A multi-mode cold source coordination regulation method and system for water storage, free cooling, AI prediction and fire pool multipurpose

Abstract

The application relates to the technical field of central air conditioning energy-saving control and cold source scheduling, and discloses a multi-mode cold source collaborative regulation method and system fusing water storage, free cooling, AI prediction and fire pool multipurpose, which comprises the following steps: collecting the vertical temperature fitting thickness of the rectangular fire pool multipurpose water body, regulating and controlling the circulating pump to limit the flow rate and reduce the frequency to inhibit when the thickness exceeds the standard; collecting meteorological and load data to calculate the future load through a time sequence prediction network and combining the state of charge to solve the scheduling sequence; implementing multi-mode smooth switching in a prediction control cycle rolling optimization mode; and implementing differential pressure pre-balancing, constant pressure switching and heat exchange anti-freezing regulation in a low-temperature topology coupling mode. The application effectively inhibits irregular water body shear mixing to maintain stable gravity stratification, eliminates heterogeneous pipe network grid-connected hydraulic impact, realizes natural cold energy cross-period active energy transfer and looks ahead to collaborative cooling supply.

Description

Technical Field

[0001] This invention belongs to the field of central air conditioning energy-saving control and cold source scheduling technology, and in particular relates to a multi-mode cold source coordinated control method and system that integrates water storage cooling, free cooling, AI prediction and fire water tank reuse. Background Technology

[0002] In the cooling systems of large industrial buildings and regional energy stations, cooling towers are typically used to obtain natural cold sources from the low-temperature outdoor environment. Water storage devices are used to store cold energy during off-peak electricity hours and release it during peak electricity hours, thereby achieving energy-saving cooling and peak shaving. As the energy efficiency requirements of cooling systems continue to increase, combining natural cold sources, mechanical refrigeration and water storage for multi-mode operation is one way to improve the energy efficiency of composite cold source systems.

[0003] In cooling scenarios where both low-temperature process terminal loads and conventional air conditioning loads exist simultaneously, existing control schemes still exhibit significant operational stability issues when reusing existing water storage facilities and switching cooling circuits at different temperature levels: 1. Irregular cross-section water bodies are prone to local shear mixing at dead corners of the pool walls due to water flow disturbances caused by load changes, increasing the thickness of the sloping temperature layer, reducing the quality of cold storage, and making it difficult to simultaneously ensure both minimum safe water storage and stable stratified cold storage; 2. When directly switching between the free cold intermediate temperature circuit and the low-temperature chilled water circuit, water hammer is easily generated due to the difference in the constant pressure reference of the two pipe networks, and water pumps may experience cavitation due to negative pressure; under severe cold conditions, excessively low primary-side inlet water temperature of the heat exchanger can also easily cause ice blockage in the heat exchange channels; 3. The cold storage water body has a large thermal inertia and exhibits a lag in response to meteorological changes, making discrete control based solely on single-point temperature thresholds insufficient. Control is difficult to adapt to changes in cooling load in advance, which easily leads to the unutilization of surplus natural cooling energy at night, while the cooling energy stored during peak daytime electricity hours is depleted prematurely. In addition to the stratification limitations brought about by the physical structure of the water storage structure, existing control strategies still have shortcomings in water flow suppression and dynamic scheduling mechanisms. For example, Chinese invention patent application CN106969449A discloses a water storage system and its usage method for reducing and utilizing the thermocline. It attempts to configure multiple electric valves in the pipeline and rely on single-point temperature detection at the inlet and outlet water pipes to adjust the valve opening to mix the water temperature. However, this type of scheme is based on the ideal assumption of steady-state flow in a regular water tank. It cannot perceive the dynamic evolution of the longitudinal temperature gradient inside the irregular pool and lacks a dynamic suppression mechanism for the flow state and shear disturbance at the water distribution orifice. It is difficult to block the diffusion and degradation of the thermocline caused by the dead corners of the irregular water body in time.

[0004] Therefore, the technical problem to be solved by this invention is how to maintain the stability of the hot and cold water stratification interface while preserving the emergency water storage function of existing irregular water storage structures, eliminate the hydraulic imbalance and freezing risks of cooling pipe networks of different temperature levels during topology switching, and make advance arrangements for cooling in conjunction with meteorological changes so that each cold source can operate smoothly and in coordination. Summary of the Invention

