Cold load comprehensive scheduling control method based on LSTM load prediction

Abstract

The invention relates to the cross technical field of artificial intelligence and intelligent control of a heating ventilation air-conditioning system, discloses a cooling load comprehensive scheduling control method based on LSTM load prediction, and aims to solve the problems that load prediction and scheduling optimization are separated, equipment energy consumption coupling modeling is insufficient and electricity price response collaboration is poor in an existing regional cooling system. The method comprises the following steps: carrying out 24-hour cold load prediction based on a six-dimensional input LSTM model; real-time electricity price and predicted load are fused to construct an economic dispatching model with the operation cost minimization as the target, and an electricity price elastic coefficient and a thermodynamic efficiency correction factor are introduced to enhance physical consistency; cooperative control of the refrigerating unit and the cold storage device is achieved based on a dynamic decision matrix of the cold storage tank reserve rate and the load grade; and dynamically adjusting the water pump frequency of the cooling water system according to the pressure difference deviation to optimize the hydraulic performance. The modules are integrated through the central cooperative controller to form a closed-loop scheduling system, and the energy efficiency and the economical efficiency are improved while the cooling load requirement is met.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of artificial intelligence and intelligent control of HVAC systems, specifically a comprehensive scheduling and control method for cooling load based on LSTM load forecasting. Background Technology

[0002] In modern urban energy systems, district cooling is a crucial technological pathway for improving building energy efficiency and reducing carbon emissions, and its operational optimization directly impacts overall energy utilization efficiency and economics. With the deepening of the "dual-carbon" strategy and the continuous improvement of the electricity market mechanism, district cooling systems not only need to meet basic cooling requirements but are also endowed with multiple functions, including responding to grid dispatch, participating in peak-valley electricity price arbitrage, and coordinating renewable energy consumption. Against this backdrop, how to achieve accurate prediction of cooling load and efficient coordinated dispatch of cooling source equipment through intelligent means has become key to improving the overall performance of district cooling systems. Traditional control strategies often rely on fixed schedules or rule-based start-stop logic, which are ill-suited to the new challenges of increased load volatility, dynamic electricity price signals, and rising equipment coupling complexity.

[0003] However, with the continuous development of related technologies and the increasingly stringent performance requirements of application scenarios, some inherent characteristics of the aforementioned technical solutions at the principle level have gradually revealed their limitations in addressing new challenges. Specifically, while existing methods have made progress in both load forecasting and electricity price response, these two aspects are often treated separately: the load forecasting module only focuses on the time-series fitting accuracy of cooling demand, while the scheduling optimization module treats the forecast results as deterministic inputs, ignoring the potential impact of forecast uncertainty on the robustness of subsequent decisions. Furthermore, the energy consumption coupling relationship between the chiller unit and the cold storage device has not been fully modeled—the COP of the chiller unit under different operating conditions is not constant, but dynamically changes with cooling water temperature, evaporator load, and compressor speed; at the same time, the heat loss, stratification efficiency, and pump power consumption of the cold storage tank during charging and discharging also significantly affect the overall energy consumption of the system. If scheduling is based solely on simplified linear performance consumption assumptions or static efficiency parameters, it is highly likely that the theoretically optimal solution will deviate significantly from actual operation. Furthermore, although real-time electricity price signals are incorporated into the objective function, their integration mechanism with load forecasting is mostly limited to weighted summation, lacking in-depth analysis of the joint distribution characteristics of electricity price and load. This makes it difficult to maintain the stability and economy of the control strategy under scenarios of drastic electricity price fluctuations or sudden load changes. The reason for this is that the existing framework has failed to construct a unified coupled optimization model with the ability to embed physical constraints, resulting in information gaps between forecasting, pricing, and equipment control. This, in turn, limits the system's adaptive adjustment capability and overall energy efficiency improvement potential under complex operating conditions.

