Intelligent Energy Consumption Control Method for Underground Reverse Osmosis Purification Stations in Coal Mines

Abstract

This invention provides an intelligent energy consumption control method for underground reverse osmosis purification stations in coal mines, relating to the field of coal mine water supply and energy consumption regulation technology. The method collects operation data of the purification station and water, electricity, and electricity price data for the entire mine, constructing a joint state description model that simultaneously characterizes the water supply status of the purification station, water consumption behavior of the entire mine, and electricity consumption behavior of the entire mine. Subsequently, based on this model, a dual prediction model for water and electricity is trained to obtain the predicted water demand sequence, electricity load sequence, and electricity price time series for each predicted period. Combining the water tank level and the adjustable output range of the generating unit, the pre-production water volume range and the post-production water volume range for each predicted period are derived. A scheduling strategy is executed in each control cycle, and the prediction model is corrected online based on operational deviations, forming a closed-loop intelligent energy consumption control system for water and electricity collaborative peak shaving and valley filling for the purification station. This can reduce energy costs, smooth peak loads, and improve water supply stability.

Description

Technical Field

[0001] This invention relates to the field of intelligent energy consumption control technology, and in particular to an intelligent energy consumption control method based on reverse osmosis purification stations in underground coal mines. Background Technology

[0002] As a key infrastructure for ensuring mine water supply and working environment, the operation of underground reverse osmosis purification stations in coal mines is subject to multiple constraints, including the strong fluctuations in underground water supply demand, the dramatic fluctuations in time-of-use electricity pricing, the coupled disturbances of underground power load, and the safety limitations of pressure difference in reverse osmosis membrane modules.

[0003] Existing water purification plants generally adopt fixed-frequency water production or fixed dispatch command modes, which makes it difficult to perceive the temporal pattern of water use behavior throughout the mine and to predict the future peak and valley trends of underground power supply and distribution load. As a result, water production load often passively overlaps with peak electricity price and peak load, leading to a significant increase in electricity costs, increased peak pressure on the power grid, and causing membrane modules to operate under high pressure differential for a long time, which accelerates their degradation and creates a series of hidden energy consumption and maintenance risks.

[0004] Furthermore, existing water purification plants generally lack rolling forecasting and flexible adjustment capabilities. They are unable to produce and store water in advance during off-peak electricity price periods, nor can they release stored water to supply water stably during peak periods. As a result, the three elements of "water production, water storage, and water use" remain strongly coupled but cannot be coordinated and regulated in a coordinated manner.

[0005] Therefore, we propose an intelligent energy consumption control method for underground reverse osmosis purification stations in coal mines. The information disclosed in the background section is only for enhancing understanding of the background of this disclosure and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing an intelligent energy consumption control method for underground reverse osmosis purification stations in coal mines, thereby solving the technical problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] The intelligent energy consumption control method for underground reverse osmosis purification stations in coal mines includes the following steps:

[0009] S1. Collect the operation data of the purification station and the water consumption, electricity consumption and electricity price data of the whole mine. Clean, align and normalize the data to construct a joint state description model that can simultaneously characterize the water supply status of the purification station, the water consumption behavior of the whole mine and the electricity consumption behavior of the whole mine.

[0010] S2. Based on the joint state description model, the water demand prediction model and the electricity load prediction model are trained respectively, and the electricity price time series information corresponding to the prediction period is generated to form a dual prediction model for water and electricity.

[0011] S3. Input the real-time updated joint state description into the water and electricity dual prediction model to obtain the future water demand prediction sequence, electricity load prediction sequence and electricity price time series sequence. Combine the water tank level and the adjustable output range of the unit to deduce the water production range that can be advanced and the water production range that can be delayed for each prediction period.

[0012] S4. Based on the predicted sequence and the water production volume range that can be advanced or delayed, a flexible load peak shaving energy consumption optimization model is constructed. The water production flow rate, water tank storage volume change and high-pressure pump power are used as optimization decision variables, and electricity cost, peak load and unit water production energy consumption are used as joint optimization objectives. The water supply pressure, water tank level, membrane module pressure difference and unit start-up and shutdown frequency operation constraints are applied.

[0013] S5. Solve the flexible load peak shaving energy consumption optimization model in each control cycle to generate a scheduling scheme that includes target water production flow, target water tank level range, high pressure pump frequency setpoint and valve opening setpoint. Distinguish between off-peak and peak periods according to electricity price and load level to form a peak shaving and valley filling scheduling strategy based on water storage capacity.

[0014] S6. Execute the scheduling scheme in each control cycle and update the joint state description based on the deviation between the actual operating data and the scheduling target. When the actual water or electricity consumption deviates from the prediction, correct the water and electricity dual prediction model online, re-execute rolling prediction, optimization modeling and scheduling generation to form a closed loop of intelligent energy consumption control for water and electricity collaborative peak shaving and valley filling.

[0015] S1 specifically includes:

[0016] The influent pressure, product water pressure, membrane pressure difference, influent flow rate, product water flow rate, water tank level, and water quality parameters of the reverse osmosis purification station in the coal mine are collected to obtain operational data reflecting the water supply status of the purification station.

[0017] Simultaneously collect water metering data for the entire mine, historical and real-time load curve data of the mine's power supply and distribution system, and time-of-use electricity price data to obtain external data reflecting the mine's water and electricity consumption behavior.

[0018] The obtained data is processed by timestamp alignment, missing data filling, outlier removal and normalization, so that the purification station operation data and the mine-wide behavior data can be directly used by the same model under a unified time reference.

[0019] The cleaned purification station operation data and the mine-wide behavior data are combined into a state vector in a unified format to form a joint state vector that reflects the water supply status of the purification station, the water use behavior of the entire mine, and the electricity use behavior of the entire mine.

[0020] Based on the joint state vector, a basic state model is constructed for subsequent use in the dual prediction model for water and electricity and the flexible load peak shaving optimization model, so that various types of operational data have a structured form that can be directly used for time series prediction and optimization solutions.

[0021] S2 specifically includes:

[0022] The temporal, shift, and seasonal features related to water use behavior are extracted from the joint state vector to provide input features for water demand forecasting.

[0023] Historical load curve features, equipment start-up and shutdown features, and load change trend features reflecting the overall power consumption behavior of the mine are extracted from the joint state vector to provide input features for the prediction of the overall power load of the mine.

[0024] Based on the input characteristics, a water demand prediction sub-model and a mine-wide electricity load prediction sub-model are trained through a time-series prediction model. Time-of-use electricity price information is mapped to electricity price characteristics consistent with the prediction period, and a unified water and electricity dual prediction model is constructed.