[0005] To address the problems in the background art, the technical solution of the present invention is as follows: A multi-mode cold source coordinated control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse, comprising the following steps: Step S101: Collect vertical temperature data along the height direction inside the water storage tank reused from the rectangular fire water tank and fit the thickness of the inclined temperature layer. Adjust the variable frequency storage and release pump to limit the outflow velocity of the water distributor to between 0.05m / s and 0.15m / s. When the thickness of the inclined temperature layer is greater than 0.5m, reduce the operating frequency of the variable frequency storage and release pump by 20% until the thickness of the inclined temperature layer converges to below 0.35m and then restore the original set flow rate. Step S102: Collect meteorological parameters, end-of-pipe chilled water supply and return temperature difference and circulation flow time series data, calculate the hourly cooling load prediction sequence for the next 24 hours by introducing a time series prediction network with attention mechanism, and solve the day-ahead operation mode scheduling sequence by combining the real-time charge state of the water storage tank. Step S103: Model predictive control is rolled to optimize with a control step size of 15 minutes, and a smooth switching is performed between the chiller direct supply mode, water storage mode, water storage release mode, combined cooling mode, natural cold source direct supply mode, natural cold source storage mode, and natural cold source and low temperature system topology coupling mode. Step S104: When the outdoor wet-bulb temperature is not higher than 2℃ and the natural cold source and low-temperature system topological coupling mode is switched, the bypass balancing valve is adjusted to reduce the pressure difference on both sides of the connecting pipeline to within 5kPa, and then the main connecting valve is opened. At the same time, the pipeline water replenishment pressure benchmark is switched to 40m high head, and the opening of the bypass balancing valve is adjusted to maintain the primary side inlet water temperature of the plate heat exchanger not lower than 15℃.

[0006] Preferably, in step S101, the fire-fighting reserve water volume retained at the bottom of the rectangular fire-fighting water tank is not less than 1 / 3 of the total volume of the tank; PT1000 temperature sensors are arranged in a vertical temperature array every 0.5m to 1.0m along the water depth direction inside the water storage tank; the Reynolds number at the outlet of the water distributor is in the laminar flow disturbance suppression range; when the liquid level sensor detects that the liquid level has reached the fire-fighting minimum water storage warning line, the valves of the cold storage and cold release circuits are cut off; when a fire alarm signal is received, the refrigeration circuit is closed and the fire-fighting dedicated direct supply pipeline is fully opened.

[0007] Preferably, in step S102, the time-series prediction network takes the historical time-series feature matrix of the past 72 hours as input. The feature matrix includes outdoor dry-bulb temperature, outdoor wet-bulb temperature, relative humidity, solar radiation intensity, wind speed, calendar time type label, end-point chilled water supply and return water temperature difference, circulation flow rate, and the vertical temperature matrix of the water storage tank. The network extracts long-short-term time correlation features through multi-layer long short-term memory units, adaptively allocates the weight distribution of each historical time step through an attention layer, and outputs the hourly cooling load prediction sequence for the next 24 hours through dimensionality reduction by a fully connected layer. The time-series prediction network performs online incremental fine-tuning every 7 days using newly added running samples to maintain the average absolute percentage error of the prediction at no more than 8%.

[0008] Preferably, in step S103, the model predictive control performs rolling calculations over a 4-hour forward-looking prediction time domain; the measured cooling load at the current moment, the real-time charge state of the water storage tank, the measured thickness of the thermocline layer, and changes in meteorological parameters are used as state feedback inputs to the optimization solver; when the trend of the charge state evolution in the forward-looking prediction time domain indicates that the effective cooling capacity of the water storage tank is insufficient to cover the cooling load demand during peak power periods, a feedforward scheduling command is issued 30 to 60 minutes in advance to smoothly switch to the combined cooling mode or start the chiller unit in advance to increase the operating frequency.

[0009] Preferably, the switching priority between modes is as follows: first, the natural cold source related mode is called; second, the water storage and cooling mode is called; and finally, the chiller unit operation mode is started. When the outdoor wet-bulb temperature is not higher than 5°C, the natural cold source direct supply mode is triggered, and the variable frequency cooling tower group supplies cooling to the terminal through the plate heat exchanger while the chiller unit is shut down. When the outdoor wet-bulb temperature is not higher than 8°C and it is during the off-peak or normal power period of the power grid, the natural cold source storage mode is triggered, and the variable frequency cooling tower group produces chilled water at 4°C to 12°C and injects it into the water storage tank for storage. When it is during the peak power period of the power grid and the charge status of the water storage tank meets the cooling demand, the water storage and cooling mode is triggered, and the cooling pump outputs chilled water at 10.0°C plus or minus 0.5°C through the plate heat exchanger for cooling. When the cooling outlet water temperature is higher than 11.0°C, the chiller unit is started first to establish a stable operating condition before the cooling circuit is cut off.

