A method and system for equipment operation scheduling based on indoor cooling load prediction

Abstract

This application provides a method and system for equipment operation scheduling based on indoor cooling load prediction, belonging to the field of building energy conservation and intelligent scheduling technology. The method includes: acquiring multi-source data information of equipment in a refrigeration station and using a preset prediction model to output hourly predicted values ​​of indoor cooling load and corresponding predicted values ​​of equipment energy consumption, thereby obtaining predicted operation data; solving the preset scheduling optimization model using an optimization algorithm to generate and execute the optimal scheduling instruction; collecting actual operation data and calculating the deviation between the actual operation data and the predicted operation data to obtain the total operation deviation and inputting it into a causal decoupling model; separating the cooling load prediction deviation component and the equipment energy efficiency attenuation deviation component from the total operation deviation; updating the scheduling optimization model and / or the prediction model based on the cooling load prediction deviation component and the equipment energy efficiency attenuation deviation component; and scheduling the refrigeration station equipment using the updated prediction model and / or scheduling optimization model for the next scheduling cycle.

Description

Technical Field

[0001] This application relates to the field of building energy conservation and intelligent scheduling technology, and in particular to a method and system for equipment operation scheduling based on indoor cooling load prediction. Background Technology

[0002] Existing refrigeration plants largely rely on manual experience or fixed-time scheduling, lacking proactive scheduling based on dynamic changes in indoor cooling load. This leads to problems such as redundant equipment operation, unreasonable load distribution, and high energy consumption. While some existing technologies incorporate cooling load forecasting and optimized scheduling, they fail to finely decouple operational deviations, making it impossible to distinguish whether the deviation is due to inaccurate cooling load forecasting or equipment energy efficiency degradation, resulting in a lack of targeted model updates. Furthermore, in scenarios such as sudden weather changes or large-scale events, the predictive models lack sufficient adaptive capabilities, leading to lag or overshoot in scheduling commands. In addition, the lack of a smooth transition mechanism between automatic and manual modes causes fluctuations in operating conditions and equipment stress.

[0003] Therefore, there is an urgent need for a method and system for equipment operation scheduling based on indoor cooling load prediction. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method and system for equipment operation scheduling based on indoor cooling load prediction.

[0005] A first aspect of this application provides a method for equipment operation scheduling based on indoor cooling load forecasting, comprising: Acquire multi-source data information of equipment in the refrigeration station, and based on the multi-source data information, use a preset prediction model to output the hourly indoor cooling load prediction value and the corresponding equipment energy consumption prediction value to obtain the predicted operation data; Based on the predicted operational data, the preset scheduling optimization model is solved using an optimization algorithm to generate the optimal scheduling instruction; Execute the optimal scheduling instruction, collect actual operating data, calculate the deviation between the actual operating data and the predicted operating data, and obtain the total operating deviation; The total operating deviation is input into the causal decoupling model, and the cooling load prediction deviation component and the equipment energy efficiency degradation deviation component are separated from the total operating deviation. Based on the cold load prediction deviation component and the equipment energy efficiency attenuation deviation component, the scheduling optimization model and / or the preset prediction model are iteratively updated. The refrigeration station equipment is scheduled based on the updated preset prediction model and / or the scheduling optimization model, and applied to the next scheduling cycle.

[0006] A second aspect of this application provides an equipment operation scheduling system based on indoor cooling load prediction, comprising: The model prediction module is used to acquire multi-source data information of the equipment in the refrigeration station. Based on the multi-source data information, a preset prediction model is used to output the hourly indoor cooling load prediction value and the corresponding equipment energy consumption prediction value to obtain the predicted operation data. The optimization scheduling module is used to solve the preset scheduling optimization model based on predicted running data and using optimization algorithms to generate the optimal scheduling instruction; The operation deviation module is used to execute the optimal scheduling instruction, collect actual operation data, calculate the deviation between the actual operation data and the predicted operation data, and obtain the total operation deviation. The causal decoupling module is used to input the total operating deviation into the causal decoupling model and separate the cooling load prediction deviation component and the equipment energy efficiency degradation deviation component from the total operating deviation. The model update module is used to iteratively update the scheduling optimization model and / or the preset prediction model based on the cold load prediction deviation component and the equipment energy efficiency attenuation deviation component. The application module is updated to schedule refrigeration station equipment based on the updated preset prediction model and / or the scheduling optimization model, and applied to the next scheduling cycle.

[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described device operation scheduling method based on indoor cooling load prediction.

[0008] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described equipment operation scheduling method based on indoor cooling load prediction.

[0009] The beneficial effects of the equipment operation scheduling method and system based on indoor cooling load prediction provided in this application are as follows: This application outputs hourly predicted values ​​of future cooling load and equipment energy consumption through a multi-source data-driven prediction model, and generates optimal scheduling instructions based on an optimization algorithm to achieve accurate operation of refrigeration station equipment. By collecting actual operating data and calculating the total operating deviation, the deviation is decomposed into a cooling load prediction deviation component and an equipment energy efficiency attenuation deviation component using a causal decoupling model, which can accurately locate the source of the deviation. Based on different deviation components, the prediction model and scheduling optimization model are iteratively updated in a targeted manner to avoid scheduling fluctuations caused by blind updates, improve the accuracy of cooling load prediction and the matching degree of equipment energy efficiency, and make the next round of scheduling more in line with actual operating conditions. While ensuring the reliability of cooling supply, the total energy consumption of the system is reduced, and intelligent scheduling with continuous self-optimization is achieved. Attached Figure Description

[0010] Figure 1 A flowchart illustrating an embodiment of the equipment operation scheduling method based on indoor cooling load prediction provided in this application; Figure 2 A structural block diagram of an equipment operation scheduling system based on indoor cooling load prediction provided in an embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0011] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0012] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1 -Appendix Figure 3 The following is an explanation using specific examples.

[0013] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for equipment operation scheduling based on indoor cooling load forecasting, provided in an embodiment of this application. The method includes: S101: Acquire multi-source data information of equipment in the refrigeration station, and based on the multi-source data information, use a preset prediction model to output the hourly indoor cooling load prediction value and the corresponding equipment energy consumption prediction value to obtain the predicted operation data.

[0014] In this embodiment, the multi-source data information refers to multi-dimensional operational data obtained from different acquisition nodes and different types of equipment in the refrigeration station. This includes outdoor meteorological data, indoor environmental parameters, and operational parameters, operating status, cumulative operating time, historical cooling load, date type, and time information of equipment such as the refrigeration unit, chilled water pumps, cooling water pumps, and cooling towers. The preset prediction model is a pre-trained machine learning or deep learning model for time-series forecasting, used to predict hourly indoor cooling load and corresponding equipment energy consumption over a future period based on historical multi-source data.

[0015] Specifically, the prediction model employs a temporal deep neural network, which is divided into an input embedding layer, a temporal feature layer, an attention fusion layer, and a dual-output prediction layer. The input embedding layer normalizes and encodes features from multiple sources; the temporal feature layer uses an LSTM or GRU hierarchical structure to capture the temporal dependency between hourly cooling load and energy consumption; the attention fusion layer assigns weights to meteorological, temporal, and equipment status features; and the dual-output prediction layer outputs the predicted values ​​of future hourly indoor cooling load and equipment energy consumption, respectively, providing a data foundation for scheduling optimization.

[0016] The training samples for the prediction model consist of historical multi-source data from the refrigeration station, including outdoor temperature and humidity, solar radiation, indoor load, equipment operating parameters, and date type. These data are divided into training, validation, and test sets in an 8:1:1 ratio, and time-series sliding window sampling and outlier removal are performed. The prediction model uses the mean squared error between predicted and actual values ​​as the loss function and is trained using the Adam optimizer. Iteration stops when the validation set error fails to decrease for 10 consecutive rounds. The inputs to the prediction model are outdoor meteorological parameters, historical load data, equipment operating status, and calendar features; the outputs are the hourly predicted indoor cooling load and corresponding equipment energy consumption, which are directly fed into the scheduling optimization model to solve for the optimal command.

