solution set into device-level control commands; All device controllers are synchronized with the network time protocol time server of the dispatch center. Each control command contains a precise execution timestamp and supports command preloading. Control commands are sent to photovoltaic inverters, wind turbine controllers, energy storage converters, heat pump controllers, electric boilers, and smart meters via Modbus TCP, IEC 61400-25, EtherCAT, BACnet / IP, Profinet, and DL / T645-2007 protocols, respectively.

[0015] This invention also provides a large-scale constrained scheduling system for a coal mine integrated energy system, used to implement the large-scale constrained scheduling method for a coal mine integrated energy system as described above, comprising: The data acquisition and prediction module is used to collect various types of data required for energy system scheduling in real time through multiple sensors and information systems, and to predict key parameters for the next 24 hours based on machine learning methods, providing input for optimized scheduling. The optimization scheduling module is used to construct a 24-hour scheduling optimization model for the coal mine integrated energy system based on the prediction results of the data acquisition and prediction module, and to solve the model using a search direction learning evolution optimization algorithm. The instruction execution and monitoring module is used to convert and verify the optimization scheme obtained by the optimization scheduling module, form executable instructions and send them to each energy device, and monitor the execution effect.

[0016] Therefore, the present invention employs the above-mentioned large-scale constrained scheduling method and system for a coal mine integrated energy system, and the beneficial technical effects are as follows: (1) This invention comprehensively considers multiple optimization objectives such as the operation and maintenance cost of electric heating equipment, the cost of purchasing electricity from the power grid, carbon emissions, the degree of dependence on the external power grid, and the waste rate of renewable energy. It generates a Pareto optimal solution set through a multi-objective optimization method, providing users with a variety of optional scheduling schemes, which facilitates flexible selection based on actual operating needs and preferences.

[0017] (2) This invention addresses the day-ahead scheduling problem of a coal mine integrated energy system, outputting the output plan of each device for the next 24 hours, covering a variety of devices such as electrochemical energy storage, circulating fluidized bed boilers, photovoltaic power generation, wind power generation, heat pumps, electric boilers, and hot water storage tanks, and meets multiple constraints such as equipment operation constraints, power balance constraints, and energy storage constraints, ensuring the safety and feasibility of the scheduling scheme.

[0018] (3) The present invention adopts a search direction learning evolutionary optimization algorithm, combined with a variable importance identification strategy, which can efficiently guide the search direction in a large-scale decision variable space, significantly improve the convergence speed and solution quality, quickly obtain a high-quality scheduling scheme under limited computing resources, and distribute the scheme to each equipment controller through a standardized communication protocol to realize the automated closed-loop operation of the coal mine integrated energy system.

[0019] (4) The present invention constructs an integrated scheduling framework from data acquisition, parameter prediction, optimization solution to instruction execution, which has complete system integration capability and independent operation capability, can adapt to the real-time scheduling needs in the complex environment of the mining area, and improves the integrity and operational robustness of the system. Attached Figure Description

[0020] Figure 1 Diagram of the operational architecture of a coal mine integrated energy system; Figure 2 A flowchart of a large-scale constrained scheduling method for a coal mine integrated energy system; Figure 3 Flowchart of a large-scale multi-objective optimization algorithm; Figure 4 This is a schematic diagram of the final scheduling plan issuance process. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0023] Example 1 This embodiment provides a large-scale constrained scheduling method for a coal mine integrated energy system, applied to a day-ahead scheduling scenario of an integrated energy system in a certain mining area. This mining area is equipped with various energy devices, including photovoltaic power generation systems, wind power generation systems, circulating fluidized bed boilers, electrochemical energy storage devices, hot water storage tanks, water source heat pumps, exhaust air source heat pumps, and electric boilers, which need to meet the electricity and heat demands for production and daily life in the mining area. Figure 1 This is a diagram of the operational architecture of the integrated energy system in a coal mine used in this embodiment, showing the coupling relationships between various energy devices and the direction of energy flow.

[0024] Figure 2 The flowchart shows a large-scale constrained scheduling method for a coal mine integrated energy system. The specific steps are as follows: Step S1 involves collecting various types of data required for energy system scheduling in real time through multiple types of sensors and information systems, and using machine learning methods to predict key parameters for the next 24 hours, providing input for optimized scheduling.