[0010] Preferably, in step S104, the initial constant pressure reference head of the free cold medium temperature loop is 10m, and the initial constant pressure reference head of the low temperature chilled water loop is 40m. When the outdoor wet-bulb temperature is not higher than 2℃ and the load at the end of the low temperature process is lower than 35% of the rated heat exchange capacity of the plate heat exchanger, the topological coupling mode between the natural cold source and the low temperature system is triggered. The opening of the bypass balancing valve is adjusted so that the absolute value of the pressure difference on both sides of the junction of the two loops drops to within 5kPa. The main connecting valve is opened and the constant pressure point switching device is triggered simultaneously to lock the constant pressure reference of the entire system's water supply to a high head of 40m. In the topological coupling mode between the natural cold source and the low temperature system, the bypass balancing valve performs closed-loop fine adjustment according to the primary side inlet temperature of the plate heat exchanger to maintain the primary side inlet water temperature of the plate heat exchanger in the range of 16.5℃ to 18.2℃, and completely disconnects the power supply of the low temperature magnetic levitation chiller unit.

[0011] Preferably, the frequency control of the variable frequency cooling tower group's fan is as follows: in the direct supply mode of the chiller unit, the water storage mode, and the combined cooling mode, the baseline approximation is calculated based on the polynomial fitting curve of the outdoor wet-bulb temperature, and the dynamic approximation target is determined by combining the real-time load rate interpolation of the system to generate the target cooling water outlet temperature; a stepped upper limit frequency constraint is applied to the fan according to the range of the target cooling water outlet temperature: the upper limit frequency is 30Hz when the target temperature is between 18℃ and 22℃, 40Hz when it is between 22℃ and 25℃, 45Hz when it is between 25℃ and 28℃, and 50Hz when it is not lower than 28℃; when the measured cooling water outlet temperature is higher than 32℃, the fan is forced to operate at full frequency of 50Hz, and when it drops below 30℃, adaptive control is restored; in the natural cold source related mode, the target cooling water outlet temperature is set to the outdoor wet-bulb temperature plus 1.5℃ and limited to the range of 4℃ to 12℃, and the fan frequency is continuously adjusted in the range of 10Hz to 50Hz.

[0012] Preferably, each operating mode follows a "start-up then stop" sequence during the transition: first, open the associated valves and equipment on the target flow path; after monitoring the secondary side outlet water temperature of the plate heat exchanger to reach the set cooling threshold and remain stable for 60 seconds, gradually reduce the load of the original operating unit to the shutdown boundary and shut it down; finally, close the isolation valve of the original flow path. Under the winter natural cold source operating condition, when the temperature difference between the cooling water supply and return water is reduced to within 3°C and the water temperature is below 3°C, the target outlet water temperature of the chilled water is increased in stages and the minimum fan operating frequency is locked.

[0013] Preferably, it also includes grid demand response control steps: when responding to grid peak shaving instructions, prioritize the dispatch of water storage cooling mode or natural cold source direct supply mode to reduce the operating load of chiller units; when responding to grid valley filling instructions, prioritize the dispatch of water storage cooling mode or natural cold source cooling mode to improve the absorption of off-peak electricity; and report the system's adjustable load capacity and the operating status of each mode to the demand response platform in real time.

[0014] A multi-mode cold source collaborative control system integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse is used to realize a multi-mode cold source collaborative control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse. It includes: a water body state sensing and flow rate control unit, a time-series load prediction and day-ahead scheduling unit, a multi-mode rolling optimization control unit, a pipeline topology and constant pressure antifreeze control unit, and a cold source heat exchange unit. The water body state sensing and flow rate control unit is used to collect vertical temperature data along the height direction inside the water storage tank reused from the rectangular fire water tank and fit the thickness of the inclined temperature layer. It controls the variable frequency storage and release pump to limit the outflow velocity of the water distribution to between 0.05 m / s and 0.15 m / s. When the inclined temperature layer thickness is greater than 0.5 m, the operating frequency of the variable frequency storage and release pump is reduced by 20% until the inclined temperature layer thickness converges to below 0.35 m, after which the original set flow rate is restored. The time-series load forecasting and day-ahead scheduling unit is used to collect meteorological parameters, end-point chilled water supply and return temperature difference and circulation flow time-series data. It calculates the hourly cooling load forecast sequence for the next 24 hours by introducing a time-series forecasting network with an attention mechanism, and solves the day-ahead operation mode scheduling sequence by combining the real-time charge state of the water storage tank. The multi-mode rolling optimization control unit is used to perform model predictive control rolling optimization with a control step size of 15 minutes, and to implement smooth switching between the chiller direct supply mode, water storage cold storage mode, water storage cold release mode, combined cooling mode, natural cold source direct supply mode, natural cold source cold storage mode, and natural cold source and low temperature system topology coupling mode. The pipeline topology and constant pressure antifreeze control unit is used to adjust the bypass balancing valve to reduce the pressure difference on both sides of the connecting pipeline to within 5 kPa when the outdoor wet bulb temperature is not higher than 2℃ and the natural cold source and low temperature system topology coupling mode is switched. Then, the main connecting valve is opened, the pipeline water makeup constant pressure reference is switched to 40m high water head, and the opening of the bypass balancing valve is adjusted to maintain the primary side inlet water temperature of the plate heat exchanger not lower than 15℃. The cold source heat exchange unit includes a medium-temperature refrigeration unit, a low-temperature magnetic levitation refrigeration unit, a plate heat exchanger, and a variable frequency cooling tower, which are used to generate cold energy and supply heat in response to control commands issued by the multi-mode rolling optimization control unit.