[0017] In this embodiment, the hourly indoor cooling load forecast is the value of the required cooling capacity of the building's interior, predicted hourly for the next 24 hours or future scheduling cycle, output by the prediction model. It serves as the basis for scheduling cooling capacity. The equipment energy consumption forecast is the estimated electrical energy and power consumption of equipment such as refrigeration units, water pumps, and cooling towers under the corresponding hourly cooling load, as they operate according to the scheduling plan. The predicted operating data is a dataset composed of the hourly cooling load forecast and the corresponding equipment energy consumption forecast, serving as the input basis for the scheduling optimization model.

[0018] S102: Based on the predicted running data, use the optimization algorithm to solve the preset scheduling optimization model and generate the optimal scheduling instruction.

[0019] In this embodiment, the optimization algorithm is an intelligent optimization algorithm used to find the optimal solution for the scheduling optimization model. It iteratively searches for a solution that satisfies the constraints and achieves the optimal objective. The preset scheduling optimization model is a pre-constructed mathematical model for optimizing the operation of refrigeration station equipment. It includes a comprehensive objective function with the goal of minimizing total energy consumption, energy efficiency models for various equipment, and constraints such as cooling capacity and equipment operation rules. This model is used to find the equipment operation scheme with the lowest energy consumption while meeting the cooling demand.

[0020] Specifically, the optimization algorithm employs a genetic algorithm, which has a four-layer structure: a population initialization layer, an evolutionary iteration layer, a fitness evaluation layer, and a parameter adaptation layer. The population initialization layer generates an initial population of scheduling schemes, including equipment operating parameters. The evolutionary iteration layer sequentially executes selection, crossover, and mutation operations to evolve the schemes. The fitness evaluation layer scores the scheduling schemes based on energy consumption and constraints. The parameter adaptation layer dynamically adjusts the algorithm's hyperparameters according to the evolutionary convergence state. Each layer executes sequentially and provides feedback layer by layer, collectively solving the scheduling optimization model. The genetic algorithm aims to minimize the total system energy consumption, satisfying cooling capacity constraints and equipment operating constraints, and outputs optimal scheduling instructions, including equipment start / stop, unit combination, load allocation, and operating parameters.

[0021] In this embodiment, the optimal scheduling instructions are the set of scheduling control instructions that minimize the total energy consumption of the system while meeting cooling demand and ensuring safe equipment operation. They represent the optimal decision result obtained from solving the scheduling optimization model. The optimal scheduling instructions include equipment start-up and shutdown times, combinations of operating units, load allocation, and operating parameters. Equipment start-up and shutdown times refer to the start-up and shutdown times of equipment such as chillers, chilled water pumps, cooling water pumps, and cooling towers, used to control the sequential operation of the equipment. Combinations of operating units refer to the number of devices of the same type operating simultaneously, such as two chillers or three chilled water pumps. Load allocation is the proportion or amount of total cooling load distributed among multiple parallel-operating devices, determining the cooling load borne by each device. Operating parameters are the control parameters for equipment operation, including the chiller load rate, pump frequency converter frequency, cooling tower fan speed, and water supply temperature setpoint.

[0022] S103: Execute the optimal scheduling instruction, collect actual operating data, calculate the deviation between the actual operating data and the predicted operating data, and obtain the total operating deviation.

[0023] In this embodiment, executing the optimal scheduling command involves real-time control of each piece of equipment in the refrigeration station according to the obtained equipment start-up and shutdown times, unit combinations, load allocation, and operating parameters, ensuring operation according to the optimal plan. Actual operating data refers to the real operating data collected by sensors, PLCs, and energy consumption metering devices during the execution of the scheduling command, including actual cooling capacity, hourly indoor cooling load, actual equipment energy consumption, number of operating units, load rate, inlet and outlet water temperatures, flow rates, and operating frequency.

[0024] In this embodiment, deviation is the difference or relative discrepancy between actual operating data and predicted operating data at the same time and under the same index, used to represent the accuracy of scheduling and prediction. Total operating deviation comprehensively represents the overall error of system operation deviating from expectations throughout the entire scheduling cycle, including total energy consumption deviation and cooling capacity deviation, and serves as the basis for subsequent deviation decoupling and model updates.

[0025] S104: Input the total operating deviation into the causal decoupling model and separate the cooling load prediction deviation component and the equipment energy efficiency degradation deviation component from the total operating deviation.

[0026] In this embodiment, the causal decoupling model is a model built based on structural equation modeling or partial least squares regression. In the inference phase, the total operating deviation is used as the only input, and the output is the cooling load prediction deviation component and the equipment energy efficiency degradation deviation component. In the model training phase, the historical total operating deviation, outdoor meteorological parameters, and cumulative equipment running time are used as training features, and the historical decoupled deviation components are used as labels to complete the training. This is used to decompose the total operating deviation into different independent components according to the cause, so as to locate whether the deviation is caused by inaccurate prediction or equipment performance degradation.

[0027] Specifically, the causal decoupling model is constructed based on structural equation modeling or partial least squares regression, and consists of an input layer, a latent variable extraction layer, a path coefficient calculation layer, and a deviation decoupling output layer. The input layer receives total operating deviation, outdoor meteorological parameters, and cumulative equipment operating time; the latent variable extraction layer performs dimensionality reduction and correlation mining on multidimensional variables; the path coefficient calculation layer constructs causal relationships between variables; and the output layer separates the cooling load prediction deviation component and the equipment energy efficiency attenuation deviation component.

[0028] The causal decoupling model uses historical operational deviations, meteorological data, and equipment runtime as training data. It achieves causal decoupling of deviations by iteratively optimizing path coefficients and load matrices by minimizing reconstruction errors. The inputs are total operational deviations (total energy consumption deviations and cooling capacity deviations), outdoor meteorological parameters, and cumulative equipment runtime; the outputs are the cooling load prediction deviation component and the equipment energy efficiency degradation deviation component, which are used to guide the hierarchical updating of the model.

[0029] In this embodiment, the overall deviation between the actual operating data and the predicted operating data within the total operating deviation scheduling cycle, including the total energy consumption deviation and the cooling capacity deviation, is the original deviation that needs to be broken down and analyzed. The cooling load prediction deviation component is the deviation portion directly caused by inaccurate indoor cooling load prediction, separated from the total operating deviation, representing the magnitude of the error in the prediction model itself. The equipment energy efficiency degradation deviation component is the deviation portion caused by energy efficiency degradation and performance decline due to long-term operation of the equipment, separated from the total operating deviation, representing the degree of equipment aging and deviation from the design energy efficiency.

[0030] S105: Based on the cold load prediction deviation component and the equipment energy efficiency attenuation deviation component, iteratively update the scheduling optimization model and / or the preset prediction model.

[0031] In this embodiment, iterative updates involve revising the model parameters, coefficients, or structure in one or more rounds based on the magnitude of the deviation, making the model output closer to actual operating conditions. A preset prediction model is used to predict future hourly cooling load and equipment energy consumption; when the deviation is large, updates are made to improve prediction accuracy. The scheduling optimization model is a mathematical optimization model used to generate optimal scheduling instructions, including an objective function, equipment energy efficiency model, and constraints; updates primarily focus on the energy efficiency model portion.

[0032] S106: Scheduling of refrigeration station equipment based on the updated preset prediction model and / or scheduling optimization model, and applying it to the next scheduling cycle.

[0033] In this embodiment, the updated preset prediction model is a cooling load and energy consumption prediction model that has undergone sliding window parameter adjustment, weight optimization, or retraining based on the cooling load prediction deviation component. This results in higher prediction accuracy and a better fit to the current load characteristics. The updated scheduling optimization model is a scheduling optimization model that has been revised based on the equipment energy efficiency degradation deviation component, correcting the equipment energy efficiency coefficient and load rate-energy consumption relationship, thus better reflecting the current actual operating performance of the equipment. The refrigeration station equipment scheduling is based on the optimized prediction model and / or scheduling optimization model, and involves re-controlling the entire process of load prediction, optimal operation scheme solving, equipment start-up and shutdown, and load allocation. The next scheduling cycle is the next control period immediately following the end of the current scheduling cycle, where the updated model is used for a new round of closed-loop optimization operation.