[0025] Meteorological sensors: collect wind speed and direction information, collect total radiation, diffuse radiation, and direct radiation on the horizontal surface, and measure its temperature and relative humidity.

[0026] Production data acquisition unit: Collects data from the coal mining face, including the face advance speed (m / h), coal mining machine operating status, and instantaneous output (t / h). The integrated belt conveyor system and mine car transport system collect belt speed and transport capacity (t / h) via PLC at a frequency of 30 seconds.

[0027] The equipment status acquisition unit collects data on the following: Solar power generation data acquisition unit collects DC voltage / current, AC power, power generation, conversion efficiency, and component temperature; wind power generation data acquisition unit collects wind speed, rotational speed, blade pitch angle, generator temperature, and vibration data; circulating fluidized bed boiler operating data acquisition unit collects steam pressure / temperature, coal feed rate, bed temperature, oxygen content, and flue gas parameters; heat pump system data acquisition unit collects evaporation / condensation temperature, compressor current, COP (Coefficient of Performance) value, and inlet / outlet water temperature information from the water source heat pump and exhaust air heat pump controllers; electric boiler heating tube status, power factor, and water temperature / level are collected; electrochemical energy storage system collects battery pack voltage / current, SOC (State of Charge), SOH (State of Health), temperature distribution, and alarm information through the BMS (Battery Management System); and thermal storage tank data acquisition unit collects temperature of each layer of the tank, inlet / outlet water temperature / flow rate, and pressure.

[0028] Energy network status acquisition unit: The power metering network collects total active / reactive power, voltage and current harmonics, and power factor, covering more than 95% of the power load; in the heat metering network, it collects supply and return water temperature, flow rate, and cumulative heat.

[0029] The video and image acquisition unit uses the U-Net deep learning segmentation algorithm to identify the coal pile outline and ensure that the material level height measurement error is <5cm; in the water tank level visual recognition, the edge detection + scale recognition algorithm is used to identify the water level and ensure that the water level measurement error is <1cm.

[0030] By integrating testing data and performing quality control, key parameters can be predicted for the next 24 hours, specifically: Data integration and processing utilizes different transmission cycles for different data types. Real-time control data: 10-100ms cycle; process monitoring data: 1-10s cycle; statistical and analysis data: 1-15 minutes cycle. In data cleaning and preprocessing, outlier detection employs a 3x3... The criteria combine logical limit checks and missing value handling: linear interpolation is used for time series data, mode imputation is used for categorical data, and first-order lag filtering is used for process data. Finally, data standardization is achieved, and all data are converted to SI (International System of Units).

[0031] The prediction system adopts a three-stage architecture of "offline training - online prediction - rolling update", which includes: Electricity / Heat Load Forecasting Model: The model employs an attention-enhanced Spatiotemporal Graph Convolutional Network (ASTGCN), consisting of a graph convolutional network module, a temporal convolutional module, an attention mechanism module, and an external factor fusion module. Specifically, the graph convolutional network module extracts spatial correlation features of electricity / heat loads in different areas of the mine (e.g., load correlation between underground working faces and surface auxiliary facilities); the temporal convolutional module captures the time-dependent features of load data (e.g., intraday peak-valley patterns and seasonal fluctuations); the attention mechanism module dynamically weights load information from different times and regions, highlighting key influencing factors; and the external factor fusion module integrates external variables such as mine production plans, ambient temperature, and seasonal variations to improve forecast accuracy. The model training data uses historical data from the past year (approximately 8760 hours), and is validated using time-series cross-validation and a sliding window method. The forecast performance requirement is that the mean absolute percentage error (MAPE) of electricity and heat load forecasts be <5%.