[0015] Compared with existing technologies, the present invention provides a multi-mode cold source coordinated control method and system that integrates water storage cooling, free cooling, AI prediction, and fire water tank reuse, which has the following advantages: 1. In a multi-mode cold source collaborative control system integrating water storage, free cooling, AI prediction, and fire water tank reuse, by monitoring the vertical temperature changes and the state of the thermocline in the water storage tank and limiting the water flow disturbance during the water distribution process, the stratification of hot and cold water in the irregular cross-section water tank can be kept stable while maintaining the fire safety storage capacity. This avoids significant shear mixing and dead-angle eddies in the tank wall during variable load operation, thereby slowing down the expansion of the thermocline and maintaining the cold storage quality of the water storage tank and the stability of cold water output during the cooling process. At the same time, when fire demand occurs, the cooling operation can be stopped in time and the direct fire supply can be restored. This allows the existing fire water tank to retain its original emergency water storage and fire water supply functions while being used for cold storage, achieving a balance between the reuse of fire-fighting facilities and cold storage.

[0016] 2. When switching between the free cooling medium-temperature circuit and the low-temperature process chilled water circuit, by reducing the pressure difference between the two sides in advance and unifying the pressure benchmark of the pipeline network, water hammer and negative pressure cavitation of the circulating water pump caused by direct connection of water systems with different pressure conditions can be avoided. This makes the switching between cooling circuits of different temperature levels smoother. At the same time, by adjusting the bypass water volume according to the primary side inlet temperature of the plate heat exchanger, the freezing of the heat exchanger inlet water temperature caused by excessively low temperature under severe cold conditions can be avoided, reducing the resulting decrease in heat exchange capacity and obstruction of pipeline flow. When conditions are met, the low-temperature magnetic levitation chiller unit can be shut down while maintaining stable cooling at the end of the low-temperature process, which helps to reduce the operating time and energy consumption of mechanical refrigeration equipment.

[0017] 3. By combining meteorological changes, terminal cooling load forecasts, and the real-time charge status of water storage tanks, the operation of the cooling source can be arranged in advance. The cooling source can be scheduled and controlled in advance based on future load changes. The model predictive control continuously corrects the cooling source output according to the actual operating status. When the expected stored cooling capacity is insufficient, mechanical refrigeration capacity can be supplemented in advance. This can reduce the supply and demand mismatch caused by the thermal inertia of the cold storage water and control lag, avoid the storage of cold capacity being exhausted too early before the peak electricity period, and reduce the situation where the refrigeration unit is forced to operate at high load for a long time during peak electricity price periods. This makes the coordination between water storage, natural cooling source and mechanical refrigeration more reasonable. Attached Figure Description

[0018] Figure 1 This is a flowchart of the multi-condition cold source rolling optimization and smooth switching method of the present invention; Figure 2 This is an architecture diagram of the functional units of the multi-mode cold source coordinated control system of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] A multi-mode cold source coordinated control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse includes the following steps: Step S101: Collect vertical temperature data along the height direction inside the water storage tank reused from the rectangular fire water tank and fit the thickness of the inclined temperature layer. Adjust the variable frequency storage and release pump to limit the outflow velocity of the water distributor to between 0.05m / s and 0.15m / s. When the thickness of the inclined temperature layer is greater than 0.5m, reduce the operating frequency of the variable frequency storage and release pump by 20% until the thickness of the inclined temperature layer converges to below 0.35m and then restore the original set flow rate. Step S102: Collect meteorological parameters, end-of-pipe chilled water supply and return temperature difference and circulation flow time series data, calculate the hourly cooling load prediction sequence for the next 24 hours by introducing a time series prediction network with attention mechanism, and solve the day-ahead operation mode scheduling sequence by combining the real-time charge state of the water storage tank. Step S103: Model predictive control is rolled to optimize with a control step size of 15 minutes, and a smooth switching is performed between the chiller direct supply mode, water storage mode, water storage release mode, combined cooling mode, natural cold source direct supply mode, natural cold source storage mode, and natural cold source and low temperature system topology coupling mode. Step S104: When the outdoor wet-bulb temperature is not higher than 2℃ and the natural cold source and low-temperature system topological coupling mode is switched, the bypass balancing valve is adjusted to reduce the pressure difference on both sides of the connecting pipeline to within 5kPa, and then the main connecting valve is opened. At the same time, the pipeline water replenishment pressure benchmark is switched to 40m high head, and the opening of the bypass balancing valve is adjusted to maintain the primary side inlet water temperature of the plate heat exchanger not lower than 15℃.