[0034] As can be seen from the above, this application outputs hourly predicted values ​​of future cooling load and equipment energy consumption through a multi-source data-driven prediction model, and generates optimal scheduling instructions based on an optimization algorithm to achieve accurate operation of the refrigeration station equipment. By collecting actual operating data and calculating the total operating deviation, the deviation is decomposed into a cooling load prediction deviation component and an equipment energy efficiency degradation deviation component using a causal decoupling model, which can accurately locate the source of the deviation. Based on different deviation components, the prediction model and the scheduling optimization model are iteratively updated in a targeted manner to avoid scheduling fluctuations caused by blind updates, improve the accuracy of cooling load prediction and the matching degree of equipment energy efficiency, and make the next round of scheduling more in line with actual operating conditions. While ensuring the reliability of cooling supply, the total energy consumption of the system is reduced, and intelligent scheduling with continuous self-optimization is achieved.

[0035] In one embodiment of this application, the preset scheduling optimization model includes: A comprehensive objective function is constructed to minimize the total operating energy consumption of the system. Energy efficiency models are established to establish the correlation between energy efficiency, load rate and operating time for the chiller, chilled water circulation pump, cooling water circulation pump and cooling tower in the refrigeration station. The total cooling capacity is not less than the predicted cooling load and the operating parameters of each equipment meet the preset operating rules as cooling constraints.

[0036] In this embodiment, the comprehensive objective function is a function constructed with the goal of minimizing the total operating energy consumption of the entire refrigeration station system, and the energy consumption of the chiller, chilled water pump, cooling water pump, and cooling tower are all included in the optimization objective. The energy efficiency models are mathematical models established for the chiller, chilled water circulation pump, cooling water circulation pump, and cooling tower, respectively, to describe the quantitative relationship between equipment energy efficiency, load rate, and cumulative operating time.

[0037] In this embodiment, the load factor is the ratio of the actual load borne by the equipment to the rated load, indicating the degree of equipment load. Running time is the continuous operating time or total cumulative operating time of the equipment, used to reflect the aging and energy efficiency degradation trend of the equipment. Cooling constraints are the limitations that must be met during optimization, including total cooling capacity not less than the predicted cooling load, and operating rules such as equipment start-up and shutdown, load factor, operating frequency, and minimum start-up / shutdown interval. Preset operating rules are boundary conditions set to ensure the safe and stable operation of the equipment, such as minimum number of operating units, maximum load factor, minimum start-up / shutdown interval, and upper and lower frequency limits.

[0038] As can be seen from the above, this embodiment constructs a comprehensive objective function with the goal of minimizing the total system operating energy consumption, and establishes energy efficiency models for the chiller, chilled water pump, cooling water pump, and cooling tower respectively. It deeply correlates equipment load rate, operating time, and energy efficiency, thus better reflecting the actual operating characteristics of the equipment. By setting constraints such as the total cooling capacity not being less than the predicted cooling load and equipment operating rules, it ensures both indoor cooling demand and comfort while avoiding equipment overload, frequent start-stops, and other violations. This scheduling optimization model balances energy-saving goals and operational safety, resulting in more reasonable load allocation and more efficient equipment combination, thereby reducing the overall energy consumption of the refrigeration station and improving the system's operational economy.

[0039] In one embodiment of this application, the scheduling optimization model and / or the preset prediction model are iteratively updated based on the cooling load prediction deviation component and the equipment energy efficiency degradation deviation component, including: If the cold load prediction deviation component is greater than or equal to the first preset threshold and the equipment energy efficiency degradation deviation component is less than the second preset threshold, then only the preset prediction model is updated. If the equipment energy efficiency attenuation deviation component is greater than or equal to the second preset threshold and the cooling load prediction deviation component is less than the first preset threshold, then only the energy efficiency model in the scheduling optimization model is updated. If the cold load prediction deviation component is greater than or equal to the first preset threshold and the equipment energy efficiency degradation deviation component is greater than or equal to the second preset threshold, then the energy efficiency model in the preset prediction model and scheduling optimization model will be jointly iterated and updated.

[0040] In this embodiment, the cooling load prediction deviation component is used to determine whether the prediction model needs to be updated. The equipment energy efficiency degradation deviation component is used to determine whether the equipment energy efficiency model needs to be updated. The first preset threshold is a primary trigger threshold for model updates, a critical value used to determine whether the cooling load prediction deviation is too large. If it exceeds the first preset threshold, it indicates that the prediction model accuracy is insufficient, and the prediction model update process needs to be initiated. The second preset threshold is an independent trigger threshold for energy efficiency model updates, a critical value used to determine whether the equipment energy efficiency degradation is severe. If it exceeds the second preset threshold, it indicates that the equipment energy efficiency model no longer conforms to the actual operating conditions, and the energy efficiency model update process needs to be initiated.

[0041] Specifically, the first preset threshold is determined based on the statistical analysis of the hourly cooling load prediction error of the refrigeration station in history. By collecting the predicted cooling load values ​​and actual cooling load data of multiple typical scheduling cycles, the hourly relative error is calculated and the error distribution curve is plotted. The critical prediction error value that ensures the safety of cooling supply and the increase in system energy consumption is within an acceptable range is selected as the first preset threshold to determine whether the cooling load prediction deviation exceeds the allowable range.

[0042] The second preset threshold is determined based on the equipment's factory energy efficiency curve, long-term operation degradation characteristics, and on-site measured energy efficiency data. The deviation between the actual energy efficiency and the design energy efficiency of equipment such as chillers and water pumps under different cumulative operating times is statistically analyzed. The critical deviation value that causes a significant increase in system energy consumption due to energy efficiency degradation is selected as the second preset threshold to determine whether the equipment performance degradation has affected the scheduling optimization effect.

[0043] In this embodiment, the joint iterative update involves simultaneously and alternately optimizing and correcting the energy efficiency model in both the cooling load prediction model and the scheduling optimization model, thereby improving the overall accuracy.

[0044] As can be seen from the above, this embodiment implements a hierarchical model update strategy based on the relationship between the magnitude of the cooling load prediction deviation component and the equipment energy efficiency degradation deviation component. When only the prediction deviation exceeds the limit, only the prediction model is updated; when only the energy efficiency degradation exceeds the limit, only the energy efficiency model is updated; when both exceed the limit, a joint update is performed, avoiding resource waste and scheduling oscillations caused by a single deviation triggering a full update. The hierarchical update mechanism is more targeted and converges faster, enabling rapid correction of corresponding model errors, maintaining prediction accuracy and energy efficiency model accuracy, ensuring long-term stable and reliable scheduling commands, maintaining efficient and energy-saving operation even under complex operating conditions, and improving system robustness and adaptability.

[0045] In one embodiment of this application, if the cooling load prediction deviation component is greater than or equal to a first preset threshold and the equipment energy efficiency degradation deviation component is less than a second preset threshold, then only the preset prediction model is updated, including: Adjust the sliding window correction parameters of the preset prediction model or retrain the preset prediction model: The sliding window correction parameters include the window size parameter and the correction intensity coefficient, where the window size parameter represents the number of historical prediction periods involved in the correction calculation, and the correction intensity coefficient represents the correction strength. Calculate the cooling load forecast deviation rate for each forecast period, and assign correction weights according to the current date type to calculate the weighted composite deviation rate; If the weighted overall deviation rate is less than the preset trigger threshold, the sliding window correction parameter adjustment process will be initiated. If the weighted overall deviation rate is greater than or equal to the preset trigger threshold, the prediction model retraining process will begin.

[0046] In this embodiment, the sliding window correction parameters are used to correct the prediction model output online. These parameters include a window size parameter and a correction strength coefficient, used to quickly suppress prediction bias without retraining the model. The window size parameter indicates the number of historical prediction periods used in the calculation during prediction correction, such as using historical bias data from the most recent 7 or 14 days. A larger window results in a smoother correction, while a smaller window provides a more sensitive response. The correction strength coefficient indicates the magnitude of the correction to the prediction result. A larger coefficient results in faster bias correction, while a smaller coefficient results in a smoother correction, avoiding abrupt changes in predicted values.

[0047] In this embodiment, the cooling load prediction deviation rate is the relative error between the predicted and actual cooling load within a single prediction period, used to represent prediction accuracy. Date types include weekdays, weekends, and holidays; load patterns differ significantly across date types, therefore different correction weights are assigned. The correction weights are weighted coefficients assigned to the deviation rate based on the date type, making the corrections more closely aligned with the load characteristics of different dates. The weighted composite deviation rate is a normalized quantitative indicator of the cooling load prediction deviation component, used to refine the assessment of the severity of the prediction deviation. The preset trigger threshold is a secondary threshold selected by the model update method, used to determine the critical value for whether to perform lightweight parameter adjustments or completely retrain the prediction model. The sliding window parameter adjustment process adjusts only the window size and correction intensity coefficient when the deviation is small, quickly and lightly correcting the prediction deviation. The prediction model retraining process retrains the prediction model using the latest historical data when the deviation is large, improving prediction accuracy.