[0004] Therefore, how to construct a comprehensive scheduling and control method that deeply integrates high-precision load forecasting, dynamic electricity price response mechanism and energy consumption characteristics of chiller-TES coupling, so as to ensure the reliability of cooling supply while taking into account the economic efficiency, energy efficiency and control robustness, has become a key challenge and a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention provides a comprehensive cooling load scheduling and control method based on LSTM load forecasting, aiming to solve technical problems in existing district cooling systems such as the disconnect between load forecasting and scheduling optimization, insufficient modeling of equipment energy consumption coupling relationships, and a lack of deep coordination in electricity price response mechanisms. To achieve the above-mentioned objectives, this invention constructs a closed-loop control architecture that integrates high-precision time-series load forecasting, dynamic electricity price signal fusion, coupled energy consumption modeling of refrigeration units and thermal energy storage (TES) devices, and coordinated optimization of the hydraulic system, ensuring a balance between operational economy, energy efficiency, and control robustness while meeting cooling load demands.

[0006] The method described in this invention comprises four core components: the first part is a 24-hour cooling load prediction model based on a Long Short-Term Memory (LSTM) neural network; the second part is an economic dispatch optimization model that integrates real-time electricity price signals and load prediction results; the third part is the coupled energy consumption modeling and collaborative control logic between the refrigeration unit and the cold storage device; and the fourth part is the variable frequency drive control strategy for the primary and secondary water pumps in the cooling water system. These four parts interact and distribute commands through a central collaborative controller, forming a complete integrated dispatch control closed loop.

[0007] Furthermore, the LSTM-based cooling load forecasting model employs a single-layer LSTM structure with a six-dimensional input dimension, consisting of ambient temperature, relative humidity, solar radiation intensity, historical 24-hour cooling load sequence, date type identifier, and current real-time electricity price. The input sequence length is fixed at 24 time steps, with each time step corresponding to a historical observation one hour prior. The number of nodes in the LSTM hidden layer is set to 64, and the output layer is a fully connected layer, directly generating hourly cooling load forecasts for the next 24 hours. During the training phase, the model uses mean squared error as the loss function and updates the network weight parameters through backpropagation. During the inference phase, the model receives standardized real-time input features and outputs a deterministic cooling load forecast curve, which serves as the sole basis for load demand in subsequent scheduling modules.

[0008] In a preferred embodiment of the present invention, the economic scheduling optimization model takes minimizing the total operating cost over 24 hours as its objective function, and the objective function expression is as follows:

[0009]

[0010] in For the first Hourly power purchase capacity of the power grid For the corresponding electricity price at that time, The electrical power of the chiller unit during operation. This is the energy efficiency penalty coefficient, used to reflect the additional energy consumption cost of the refrigeration unit under suboptimal operating conditions. The constraints of this optimization model include upper and lower limits for the cold storage tank capacity, cold load supply and demand balance constraints, and equipment start-up and shutdown logic constraints. Specifically, the cold storage tank at any given time... Storage cold energy satisfy The total cooling capacity of the system equals the predicted cooling load demand, i.e. ,in To provide cooling capacity for the chiller unit This refers to the amount of cold released from the cold storage tank.

[0011] Furthermore, this invention introduces an electricity price elasticity coefficient. With thermodynamic efficiency correction factor Physical consistency enhancements are applied to the economic dispatch model. The electricity price elasticity coefficient is mentioned. This is used to quantify the adjustable potential of user-side cooling load under different electricity price levels; its value is calibrated offline based on the electricity price-load response relationship in historical operating data. The thermodynamic efficiency correction factor... This is used to correct the actual Coefficient of Performance (COP) of the refrigeration unit under non-rated operating conditions. This factor is a function of the cooling water inlet temperature, evaporator load rate, and compressor speed, and is obtained online through table lookup or piecewise linear interpolation. The two correction parameters mentioned above are embedded in the objective function. The calculation process ensures that scheduling decisions not only consider electricity costs but also reflect the energy consumption characteristics of equipment under real operating conditions.