[0032] Renewable Energy Output Prediction Model: This model includes a photovoltaic power generation output prediction module and a wind power generation output prediction module. Each module employs a specific fusion model, as detailed below: (1) Photovoltaic power output prediction module: A hybrid model integrating a physical model and an LSTM neural network is adopted. This model consists of two parts: a physical model baseline module and an LSTM residual correction module. The physical model baseline module is based on basic parameters such as photovoltaic module parameters, solar irradiance, and ambient temperature. It outputs the photovoltaic power output baseline prediction value through the physical calculation formula. The LSTM residual correction module takes the physical model prediction residual sequence, cloud cover prediction data, and module cleanliness data as inputs. It captures the time dependence of the residual sequence through the gating unit of the LSTM network and dynamically outputs the correction coefficient of the physical model to achieve accurate correction of the baseline prediction value and improve the accuracy of photovoltaic power output prediction.

[0033] (2) Wind power output prediction module: The module adopts the method of fusion of numerical weather prediction (NWP) downscaling and extreme gradient boosting (XGBoost). The module consists of an NWP downscaling module, a feature extraction module, and a prediction interval generation module. The NWP downscaling module uses ECMWF high-resolution forecast data and downloaded relevant parameters as data sources. It calculates the local wind flow field in the mining area through OpenFOAM software, completes terrain correction and roughness correction, and transforms the macro weather forecast data into micro meteorological data of the area where the wind turbines are located in the mining area. The feature extraction module uses the XGBoost algorithm to mine the nonlinear correlation features between micro meteorological data (such as wind speed and wind direction) and wind power output. The prediction interval generation module combines XGBoost and quantile regression forest (QRF) algorithms to output the prediction interval of wind power output and quantify the prediction uncertainty.

[0034] Associated Energy Availability Prediction Model: This model includes a mine water inflow heat prediction module and a mine exhaust air heat prediction module. The two modules employ targeted coupling models, as described below: (1) Mine water inflow heat prediction module: A hydrological-thermal coupling model is established, which consists of a hydrological parameter calculation module, a geothermal gradient module, an LSTM network correction module, and a heat calculation module. The hydrological parameter calculation module is used to obtain basic hydrological data such as flow rate and velocity of mine water inflow; the geothermal gradient module calculates the geothermal distribution at different depths based on geological data of the mining area to determine the benchmark value of water inflow temperature; the LSTM network correction module combines historical water inflow temperature data and geological structure change data to correct the benchmark temperature value and improve the accuracy of temperature prediction; the heat calculation module calculates the available heat of mine water inflow based on the corrected water inflow temperature and flow rate using the heat calculation formula.

[0035] (2) Mine exhaust air heat prediction module: A ventilation network-thermodynamic joint model is established, which consists of a graph neural network (GNN) air volume prediction module, an air temperature prediction module, and an available heat calculation module. The graph neural network (GNN) air volume prediction module takes the mine ventilation network topology and fan operating parameters as input, learns the dynamic characteristics of the ventilation network (such as air volume distribution law, resistance change, etc.) through the GNN network, and outputs the air volume prediction value of each ventilation node; the air temperature prediction module calculates the air temperature baseline value based on the heat balance equation, and outputs the accurate air temperature prediction value by combining the neural network to correct interference factors such as ambient temperature and ventilation resistance; the available heat calculation module calculates the available heat of the mine exhaust air based on the air volume and air temperature prediction results and the principle of heat exchange.

[0036] Power Grid Interaction Parameter Prediction Model: This model includes a time-of-use electricity price prediction module and a power purchase limit prediction module, and is equipped with a rolling update mechanism for the prediction model. Details are as follows: (1) Time-of-use electricity price prediction module: A multi-factor driven hybrid model of Prophet and Extreme Gradient Boosting (XGBoost) is adopted. This model consists of a Prophet trend prediction module, an XGBoost feature enhancement module, and a fusion output module. The Prophet trend prediction module is used to capture the long-term trend, seasonal fluctuations, and holiday effects of time-of-use electricity prices and output the basic predicted value of electricity prices. The XGBoost feature enhancement module takes multiple factors such as electricity market supply and demand data, grid dispatching policies, and fuel prices as inputs, explores the nonlinear correlation between each factor and electricity prices, and corrects the basic predicted value. The fusion output module combines the prediction results of the two modules to output accurate time-of-use electricity price prediction values.