[0025] A unified output structure is defined for the dual forecasting model for water and electricity, which can simultaneously output the water demand forecast sequence, the total mine electricity load forecast sequence, and electricity price time series information for multiple forecast periods in the future.

[0026] The parameters of the water and electricity dual prediction model are solidified, the interface is configured, and online operation is adapted, so that the water and electricity dual prediction model has the ability to receive the joint state vector and output the water and electricity prediction sequence in real time within the control cycle.

[0027] S3 specifically includes:

[0028] Within each control cycle, the constructed joint state vector is updated using the currently collected purification station operation data and whole mine behavior data to form the latest state vector for prediction.

[0029] The latest state vector is input into the water and electricity dual prediction model for the water demand-related part to generate a water demand prediction sequence covering multiple future time periods.

[0030] The latest state vector is input into the load forecasting-related part of the water and electricity dual forecasting model to generate the corresponding full mine electricity load forecasting sequence, and at the same time outputs the electricity price time series sequence corresponding to the forecast period.

[0031] Based on the water demand forecast sequence, the current liquid level and effective volume of the water tank in the purification station, and the adjustable output range of the reverse osmosis unit, the range of water production volume that can be produced in advance and the range of water production that can be delayed in the future forecast period are derived.

[0032] The output water demand forecast sequence, the total mine electricity load forecast sequence, the electricity price time series sequence, and the adjustable output range of the purification station are combined to form the rolling forecast input set required for subsequent flexible load peak shaving energy consumption optimization.

[0033] S4 specifically includes:

[0034] Based on the rolling forecast input set, the decision variables in the flexible load peak shaving energy consumption optimization model are defined, including the water production flow set value of the purification station, the change in water tank storage capacity and the target output power of the high-pressure pump in each forecast period.

[0035] A joint optimization objective is constructed, which takes the cumulative electricity cost, the peak total load of the entire mine and the energy consumption per unit water production of the purification station as multi-objective optimization items, to achieve the comprehensive objective of maximizing flexible load peak shaving and energy-saving benefits.

[0036] The optimization model incorporates safety differential pressure constraints for reverse osmosis membrane modules, minimum water supply pressure constraints, minimum water tank level constraints, and start-up and shutdown frequency limits for purification station units to ensure that the optimization results meet the requirements for water supply safety and equipment safety.

[0037] Based on the water production range that can be produced in advance and the water production range that can be delayed, a dynamic time-series constraint is established for the water production load of the purification station in each predicted period, so that the optimization process can simultaneously meet the future water demand and the scheduling conditions for peak shaving and valley filling.

[0038] The constructed multi-objective, multi-constraint optimization model is encapsulated so that it can be solved based on the rolling prediction input in each control cycle to obtain the optimal scheduling scheme covering the prediction time domain.

[0039] S5 specifically includes:

[0040] Within the current control cycle, the encapsulated flexible load peak shaving energy consumption optimization model is invoked to obtain the optimal scheduling scheme covering the entire prediction time domain.

[0041] The target water production flow rate of the purification station for each predicted period is obtained from the optimal scheduling scheme to guide the operation strategy of the subsequent reverse osmosis unit and high-pressure pump.

[0042] Based on the optimal scheduling scheme, the target range of water tank storage capacity for the corresponding time period is generated, and the high-load water production action during off-peak hours and the low-load water production action during peak hours are converted into executable water storage commands.

[0043] The strategy of increasing the water production load of the purification station during off-peak hours, reducing the water production load during peak hours, and relying on water tank storage for water supply will generate a complete set of flexible peak shaving control instructions.

[0044] The flexible peak shaving control instruction set is encapsulated in an executable format as structured scheduling instructions for high-pressure pump frequency setting value, valve opening setting value and water production flow rate setting value, and used as input to the execution layer.

[0045] S6 specifically includes:

[0046] The generated structured scheduling instructions are sent to the high-pressure pumps, valves and reverse osmosis units of the purification station, so that they can perform flexible peak shaving operation according to the instructions within the current control cycle;

[0047] Collect actual operating data after the execution of scheduling instructions, compare it with the target water production load and target water storage range, and calculate the tracking deviation of the current control cycle;

[0048] When the tracking deviation exceeds the preset threshold, it triggers the need to update the joint state description model to reflect sudden changes in water demand, overall mine load, or electricity price.

[0049] Based on the situation where the preset threshold is exceeded, online parameter correction is performed on the water and electricity dual prediction model, so that more accurate prediction results can be generated based on the corrected state vector in the next control cycle;

[0050] After completing the model correction, steps S3–S5 are re-executed to form new flexible peak shaving scheduling instructions, thereby realizing a closed-loop intelligent energy consumption control for water and electricity coordinated peak shaving at the purification station.

[0051] The beneficial effects of this invention are as follows:

[0052] This invention utilizes a dual water-electricity prediction model to anticipate future water demand, overall mine electricity load, and time-of-use electricity price trends. Combined with a flexible water production range that allows for both advance and delayed production, it enables rolling optimization of the reverse osmosis unit's water production load. This allows the purification station to proactively avoid peak periods of high electricity prices and high loads, increasing water production and storage during off-peak hours, and releasing stored water for supply during peak periods, achieving peak-shaving and valley-filling operation. This strategy not only significantly reduces peak-period energy consumption and operating costs but also significantly alleviates the pressure on the mine's power supply system during peak hours, improving the overall stability and safety margin of the power grid.

[0053] This invention comprehensively considers pump efficiency curves, unit water production energy consumption characteristics, electricity price weights, and load superposition effects in its optimization model. It performs rolling optimization with the goal of minimizing overall energy consumption, achieving a globally optimal configuration of water production flow rate, pump power, and water storage strategy. Compared to traditional fixed-frequency operation, this reduces energy consumption per unit mass of water produced and minimizes additional energy losses caused by high-power pump operation, thus saving operating costs.

[0054] This invention uses safety constraint modeling to strictly limit membrane pressure difference, minimum water supply pressure, pump start-up and shutdown frequency, and gradient changes. This allows the optimized operating strategy to naturally maintain the membrane module under conditions of smoother pressure and more balanced load over the long term, significantly reducing membrane fouling, membrane fatigue, and cleaning frequency, thereby extending membrane life, reducing maintenance costs, and reducing the risk of water supply interruption due to maintenance shutdowns.

[0055] This invention establishes an online parameter correction mechanism to dynamically update normalized parameters, statistical features, and prediction model parameters, enabling the prediction model and state model to quickly adapt to new working conditions and achieve the adaptive capability of "prediction automatically evolving with changes in working conditions".