[0012] As another key feature of this invention, the collaborative control logic between the refrigeration unit and the cold storage device is implemented based on a dynamic decision matrix. This decision matrix takes the current cold storage capacity reserve rate of the cold storage tank and the predicted load demand level as input variables, and outputs corresponding equipment action commands. The cold storage capacity reserve rate is defined as the ratio of the current cold storage capacity to the maximum cold storage capacity; the load demand level is divided into three categories: peak, medium load, and low load, determined based on the deviation between the predicted cooling load and the historical average for the same period. When the cold storage capacity reserve rate is below 30% and the predicted load is during a peak period, the system starts the standby chiller unit and simultaneously opens the cold storage tank's cooling valve; when the cold storage capacity reserve rate is between 30% and 70% and the load is at a medium load level, the main chiller unit operates in variable frequency mode, while the cold storage tank provides auxiliary cooling compensation; when the cold storage capacity reserve rate is above 70% and it is during a low electricity price period, the system stops all chiller units, relying solely on the cold storage tank for cooling, and, under conditions where renewable energy access is available, starts a photovoltaic-driven cold storage circulation pump for supplementary cold storage. This decision logic is executed periodically by the central collaborative controller, with an execution cycle of 15 minutes.

[0013] Furthermore, this invention provides refined modeling and control of the hydraulic characteristics of the cooling water system. The cooling water system includes a primary-side cooling water pump and a secondary-side chilled water pump, both equipped with variable frequency drives. The central coordinating controller operates based on the measured values ​​of the supply and return water pressure difference. With set value The deviation is used to calculate the pump drive frequency command. Its control law is

[0014]

[0015] in , This is the measured value of the supply and return water pressure difference. Set the differential pressure value. and The proportional gain and integral gain are pre-tuned based on the pipeline impedance characteristics and pump performance curves. This control law ensures that the pressure difference between the supply and return water is maintained within the optimal set range, thereby minimizing pump energy consumption while meeting the terminal cooling load flow requirements. The pressure difference setpoint... It is not a fixed constant, but is dynamically adjusted according to the current cooling load level and the cooling release ratio of the cold storage tank to adapt to changes in the system's hydraulic conditions.

[0016] As a system integration method of the present invention, the central collaborative controller establishes a communication connection with the LSTM prediction module, the economic scheduling solver, the equipment status monitoring unit, and the frequency converter via an industrial Ethernet. The LSTM prediction module uploads an updated 24-hour cooling load prediction curve to the central collaborative controller every 15 minutes; the economic scheduling solver completes rolling optimization calculations within 5 minutes based on the latest prediction results and real-time electricity price signals, and sends the optimal equipment start-up and shutdown plan and cold storage strategy to the execution layer; the equipment status monitoring unit collects parameters such as chiller unit operating current, cold storage tank inlet and outlet temperatures, pump frequency, and valve opening in real time, and feeds them back to the central collaborative controller to verify the execution effect of scheduling instructions; if the deviation between the actual cooling capacity and the predicted demand exceeds a preset threshold, a model retraining mechanism or a scheduling strategy fine-tuning process is triggered.

[0017] Furthermore, the input features of the LSTM prediction model undergo standardization before entering the network. The standardization parameters are obtained based on historical data from the past 30 days and are automatically updated daily at midnight. Date type identifiers are represented using one-hot encoding, including weekdays, weekends, and public holidays. Real-time electricity price signals are based on hourly updated market clearing prices; if a price is missing, it is filled using a strategy that preserves the previous valid value. The model outputs a cooling load forecast in tons of cooling (RT) and is directly used in the economic dispatch model. The calculations are performed without any post-processing, smoothing, or correction.

[0018] As the physical basis for the present invention, the cold storage tank is equipped with a temperature stratification monitoring array, with no fewer than eight temperature sensors arranged along the height of the tank, for real-time assessment of the hot and cold water stratification status and effective cold storage capacity; the chiller unit is equipped with a variable frequency compressor and an electronic expansion valve, supporting continuous adjustment within a load range of 10% to 100%; the cooling tower fan also adopts variable frequency control, and its speed is dynamically adjusted according to the difference between the condenser outlet water temperature and the wet bulb temperature to maintain optimal condensing pressure. The above hardware configuration collectively supports the engineering implementation of the refined scheduling and control strategy proposed in this invention.