[0037] (2) Electricity Purchase Cap Prediction Module: A model combining grid dispatch instruction parsing and load forecasting back-calculation is adopted. This model consists of a dispatch instruction parsing module, a load forecasting back-calculation module, and a probability distribution output module. The dispatch instruction parsing module is used to parse the dispatch instructions issued by the grid and extract the relevant constraints for electricity purchase; the load forecasting back-calculation module combines the future load forecast results of the mine to back-calculate the upper limit of electricity purchase demand for different time periods; the probability distribution output module outputs the probability distribution of the upper limit of electricity purchase for each time period in the next 24 hours based on historical electricity purchase data and the fluctuation of dispatch constraints, quantifying the uncertainty of electricity purchase constraints.

[0038] (3) Rolling update mechanism for prediction models: An online learning framework is adopted, which consists of a trigger condition judgment module, a model update module, and a prediction quality monitoring module. The trigger condition judgment module monitors three update trigger conditions in real time (automatic trigger every 4 hours, prediction error exceeding the threshold (MAPE>10%) in the last 24 hours, and receipt of major event information). An update is triggered when any condition is met. The model update module adjusts the parameters of each prediction model online based on the latest historical data and real-time monitoring data to ensure model adaptability. The prediction quality dashboard is driven by the prediction quality monitoring module, which monitors three core indicators in real time: point prediction accuracy, interval prediction accuracy, and peak prediction capability. At the same time, it quantifies the uncertainty of each prediction module and transmits it to the optimization layer to provide support for mine comprehensive energy optimization decision-making.

[0039] Step S2: Based on the prediction results of step S1, a 24-hour scheduling optimization model for the coal mine integrated energy system is constructed, and the model is solved using a search direction learning evolution optimization algorithm to obtain the optimization scheme.

[0040] To meet the diverse electricity and heat demands of integrated energy systems in coal mines, it is necessary to adjust the output of operating equipment. During this adjustment process, specific equipment and operating scenarios must meet corresponding constraints, including: electrochemical energy storage capacity limitations, hot water storage tank capacity limitations, electrical load balancing, thermal load balancing, coal gangue power generation limitations, and waste heat recovery limitations. The equipment operation plan must optimize costs related to the operation and maintenance of electric heating equipment and grid purchases, carbon emissions, external grid dependence, and the system's renewable energy waste rate.

[0041] 1) To facilitate problem switching under multiple decision variables and multiple optimization objectives, two sets of variables are defined to represent the optimization objective and the equipment state, as follows: (1); (2); in, Indicate whether to consider the first The indicator variables of the optimization objective, These represent economic targets, carbon dioxide emissions, external grid dependence, and the minimum rate of wind and solar curtailment, respectively. Indicates device The indicator variable for the working status, express Time period Indicates different devices, These represent electrochemical energy storage, circulating fluidized bed boiler, photovoltaic power generation equipment, wind power generation equipment, water source heat pump, electric boiler, hot water storage tank, and exhaust air source heat pump, respectively. Based on the above settings, the optimization objectives, constraints, and specific model design of the coal mine integrated energy system are as follows.

[0042] The system's equipment should meet the upper and lower limits of equipment output constraints, namely: (3); in, They represent The actual output of photovoltaic power generation, wind turbines, circulating fluidized bed boilers, coal gangue power generation, water source heat pumps, and electric boilers during the specified time period. They represent The maximum output of photovoltaic power generation, wind turbines, circulating fluidized bed boilers, coal gangue power generation, water source heat pumps, and electric boilers during the specified time period. This indicates the total number of time periods within the scheduling cycle.

[0043] The system imposes constraints on external power purchases: (4); in, express The actual power purchased by the time-period system from the external power grid. express The maximum power that the system is allowed to purchase from the external power grid during a given time period.