[0056] This invention utilizes water balance relationships to construct a target water storage trajectory. Through closed-loop coupling between predicted water demand, predicted load changes, and water storage capacity, it achieves synchronous optimization of the "water production, water storage, and water use" chain. The system can automatically assess whether the current water storage is sufficient to support peak shaving strategies. When water storage is insufficient, it preemptively produces water to replenish the liquid; when water storage is sufficient, it proactively reduces the water production load, forming a predictive and collaborative operation mode, thereby improving the balance between water supply stability and energy economy.

[0057] This invention employs a closed-loop operation mode of status acquisition, rolling prediction, optimal scheduling, execution feedback, online correction, and re-prediction, enabling the system to possess a self-learning capability similar to its own operating conditions. Each control cycle utilizes the latest actual operating data to optimize subsequent predictions and scheduling, forming a continuously learning and self-evolving intelligent energy consumption control system. This allows the purification station to maintain optimal control performance during long-term operation and adapt to changes in different seasons, shifts, and mining conditions. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the intelligent energy consumption control method for underground reverse osmosis purification stations in coal mines according to the present invention. Detailed Implementation

[0059] 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.

[0060] Example: Figure 1 As shown, this embodiment provides an intelligent energy consumption control method for underground reverse osmosis purification stations in coal mines, including the following steps:

[0061] S1. Collect the inlet and outlet water pressure, flow rate, water tank level, water quality parameters, and operating status of each actuator of the purification station. At the same time, collect the water metering data, electricity load data, and time-of-use electricity price information of the entire mine. Clean, align, and normalize the above data to construct a joint state description model that can simultaneously characterize the water supply status of the purification station, the water use behavior of the entire mine, and the electricity use behavior of the entire mine.

[0062] S2. Based on the joint state description model, extract the time-series features, shift features, and environmental features related to water use behavior, and train a water demand prediction model for predicting future target water supply; and extract the load features, equipment start-up and shutdown features, and electricity price features related to electricity use behavior, and train a load prediction model for predicting future electricity load of the entire mine, while generating time-series electricity price information corresponding to each prediction period to form a water and electricity dual prediction model.

[0063] S3. In each control cycle, the real-time updated joint state description is input into the water-electricity dual prediction model to obtain the water demand prediction sequence, electricity load prediction sequence and electricity price time series sequence for each prediction period in the future; and combined with the current water tank level, effective volume and adjustable output range of the unit, the future water demand is converted into the water production volume range that can be produced in advance and the water production volume range that can be delayed for each prediction period, forming the time series constraints for flexible scheduling.

[0064] S4. Based on the water demand forecast sequence, electricity load forecast sequence, electricity price time series sequence, and water production intervals that can be advanced or delayed, a flexible load peak shaving energy consumption optimization model is constructed. The water production flow rate, water tank storage changes, and high-pressure pump power in each forecast period are used as optimization decision variables. Electricity cost, peak load of the entire mine, and unit water production energy consumption are used as joint optimization objectives. Operational constraints such as water supply pressure, water tank level, membrane module pressure difference, and unit start-up and shutdown frequency are applied to obtain the optimal energy consumption operation scheme that meets water supply security.

[0065] S5. Solve the flexible load peak shaving energy consumption optimization model within the control cycle to generate a scheduling scheme covering the prediction time domain, including the target water production flow rate, the target water tank level range, the high-pressure pump frequency setting value, and the valve opening setting value; and distinguish between off-peak and peak periods according to the predicted electricity price and load level, increase the water production load and enhance water storage during off-peak periods, and reduce the water production load and rely on water storage for water supply during peak periods, forming a peak shaving and valley filling scheduling strategy based on the water tank energy storage capacity;

[0066] S6. In each actual control cycle, the scheduling scheme is sent to the execution layer of the purification station, and the deviation is calculated based on the actual operating data and the scheduling target. When the actual water demand, electricity load or electricity price information deviates from the prediction, the joint state description is updated and the parameters of the water and electricity dual prediction model are corrected online. Based on the corrected prediction results, rolling prediction, optimization modeling and scheduling generation are re-executed to realize the closed loop of intelligent energy consumption control for water and electricity collaborative peak shaving and valley filling for the purification station.

[0067] S1 specifically includes the following sub-steps:

[0068] S110. Collection of purification station operation data: Real-time collection of operation data of the reverse osmosis purification station in the coal mine, including: inlet water pressure. Water production pressure Difference pressure before membrane Inlet flow rate Water production flow rate Water tank level Water quality parameters such as conductivity, as well as the operating status of each high-pressure pump and valve, are collected to form a raw operating dataset reflecting the water supply status of the purification station. The above parameters are collected according to a unified timestamp, with a sampling period of, for example, 1 second, which can be adjusted according to on-site needs.

[0069] S120. Collection of water and electricity consumption data for the entire mine: Synchronously collect water metering values ​​for the entire mine. Historical and real-time load curves of the entire mine's power supply and distribution system And the time-of-use electricity pricing published by the energy management system This yields the original external dataset related to the external coupling behavior of the purification station.

[0070] S130. Time Alignment and Data Cleaning: Alignment and cleaning of the purification station operation data (S110) and external data (S120) under a unified time reference are performed, including:

[0071] Time synchronization processing:

[0072] Interpolate data from different sampling periods to arrange all parameters in the same time series;

[0073] Outlier removal:

[0074] A determination method combining statistical thresholds and operating condition thresholds is adopted, wherein the statistical thresholds satisfy: ;in For the current sample point, For moving average, is the sliding standard deviation, and k is the outlier coefficient (e.g., 3);

[0075] The operating thresholds are set based on the cleanroom operation specifications, for example: ;

[0076] in This represents the permeate pressure at time t in the reverse osmosis purification station. This pressure is measured in real time by a pressure sensor at the permeate end and is a core physical quantity reflecting the operating status of the reverse osmosis membrane module. It is used to reflect concentration polarization, membrane fouling, and overall hydraulic load. This indicates the lower limit of permeate pressure; it is set according to equipment technical parameters or operating experience to ensure normal water permeation of the membrane module and to prevent the permeate pump from operating under low pressure, and is used to constrain the permeate pressure to not be lower than the safety lower limit. This indicates the upper limit of the permissible pressure for the produced water. This value is determined by the maximum permissible pressure difference of the membrane module, the rated pressure of the pump, and the pressure rating of the supporting pipeline. It is used to prevent the membrane module from being damaged or accelerated to decay due to excessive pressure.

[0077] Data that violates the above conditions should be repaired by elimination or interpolation.