[0019] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0020] This invention organically integrates high-precision LSTM load forecasting, dynamic electricity price response mechanisms, equipment coupled energy consumption modeling, and hydraulic system optimization control to construct a comprehensive cooling load scheduling and control system with strong physical constraint embedding capabilities and multi-source information fusion characteristics. During operation, this system consistently uses predicted cooling load as the scheduling benchmark, real-time electricity price as the economic guide, actual equipment energy consumption characteristics as the constraint boundary, and hydraulic stability as the guarantee condition, thereby achieving globally optimal control of the district cooling system in complex and ever-changing operating environments. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall system flow of the present invention.

[0022] Figure 2 This is a schematic diagram of the network structure and input-output relationship of the LSTM cooling load prediction model in this invention.

[0023] Figure 3 This is a schematic diagram of the decision-making process of the economic scheduling optimization model and equipment collaborative control logic of the present invention.

[0024] Figure 4 This is a schematic diagram of the hydraulic regulation process of the frequency conversion drive control strategy for the primary and secondary water pumps in the cooling water system of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] This invention provides a comprehensive cooling load scheduling and control method based on LSTM load forecasting. Its overall architecture comprises a central coordinating controller, an LSTM cooling load forecasting module, an economic scheduling optimization solver, equipment execution units, and a hydraulic regulation system. All components achieve high-speed data interaction via industrial Ethernet, forming a closed-loop feedback and feedforward coordination mechanism. The specific implementation of this invention will be described in detail below from five dimensions: system-level deployment, model construction details, optimization algorithm implementation, equipment control logic, and hardware configuration.

[0027] The central coordinating controller, serving as the decision-making hub of the entire system, employs an embedded industrial computer platform equipped with a real-time operating system and possesses multi-threaded task scheduling capabilities. This controller periodically executes core functions such as generating scheduling instructions, monitoring equipment status, handling anomalies, and updating model parameters. Its internal software architecture is divided into a communication interface layer, a data preprocessing layer, a strategy execution layer, and a security verification layer. The communication interface layer is responsible for establishing TCP / IP connections with the LSTM prediction module, the economic scheduling solver, the chiller controller, the cold storage tank monitoring unit, and the water pump frequency converter. The data preprocessing layer filters, aligns, and synchronizes the acquired raw sensor signals. The strategy execution layer parses the scheduling instructions according to preset control logic and sends them to the execution devices. The security verification layer continuously compares the deviation between the actual cooling capacity and the predicted demand; when the deviation exceeds a set threshold, it triggers a fine-tuning of the scheduling strategy or a model retraining process.

[0028] The LSTM cooling load forecasting module is deployed on a standalone server node, running a deep learning inference engine. This module receives historical environmental and load data streams from the Building Automation System (BAS), including ambient temperature, relative humidity, solar radiation intensity, historical 24-hour cooling load sequences, date type identifiers, and the current real-time electricity price. All input features are standardized before entering the neural network. The mean and standard deviation used for standardization are dynamically calculated based on historical data within a 30-day rolling window and are automatically updated daily at midnight. The date type identifier is represented using one-hot encoding and includes three categories: weekdays, weekends, and public holidays, corresponding to the three-dimensional vectors [1,0,0], [0,1,0], and [0,0,1], respectively. The real-time electricity price signal is based on the hourly updated market clearing price. If the electricity price data is missing at a certain moment, a strategy of preserving the previous valid value is used to fill in the missing data, ensuring the integrity of the input sequence.