[0044] For the system to achieve supply and demand balance, it is necessary to satisfy the power balance of the power grid and the heat balance of the heating network, as expressed below: (5); (6); in, They represent Total electricity load in mining areas 1 and 2 during the specified time period; They represent Heat load of mining areas 1 and 2 during the specified time period; These represent the actual power per unit time for wind power generation, photovoltaic power generation, circulating fluidized bed boiler coal gangue power generation, electrochemical energy storage device energy storage, electrochemical energy storage device discharge, exhaust air source heat pump heating, water source heat pump heating, electric boiler heating, hot water storage tank heat storage, and hot water storage tank heat release, respectively. They represent The energy storage / discharge status of the time-limited electrochemical energy storage device (hot water storage tank), where 1 indicates active and 0 indicates inactive. These represent the energy efficiency ratios of the exhaust air source heat pump, the water source heat pump, and the electric boiler, respectively.

[0045] The operating constraints of the system's electric thermal energy storage equipment are as follows: (7); in, They represent Maximum energy storage / discharge power of time-limited electrochemical energy storage device (hot water storage tank); They represent The energy storage / discharge status of the electrochemical energy storage device (hot water tank) during a given time period, where 1 indicates active and 0 indicates inactive; These respectively represent electrochemical energy storage devices and hot water storage tanks. Energy stored over a period of time; These represent the maximum capacity of the electrochemical energy storage device (hot water tank); These represent the upper and lower limits of the energy stored in the electrochemical energy storage device (hot water tank); These represent the energy storage / discharge efficiency and static energy efficiency of the electrochemical energy storage device (hot water tank), respectively.

[0046] The daily power output constraint for coal gangue power generation can be expressed as: (8); in, express The actual output of coal gangue power generation from a circulating fluidized bed boiler during a specific time period. This indicates the maximum daily output of coal gangue power generation using a circulating fluidized bed boiler.

[0047] The daily heat extraction constraint can be expressed as: (9); in, express The actual heat extraction power of water inrush heat extraction during a given period This indicates the maximum daily heat extraction from water flow.

[0048] 2) The optimization objectives of the system are economic cost, carbon emissions, external grid dependence, and renewable energy abandonment rate.

[0049] System optimization aims to minimize equipment operation and maintenance costs and transaction costs for external electricity purchases, i.e., to minimize economic costs. This is defined as follows: (10); in, This represents the total economic cost. This indicates the cost of purchasing electricity from external sources. This represents the total cost of operation and maintenance for all equipment.

[0050] The settlement of electricity purchased from outside the system follows the time-of-use pricing regulations: (11); in, Indicates the day Purchased power during the time period express Electricity purchase price during certain time periods.

[0051] The calculation method for the operation and maintenance costs of the system's core energy equipment is as follows: (12); in, The figures represent the unit time and unit power operation and maintenance costs (RMB / MW) for wind power, photovoltaic, circulating fluidized bed boilers, electrochemical energy storage devices, exhaust air source heat pumps, water source heat pumps, electric boilers, and hot water storage tanks, respectively.

[0052] Through system optimization, the goal is to minimize the system's carbon emissions, i.e.: (13); in, Indicates system carbon emissions, , These represent the carbon emission factors (kgCO2 / MWh) of coal gangue power generation and local grid electricity purchases, respectively.

[0053] To improve system stability and minimize dependence on the external power grid, namely: (14); in, It indicates the degree of dependence on the external power grid, expressed as the proportion of externally purchased electricity in the total electricity supply.

[0054] In addition to the above objectives, reducing the waste rate of renewable energy can not only improve economic efficiency but also reduce carbon emissions, namely: (15); in, This indicates the rate of abandonment of renewable energy.

[0055] In summary, the operational optimization model for the integrated energy system of a mine with combined electricity and heat can be expressed as follows: (16); in, , , , These represent the weights for economic costs, carbon emissions, grid dependence, and renewable energy abandonment rates, respectively.