[0078] Missing point imputation: occasional missing points are imputed using linear interpolation or nearest neighbor-based methods;

[0079] Noise smoothing: The pressure and flow signals are smoothed using moving average or Kalman filtering;

[0080] Normalization: All data are normalized according to a unified rule. The formula is as follows:

[0081]

[0082] in For the i-th type of original monitoring features, and These are the minimum and maximum values ​​of the feature within the defined statistical window, respectively. The normalized feature values ​​are normalized to [0,1]. After the above steps, a cleaned dataset that can be directly used for modeling is obtained.

[0083] S140, Joint State Vector Construction: The operational data of the purification station after S130 cleaning and the overall mine behavior data are vectorized and concatenated in a unified dimensional manner to construct a single-moment joint state vector reflecting the coupling relationship between the two types of data. Its general form is:

[0084]

[0085] All of the above components are normalized values ​​with fixed dimensions, and are used as standard inputs for subsequent prediction and optimization models. Let be the joint state vector at the current moment.

[0086] S150, Basic state modeling for prediction and optimization: Single-time state vector constructed based on S140 This forms the foundational state model for the water-electricity dual forecasting model and the flexible load peak shaving optimization model, specifically including:

[0087] Feature vector structuring: The inputs are rearranged according to their timing format to give them a structured form for use as model inputs.

[0088] Normalization feature solidification: Maintain the normalization rules adopted by S130 to ensure that all features have uniform dimensions and stable numerical scale;

[0089] Model input interface binding: binding the structured input interface It is solidified into the standard input vector format received by the prediction model and the optimization model;

[0090] Temporal window sequence construction: Combining state vectors from several consecutive time points into a temporal state window sequence:

[0091]

[0092] The window length n is set according to the time requirements of the prediction model. As the final time series input to the hydro-electricity dual prediction model, A time-series window of a state vector with time t as the endpoint and length n;

[0093] Encapsulation of the basic state model: This involves encapsulating the above structured vectors. With timing window It is encapsulated as a continuously updatable data structure, enabling it to automatically update within each control cycle and provide consistent input to the water-electricity dual prediction model and the optimization model.

[0094] S2 specifically includes the following sub-steps:

[0095] S210. Water demand feature extraction: Single-time state vector constructed based on S150 and its corresponding time window sequence Structured features are extracted from components in the window sequence that are highly correlated with the overall mine water use behavior to form a feature set for subsequent water demand prediction sub-models. .

[0096] in A continuous time window of length n, containing the state vectors of the most recent n time steps. It can reflect the short-term fluctuation trend of water use behavior over time.

[0097] Specifically, including but not limited to the following feature extraction methods:

[0098] (1) Characteristics of the mean and standard deviation of the sliding window: Calculate the total water consumption of the mine within the window length n. The moving mean and moving standard deviation are used to characterize recent water usage levels and the intensity of fluctuations. These statistics can reflect the overall trend and short-term dispersion of mine water usage and have strong discriminative power in the frequently changing working environment of mines.

[0099] (2) Characteristics of time difference and lag terms: based on Construct differential features reflecting the rate of change in water use at adjacent time points within the time frame (e.g., The data includes water usage data from the previous time step, as well as several lag terms (such as water usage from the previous three time steps and five time steps ago). These features are used to capture short-term acceleration trends and historical inertia characteristics of water usage behavior, which helps improve the sensitivity of prediction models to sudden changes in operating conditions.

[0100] (3) Periodic coding features: Based on the daily shift schedule, day and night operation mode or periodic intermittent operation characteristics of the mine, the current time is mapped to a periodic code (such as day / night shift code, periodic code based on week number, etc.) to enhance the model's ability to characterize periodic water use behavior and enable the prediction model to identify the "periodic peak-valley" structure.

[0101] (4) Construction of composite features: Based on at least one or more combinations of the above-mentioned sliding statistical features, difference and lag features, and periodic coding, construct composite features that reflect the comprehensive trend of future water use behavior, thereby forming a feature set for characterizing future water demand behavior. This feature set provides structured input for subsequent water demand prediction sub-models.

[0102] S220, Feature extraction of total mine power load: Based on the single-time state vector s(t) constructed from S150 and its corresponding time-series window sequence Select load curves from the entire mine's power supply and distribution system. The components are used to extract a structured feature set reflecting the overall mine's electricity load variation trend, which is then used as the main input for constructing the load forecasting sub-model. The time-series window sequence... It contains historical load information from the most recent n times, which can be used to characterize the short-term dynamic characteristics of load changes over time.

[0103] Specifically, including but not limited to the following feature extraction methods:

[0104] Load sliding statistical characteristics: Calculate the total mine load within a window length n. The moving average and moving standard deviation, for example:

[0105] The moving average is used to characterize the recent trend of the overall load level in a mine.

[0106] The sliding standard deviation reflects the intensity of load fluctuations over a short period of time, and helps to identify whether the load is in a stable or drastically fluctuating range.

[0107] Load change gradient and differential characteristics: To reflect the current upward or downward trend of the load, load change parameters (such as...) are constructed from a window. This allows for the introduction of multiple lag terms (such as load values ​​from the previous time step, three time steps ago, and five time steps ago). These features are used to capture the transient rate of change and historical inertia of the load. (This represents the load increment at time t).

[0108] Periodic fluctuation coding characteristics: Based on the time-of-use electricity price table provided by the energy management system, the current moment is mapped to a periodic code representing peak, flat, or low-valley intervals to reflect the load cycle variation driven by electricity prices. Simultaneously, daily or weekly cycle characteristics can be constructed based on intraday time or week number to enhance the modeling capability of "peak-flat-valley" load distribution patterns.

[0109] Construction of Combined Trend Features: Based on at least one or more combinations of moving average, standard deviation, difference components, lag terms, and periodic coding, a combined feature vector is constructed to characterize future load change trends. This combined feature can comprehensively reflect the stability, abrupt changes, and periodicity of the load, which is beneficial to improving the predictive model's ability to identify load change trends.

[0110] By structurally combining the aforementioned sliding statistical characteristics, gradient and differential characteristics, periodic coding, and combined trend characteristics, a feature set is formed to characterize future load change trends. As the main input to the load forecasting sub-model, it provides multi-dimensional structured information for the subsequent rolling forecasting process.

[0111] S230, Water-Electricity Dual Prediction Model Training and Construction: The feature set formed in S210... The feature set formed in S220 As input, historical water consumption sequences and historical load sequences are used as supervision labels to train a water demand prediction sub-model and a mine-wide load prediction sub-model. The prediction sub-models can employ time-series prediction models based on recurrent neural networks (such as LSTM or GRU) or ARIMA structures based on statistical learning, which are conventional models available in this field. Simultaneously, time-of-use electricity pricing will be implemented. The time series is expanded and aligned with the forecast output interval, so that the water-electricity dual forecast model can simultaneously output the water demand forecast sequence, the total mine electricity load forecast sequence, and the electricity price time series sequence for multiple future time periods.