[0029] The LSTM model employs a single-layer structure with 64 hidden layer nodes. The input sequence length is fixed at 24 time steps, with each time step corresponding to a historical observation from one hour prior; therefore, the input tensor dimension is (24, 6). The output layer is a fully connected layer, directly generating hourly cooling load forecasts for the next 24 hours, in tons of cooling (RT). During model training, mean squared error is used as the loss function, and backpropagation is performed using the Adam optimizer. The initial learning rate is set to 0.001, and an early stopping mechanism is employed to prevent overfitting. During inference, the model receives an updated input feature sequence every 15 minutes, outputting a deterministic 24-hour cooling load forecast curve, which is then uploaded to the central collaborative controller via industrial Ethernet. This forecast curve serves as the basis for the cooling load demand in the subsequent economic scheduling module. The sole basis for this is that it undergoes no post-processing smoothing or correction.

[0030] The economic scheduling optimization solver runs locally on the central collaborative controller and is modeled using a mixed-integer linear programming (MILP) framework. Its objective function is to minimize the total operating cost over 24 hours, expressed as follows:

[0031]

[0032] in For the first Hourly power purchase capacity of the power grid For the corresponding electricity price at that time, The electrical power of the chiller unit during operation. This is the energy efficiency penalty factor. This factor is used to quantify the additional energy consumption cost of a refrigeration unit under non-optimal operating conditions. Its value range is determined according to the performance curve provided by the equipment manufacturer, and is usually between 0.8 and 1.2.

[0033] The constraints of the optimization model include upper and lower limits for the cold storage tank capacity, cold load supply and demand balance constraints, and equipment start-up and shutdown logic constraints. The cold storage tank at any given time... Storage cold energy satisfy The total cooling capacity of the system equals the predicted cooling load demand, i.e. ,in To provide cooling capacity for the chiller unit The total cooling capacity is comprised of the cooling output from the cold storage tank and must be exactly equal to the predicted cooling load demand. That is, satisfying:

[0034]

[0035] To further enhance the physical consistency of the model, an electricity price elasticity coefficient is introduced. With thermodynamic efficiency correction factor Electricity price elasticity coefficient Used to quantify the adjustable potential of user-side cooling load under different electricity price levels, its value is calibrated offline based on the electricity price-load response relationship in historical operating data, and is usually stored in the controller database as a piecewise constant. Thermodynamic efficiency correction factor This factor is used to correct the actual coefficient of performance (COP) of the refrigeration unit under non-rated operating conditions. It is a function of the cooling water inlet temperature, evaporator load rate, and compressor speed. In engineering implementation, The data is obtained online through table lookup or piecewise linear interpolation, and its basic data comes from the factory performance test report of the chiller unit.

[0036] The economic dispatch solver receives updated cooling load forecast curves and real-time electricity price signals every 15 minutes, completes rolling optimization calculations within 5 minutes, and distributes the optimal equipment start-up and shutdown plan, cold storage strategy, and chiller load allocation scheme to the execution layer. This solution process uses the commercial solver Gurobi or the open-source solver CBC, supporting hot start to accelerate convergence.

[0037] The coordinated control logic between the refrigeration unit and the cold storage device is implemented by a dynamic decision matrix built into the central coordinated controller. This decision matrix takes the current cold storage capacity reserve rate of the cold storage tank and the predicted load demand level as input variables, and outputs corresponding equipment action commands. The cold storage capacity reserve rate is defined as the ratio of the current cold storage capacity to the maximum cold storage capacity; its real-time value is calculated from the temperature difference between the inlet and outlet of the cold storage tank and the integral of the flow rate. The load demand level is divided into three categories: peak, medium load, and low load. The determination is based on the deviation of the predicted cooling load from the historical average for the same period: if the deviation is greater than 20%, it is considered peak; if the deviation is between -10% and 20%, it is considered medium load; and if the deviation is less than -10%, it is considered low load.

[0038] When the cooling capacity reserve rate is below 30% and the predicted load is during peak hours, the system starts the standby chiller units and simultaneously opens the cooling storage tank's discharge valve to ensure cooling reliability. When the cooling capacity reserve rate is between 30% and 70% and the load is at a medium load level, the main chiller units operate in variable frequency mode, while the cooling storage tank provides auxiliary cooling capacity compensation, achieving efficient operation under partial load. When the cooling capacity reserve rate is above 70% and it is during a low electricity price period, the system stops all chiller units and relies solely on the cooling storage tank for cooling. If renewable energy access is available, a photovoltaic-driven cooling storage circulation pump is started to supplement cooling storage. This decision-making logic is executed periodically by the central coordination controller every 15 minutes to ensure that the scheduling strategy and system status are updated synchronously.