[0056] The optimization model constructed is solved using a large-scale constrained scheduling method for integrated energy systems in coal mines, which employs search direction learning. Figure 3 As shown, the specific steps include: Initialize the population to have a size of NP and a dimension of . population In the environmental pool Individuals are selected, and Search Direction Learning (SDL) is performed between crossover and mutation operations. To balance the effectiveness and diversity of the guidance, the SDL module adopts a phased strategy. The early stage (proportion of function evaluation times) uses traditional differential mutation for exploration; the later stage utilizes random directions from the learned direction library SD to guide evolution. (17); in, express In the middle of the generation Improved individual information express In the middle of the generation Individual information after individual cross-operation It is a random value in the range [0, 1], which controls the random learning scale. They are for individuals Two individuals are randomly selected to provide learning direction. This refers to the learning direction of the direction learning library. Indicates the maximum number of evaluations. This indicates the current evaluation count. The mainstream direction is extracted from the population distribution using principal component analysis (PCA), which essentially finds the direction with the largest data variance, representing the main evolutionary trend of the population. First, the population center is calculated, and each individual is centered to eliminate translation invariance. ,in Represents the expected value of the population. Indicates population size (i.e., the number of individuals). express generation Information on individual decision variables, Represent the difference between an individual's value and the expected value; calculate the covariance matrix to capture the correlation between variables: ,in, It is a D×D covariance matrix, reflecting the linear correlation between decision variables. The covariance matrix is ​​then subjected to eigenvalue decomposition. ,in This is an eigenvalue diagonal matrix, and its diagonal elements are... , That is, eigenvalues. The eigenvector matrix, yes The transpose of the vector retains the first D / 2 principal components (eigenvectors), whose importance is determined by the normalized eigenvalues. Measurement, among which These are the normalized eigenvalues. These are eigenvalues, whose values ​​represent the importance of a dimension. It is a representative of the eigenvectors.

[0057] Calculate the weighted alignment score of each individual with the top D / 2 principal components. Divide the population into two groups based on the alignment score: high alignment (A1) and low alignment (A2). The mainstream direction vector is defined as the vector pointing from random individuals in A2 to random individuals in A1, and a Gaussian perturbation is introduced to enhance robustness. Store it in the search direction library SD, where It is the mainstream learning direction for generation. The Gaussian perturbation factor is... These are random individuals from A1 and A2, used to provide the mainstream direction.

[0058] The branch search direction is for each constraint According to the degree of constraint violation Come to learn. For the first The individual on the first The degree of constraint violation is used to repair infeasible solutions. A learning rate reflecting the severity of overall constraint violations is calculated: (18); in, The learning rate indicates the severity of constraint violation. It is the minimum value (i.e., 0.000001, to avoid the denominator being 0). For the first among all individuals The constraint with the lowest violation Individual violation information, For the first among all individuals The constraint with the highest violation Individual violation information, A larger value indicates a more severe constraint violation, requiring stronger guidance. Based on the constraint violation values, the population is divided into two groups: one with low violation (B1) and the other with high violation (B2). The tributary direction vector is defined as a vector pointing from a random individual in B2 to a random individual in B1, multiplied by the learning rate and a random perturbation. ,in, To constrain The tributary study direction, These are random individuals from B1 and B2, used to provide branch directions, which are stored in the search direction library SD. This direction approximates the negative gradient direction of the constraint function, guiding the population to move towards the feasible region.

[0059] Variable importance identification resolves the conflict between objective optimization and constraint satisfaction by selectively activating variables, thereby reducing computational overhead. This is achieved through the selective activation of each variable. For each constraint gradient estimation : (19); in, It is a gradient estimate obtained through calculation. yes The The values ​​of the variables, It is the first in the population The average constraint violation rate of each constraint. It is the first The first individual The degree of constraint violation of a constraint. It is the first in the population The mean of each variable is used. The square root of the sum of squares is calculated to measure the combined impact of the variables on overall feasibility. ,in Indicates the number of constraints. Used to measure the importance of variables to overall feasibility. The larger the variable, the more critical it is to the feasibility of the constraint.

[0060] Identify variables that have a significant impact on the optimization objective: (20); in, Represents variance. For the first There are two objective functions, where NP represents the population size. Indicates the number of targets. Indicates the first The first generation Individual target value, No. generation Individuals in the goal The values ​​of the decision variables in the dimension. The larger the value, the more important the variable is to the objective optimization; then the original importance score is calculated: (twenty one); in, It is a variable Indicators of importance in mainstream learning directions It is a variable Under constraints The importance of the sub-disciplines of learning is assessed by normalizing the original importance scores: (twenty two); Determine the activation variable through normalization and scaling: (twenty three); in, Probability scaling The importance parameter of the variable in the direction of the tributary. It is probabilistically scaled. The importance of variables in the mainstream direction. This indicates taking the minimum value, parameter The value is set to 0.5, which directly regulates the expected participation rate and ensures computational efficiency by controlling sparsity. The median scaling factors for objective importance and constraint importance provide robust normalization, preventing extreme values ​​from dominating the probability distribution while preserving effective discrimination between variables. Finally, probabilistic selection is used to activate the decision variables. (twenty four); in, Representing variables respectively Whether or not one studies within the mainstream or constrained learning direction. It is a random value in [0, 1]. Each objective and constraint maintains an independent mask. In direction learning, the decision variables are optimized independently based on the masks of the objective and constraints.