[0021] Preferably, in step S101, the fire-fighting reserve water volume retained at the bottom of the rectangular fire-fighting water tank is not less than 1 / 3 of the total volume of the tank; PT1000 temperature sensors are arranged in a vertical temperature array every 0.5m to 1.0m along the water depth direction inside the water storage tank; the Reynolds number at the outlet of the water distributor is in the laminar flow disturbance suppression range; when the liquid level sensor detects that the liquid level has reached the fire-fighting minimum water storage warning line, the valves of the cold storage and cold release circuits are cut off; when a fire alarm signal is received, the refrigeration circuit is closed and the fire-fighting dedicated direct supply pipeline is fully opened.

[0022] Preferably, in step S102, the time-series prediction network takes the historical time-series feature matrix of the past 72 hours as input. The feature matrix includes outdoor dry-bulb temperature, outdoor wet-bulb temperature, relative humidity, solar radiation intensity, wind speed, calendar time type label, end-point chilled water supply and return water temperature difference, circulation flow rate, and the vertical temperature matrix of the water storage tank. The network extracts long-short-term time correlation features through multi-layer long short-term memory units, adaptively allocates the weight distribution of each historical time step through an attention layer, and outputs the hourly cooling load prediction sequence for the next 24 hours through dimensionality reduction by a fully connected layer. The time-series prediction network performs online incremental fine-tuning every 7 days using newly added running samples to maintain the average absolute percentage error of the prediction at no more than 8%.

[0023] Preferably, in step S103, the model predictive control performs rolling calculations over a 4-hour forward-looking prediction time domain; the measured cooling load at the current moment, the real-time charge state of the water storage tank, the measured thickness of the thermocline layer, and changes in meteorological parameters are used as state feedback inputs to the optimization solver; when the trend of the charge state evolution in the forward-looking prediction time domain indicates that the effective cooling capacity of the water storage tank is insufficient to cover the cooling load demand during peak power periods, a feedforward scheduling command is issued 30 to 60 minutes in advance to smoothly switch to the combined cooling mode or start the chiller unit in advance to increase the operating frequency.

[0024] Preferably, the switching priority between modes is as follows: first, the natural cold source related mode is called; second, the water storage and cooling mode is called; and finally, the chiller unit operation mode is started. When the outdoor wet-bulb temperature is not higher than 5°C, the natural cold source direct supply mode is triggered, and the variable frequency cooling tower group supplies cooling to the terminal through the plate heat exchanger while the chiller unit is shut down. When the outdoor wet-bulb temperature is not higher than 8°C and it is during the off-peak or normal power period of the power grid, the natural cold source storage mode is triggered, and the variable frequency cooling tower group produces chilled water at 4°C to 12°C and injects it into the water storage tank for storage. When it is during the peak power period of the power grid and the charge status of the water storage tank meets the cooling demand, the water storage and cooling mode is triggered, and the cooling pump outputs chilled water at 10.0°C plus or minus 0.5°C through the plate heat exchanger for cooling. When the cooling outlet water temperature is higher than 11.0°C, the chiller unit is started first to establish a stable operating condition before the cooling circuit is cut off.

[0025] Preferably, in step S104, the initial constant pressure reference head of the free cold medium temperature loop is 10m, and the initial constant pressure reference head of the low temperature chilled water loop is 40m. When the outdoor wet-bulb temperature is not higher than 2℃ and the load at the end of the low temperature process is lower than 35% of the rated heat exchange capacity of the plate heat exchanger, the topological coupling mode between the natural cold source and the low temperature system is triggered. The opening of the bypass balancing valve is adjusted so that the absolute value of the pressure difference on both sides of the junction of the two loops drops to within 5kPa. The main connecting valve is opened and the constant pressure point switching device is triggered simultaneously to lock the constant pressure reference of the entire system's water supply to a high head of 40m. In the topological coupling mode between the natural cold source and the low temperature system, the bypass balancing valve performs closed-loop fine adjustment according to the primary side inlet temperature of the plate heat exchanger to maintain the primary side inlet water temperature of the plate heat exchanger in the range of 16.5℃ to 18.2℃, and completely disconnects the power supply of the low temperature magnetic levitation chiller unit.