[0039] The hydraulic characteristics modeling and control of the cooling water system are performed by the hydraulic regulation submodule in the central coordinating controller. This system includes primary-side cooling water pumps and secondary-side chilled water pumps, both equipped with variable frequency drives. The hydraulic regulation submodule operates based on the measured values ​​of the supply and return water pressure difference. With set value The deviation is used to calculate the pump drive frequency command. Its control law is:

[0040]

[0041] in , This is the measured value of the supply and return water pressure difference. Set the differential pressure value. and The proportional gain and integral gain are pre-tuned. The control law is implemented in discretized form, and the integral term is approximated using the trapezoidal rule. and The values ​​are pre-tuned based on the pipeline impedance characteristics and pump performance curves, and stored in the controller parameter library. Differential pressure setpoint. It is not a fixed constant, but rather dynamically adjusted based on the current cooling load level and the cooling rate released from the chilled water storage tank. Specifically, when the cooling rate released from the chilled water storage tank increases, the required chilled water flow rate at the terminal decreases, and the system cools down accordingly. To avoid overpressurization; conversely, when the chiller unit undertakes the main cooling task, The flow rate is appropriately increased to ensure supply. This dynamic adjustment mechanism ensures that the pressure difference between the supply and return water is maintained within the optimal setting range, thereby minimizing pump energy consumption while meeting the flow rate requirements of the terminal cooling load.

[0042] In terms of hardware configuration, the cold storage tank is equipped with no fewer than eight temperature sensors arranged along its height, forming a temperature stratification monitoring array. This array is used to assess the stratification of hot and cold water in real time, calculate the effective cold storage capacity, and identify thermal mixing phenomena. The effective cold storage capacity is calculated based on the integral of the temperature at each measuring point with the enthalpy of the corresponding volume element, using the following formula:

[0043]

[0044] in, The density of water, For specific heat capacity, For the first Layer volume, This is the measured temperature of this layer. The reference temperature is 15°C. This calculation result is used to correct the state estimation of the cold storage tank and improve the scheduling accuracy.

[0045] The chiller unit is equipped with a variable frequency compressor and an electronic expansion valve, supporting continuous adjustment within a load range of 10% to 100%. The compressor speed is determined by the central control unit based on the economic dispatch results. The command is dynamically set, and the opening of the electronic expansion valve is adjusted in a closed loop based on feedback from the evaporator outlet superheat to ensure that the refrigerant flow matches the load. The cooling tower fan also uses frequency conversion control, and its speed is dynamically adjusted according to the difference between the condenser outlet water temperature and the wet-bulb temperature. When this difference is greater than a set threshold, the fan speed increases to enhance heat dissipation; when the difference is less than the threshold, the speed decreases to save fan energy. This control strategy maintains the condensing pressure within the optimal range, improving the overall energy efficiency of the unit.

[0046] To verify the technical effectiveness of this invention, a 30-day field comparative experiment was conducted. The experimental subject was a regional cooling station, equipped with two centrifugal chillers with a rated cooling capacity of 2000 RT, a cold storage tank with an effective capacity of 8000 RT·h, and a supporting cooling water system. During the experiment, the system alternately operated under the control method described in this invention and a traditional rule-based scheduling strategy, with each strategy running continuously for 15 days. Experimental data records included total power consumption, electricity costs, average COP, cooling deviation rate, and the number of equipment start-ups and shutdowns.