[0061] Step S3: Select a solution from the Pareto solution set as the final scheduling scheme, and transmit it to different devices through various protocols to guide their operation, such as... Figure 4 As shown, it includes the following steps: The 24-hour scheduling scheme output from the optimized model is converted into device-level control commands. All device controllers are synchronized with the dispatch center's NTP (Network Time Protocol) time server. Commands include precise execution timestamps and support command preloading (sent to the device buffer 5-10 minutes in advance). Modbus TCP is used to send the scheme to the photovoltaic inverters, IEC 61400-25 protocol to the wind turbine controllers, EtherCAT to the energy storage PCS, BACnet / IP to the heat pump controllers, Profinet protocol to schedule electric boiler heat production, and DL / T645-2007 protocol to control smart meters. By controlling each controller to adjust the output of each device and energy trading, the efficient operation of the coal mine's integrated energy system is achieved.

[0062] The final scheduling scheme obtained by the method has the following equipment electrical power output (kW) as shown in Table 1 and the final scheduling scheme has the following equipment thermal power output (kW) as shown in Table 2.

[0063] Table 1. Equipment power output at 24-hour intervals

[0064] Table 2 Equipment thermal power output at 24 hours

[0065] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0066] Therefore, the present invention adopts the above-mentioned large-scale constrained scheduling method and system for integrated coal mine energy systems. By integrating multi-source data acquisition, key parameter prediction, multi-objective optimization solution and equipment-level instruction execution into an integrated framework, it can quickly obtain the optimal scheduling scheme that takes into account economy, environmental protection and reliability under the condition of meeting complex operating constraints, and effectively improve the intelligent operation level of integrated coal mine energy systems.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A large-scale constrained scheduling method for a coal mine integrated energy system, characterized in that, Includes the following steps: Step S1: Collect various types of data required for energy system scheduling in real time through multiple types of sensors and information systems, and predict key parameters for the next 24 hours based on machine learning methods. Step S2: Based on the prediction results of step S1, construct a 24-hour scheduling optimization model for the coal mine integrated energy system, and use the search direction learning evolution optimization algorithm to solve the model and obtain the optimization scheme. Step S3: Convert and verify the optimization scheme obtained in step S2, form executable instructions and send them to each energy device, and monitor the execution effect; In step S1, the various types of sensors include meteorological sensors, production data acquisition units, equipment status acquisition units, energy network acquisition units, and video image acquisition units; Among them, the video image acquisition unit is equipped with a machine vision-based physical state recognition system to identify the coal bunker level and water level. Key parameter forecasts include electricity load forecast, heat load forecast, renewable energy output forecast, associated energy availability forecast, and grid interaction parameter forecast. Among them, the electrical load prediction and heat load prediction adopt the spatiotemporal graph convolutional network model enhanced by attention mechanism; Photovoltaic power generation output prediction employs a method that integrates physical models with long short-term memory neural networks; Wind power output prediction adopts a method combining numerical weather prediction downscaling and a limit gradient enhancement model; The prediction of the available associated energy resources adopts a method combining a hydro-thermal coupling model and a graph neural network. The prediction of power grid interaction parameters adopts a hybrid model that combines the Prophet model and the limit gradient boosting model; In step S2, the search direction learning evolutionary optimization algorithm includes the following steps: Initialize the population and introduce a search direction learning module between crossover and mutation operations; Principal component analysis is used to extract the mainstream search direction from the population distribution. The search direction of the branch is learned based on the degree of constraint violation, which is used to guide infeasible solutions to move towards the feasible region; Decision variables are selectively activated using a variable importance identification method, and masks are maintained independently for both the optimization objective and constraints to guide evolution in a specific direction. The learning process for mainstream search directions includes: Calculate the population center and centralize each individual; Calculate the covariance matrix and perform eigenvalue decomposition, retaining the first half of the principal components; The population is divided into a first group and a second group based on the alignment score of each individual with the principal components, with the alignment score of the first group being higher than that of the second group; The mainstream direction vector is defined as the vector from an individual randomly selected from the second group to an individual randomly selected from the first group, and after introducing Gaussian perturbation, it is stored in the search direction library. The learning process for tributary search includes: Calculate the degree of violation of each constraint for each individual, and calculate the learning rate based on the degree of violation; The population is divided into a third group and a fourth group based on the constraint violation value, with the constraint violation value of the third group being lower than that of the fourth group. The tributary direction vector is defined as the vector from an individual randomly selected from the fourth group to an individual randomly selected from the third group. After multiplying by the learning rate and introducing random perturbation, it is stored in the search direction library.