[0026] Preferably, the frequency control of the variable frequency cooling tower group's fan is as follows: in the direct supply mode of the chiller unit, the water storage mode, and the combined cooling mode, the baseline approximation is calculated based on the polynomial fitting curve of the outdoor wet-bulb temperature, and the dynamic approximation target is determined by combining the real-time load rate interpolation of the system to generate the target cooling water outlet temperature; a stepped upper limit frequency constraint is applied to the fan according to the range of the target cooling water outlet temperature: the upper limit frequency is 30Hz when the target temperature is between 18℃ and 22℃, 40Hz when it is between 22℃ and 25℃, 45Hz when it is between 25℃ and 28℃, and 50Hz when it is not lower than 28℃; when the measured cooling water outlet temperature is higher than 32℃, the fan is forced to operate at full frequency of 50Hz, and when it drops below 30℃, adaptive control is restored; in the natural cold source related mode, the target cooling water outlet temperature is set to the outdoor wet-bulb temperature plus 1.5℃ and limited to the range of 4℃ to 12℃, and the fan frequency is continuously adjusted in the range of 10Hz to 50Hz.

[0027] Preferably, each operating mode follows a "start-up then stop" sequence during the transition: first, open the associated valves and equipment on the target flow path; after monitoring the secondary side outlet water temperature of the plate heat exchanger to reach the set cooling threshold and remain stable for 60 seconds, gradually reduce the load of the original operating unit to the shutdown boundary and shut it down; finally, close the isolation valve of the original flow path. Under the winter natural cold source operating condition, when the temperature difference between the cooling water supply and return water is reduced to within 3°C and the water temperature is below 3°C, the target outlet water temperature of the chilled water is increased in stages and the minimum fan operating frequency is locked.

[0028] Preferably, it also includes grid demand response control steps: when responding to grid peak shaving instructions, prioritize the dispatch of water storage cooling mode or natural cold source direct supply mode to reduce the operating load of chiller units; when responding to grid valley filling instructions, prioritize the dispatch of water storage cooling mode or natural cold source cooling mode to improve the absorption of off-peak electricity; and report the system's adjustable load capacity and the operating status of each mode to the demand response platform in real time.

[0029] A multi-mode cold source collaborative control system integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse is used to realize a multi-mode cold source collaborative control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse. It includes: a water body state sensing and flow rate control unit, a time-series load prediction and day-ahead scheduling unit, a multi-mode rolling optimization control unit, a pipeline topology and constant pressure antifreeze control unit, and a cold source heat exchange unit. The water body state sensing and flow rate control unit is used to collect vertical temperature data along the height direction inside the water storage tank reused from the rectangular fire water tank and fit the thickness of the inclined temperature layer. It controls the variable frequency storage and release pump to limit the outflow velocity of the water distribution to between 0.05 m / s and 0.15 m / s. When the inclined temperature layer thickness is greater than 0.5 m, the operating frequency of the variable frequency storage and release pump is reduced by 20% until the inclined temperature layer thickness converges to below 0.35 m, after which the original set flow rate is restored. The time-series load forecasting and day-ahead scheduling unit is used to collect meteorological parameters, end-point chilled water supply and return temperature difference and circulation flow time-series data. It calculates the hourly cooling load forecast sequence for the next 24 hours by introducing a time-series forecasting network with an attention mechanism, and solves the day-ahead operation mode scheduling sequence by combining the real-time charge state of the water storage tank. The multi-mode rolling optimization control unit is used to perform model predictive control rolling optimization with a control step size of 15 minutes, and to implement smooth switching between the chiller direct supply mode, water storage cold storage mode, water storage cold release mode, combined cooling mode, natural cold source direct supply mode, natural cold source cold storage mode, and natural cold source and low temperature system topology coupling mode. The pipeline topology and constant pressure antifreeze control unit is used to adjust the bypass balancing valve to reduce the pressure difference on both sides of the connecting pipeline to within 5 kPa when the outdoor wet bulb temperature is not higher than 2℃ and the natural cold source and low temperature system topology coupling mode is switched. Then, the main connecting valve is opened, the pipeline water makeup constant pressure reference is switched to 40m high water head, and the opening of the bypass balancing valve is adjusted to maintain the primary side inlet water temperature of the plate heat exchanger not lower than 15℃. The cold source heat exchange unit includes a medium-temperature refrigeration unit, a low-temperature magnetic levitation refrigeration unit, a plate heat exchanger, and a variable frequency cooling tower, which are used to generate cold energy and supply heat in response to control commands issued by the multi-mode rolling optimization control unit.