[0048] Specifically, the preset trigger threshold is determined by statistical analysis of historical multi-cycle cooling load prediction data of the refrigeration station. First, prediction deviation rate samples are collected under multiple typical seasons and different date types. The deviation range that can meet the accuracy requirements by using only sliding window correction and the deviation critical value that must be retrained to suppress the prediction model are statistically analyzed. Based on the cooling reliability and the fluctuation range of system energy consumption, the critical relative error value that distinguishes between slight deviation and severe deviation is selected as the preset trigger threshold. At the same time, the computational resource consumption and prediction real-time performance are taken into account. This preset trigger threshold can avoid frequent retraining and can update the prediction model in time when the deviation is too large.

[0049] As can be seen from the above, this embodiment achieves refined management of prediction bias by adjusting sliding window correction parameters or retraining the prediction model. By calculating the prediction bias rate and assigning weighted weights according to weekdays, weekends, and holidays, it is possible to distinguish the differences in load patterns across different date types. Based on the weighted comprehensive bias rate and a preset trigger threshold, parameter fine-tuning or model retraining is selected, allowing for rapid correction of small biases and thorough optimization of large biases, balancing correction efficiency and prediction accuracy. This avoids the computational overhead of frequent retraining while ensuring that the cold load prediction continuously closely reflects actual load changes, providing high-quality data support for subsequent precise scheduling.

[0050] In one embodiment of this application, if the weighted overall deviation rate is less than a preset trigger threshold, the sliding window correction parameter adjustment process is initiated, including: Based on the actual cooling load value and the predicted cooling load value of the prediction period based on the window size parameter, calculate the hourly deviation rate sequence of each period; Calculate the mean and standard deviation of the hourly deviation rate sequence, and match the corresponding target deviation tolerance interval from the preset deviation tolerance table according to the date type and time period of the current prediction period; If the mean is greater than or equal to the upper limit of the target deviation tolerance interval, the correction strength coefficient is increased based on the preset first step length; if the mean is less than the lower limit of the target deviation tolerance interval, the correction strength coefficient is decreased based on the preset first step length. If the mean is within the target deviation tolerance range and the standard deviation is greater than the preset fluctuation tolerance threshold, then the window size parameter is adjusted based on the preset second step size. The sliding window correction formula is updated using the adjusted window size parameters and the correction intensity coefficient, and the updated parameters are stored in the database.

[0051] In this embodiment, the hourly deviation rate sequence is a set of deviation data formed by statistically analyzing the relative deviation between the predicted and actual cooling load values ​​at each moment within a sliding window, arranged chronologically. The mean is the average value of the hourly deviation rate sequence, representing the magnitude and direction of the overall prediction deviation. The standard deviation is the dispersion of the hourly deviation rate sequence, reflecting the stability of the prediction results; a larger standard deviation indicates more severe prediction fluctuations. Date types include weekdays, weekends, and holidays. Time periods include daytime, nighttime, and peak load periods, etc., with different load characteristics and deviation tolerances for different time periods. The deviation tolerance table is a pre-established comparison table based on cooling comfort and energy consumption requirements, with different allowable deviation ranges corresponding to different date types and time periods. The target deviation tolerance interval is the range of prediction deviations that the system can accept, obtained by matching the deviation tolerance table.

[0052] In this embodiment, the preset first step size is used to adjust the step size of the correction intensity coefficient, thereby controlling the degree of deviation correction. The preset fluctuation tolerance threshold is a critical value for judging whether the prediction fluctuation is too large; exceeding it indicates insufficient prediction stability. The preset second step size is used to adjust the step size parameter, thereby balancing the correction response speed and stability. The sliding window correction formula is a calculation formula that uses historical deviations within the window to correct the current prediction value online.

[0053] Specifically, the preset first step length is determined by statistical analysis of the historical load prediction correction effect of the refrigeration station. Simulation tests are first conducted on the adjustment step length of different correction intensity coefficients to analyze the boundary values ​​where the correction response is too slow when the step length is too small and the prediction value oscillates and overshoots when the step length is too large. A moderate adjustment step length that can quickly converge the deviation without causing prediction fluctuations is selected as the preset first step length. This preset first step length takes into account both correction efficiency and prediction stability and is adapted to the load change characteristics of different seasons and dates.

[0054] The preset second step size is determined based on the load fluctuation characteristics of the refrigeration station and the statistical stability of the sliding window correction. Through multi-cycle testing of the convergence speed and fluctuation amplitude of the prediction deviation under different window size adjustment ranges, the adjustment step size that can quickly improve the prediction stability without causing frequent jumps in window parameters is selected as the preset second step size. This allows the window size to be adjusted smoothly while quickly adapting to the temporal variation of the indoor cooling load, taking into account both the correction response speed and the stability of the prediction output.

[0055] The preset fluctuation tolerance threshold is determined based on the normal fluctuation range of the hourly cooling load prediction deviation in the history of the refrigeration station: First, the hourly deviation rate sequence of multiple typical scheduling cycles is collected, the standard deviation of the deviation in each cycle is calculated and the distribution curve is plotted. Combining the requirements for indoor cooling comfort and scheduling stability, the standard deviation critical value that can be met in more than 95% of normal operating conditions is selected as the preset fluctuation tolerance threshold. When the actual standard deviation is greater than the preset fluctuation tolerance threshold, it is determined that the prediction fluctuation is too large, so as to ensure that the adjustment logic can identify abnormal fluctuations without frequently adjusting the window parameters due to fluctuations in normal operating conditions.

[0056] As can be seen from the above, this embodiment determines the deviation characteristics by analyzing the mean and standard deviation of the hourly deviation rate sequence, matches the target deviation tolerance range based on the date type and time period, and achieves adaptive adjustment of the sliding window parameters. When the mean deviation exceeds the limit, the correction intensity coefficient is adjusted; when the fluctuation is too large, the window size is optimized, making the correction strategy more closely match the load change characteristics. The adjusted parameters are stored in the database in real time and applied to subsequent forecasts, which can continuously suppress forecast deviations and improve the stability of hourly cold load forecasts. This mechanism can automatically optimize the correction logic without manual intervention, reducing operation and maintenance costs while improving the reliability of the initial data for scheduling instructions.

[0057] In one embodiment of this application, if the equipment energy efficiency degradation deviation component is greater than or equal to a second preset threshold and the cooling load prediction deviation component is less than a first preset threshold, then only the energy efficiency model in the scheduling optimization model is updated, including: The coefficients of the energy efficiency model are iteratively updated using the least squares method based on historical operational data. Identify target devices that are scheduled to run in the optimal scheduling instruction but whose actual energy efficiency is less than the predicted energy efficiency; Based on the target equipment's operational sensor data and historical energy efficiency records, analyze the target equipment's cumulative operating time, load rate range, and actual energy efficiency degradation pattern. Obtain the energy efficiency degradation rate in the actual energy efficiency degradation law, and calculate the energy efficiency correction coefficient based on the energy efficiency degradation rate and equipment aging factor; The energy efficiency correction coefficient is used to iteratively update the model coefficients of the corresponding equipment in the energy efficiency model.

[0058] In this embodiment, the least squares method is a classic data fitting method used to solve and correct the coefficients in the energy efficiency model based on historical operating data, so as to minimize the error between the calculated value of the energy efficiency model and the actual energy efficiency value. The target equipment is a chiller, water pump or cooling tower that is put into operation in this round of optimal scheduling, but whose actual operating energy efficiency is significantly lower than the energy efficiency predicted by the energy efficiency model and which exhibits energy efficiency degradation.