2. The large-scale constrained scheduling method for a coal mine integrated energy system according to claim 1, characterized in that, In step S1, the prediction system adopts a three-stage architecture of offline training, online prediction, and rolling update. The model update triggering condition includes any of the following: Time-triggered, meaning an update is automatically triggered every 4 hours; Error trigger: The average absolute percentage error of the predictions over the most recent 24 hours is greater than 10%. Event triggering occurs when equipment maintenance plan information, policy adjustment instructions, or extreme weather warning information are received.

3. The large-scale constrained scheduling method for a coal mine integrated energy system according to claim 2, characterized in that, Methods for identifying variable importance include: Estimate the gradient of each decision variable with respect to each constraint, and calculate the combined impact of each decision variable on feasibility; Analysis of variance was used to identify decision variables that significantly affect the optimization objective. Calculate the original importance score for each decision variable and normalize it; The decision variables are activated through a probabilistic selection mechanism, and the masks are maintained independently for the optimization objective and constraints respectively.

4. The large-scale constrained scheduling method for a coal mine integrated energy system according to claim 1, characterized in that, In step S2, the 24-hour scheduling optimization model for the coal mine integrated energy system includes the following constraints: Equipment output upper and lower limits constraints, external power purchase constraints, electric power balance constraints, thermal power balance constraints, electric thermal energy storage equipment operation constraints, total coal gangue power generation constraints, and total water inrush heat extraction constraints. The optimization objectives of the model include minimizing economic costs, minimizing carbon emissions, minimizing dependence on external power grids, and minimizing the rate of renewable energy abandonment.

5. A large-scale constrained scheduling method for a coal mine integrated energy system according to claim 1, characterized in that, The process of converting the optimized scheme into executable instructions in step S3 includes: Convert the scheduling scheme selected from the Pareto solution set into device-level control commands; All device controllers are synchronized with the network time protocol time server of the dispatch center. Each control command contains a precise execution timestamp and supports command preloading. Control commands are sent to photovoltaic inverters, wind turbine controllers, energy storage converters, heat pump controllers, electric boilers, and smart meters via Modbus TCP, IEC 61400-25, EtherCAT, BACnet / IP, Profinet, and DL / T645-2007 protocols, respectively.

6. A large-scale constrained scheduling system for a coal mine integrated energy system, characterized in that, A method for implementing a large-scale constrained scheduling of a coal mine integrated energy system as described in any one of claims 1-5 includes: The data acquisition and prediction module is used to collect various types of data required for energy system scheduling in real time through multiple sensors and information systems, and to predict key parameters for the next 24 hours based on machine learning methods, providing input for optimized scheduling. The optimization scheduling module is used to construct a 24-hour scheduling optimization model for the coal mine integrated energy system based on the prediction results of the data acquisition and prediction module, and to solve the model using a search direction learning evolution optimization algorithm. The instruction execution and monitoring module is used to convert and verify the optimization scheme obtained by the optimization scheduling module, form executable instructions and send them to each energy device, and monitor the execution effect.

Images