[0030] Example 1: Currently, a cooling energy station in an industrial park in East China is operating under high temperature and high humidity conditions in summer. The outdoor dry-bulb temperature reaches 36.5℃, and the outdoor wet-bulb temperature... Located in the temperature range of 24.2℃ to 25.1℃, the energy station is equipped with two medium-temperature centrifugal chillers with a rated cooling capacity of 3516kW each, one low-temperature magnetic levitation chiller with a rated cooling capacity of 1407kW, one plate heat exchanger with a rated heat exchange capacity of 2800kW, several variable frequency cooling water pumps and variable frequency chilled water pumps, and four units with a rated circulating water volume of 800m³ / h. 3 A variable frequency cooling tower with a capacity of / h; the energy storage side will have an effective dimension of 30m in length, 15m in width, and 6m in depth, with a total volume of 2700m³. 3 The existing rectangular reinforced concrete fire water tank was converted into a water-cooled storage tank. The outer wall of the tank was covered with a 100mm thick polyurethane insulation layer. The water depth of 2m at the bottom of the tank corresponds to a volume of 900m³. 3 Designated as the minimum fire-fighting reserve water volume and maintained hydraulically isolated, the upper 4m effective cold storage water depth corresponds to an effective cold storage volume of 1800m³. 3 Low-speed radial water distributors are symmetrically installed at the top and bottom of the pool. A layer of PT1000 temperature sensors is arranged every 0.5m along the water depth to form a vertical temperature array, which is used to detect the vertical temperature distribution of the water and fit the thickness of the thermocline. The fire-fighting reserve zone formed by a 2m water depth at the bottom of the pool maintains hydraulic isolation by limiting flow velocity and utilizing the temperature and density distribution of the water, thus forming a dynamic isolation interface. A low-velocity radial water distributor is installed at the bottom boundary of the effective cold storage working zone, 2m below the pool bottom, limiting the outflow velocity of the distributor to the range of 0.05m / s to 0.15m / s. The fluid dynamic head is less than 0.001m head. Utilizing the gravity settling characteristic that the density of water at 4℃ is greater than that at room temperature, a stable buoyancy damping interface is formed above the 2m dead water zone at the bottom. The circulating water flow energy cannot penetrate this gravity damping interface and diffuse downwards, thus preventing the lower 900m... 3 The fire-fighting reserve stagnant water mixes with the momentum of the upper circulating water; the physical intake of the fire-fighting dedicated direct supply pipeline is independently installed at a depth of 0.3m at the bottom of the pool, normally in an electrically isolated and normally closed state, and is completely physically decoupled from the upper cold storage circulation network in the flow path; the vertical temperature sensor array runs through the entire water depth of the pool, continuously collecting the overall water temperature from the top to the bottom of the pool to monitor the stability of the buoyancy self-locking interface; the cooling terminal simultaneously bears the comfort air conditioning load of the workshop and the air conditioning load of the condensation bath process, which has strict requirements for the stability of the chilled water supply temperature. The process terminal requires the supply water temperature to be constant at 7℃ and the transient temperature fluctuation range to be limited to within ±0.2℃. During the daytime peak electricity price period, when the peak production load of the entire plant is superimposed, the large temperature difference and large flow rate of cooling will cause shear mixing in the dead corner of the rectangular pool, and cause the cooling supply and the precise temperature control of the process to be out of balance.