[0059] In this embodiment, cumulative runtime is the total operating time of the equipment since its commissioning; the longer the runtime, the more significant the energy efficiency degradation. The load rate range is the ratio of the actual load to the rated load during equipment operation; the degree of energy efficiency degradation varies for the same equipment under different load rates. The actual energy efficiency degradation pattern is obtained through measured data, showing the downward trend of equipment energy efficiency with changes in runtime and load rate. The energy efficiency degradation rate is the percentage decrease in equipment energy efficiency per unit runtime or load rate change, used to quantify the degree of degradation. The equipment aging factor is an empirical coefficient set based on equipment type, service life, and maintenance conditions, used to represent the degree of impact of aging on energy efficiency. The energy efficiency correction coefficient is a correction value calculated from the energy efficiency degradation rate and the aging factor, used to correct the original energy efficiency model coefficients. The energy efficiency model coefficients are polynomial or function coefficients in the energy efficiency model that characterize the relationship between energy efficiency and load rate and runtime.

[0060] As can be seen from the above, this embodiment accurately identifies target equipment whose actual energy efficiency is less than the predicted energy efficiency value by iteratively updating the energy efficiency model coefficients using the least squares method to address energy efficiency degradation deviations. Based on the equipment's cumulative operating time, load rate range, and energy efficiency degradation pattern, the energy efficiency degradation rate and correction coefficient are calculated to dynamically correct the equipment's energy efficiency model, making the model more closely reflect the actual performance of the equipment after aging. The updated energy efficiency model significantly improves the accuracy of energy consumption prediction, avoids scheduling deviations from the optimal range due to equipment performance degradation, reduces ineffective energy consumption and inefficient equipment operation, extends equipment lifespan, and improves the long-term energy-saving benefits of the refrigeration station.

[0061] In one embodiment of this application, if the cooling load prediction deviation component is greater than or equal to a first preset threshold and the equipment energy efficiency degradation deviation component is greater than or equal to a second preset threshold, then the energy efficiency model in the preset prediction model and the scheduling optimization model is jointly iteratively updated, including: Obtain the total energy consumption deviation and cooling capacity deviation from the total operating deviation; An alternating optimization strategy is adopted, with the objective function of minimizing the joint prediction error. The sliding window correction parameters of the preset prediction model and the coefficients of the energy efficiency model are adjusted alternately. Specifically, the sliding window correction parameters are updated when the coefficients of the energy efficiency model are fixed, and the coefficients of the energy efficiency model are updated when the sliding window correction parameters are fixed. This process is repeated until the joint prediction error converges. The joint prediction error is the weighted sum of the total energy consumption deviation and the cooling capacity deviation. The cold load prediction deviation component and the equipment energy efficiency degradation deviation component are stored as new features in the database for online meta-learning.

[0062] In this embodiment, the total energy consumption deviation is the difference between the actual total operating energy consumption and the predicted total energy consumption, representing the overall deviation in system energy consumption. The cooling capacity deviation is the difference between the actual cooling capacity and the predicted cooling capacity, representing the deviation in the degree to which cooling demand is met.

[0063] In this embodiment, the alternating optimization strategy is a phased collaborative optimization method that alternately fixes one model while optimizing another, avoiding instability caused by simultaneous adjustment of multiple parameters. The joint prediction error is a comprehensive error index obtained by adding the total energy consumption deviation and the cooling capacity deviation according to their weights, used to uniformly measure the overall scheduling effect. The sliding window correction parameters include window size parameters and correction intensity coefficients, used for online fine-tuning of the prediction model output. The energy efficiency model coefficients are mathematical model coefficients describing the relationship between equipment energy efficiency, load rate, and runtime.

[0064] In this embodiment, convergence occurs when the joint prediction error no longer decreases significantly, reaching a stable optimal state, at which point iteration can be stopped. Online meta-learning utilizes newly added deviation features for continuous incremental learning, constantly improving the adaptability of the energy efficiency model in the prediction model and scheduling optimization model under new operating conditions. The new features are the cold load prediction deviation component and the equipment energy efficiency degradation deviation component, which are added as new input features for model training, enhancing the model's expressive power.

[0065] As can be seen from the above, this embodiment employs an alternating optimization strategy when both deviations exceed the limit, aiming to minimize the joint prediction error. It alternately optimizes the prediction model window parameters and energy efficiency model coefficients, avoiding coupling errors caused by single model updates. Through iterative iteration, the joint error gradually converges, achieving synergistic improvement between the prediction and scheduling models. Simultaneously, the deviation component is used as a new feature in online meta-learning, continuously enriching the model feature library and enhancing the system's adaptability under complex operating conditions. The joint update method provides more thorough correction and stronger global optimality, quickly pulling back from deviated operating conditions and ensuring simultaneous improvement in cooling stability and energy-saving performance.

[0066] In one embodiment of this application, a device operation scheduling method based on indoor cooling load prediction further includes: When an outdoor meteorological abrupt event is detected, based on the temperature change information in the weather forecast, the update frequency of the cooling load forecast is temporarily increased, and the correction intensity coefficient in the sliding window correction parameter of the preset forecast model is increased to a preset multiple. When a large-scale event is identified, the corresponding cold load forecast value for the area is increased in advance based on the event time, number of participants, and event area, and the minimum number of operating devices during the event period is adjusted.

[0067] In this embodiment, an outdoor meteorological abrupt event refers to an abnormal meteorological condition characterized by drastic changes in outdoor temperature, humidity, and solar radiation within a short period, which directly leads to significant fluctuations in indoor cooling load. Examples include sudden temperature increases, cold waves, and strong sunlight. Temperature change information refers to the predicted data in the weather forecast indicating a significant rise or fall in outdoor temperature over a future period, including the magnitude of the temperature change, the rate of change, and the duration. The cooling load forecast update frequency is the time interval at which the forecast model recalculates and outputs the cooling load; increasing the frequency allows for faster tracking of load abrupt changes. The correction intensity coefficient is a parameter used to control the magnitude of the correction for forecast deviations; a larger coefficient results in faster correction. The preset multiplier is the correction intensity amplification factor set for meteorological abrupt changes, used to quickly mitigate forecast errors caused by load abrupt changes.

[0068] In this embodiment, large-scale events refer to activities such as conferences, exhibitions, and performances held within the building, where the population density is significantly higher than usual, causing a short-term surge in indoor cooling load. Event time, number of participants, and event area refer to the start and end times of the event, the expected number of people present, and the building area where the load surge occurs, respectively, serving as the basis for cooling load correction. The minimum number of operating equipment constraints are the minimum number of refrigeration units, water pumps, and other equipment that must be put into operation during optimized scheduling to ensure peak cooling capacity.

[0069] As can be seen from the above, this embodiment, by setting up a special scheduling enhancement mechanism for sudden changes in outdoor weather and large-scale events, increases the prediction update frequency and strengthens the correction intensity when temperatures change abruptly. This enables rapid response to sudden changes in cooling load, avoiding insufficient cooling or excessive energy consumption. In the case of large-scale events, the predicted cooling load value for the corresponding area is adjusted upwards in advance, and the minimum number of operating equipment is adjusted to achieve forward load reserves, ensuring reliable peak cooling. This mechanism enhances adaptability to emergencies and special scenarios, reduces the decrease in comfort and energy waste caused by scheduling lag, and improves the operational stability and intelligence level of the refrigeration station in complex scenarios.

[0070] In one embodiment of this application, when an outdoor meteorological abrupt change event is detected, based on the temperature change information in the weather forecast, the update frequency of the cooling load forecast is temporarily increased, and the correction intensity coefficient in the sliding window correction parameters of the preset prediction model is increased to a preset multiple, including: Obtain the future temperature change curve from the weather forecast and calculate the rate of temperature change per unit time. If the absolute value of the temperature change rate is greater than or equal to the first change threshold, it is determined to be a level one meteorological abrupt event. The update frequency of the cold load forecast is increased from the initial frequency to the first frequency, and the correction intensity coefficient is increased to the first intensity multiple. If the absolute value of the temperature change rate is greater than or equal to the second change threshold and less than the first change threshold, it is determined to be a level two meteorological abrupt event. The update frequency of the cold load forecast is increased from the initial frequency to the second frequency, and the correction intensity coefficient is increased to the second intensity multiple.

[0071] The preset magnification includes a first intensity magnification and a second intensity magnification; Among them, the first change threshold is greater than the second change threshold; the first intensity ratio is greater than the second intensity ratio.