[0047] In one specific embodiment, the system operates using the method described in this invention. The LSTM model updates the prediction curve every 15 minutes, the economic scheduling solver continuously optimizes the 24-hour plan, and the hydraulic control system dynamically adjusts the differential pressure setpoint. During periods of low electricity prices (00:00–08:00), the system prioritizes cooling by utilizing the cold storage tanks and initiates the photovoltaic cold storage cycle; during periods of high electricity prices (10:00–14:00 and 18:00–22:00), the system determines whether to activate standby units based on the cold storage reserve rate, while maximizing the utilization of cold storage and release capacity to reduce peak electricity demand.

[0048] In the comparative example, the system adopts a traditional scheduling strategy: the cooling load forecast uses the moving average method, the scheduling decision is based on the fixed electricity price threshold to trigger the charging and discharging of cold storage, the water pump differential pressure is set to a constant value, and the start-up and shutdown of the chiller units follow a simple load percentage rule (e.g., if the load is >70%, the second unit is started).

[0049] The experimental results are summarized in the table below:

[0050] index Example (of the present invention) Comparative Analysis (Traditional Strategy) Total power consumption (MWh) 182.4 215.7 Electricity expenses (yuan) 138,620 169,450 Average COP 5.82 5.14 Cooling deviation rate (%) 2.1 4.7 Chiller start-up and shutdown times 28 45

[0051] Data shows that, while ensuring cooling accuracy, this invention significantly reduces energy consumption and operating costs, improves system energy efficiency, and reduces mechanical wear caused by frequent equipment start-ups and shutdowns.

[0052] Furthermore, this invention supports deep integration with renewable energy systems. When a site is equipped with a photovoltaic power generation system, the central coordinating controller can receive photovoltaic power prediction signals and, during periods of low electricity prices and sufficient sunlight, prioritize scheduling photovoltaic power to drive cold storage circulation pumps for supplementary cold storage, achieving coordinated optimization of on-site green electricity consumption and cold energy storage. This function is achieved by extending the objective function of the economic dispatch model and introducing a photovoltaic curtailment penalty term, the weighting coefficient of which can be dynamically adjusted according to policy guidance.

[0053] At the software implementation level, the LSTM prediction module is deployed in TensorFlow Lite format to ensure low-latency inference on edge computing devices; the economic scheduling solver is encapsulated as a dynamic link library (DLL) for use by the central coordinating controller's main program; and the hydraulic control logic is coded in the form of a state machine to ensure basic differential pressure stability under abnormal operating conditions such as communication interruptions. All software modules have passed IEC 61131-3 standard functional safety certification, meeting the reliability requirements of industrial control systems.

[0054] In summary, this invention organically integrates high-precision LSTM load forecasting, dynamic electricity price response mechanisms, equipment coupled energy consumption modeling, and hydraulic system optimization control to construct a comprehensive cooling load scheduling and control system with strong physical constraint embedding capabilities and multi-source information fusion characteristics. During operation, this system consistently uses predicted cooling load as the scheduling benchmark, real-time electricity price as the economic guide, actual equipment energy consumption characteristics as the constraint boundary, and hydraulic stability as the guarantee condition, thereby achieving globally optimal control of the district cooling system in complex and ever-changing operating environments. Those skilled in the art can make adaptive adjustments to specific parameters, hardware selection, or control cycles based on the above embodiments without departing from the core ideas of this invention; all such adjustments should be considered within the scope of protection of this invention.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A comprehensive cooling load scheduling and control method based on LSTM load forecasting, characterized in that, The steps include: constructing a single-layer long short-term memory neural network (LSTM) model, using ambient temperature, relative humidity, solar radiation intensity, historical 24-hour cooling load sequence, date type identifier, and current real-time electricity price as six-dimensional input features. The input sequence length is 24 time steps, the number of hidden layer nodes is 64, and the output is the hourly cooling load prediction value for the next 24 hours. Based on the predicted cooling load and real-time electricity price signal, an economic dispatch optimization model is established with the objective of minimizing the total 24-hour operating cost. The objective function is: · in For the first Hourly power purchase capacity of the power grid For the corresponding electricity price at that time, The electrical power of the chiller unit during operation. Energy efficiency penalty coefficient; Thermodynamic efficiency correction factor is embedded in the economic scheduling optimization model. This factor is a function of cooling water inlet temperature, evaporator load rate and compressor speed, and is used to correct the actual energy efficiency ratio of the chiller under non-rated operating conditions; Based on the current cold storage capacity reserve rate of the cold storage tank and the predicted load demand level, a dynamic decision matrix is ​​used to generate coordinated control commands for the refrigeration unit and the cold storage device. Variable frequency drive control is implemented for the primary cooling water pump and the secondary chilled water pump in the cooling water system. The pump drive frequency command is calculated by proportional-integral control law based on the deviation between the measured value of the supply and return water pressure difference and the dynamic set value.