[0112] S240. Prediction Model Output Structure Design: Based on the water-electricity dual prediction model trained in S230, a unified output structure is defined to enable the model to simultaneously output the water demand prediction sequence in the future prediction time domain. Total Mine Electricity Load Forecast Sequence and electricity price time series The above three types of prediction sequences adopt a unified time indexing rule, which facilitates the subsequent multi-sequence coupling calculation of the flexible load peak shaving optimization model.

[0113] in This represents the predicted water demand for the k-th control cycle at the current forecast start time t. This represents the predicted electricity load for the entire mine in the kth control cycle from the current prediction start time t. This represents the time-of-use electricity price corresponding to the kth future control cycle; k is the index of the future prediction step, with values ​​of 1, 2, ..., H, where H is the prediction time domain length, i.e., the total number of control cycles to be predicted.

[0114] Deployment preparation for the S250 and hydro-electric dual prediction models: Deploy and solidify the prediction models generated by S240 online, including configuring the model input interface so that the model can automatically receive the latest updates in each control cycle. and Configure the model output interface so that the model can directly generate complete three-class prediction sequences during runtime; and periodically update the moving average, normalized parameters or some learnable parameters within the model based on real-time collected data to maintain the prediction accuracy of the model during long-term operation, so that the hydro-electric dual prediction model can stably provide reliable input for subsequent rolling predictions of S310–S350.

[0115] S3 specifically includes the following sub-steps:

[0116] S310 Real-time Joint State Vector Update: Within each control cycle, the latest collected purification station operation data and overall mine behavior data are processed according to the cleaning rules of S130 and the structured rules of S150 to update the state vector at a single moment. Simultaneously update its corresponding time window sequence. To satisfy:

[0117]

[0118] Where n is the window length, ensuring that the prediction model always uses the most complete state information corresponding to the current time as input.

[0119] S320, Water Demand Forecast Sequence Generation: Update S310 and The water demand forecasting sub-model in the water-electricity dual forecasting model deployed on the S250 is input, and based on the temporal correlation learned from the window input and historical data, a water demand forecasting sequence covering the forecast time domain H is generated:

[0120]

[0121] The prediction sequence is used to represent the expected water supply demand for multiple control periods in the future.

[0122] S330, Generation of the Mine-wide Electricity Load and Price Forecast Sequence: This involves generating the S310 sequence. and The load forecasting sub-model in the dual forecasting model is input, and combined with the electricity price time series data obtained from S120 and cleaned by S130, the total mine electricity load forecast sequence and electricity price sequence for future periods are generated by aligning them with the forecast time domain:

[0123]

[0124] The former is used to characterize the future load change trend of the entire mine, while the latter is used to represent the time-of-use electricity price level for each future period.

[0125] Derivation of the adjustable output range of S340 and the purification station: Based on the water demand prediction sequence of S320 and the real-time water tank level of S310. With effective volume Combined with the minimum adjustable power of the reverse osmosis unit in the purification station With maximum adjustable power This involves deriving the range of water production that can be brought forward and the range that can be postponed for each predicted period; and using the water balance relationship:

[0126]

[0127] in It is the water tank level at the current time t; It is the predicted water tank level in the kth future control cycle derived from the water balance at the current time t; It is the water production (or water output) in the i-th control cycle in the future, and its unit is flow rate unit. It is usually determined by the decision variables in the optimization scheduling module S410–S450. This is the predicted water consumption at the current time t for the i-th future control cycle; i is the summation index, representing the future control cycle from 1 to k; this formula is used to describe the predicted future liquid level. With initial liquid level The water balance relationship between them reflects the cumulative impact of future water production and predicted water consumption on liquid level changes.

[0128] Under the condition of meeting the liquid level safety constraints Under the premise of:

[0129] Pre-production water volume range: Allows for increased water production without causing the liquid level to exceed the upper limit in the future;

[0130] Delayed water production range: This allows for a reduction in water production or reliance on water storage for supply without causing the water level to fall below the lower limit in the future. This range serves as the time-series constraint input for subsequent peak shaving and valley filling optimization modeling. This is the safe lower limit of the water tank level, which is usually determined by the minimum suction height of the water pump, the minimum water replenishment requirements of the reverse osmosis unit, and the on-site operation safety standards. Falling below this value will lead to the risk of water supply interruption. This is the safe upper limit of the water tank level, which is determined by the water tank's design volume, overflow height, and safe operating procedures. Exceeding this value may cause overflow or trigger an emergency shutdown.

[0131] Among them, the water production range that can be prepared in advance With the range of water production that can be delayed These represent the range of water production that can be increased or decreased in the k-th predicted time period, provided that the upper and lower limits of the water tank level are met.

[0132] S350. Summarizing the rolling forecast results for optimization modeling: This involves summarizing the water demand forecast sequence output from S320 above. Load forecast sequence output by S330 With electricity price series The data, along with the water production intervals that can be advanced or delayed derived from S340, are summarized to form a set of rolling prediction inputs that cover the entire prediction time domain and have a unified structure.

[0133]

[0134] in This represents the flexible adjustable output range for the k-th future control cycle derived from S340, covering the range of water production that can be advanced and the range of water production that can be delayed. It is used to characterize the range of water production that the purification station can increase or decrease during this period. It is used as all the inputs to the flexible load peak shaving energy consumption optimization model (S410–S450) to ensure that the optimization model can make rolling decisions based on the latest state.

[0135] S4 specifically includes the following sub-steps:

[0136] S410, Definition of Optimized Decision Variables: Rolling Prediction Input Set Based on S350 To define the optimization decision variables for the flexible load peak shaving energy consumption optimization model, the following are defined:

[0137] Water production flow rate setpoint sequence for the purification plant in future time periods: ;

[0138] Water tank storage volume change sequence (determined endogenously by water balance):

[0139]

[0140] The target power or frequency setpoint of the high-pressure pump is correspondingly used, and the above sequence is combined into a decision variable vector for the optimization model:

[0141]

[0142] in This represents the target power / frequency setting value of the high-pressure pump in the k-th time period in the future; ensuring that all decisions participate in the optimization calculation with a unified time index structure.