[0072] In this embodiment, the future temperature change curve is continuous data showing the change of outdoor temperature over time over a future period, provided by a weather forecast, used to determine the magnitude of meteorological fluctuations. The rate of temperature change per unit time is the magnitude of outdoor temperature change per unit time (e.g., every 15 minutes, every hour), used to represent the degree of temperature abrupt change. The first and second change thresholds correspond to the critical values ​​of the rate of temperature change for Level 1 and Level 2 meteorological abrupt events, respectively. A larger first change threshold indicates a more severe meteorological abrupt change. A Level 1 meteorological abrupt event is an extreme meteorological condition characterized by drastic temperature fluctuations within a short period, which can trigger a significant change in cooling load. A Level 2 meteorological abrupt event is a meteorological condition where the temperature change is relatively significant, but the magnitude is smaller than that of a Level 1 abrupt event, and it still significantly affects the cooling load.

[0073] Specifically, the first change threshold is determined based on historical extreme weather data and cooling load abrupt change response characteristics of the building's location: hourly outdoor temperature data in recent summers is collected, meteorological events that have caused drastic fluctuations in indoor cooling load, insufficient cooling, or sudden increases in energy consumption are screened out, and the temperature change amplitude per unit time during the corresponding period is statistically analyzed; combined with the maximum adjustment rate of the refrigeration station and the allowable range of terminal comfort, the critical rate of temperature change that can cause the cooling load to deviate significantly from the conventional prediction is set as the first change threshold, which is used to identify extreme meteorological abrupt changes that require the highest level of response.

[0074] The second change threshold is determined based on statistical analysis of regular meteorological fluctuation data and the normal range of cooling load variation in the building's location. First, hourly temperature data from the same period over many years is collected, and the normal distribution range of temperature change per unit time under non-extreme weather conditions is statistically analyzed. The temperature change rate that significantly exceeds daily fluctuations but does not reach extreme abrupt changes is selected as a benchmark. Then, combined with the refrigeration station's load regulation capacity, the allowable range of indoor comfort, and the response sensitivity of the prediction model, the critical change rate that can trigger the identification of secondary meteorological abrupt changes is determined, i.e., the second change threshold. This second change threshold is lower than the first change threshold and is used to distinguish between general meteorological fluctuations and moderate meteorological abrupt changes, ensuring early intervention and adjustment when significant load changes occur.

[0075] In this embodiment, the initial frequency is the default prediction update frequency of the cooling load model under normal and stable weather conditions, such as updating once per hour. The first frequency and the second frequency correspond to the improved prediction update frequencies under Level 1 and Level 2 meteorological abrupt events, respectively, with the first frequency being higher and responding faster.

[0076] Specifically, the first frequency is determined based on the fastest rate of change of cooling load under a Level 1 meteorological abrupt change and the adjustable response time of the refrigeration station equipment. Through statistical analysis of historical meteorological and load data, the minimum time interval for significant fluctuations in cooling load under extreme temperature change scenarios is analyzed. Combined with the actual adjustment inertia and control cycle of equipment such as refrigeration units and water pumps, the reciprocal of the update cycle—which can quickly track load abrupt changes without causing excessive system operation due to overly frequent updates—is selected as the first frequency. This frequency is the highest-level predictive update frequency, numerically greater than the second frequency and the initial frequency, and is used to cope with the most severe meteorological and load fluctuations.

[0077] The second frequency is determined by combining the rate of change of cooling load under moderate meteorological fluctuations with the routine control cycle of the refrigeration station. Historical typical meteorological change data are collected, and the time intervals in which the cooling load deviates significantly but does not change drastically are statistically analyzed. Combined with the routine control cycle of the refrigeration station and the data acquisition frequency, an update frequency greater than the initial frequency and less than the first frequency is selected as the second frequency. This ensures timely tracking of load changes without causing excessive consumption of equipment and computing resources, thus guaranteeing a balance between prediction accuracy and system operational stability.

[0078] In this embodiment, the correction intensity coefficient is the force coefficient for online correction of prediction deviations; the larger the coefficient, the faster the correction. The first intensity multiplier and the second intensity multiplier are the factors that amplify the correction force when there are sudden weather changes; the first intensity multiplier is larger and is used to cope with more severe load fluctuations.

[0079] Specifically, the first intensity multiplier is determined based on the maximum deviation of the cooling load under a Level 1 meteorological abrupt change and the predictive correction response characteristics. Through simulation and field measurements of historical extreme temperature change conditions, the maximum deviation of the predicted value without enhanced correction is statistically analyzed. Using the minimum amplification factor that can quickly pull the deviation back to the allowable range without predictive oscillation overshoot as a benchmark, combined with indoor comfort constraints and the refrigeration station's adjustment margin, the specific value of the first intensity multiplier is determined. This ensures rapid correction under extreme weather conditions, guaranteeing prediction accuracy and scheduling stability.

[0080] The second intensity multiplier is determined comprehensively based on the degree of cold load deviation corresponding to the secondary meteorological abrupt change and the requirement for predictive correction stability. By statistically analyzing the predictive deviations under historical moderate temperature change conditions, a multiplier that can converge the deviation to the allowable range within several update cycles without causing drastic fluctuations in the predicted value is selected as the benchmark. The second intensity multiplier is greater than 1 and less than the first intensity multiplier, which achieves effective correction for moderate load fluctuations while avoiding excessive correction that would cause frequent fluctuations in dispatching instructions.

[0081] As can be seen from the above, this embodiment determines the level of meteorological abrupt changes by classifying the rate of temperature change, and sets different prediction update frequencies and correction intensity ratios accordingly to achieve graded response and accurate regulation. Under Level 1 abrupt changes, higher frequency updates and stronger corrections can quickly smooth out severe load fluctuations. Under Level 2 abrupt changes, moderate adjustments are used to balance stability and efficiency, avoiding excessive corrections that could cause scheduling oscillations. This graded strategy ensures cooling security under extreme weather conditions while maintaining system stability under normal abrupt changes, improving the robustness of the prediction model in meteorological fluctuation environments, and ensuring that scheduling commands are always highly matched with actual cooling load demand.

[0082] In one embodiment of this application, before generating the optimal scheduling instruction by solving the scheduling optimization model using an optimization algorithm based on predicted running data, the method further includes: In the process of solving the scheduling optimization model using a genetic algorithm, the hyperparameters are adjusted according to the current evolutionary state; among them, the hyperparameters include population size, crossover probability, and mutation probability. If the fitness function value improvement rate is less than the preset improvement threshold for three consecutive generations, the crossover probability is increased based on the first adjustment amount. If the quality variance of the solution is greater than the preset variance threshold, the mutation probability is increased based on the second adjustment.

[0083] In this embodiment, the evolutionary state refers to the convergence degree and solution distribution of the genetic algorithm during the iterative optimization process, represented by the fitness improvement rate and the variance of solution quality. Hyperparameters are key parameters controlling the optimization behavior of the genetic algorithm, including population size, crossover probability, and mutation probability. Population size is the number of candidate scheduling schemes participating in the evolution in each generation iteration.

[0084] In this embodiment, the crossover probability controls the probability of information exchange between individuals from two generations, affecting the algorithm's global search capability. The mutation probability controls the probability of random mutations in an individual's genes, affecting the algorithm's ability to escape local optima. The fitness function value is a function used to evaluate the quality of a scheduling scheme, generally related to energy consumption and the degree of constraint satisfaction; a higher value indicates a better scheme. The fitness function improvement rate is the increase in the optimal fitness over several generations, representing the algorithm's convergence speed.

[0085] In this embodiment, the preset improvement threshold is the critical value used to determine whether the optimization algorithm has fallen into premature convergence or stagnation. The first adjustment amount is a fixed increment used to increase the crossover probability and improve the global search capability.

[0086] The preset improvement threshold was determined through simulation testing of the genetic algorithm optimization process under multiple typical load conditions of the refrigeration station. Fitness improvement rate data of the algorithm for three consecutive generations under both normal convergence and premature convergence stagnation states were collected, and the improvement rate distribution interval was plotted. The critical value that can effectively distinguish between normal convergence and premature convergence stagnation was selected as the preset improvement threshold, so that evolutionary stagnation can be identified in a timely manner without triggering parameter adjustments due to normal small fluctuations.