2. The integrated cooling load scheduling and control method according to claim 1, characterized in that, The input features of the LSTM model are standardized before entering the network, and the standardized parameters are updated daily based on historical data within the past 30-day rolling window; the date type identifier uses one-hot encoding to represent three categories: weekdays, weekends, and statutory holidays; when the real-time electricity price signal is missing, it is filled with the previous valid value.

3. The integrated cooling load scheduling and control method according to claim 1, characterized in that, The constraints of the economic scheduling optimization model include: the cold storage capacity stored in the cold storage tank. satisfy The total cooling capacity of the system equals the predicted cooling load demand, i.e. ,in To provide cooling capacity for the chiller unit This refers to the amount of cold released from the cold storage tank.

4. The integrated cooling load scheduling and control method according to claim 1, characterized in that, The economic dispatch optimization model further introduces the electricity price elasticity coefficient. This coefficient is calibrated offline based on the electricity price-load response relationship in historical operating data and is used to quantify the adjustable potential of user-side cooling load under different electricity price levels.

5. The integrated cooling load scheduling and control method according to claim 1, characterized in that, In the dynamic decision matrix, the cold storage reserve rate is defined as the ratio of the current cold storage capacity to the maximum cold storage capacity. The predicted load demand is divided into three categories: peak, medium load, and off-peak, based on the deviation between the predicted cooling load and the historical average for the same period. When the cooling capacity reserve rate is less than 30% and the predicted load is during the peak period, the standby chiller unit is started and the cooling storage tank's cooling valve is opened. When the cooling capacity reserve rate is between 30% and 70% and the load is medium load, the main chiller unit operates in variable frequency mode and the cooling storage tank provides auxiliary cooling. When the cooling capacity reserve rate is higher than 70% and the electricity price is during the off-peak period, all chiller units are shut down and only the cooling storage tank provides cooling.

6. The integrated cooling load scheduling and control method according to claim 5, characterized in that, During periods of low electricity prices and when renewable energy access is available, photovoltaic-driven cold storage circulation pumps are activated to supplement cold storage.

7. The integrated cooling load scheduling and control method according to claim 1, characterized in that, The water pump drive frequency command By control law The calculation shows that, among which , This is the measured value of the supply and return water pressure difference. Set the differential pressure value. and For pre-tuned proportional gain and integral gain.

8. The integrated cooling load scheduling and control method according to claim 7, characterized in that, The differential pressure set value The system is dynamically adjusted based on the current cooling load level and the cooling output ratio of the cold storage tank: when the cooling output ratio of the cold storage tank increases, the system is reduced. When the chiller unit undertakes the main cooling task, improve .

9. The integrated cooling load scheduling and control method according to claim 1, characterized in that, The cold storage tank is equipped with no fewer than eight temperature sensors along its height, forming a temperature stratified monitoring array for calculating the effective cold storage capacity. ,in For the density of water, For specific heat capacity, For the first Layer volume, This is the measured temperature of this layer. This is a reference temperature.

10. The integrated cooling load scheduling and control method according to claim 1, characterized in that, The chiller unit is equipped with a variable frequency compressor and an electronic expansion valve, supporting continuous adjustment from 10% to 100% load range; the cooling tower fan adopts variable frequency control, and its speed is dynamically adjusted according to the difference between the condenser outlet water temperature and the wet bulb temperature to maintain the optimal condensing pressure.

Images