[0143] S420. Construction of the Optimization Objective Function: Under the control objective of flexible load peak shaving, electricity cost, peak load suppression of the entire mine, and unit water production energy consumption are taken as joint optimization objectives to construct a comprehensive cost function:

[0144]

[0145] in It is a comprehensive optimization objective function used to measure the comprehensive performance indicators of electricity cost, peak load suppression capability, and unit water production energy consumption in the future forecast time domain. The optimization objective is to make minimize.

[0146] Specific Electricity costs:

[0147]

[0148] To suppress peak load across the entire mine:

[0149]

[0150] Energy consumption per unit of water production:

[0151]

[0152] in From S330; This represents the basic electrical load of the mine (excluding the load of the purification station) predicted in the kth control cycle in the future, which is used to construct the peak load index; , It is a multi-objective balance coefficient; the objective function simultaneously measures electricity cost, peak load and energy consumption level; this objective structure ensures that the water purification station can increase water production load during off-peak hours and reduce water production load during peak hours, thus achieving peak shaving and valley filling.

[0153] S430. Modeling of operational safety and water supply stability constraints: To ensure that the optimization results meet the requirements for long-term stable operation and water supply safety of the purification plant, the following constraints are established:

[0154] Membrane module safety differential pressure constraints:

[0155]

[0156] in It is the transmembrane pressure difference of the reverse osmosis membrane module in the kth control cycle in the future; It is the maximum safe threshold allowed by the membrane pressure difference.

[0157] Minimum water supply pressure constraint:

[0158]

[0159] in This is the water production side pressure in the kth control cycle. It is the minimum water supply pressure required for safe operation, and the lower limit is derived from the on-site water supply demand or the minimum pressure requirement of the water pump.

[0160] Water tank level safety constraints:

[0161]

[0162] Unit start-up and shutdown constraints (to avoid frequent start-ups and shutdowns):

[0163]

[0164] It is the maximum allowable variation in power / frequency between adjacent cycles of the high-pressure pump.

[0165] Water production flow rate adjustable range constraints:

[0166]

[0167] in It is the lower limit of the water production flow rate of the purification station (minimum allowable water production). This is the upper limit of the water production flow rate of the purification station (maximum allowable water production); the above constraints ensure that while pursuing energy consumption reduction, the optimization does not trigger risks such as membrane damage, water tank shortage or system oscillation.

[0168] S440, Water-Electricity Coordination Timing Constraints Introduced: Flexible Adjustable Output Range Derived from S340 Establish dynamic time constraints for water-electricity synergy:

[0169]

[0170] in Source: Water tank level can be used to pre-determine water production volume; water tank level can be used to postpone water production volume constraints; predict water demand. The system's adjustable power / flow range ensures that the water production load of the purification station meets the rolling adjustment space of "meeting water supply demand and participating in peak shaving and valley filling" in each predicted period.

[0171] S450, Flexible Load Peak Shaving Energy Consumption Optimization Model Encapsulation: The objective function, decision variable vector, and all constraints constructed in S410–S440 are uniformly encapsulated to form a flexible load peak shaving energy consumption optimization model:

[0172]

[0173] The optimization model is configured to use a rolling predictive input set based on S350 in each control cycle. The solution is performed to output the optimal water production load allocation scheme, the target power sequence of the high-pressure pump, and the corresponding valve control quantities covering the entire prediction time domain H, providing the optimal solution for the subsequent generation of the flexible peak shaving scheduling strategy of S510–S550.

[0174] S5 specifically includes the following sub-steps:

[0175] S510, Optimization Model Rolling Solution: Within the current control cycle, the rolling prediction input set formed in S350 is used... The input is fed into the flexible load peak-shaving energy consumption optimization model of the S450 package, and the solver is invoked to perform rolling optimization calculations in the prediction time domain H to obtain the optimal control sequence for multiple future time periods:

[0176]

[0177] in It is the optimal water production flow rate setting value (optimal water production) for the kth control cycle in the future. It is the optimal power (or optimal frequency) setting value of the high-pressure pump in the kth control cycle in the future.

[0178] The above optimal sequence serves as the target control solution for this cycle, guiding the generation of subsequent scheduling strategies.

[0179] S520, Target Water Production Load Generation: Extract the optimal water production flow setpoint for the future time period from the optimal control sequence obtained in S510.

[0180]

[0181] Set the target water production flow rate for the current control cycle as follows:

[0182]

[0183] in It is the target water production flow rate setpoint for the current control cycle t, that is, the actual water production command issued to the purification station in this cycle; so that the purification station can start to implement water production load adjustment according to the optimal peak shaving strategy.

[0184] S530, Target water tank storage strategy generation: Optimal water production flow rate obtained from S520. Water demand forecast series with S320 The target water tank level trajectory for future time periods is calculated based on the water balance relationship:

[0185]

[0186] in It is the target water tank level for the kth control cycle in the future. It is calculated by the system in the current cycle based on the future water production plan and predicted water consumption. It is used to guide the water storage / release strategy to achieve water storage during low periods and water release during peak periods. It is the i-th step prediction of the optimal water production flow rate obtained by the optimization solver in S510.

[0187] And set the water storage target for the next control cycle as follows:

[0188]

[0189] The formula means "rolling target forward": the liquid level target value for this cycle adopts the predicted target liquid level for the first step in the future; to ensure that water storage rises during low periods and water storage is released during peak periods, thereby providing water volume support for flexible peak shaving.

[0190] S540, Flexible Peak Shaving Strategy Generation: Based on the optimal water production flow target of S520 and the water storage target of S530, combined with the optimal high-pressure pump power / frequency sequence generated by S510. Generate a complete flexible peak-shaving control strategy, including:

[0191] Off-peak water production acceleration strategy: When When in a low range, execute: To increase water production capacity and increase water storage; This is the target water production flow rate setpoint for the current control cycle; This is the predicted water consumption for the current control cycle;

[0192] Peak water production load reduction strategy: When or electricity price When the peak range is reached, execute: It also utilizes the water tank's water storage differential for water supply to achieve peak shaving and valley filling.

[0193] High-pressure pump coordinated control strategy: Set the optimal pump power / frequency as follows:

[0194]

[0195] in It is the target pump power / pump frequency setpoint for the current control cycle. The optimal pump power / pump frequency setpoint for the next cycle in the optimal control sequence; ensuring that the pump power is consistent with the water production load; the above strategies together constitute an executable control instruction framework for flexible load to participate in mine peak shaving scheduling.