[0087] The first adjustment amount is determined based on the convergence characteristics and parameter stability of the genetic algorithm. Through multi-condition simulation tests with different crossover probability increments, the boundary intervals where too small an adjustment amount cannot effectively improve evolutionary stagnation and too large an adjustment amount causes the algorithm's search to oscillate and diverge are analyzed. A smaller increment that can significantly improve the global search efficiency while ensuring smooth convergence of the iterative process is selected as the first adjustment amount. This allows the optimization algorithm to moderately enhance the crossover operation when the fitness improvement is slow, while avoiding excessive parameter jumps that affect the optimization stability.

[0088] In this embodiment, the quality variance of the solution represents the dispersion of the fitness distribution of all candidate solutions in the current population. A large variance indicates a dispersed solution distribution, while a small variance indicates that the solutions tend to be concentrated. The preset variance threshold is a critical value used to determine whether the solution is too homogeneous and prone to getting trapped in local optima. The second adjustment amount is used to increase the increment of the mutation probability, improve population diversity, and avoid local optima.

[0089] The preset variance threshold is determined statistically based on the fitness distribution characteristics of individuals in the genetic algorithm population. The variance of the solution quality during the optimization process under multiple typical working conditions is sampled, and the variance boundary points of uniform solution distribution and homogeneous solutions are analyzed. Combined with the stability requirements of global optimal solution, a critical variance value that can indicate insufficient population diversity and susceptibility to local optima is selected as the preset variance threshold.

[0090] The second adjustment amount is determined comprehensively based on the population diversity regulation requirements and solution stability of the genetic algorithm. Through multiple simulation tests on the optimization performance of the algorithm under different mutation probability increments, the critical intervals were analyzed, showing that too small an adjustment amount cannot effectively improve population diversity, while too large an adjustment amount will destroy the excellent solution structure. A smaller increment that can significantly increase the solution space exploration capability without causing the iteration process to oscillate and diverge was selected as the second adjustment amount. This allows the optimization algorithm to moderately enhance the mutation operation when the solution quality variance is too large, balancing the escape from local optima and solution stability.

[0091] As can be seen from the above, this embodiment adaptively adjusts the population size, crossover probability, and mutation probability based on the evolutionary state during the genetic algorithm solution process. When three consecutive generations of improvement are insufficient, the crossover probability is increased to enhance the global search; when the solution variance is too large, the mutation probability is increased to improve the local optimization ability, effectively avoiding the problems of premature convergence and slow optimization. The dynamic hyperparameter mechanism makes the optimization process more efficient and the solution quality higher, enabling it to approach the optimal scheduling scheme faster, improving the optimization accuracy of equipment start-up and shutdown, unit combination, and load allocation, and obtaining a scheduling strategy with lower energy consumption and more stable operation in a shorter time.

[0092] In one embodiment of this application, a device operation scheduling method based on indoor cooling load prediction further includes: The system allows for bidirectional switching between automatic optimization mode and manual control mode, and includes a mode switching buffer mechanism. When switching from automatic optimization mode to manual control mode, a preset buffer transition period is entered to keep the current equipment operating status unchanged. After key operating parameters such as water supply temperature, return water temperature, and equipment operating frequency stabilize, the equipment control is then handed over to the operator. When switching from manual control mode to automatic optimization mode, the current actual operating conditions at the time of switching are used as the initial value of the optimization algorithm. A step-by-step adjustment strategy is adopted to limit the maximum adjustment range of a single control command, gradually converging to the optimal scheduling condition, so as to achieve a smooth and disturbance-free transition of scheduling commands.

[0093] In this embodiment, the automatic optimization mode is a control mode in which the system automatically generates and issues optimal scheduling commands based on a cooling load prediction and scheduling optimization model, thereby achieving fully automatic energy-saving operation of the equipment. The manual control mode is a manual control mode in which maintenance personnel manually set parameters such as equipment start / stop, frequency, and load rate through a host computer and control cabinet. The mode switching buffer mechanism is a smooth transition mechanism set up to avoid sudden changes in control commands during mode switching, which could cause equipment shocks or cooling fluctuations.

[0094] In this embodiment, the buffer transition period is a specially designed transition time to maintain the same operating status when switching from automatic mode to manual mode, used for stabilizing operating conditions and handing over control. Stable operating conditions mean that key parameters such as supply water temperature, return water temperature, load, and equipment frequency do not fluctuate drastically, and the system operates smoothly. Handing over control means that after the buffer period, control of the equipment is formally transferred to the operator, allowing manual intervention.

[0095] In this embodiment, the initial optimization value is used as the starting point for the optimization algorithm when switching back from manual to automatic mode, based on the current actual operating parameters, rather than starting the calculation from zero. The step-by-step adjustment strategy breaks down the total adjustment amount of the scheduling command into multiple small-amplitude, progressive adjustments, rather than a one-time adjustment. The single adjustment range is the maximum allowable range of change for each control command, used to limit the adjustment speed and fluctuations. The smooth transition of the scheduling command involves gradually converging from manual operation to automatic optimized operation while ensuring stable cooling.

[0096] As can be seen from the above, this embodiment establishes a bidirectional switching mechanism between automatic optimization and manual control, and is equipped with a buffer transition period and a stepped adjustment strategy. When switching from automatic to manual, the system maintains stable operating conditions before transferring control, avoiding sudden switching that could cause hydraulic imbalance and equipment shock. When switching from manual to automatic, the transition is smooth, starting from the current operating condition, limiting the magnitude of single adjustments and ensuring continuous and uninterrupted scheduling commands. This mode switching mechanism enhances the system's operational flexibility, providing both automatic energy-saving efficiency and manual emergency intervention capabilities. It ensures the safety and stability of the refrigeration station throughout its daily operation and maintenance, improving system availability and field applicability.

[0097] Corresponding to the equipment operation scheduling method based on indoor cooling load prediction in the above embodiment, Figure 2 This is a structural block diagram of an equipment operation scheduling system based on indoor cooling load forecasting, provided as an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The equipment operation scheduling system 20 based on indoor cooling load prediction includes: a model prediction module 21, an optimization scheduling module 22, an operation deviation module 23, a causal decoupling module 24, a model update module 25, and an update application module 26.

[0098] Among them, the model prediction module 21 is used to acquire multi-source data information of the equipment in the refrigeration station, and based on the multi-source data information, it uses a preset prediction model to output the hourly indoor cooling load prediction value and the corresponding equipment energy consumption prediction value to obtain the predicted operation data. The optimization scheduling module 22 is used to solve the preset scheduling optimization model based on the predicted running data and generate the optimal scheduling instruction. The operation deviation module 23 is used to execute the optimal scheduling instruction, collect actual operation data, calculate the deviation between the actual operation data and the predicted operation data, and obtain the total operation deviation. The causal decoupling module 24 is used to input the total operating deviation into the causal decoupling model and separate the cooling load prediction deviation component and the equipment energy efficiency degradation deviation component from the total operating deviation. The model update module 25 is used to iteratively update the scheduling optimization model and / or the preset prediction model based on the cold load prediction deviation component and the equipment energy efficiency attenuation deviation component. Update application module 26 to schedule refrigeration station equipment based on the updated preset prediction model and / or scheduling optimization model, and apply it to the next scheduling cycle.

[0099] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The functions of the model prediction module 21, optimization scheduling module 22, operation deviation module 23, causal decoupling module 24, model update module 25, and update application module 26 are shown.

[0100] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0101] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0102] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0103] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the equipment operation scheduling method based on indoor cooling load prediction provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0104] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0105] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., provided on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0106] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. 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 implementations should not be considered beyond the scope of this application.

[0107] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0108] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0109] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0110] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0111] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for equipment operation scheduling based on indoor cooling load forecasting, characterized in that, include: Acquire multi-source data information of equipment in the refrigeration station, and based on the multi-source data information, use a preset prediction model to output the hourly indoor cooling load prediction value and the corresponding equipment energy consumption prediction value to obtain the predicted operation data; Based on the predicted operational data, the preset scheduling optimization model is solved using an optimization algorithm to generate the optimal scheduling instruction; Execute the optimal scheduling instruction, collect actual operating data, calculate the deviation between the actual operating data and the predicted operating data, and obtain the total operating deviation; The total operating deviation is input into the causal decoupling model, and the cooling load prediction deviation component and the equipment energy efficiency degradation deviation component are separated from the total operating deviation. Based on the cold load prediction deviation component and the equipment energy efficiency attenuation deviation component, the scheduling optimization model and / or the preset prediction model are iteratively updated. The refrigeration station equipment is scheduled based on the updated preset prediction model and / or the scheduling optimization model, and applied to the next scheduling cycle.

2. The equipment operation scheduling method based on indoor cooling load prediction according to claim 1, characterized in that, The preset scheduling optimization model includes: A comprehensive objective function is constructed to minimize the total operating energy consumption of the system, and energy efficiency models are established for the correlation between energy efficiency, load rate and operating time for the chillers, chilled water circulation pumps, cooling water circulation pumps and cooling towers in the refrigeration station. The total cooling capacity is not less than the predicted cooling load value and the operating parameters of each device meet the preset operating rules as cooling constraints.

3. The equipment operation scheduling method based on indoor cooling load prediction according to claim 2, characterized in that, The iterative update of the scheduling optimization model and / or the preset prediction model based on the cold load prediction deviation component and the equipment energy efficiency degradation deviation component includes: If the cold load prediction deviation component is greater than or equal to the first preset threshold and the equipment energy efficiency degradation deviation component is less than the second preset threshold, then only the preset prediction model is updated. If the energy efficiency attenuation deviation component of the equipment is greater than or equal to the second preset threshold and the cooling load prediction deviation component is less than the first preset threshold, then only the energy efficiency model in the scheduling optimization model is updated. If the cold load prediction deviation component is greater than or equal to the first preset threshold and the equipment energy efficiency attenuation deviation component is greater than or equal to the second preset threshold, then the preset prediction model and the energy efficiency model in the scheduling optimization model are jointly iteratively updated.

4. The equipment operation scheduling method based on indoor cooling load prediction according to claim 3, characterized in that, If the cold load prediction deviation component is greater than or equal to a first preset threshold and the equipment energy efficiency degradation deviation component is less than a second preset threshold, then only the preset prediction model is updated, including: Adjust the sliding window correction parameters of the preset prediction model or retrain the preset prediction model: The sliding window correction parameters include a window size parameter and a correction intensity coefficient, wherein the window size parameter represents the number of historical prediction periods involved in the correction calculation, and the correction intensity coefficient represents the correction strength; Calculate the cooling load forecast deviation rate for each forecast period, and assign correction weights according to the current date type to calculate the weighted composite deviation rate; If the weighted overall deviation rate is less than the preset trigger threshold, the sliding window correction parameter adjustment process will be initiated. If the weighted overall deviation rate is greater than or equal to the preset trigger threshold, the prediction model retraining process will begin.

5. The equipment operation scheduling method based on indoor cooling load prediction according to claim 4, characterized in that, If the weighted overall deviation rate is less than a preset trigger threshold, the sliding window correction parameter adjustment process will begin, including: Based on the actual cooling load value and the predicted cooling load value of the prediction period according to the window size parameter, calculate the hourly deviation rate sequence of each period; Calculate the mean and standard deviation of the hourly deviation rate sequence, and match the corresponding target deviation tolerance interval from the preset deviation tolerance table according to the date type and time period of the current prediction period; If the mean is greater than or equal to the upper limit of the target deviation tolerance interval, the correction strength coefficient is increased based on the preset first step length; if the mean is less than the lower limit of the target deviation tolerance interval, the correction strength coefficient is decreased based on the preset first step length. If the mean is within the target deviation tolerance range and the standard deviation is greater than the preset fluctuation tolerance threshold, then the window size parameter is adjusted based on the preset second step size. The sliding window correction formula is updated using the adjusted window size parameters and correction intensity coefficient, and the updated parameters are stored in the database.

6. The equipment operation scheduling method based on indoor cooling load prediction according to claim 3, characterized in that, If the energy efficiency degradation deviation component of the equipment is greater than or equal to the second preset threshold and the cooling load prediction deviation component is less than the first preset threshold, then only the energy efficiency model in the scheduling optimization model is updated, including: The coefficients of the energy efficiency model are iteratively updated using the least squares method based on historical actual operating data. Identify target devices that are scheduled to run in the optimal scheduling instruction but whose actual energy efficiency is less than the predicted energy efficiency; Based on the operating sensor data and historical energy efficiency records of the target device, the cumulative operating time, load rate range and actual energy efficiency degradation pattern of the target device are analyzed. Obtain the energy efficiency degradation rate in the actual energy efficiency degradation law, and calculate the energy efficiency correction coefficient based on the energy efficiency degradation rate and the equipment aging factor; The energy efficiency correction coefficient is used to iteratively update the model coefficients of the corresponding equipment in the energy efficiency model.

7. The equipment operation scheduling method based on indoor cooling load prediction according to claim 3, characterized in that, If the cold load prediction deviation component is greater than or equal to the first preset threshold and the equipment energy efficiency degradation deviation component is greater than or equal to the second preset threshold, then a joint iterative update is performed on the preset prediction model and the energy efficiency model in the scheduling optimization model, including: Obtain the total energy consumption deviation and cooling capacity deviation from the total operating deviation; An alternating optimization strategy is adopted, with the objective function of minimizing the joint prediction error. The sliding window correction parameters of the preset prediction model and the coefficients of the energy efficiency model are adjusted alternately. Specifically, the sliding window correction parameters are updated when the coefficients of the energy efficiency model are fixed, and the coefficients of the energy efficiency model are updated when the sliding window correction parameters are fixed. This process is repeated until the joint prediction error converges. The joint prediction error is the weighted sum of the total energy consumption deviation and the cooling capacity deviation. The cold load prediction deviation component and the equipment energy efficiency attenuation deviation component are stored as new features in the database for online meta-learning.

8. The equipment operation scheduling method based on indoor cooling load prediction according to claim 1, characterized in that, Also includes: When an outdoor meteorological abrupt event is detected, based on the temperature change information in the weather forecast, the update frequency of the cooling load forecast is temporarily increased, and the correction intensity coefficient in the sliding window correction parameter of the preset prediction model is increased to a preset multiple. When a large-scale event is identified, the corresponding cold load forecast value for the area is increased in advance based on the event time, number of participants, and event area, and the minimum number of operating devices during the event period is adjusted.

9. A method for equipment operation scheduling based on indoor cooling load prediction according to claim 8, characterized in that, When an outdoor meteorological abrupt change event is detected, based on the temperature change information in the weather forecast, the update frequency of the cooling load forecast is temporarily increased, and the correction intensity coefficient in the sliding window correction parameters of the preset prediction model is increased to a preset multiple, including: Obtain the future temperature change curve from the weather forecast and calculate the rate of temperature change per unit time. If the absolute value of the temperature change rate is greater than or equal to the first change threshold, it is determined to be a first-level meteorological sudden event, the update frequency of the cold load forecast is increased from the initial frequency to the first frequency, and the correction intensity coefficient is increased to the first intensity multiple. If the absolute value of the temperature change rate is greater than or equal to the second change threshold and less than the first change threshold, it is determined to be a level two meteorological sudden event. The update frequency of the cold load forecast is increased from the initial frequency to the second frequency, and the correction intensity coefficient is increased to the second intensity multiple. The preset magnification includes a first intensity magnification and a second intensity magnification; Wherein, the first change threshold is greater than the second change threshold; and the first intensity multiplier is greater than the second intensity multiplier.

10. A system for scheduling equipment operation based on indoor cooling load forecasting, characterized in that, include: The model prediction module is used to acquire multi-source data information of the equipment in the refrigeration station. Based on the multi-source data information, a preset prediction model is used to output the hourly indoor cooling load prediction value and the corresponding equipment energy consumption prediction value to obtain the predicted operation data. The optimization scheduling module is used to solve the preset scheduling optimization model based on predicted running data and using optimization algorithms to generate the optimal scheduling instruction; The operation deviation module is used to execute the optimal scheduling instruction, collect actual operation data, calculate the deviation between the actual operation data and the predicted operation data, and obtain the total operation deviation. The causal decoupling module is used to input the total operating deviation into the causal decoupling model and separate the cooling load prediction deviation component and the equipment energy efficiency degradation deviation component from the total operating deviation. The model update module is used to iteratively update the scheduling optimization model and / or the preset prediction model based on the cold load prediction deviation component and the equipment energy efficiency attenuation deviation component. The application module is updated to schedule refrigeration station equipment based on the updated preset prediction model and / or the scheduling optimization model, and applied to the next scheduling cycle.

Images