[0196] S550, Structured Output of Scheduling Commands: The flexible peak-shaving scheduling strategy generated by S540 is encapsulated into structured control commands that can be directly executed by the equipment, specifically including:

[0197] High-pressure pump target frequency / power setting: Target water production flow rate for reverse osmosis units: ; Opening setting values ​​of relevant key valves This is obtained through internal function mapping:

[0198]

[0199] This ultimately results in a complete set of scheduling instructions adapted to the operational requirements of the current control cycle:

[0200]

[0201] It is then sent to the purification station's execution mechanism to achieve real-time control of flexible peak-shaving operation.

[0202] S6 specifically includes the following sub-steps:

[0203] S610, Issuance and Execution of Scheduling Instructions: Within the current control cycle, the scheduling instruction set generated by S550 will be executed.

[0204]

[0205] The control bus sends commands to the actuators in the underground reverse osmosis purification station of the coal mine, including the high-pressure pump frequency converter, the reverse osmosis unit controller, and the electric actuators of related regulating valves; each actuator sets its corresponding target water production flow rate. High-pressure pump target power / frequency setting value and valve opening setting value The system is run to ensure that the flexible peak-shaving scheduling strategy is actually executed within the current control cycle.

[0206] S620. Actual Operation Data Acquisition and Tracking Deviation Calculation: During the execution of dispatch instructions, actual operation data within the current control cycle is collected in real time, including actual water production flow. Actual water tank level Actual high-pressure pump power Actual water consumption and the actual total electricity load of the entire mine And compare it with the corresponding target value to calculate the tracking deviation, for example:

[0207]

[0208]

[0209] in Given by S530. Through the above deviation calculation, the actual tracking status of the purification station execution layer towards the flexible peak shaving scheduling target within the current control cycle is obtained.

[0210] S630, Deviation Exceedance and Operating Condition Anomaly Detection: Based on the tracking deviation calculated in S620, combined with the preset deviation threshold and operating condition safety threshold, the operating status of the current control cycle is determined: When the following conditions are met: or At that time, it was considered that the tracking deviation of the purification station from the target water production load or the target water storage trajectory exceeded the limit;

[0211] in It is the water production flow tracking deviation in the current control cycle t; It is the maximum permissible deviation threshold for water production flow rate deviation; It is the water storage tracking deviation in the current control period t; It is the maximum allowable tracking deviation threshold for the water tank level;

[0212] When the actual total electricity load of the mine Compared with the predicted load The difference, or the actual water consumption Compared with forecasted water demand When the difference exceeds the set prediction error threshold, the external operating conditions are considered to have changed significantly. When any of the above conditions are met, an online correction requirement is triggered for the state descriptions on which the prediction model and optimization model depend, and the current control cycle is marked as an "abnormal cycle that requires updating model parameters".

[0213] Online parameter correction for S640, the dual prediction model, and the basic state model: When an excessive deviation or abnormal operating condition is detected in S630, online parameter correction is performed on the basic state model of S150 and the water-electricity dual prediction model deployed in S250 based on the actual operating data collected by S620, including:

[0214] Normalization parameter update: Update the normalization formula based on the latest actual data.

[0215]

[0216] In and This allows the feature scale to be adapted to the current operating conditions.

[0217] Statistical Features and Moving Mean Updates: Moving Mean for Anomaly Detection and Feature Extraction Standard deviation And update the window statistics so that the basic state model can reflect the latest data distribution;

[0218] Incremental adjustment of prediction model parameters: Using the actual water consumption and load data of the most recent control cycles, incremental learning or parameter fine-tuning is performed on the water demand prediction sub-model and the load prediction sub-model to maintain the adaptability of the prediction model to new operating conditions.

[0219] Through the above online corrections, the basic state model and the water-electricity dual prediction model maintain high prediction accuracy and input-output matching even after changes in operating conditions.

[0220] S650, Closed-Loop Optimization Recalculation and Control Strategy Rolling Update: After completing the online parameter correction of S640, the rolling prediction steps of S310–S350 are re-executed to generate a new rolling prediction input set based on the updated basic state model and the hydropower dual prediction model. The optimal control sequence is obtained by resolving the problem using the flexible load peak-shaving energy consumption optimization model packaged in S450. With the new scheduling instruction set CMD(t), the target water production load, target water storage trajectory, and flexible peak shaving strategy are updated through steps S510–S550, thereby continuing to achieve water-electricity coordinated peak shaving and valley filling intelligent energy consumption control in the next control cycle. Through the above-mentioned rolling closed-loop mechanism of "prediction-optimization-execution-correction-re-prediction-re-optimization", the underground reverse osmosis purification station in the coal mine maintains its adaptive capability and energy consumption optimization level in response to changes in external operating conditions during long-term operation.

[0221] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0222] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0223] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. The intelligent energy consumption control method based on the coal mine underground reverse osmosis purification station, characterized in that, The method comprises the following steps: S1, collecting purification station operation data and mine-wide water, electricity and electricity price data, cleaning, aligning and normalizing the data, and constructing a joint state description model capable of representing the purification station water supply state, mine-wide water behavior and mine-wide electricity behavior; S2, based on the joint state description model, training a water demand prediction model and an electricity load prediction model respectively, and generating time series information corresponding to the prediction period, forming a water and electricity double prediction model; S3, inputting the real-time updated joint state description into the water and electricity double prediction model to obtain future water demand prediction sequence, electricity load prediction sequence and electricity price time series sequence, combining water tank liquid level and unit adjustable output range to deduce the advance water production quantity interval and the delay water production quantity interval in each prediction period; S4, based on the prediction sequence and the advance and delay water production quantity interval, constructing a flexible load peak shaving energy consumption optimization model, taking water production flow, water tank water storage quantity change and high pressure pump power as optimization decision variables, taking electricity cost, peak load and unit water production energy consumption as joint optimization objectives, and applying water supply pressure, water tank liquid level, membrane component pressure difference and unit start-stop frequency operation constraint; S5, solving the flexible load peak shaving energy consumption optimization model in each control period to generate a scheduling scheme including target water production flow, target water tank liquid level interval, high pressure pump frequency set value and valve opening degree set value, and forming a water storage capacity peak clipping and valley filling type scheduling strategy according to electricity price and load level to distinguish low valley period and peak period.

2. The intelligent energy consumption control method based on the coal mine underground reverse osmosis purification station according to claim 1, characterized in that, Further comprising: S6, executing the scheduling scheme in each control period, and updating the joint state description according to the deviation of actual operation data and scheduling target; when the actual water or electricity consumption deviates from the prediction, online correcting the water and electricity double prediction model, re-executing rolling prediction, optimization modeling and scheduling generation to form a water and electricity collaborative peak shaving and valley filling type intelligent energy consumption control closed loop.

3. The intelligent energy consumption control method based on the coal mine underground reverse osmosis purification station according to claim 1, characterized in that, S1 specifically comprises: Collecting the water inlet pressure, water production pressure, pre-membrane pressure difference, water inlet flow, water production flow, water tank liquid level and water quality parameters of the coal mine underground reverse osmosis purification station to obtain operation data reflecting the water supply state of the purification station; Synchronously collecting mine-wide water metering data, mine power supply and distribution system historical and real-time load curve data and time-of-use electricity price data to obtain external data reflecting mine-wide water behavior and mine-wide electricity behavior; Performing timestamp alignment, missing data filling, outlier removal and normalization processing on the obtained data, so that the purification station operation data and mine-wide behavior data can be directly used by the same model under the same time reference.

4. The intelligent energy consumption control method based on the coal mine underground reverse osmosis purification station according to claim 3, characterized in that, S1 further comprises: Combining the cleaned purification station operation data and mine-wide behavior data into a state vector in a unified format to form a joint state vector reflecting the purification station water supply state, mine-wide water behavior and mine-wide electricity behavior; Based on the joint state vector, a basic state model is constructed for subsequent water and electricity double prediction model and flexible load peak shaving optimization model calling, so that various types of operation data have a structured form that can be directly used for time series prediction and optimization solution.

5. The intelligent energy consumption control method based on the coal mine underground reverse osmosis purification station according to claim 1, characterized in that, S2 specifically comprises: Extracting time series features, shift features and seasonal features related to water behavior in the joint state vector to provide input features for water demand prediction; The historical load curve characteristics, equipment start-stop characteristics and load change trend characteristics reflecting the whole mine power consumption behavior are extracted from the joint state vector to provide input characteristics for the whole mine power load prediction; According to the input characteristics, a water and electricity double prediction model is constructed by training a water demand prediction sub-model and a whole mine power load prediction sub-model through a time series prediction model, and mapping the time-of-use electricity price information into a price characteristic consistent with the prediction period; A unified output structure is defined for the water and electricity double prediction model, which can simultaneously output the water demand prediction sequence, the whole mine power load prediction sequence and the price time sequence information in the future multiple prediction periods; The water and electricity double prediction model is parameterized, the calling interface is configured and the online operation is adapted, so that the water and electricity double prediction model has the ability to receive the joint state vector in real time within the control period and output the water and electricity prediction sequence.

6. The intelligent energy consumption control method based on the coal mine underground reverse osmosis purification station of claim 1, characterized in that, S3 specifically includes: In each control period, the joint state vector constructed is updated with the current collected purification station operation data and the whole mine behavior data to form the latest state vector for prediction; The latest state vector is input into the part related to water demand of the water and electricity double prediction model to generate a water demand prediction sequence covering multiple future periods; The latest state vector is input into the part related to load prediction of the water and electricity double prediction model to generate a corresponding whole mine power load prediction sequence, and simultaneously output a price time sequence corresponding to the prediction period; According to the water demand prediction sequence, the current liquid level and effective volume of the purification station water tank and the adjustable output range of the reverse osmosis unit, the pre-advance water production interval and the delayed water production interval in the future prediction period are derived; The output water demand prediction sequence, whole mine power load prediction sequence, price time sequence and purification station adjustable output interval are combined to form a rolling prediction input set required for subsequent flexible load peak shaving energy optimization.

7. The intelligent energy consumption control method based on the coal mine underground reverse osmosis purification station of claim 1, characterized in that, S4 specifically includes: Based on the rolling prediction input set, define the decision variables in the flexible load peak shaving energy optimization model, including the water production flow set value of the purification station in each prediction period, the water tank water storage volume change and the high-pressure pump target output power; Construct a joint optimization objective, taking the cumulative electricity cost, the whole mine total load peak and the purification station unit water production energy consumption as multi-objective optimization items to realize the comprehensive goal of maximizing the flexible load peak shaving and energy saving benefits; Introduce the reverse osmosis membrane component safety pressure difference constraint, the minimum water supply pressure constraint, the minimum water tank liquid level constraint and the purification station unit start-stop frequency limit in the optimization model to ensure that the optimization result meets the requirements of water supply safety and equipment safety; According to the pre-advance water production interval and the delayed water production interval, establish dynamic time sequence constraints of the purification station water production load in each prediction period to make the optimization process meet the scheduling conditions of future water demand and peak load shifting at the same time; Encapsulate the constructed multi-objective and multi-constraint optimization model to make it able to be solved according to the rolling prediction input in each control period to obtain an optimal scheduling scheme covering the prediction time domain.

8. The intelligent energy consumption control method based on the coal mine underground reverse osmosis purification station of claim 1, characterized in that, S5 specifically includes: In the current control period, the encapsulated flexible load peak shaving energy optimization model is called to obtain an optimal scheduling scheme covering the entire prediction time domain; The target water production flow of the purification station in each prediction period is obtained from the optimal scheduling scheme to guide the operation strategy of the reverse osmosis unit and the high-pressure pump; According to the optimal scheduling scheme, the water storage target interval of the water tank in the corresponding period is generated, and the high-load water production action in the low valley period and the low-load water production action in the peak period are converted into executable water storage instructions.

9. The intelligent energy consumption control method based on the coal mine underground reverse osmosis purification station of claim 8, characterized in that, S5 further comprises: The strategy of increasing the purification station water production load in the low valley period, reducing the water production load in the peak period, and relying on water storage for water supply is used to generate a complete flexible peak shaving control instruction set; The flexible peak shaving control instruction set is packaged into a structured scheduling instruction of the high-pressure pump frequency setting value, the valve opening degree setting value, and the water production flow setting value in an executable format as the input of the execution layer.

10. The intelligent energy consumption control method based on the coal mine underground reverse osmosis purification station according to claim 2, S6 specifically comprises: The generated structured scheduling instruction is sent to the high-pressure pump, valve, and reverse osmosis unit of the purification station, so that they perform flexible peak shaving operation according to the instruction in the current control period; Actual operation data after the execution of the scheduling instruction is collected and compared with the target water production load and the target water storage interval to calculate the tracking deviation of the current control period; When the tracking deviation exceeds the preset threshold, the update demand of the joint state description model is triggered to reflect the sudden changes of the water demand, the mine load, or the electricity price; According to the situation of exceeding the preset threshold, online parameter correction is performed on the water and electricity prediction model, and more accurate prediction results can be generated based on the corrected state vector in the next control period; After the model correction is completed, steps S3-S5 are re-executed to form a new flexible peak shaving scheduling instruction, and the water and electricity collaborative peak shaving intelligent energy consumption control closed loop of the purification station is realized.

Images