[0031] At 00:00 daily, the time-series prediction network deployed in the AI ​​optimization decision server automatically collects outdoor dry-bulb temperature, outdoor wet-bulb temperature, relative humidity, solar radiation intensity, wind speed, calendar time type label, end-point chilled water supply and return temperature difference, circulation flow rate, and the vertical temperature matrix of the water-cooled storage tank over the past 72 hours. It aligns the data according to the historical time steps corresponding to each data point, constructing a 72-time-step multiplied by a 12-dimensional historical time-series feature matrix. The vertical temperature matrix of the water-cooled storage tank maintains the correspondence between layer number and temperature value according to a fixed order of temperature measurement layers and is synchronously written with the other inputs of the same historical time step. Each historical time step uses the same feature arrangement order. The historical time-series feature matrix is ​​input into the time-series prediction network, which consists of multiple layers of long short-term memory units and an attention mechanism. The long short-term memory units extract dependencies across multiple time scales, the attention layer adaptively distributes the weights of key historical feature time steps, and the fully connected layer reduces the dimensionality to output the hourly cooling load prediction sequence for the next 24 hours. The prediction mean absolute percentage error is kept within 8%. Specifically, a sequence structure consisting of two layers of long short-term memory (LSM) network and a single layer of self-attention is employed. The first LSM network contains 128 hidden neurons, and the second LSM network contains 64 hidden neurons. Both the forget gate and the input gate use the Sigmoid activation function, and the state update uses the Tanh activation function. The self-attention layer calculates the dot product correlation score of each hidden state vector output by the second LSM network, and obtains the dimensionless scalar weight coefficients for each historical time step through a normalized exponential function mapping. The hidden state vectors at each time step are weighted and summed to generate a context state vector representing the key dynamic features of the load. The fully connected layer contains 32 neurons and finally outputs a 24-dimensional hourly load prediction value through linear dimensionality reduction. The network training uses the mean squared error function as the loss metric, and the optimizer uses the adaptive moment estimation algorithm. The initial learning rate is set to 0.001, the batch size is set to 32, and the maximum number of training iterations is set to 100. An early stopping mechanism is triggered if the loss on the validation set does not decrease after 10 consecutive iterations to prevent overfitting. Subsequently, the server is optimized based on the daily operating electricity cost for the next 24 hours. With total carbon emissions Minimization is a dual optimization objective. Combining the peak-valley time-of-use electricity pricing period of the power grid, the available volume constraint of the water-storage cooling pool, the operating load rate range of the chiller units, and the minimum duration constraint of no less than 30 minutes for each mode, a multi-objective particle swarm optimization algorithm is used to obtain the day-ahead operating mode scheduling sequence. The objective function for daily operating electricity cost is the time integral value of the product of the sum of the chiller unit input power, circulating water pump input power, and cooling tower fan input power for each time period within the next 24 hours, multiplied by the corresponding peak-valley electricity price per unit price. The objective function for total carbon emissions is the sum of the total power consumption of the entire system for each time period and the real-time power grid output. The product integral value of the carbon emission factor for power supply; the number of particles in the multi-objective particle swarm optimization algorithm is set to 50, the maximum number of generations is set to 200, the inertia weight in particle velocity update adopts an adaptive linear decreasing mechanism to smoothly decrease from the initial 0.9 to 0.4, and the individual cognitive learning factor and the social group learning factor are both set to 1.5; the algorithm stores the Pareto optimal solutions generated in each generation by maintaining an external archive of non-dominated solutions, and uses a weighted proximity determination method to extract a compromise solution that takes into account both operating cost and carbon emissions from the final Pareto front solution set as the day-ahead operating mode scheduling sequence.

[0032] During the off-peak electricity price period from 00:00 to 07:00, the system switches to the M2 water-based cold storage mode according to the day-ahead operation mode scheduling sequence. This involves starting a single medium-temperature centrifugal chiller unit, the cooling water circulation loop, and the variable frequency cold storage pump. The centrifugal chiller's supply water temperature is set to 4.2℃, and the prepared low-temperature chilled water is injected into the rectangular water-based cold storage tank through a top low-speed radial distributor. During the cold storage process, the edge controller dynamically adjusts the output frequency of the variable frequency cold storage pump based on the measured circulation flow rate, controlling the outflow velocity from the top distributor to 0.08 m / s. Based on this velocity, the distributor orifice characteristic size, and the water's kinematic viscosity, the Reynolds number at the distributor's outflow orifice is calculated to ensure the Reynolds number is within the laminar flow disturbance suppression range. A vertical temperature sensor array is used to fit the vertical temperature gradient distribution curve in real time, detecting the thickness of the oblique temperature layer formed at the hot-cold interface inside the tank. The thickness is stabilized between 0.32m and 0.38m, within the stable thickness threshold of 0.5m, eliminating dead-angle eddies and hydraulic mixing in the rectangular pool wall. As the cooling process progresses, the optimization decision server continuously calculates the real-time state of charge of the water-cooled pool based on the vertical temperature distribution. Its calculation is expressed as: ,in, The total number of temperature measurement layers. For water density, The specific heat capacity of water at constant pressure. For the first The volume of water corresponding to the thermometer layer. The upper limit temperature of the reference return water is set at 11℃. For the current moment, the first The measured temperature value of the temperature sensing layer, where the subscript is... Indicates the temperature measurement layer number, superscript Indicates the current moment. To determine the maximum effective cooling capacity of the water-cooled storage tank, the system performs non-negative truncation filtering on the temperature difference between each temperature measurement layer during the actual calculation process. When the measured temperature of a certain temperature measurement layer is higher than or equal to the upper limit of the reference return water temperature of 11℃, it is determined that the water in that layer has no effective cooling potential, and the calculated temperature difference corresponding to that temperature measurement layer is automatically truncated to zero and does not participate in the negative value accumulation. The controller only performs temperature difference and volume weighted integration on temperature measurement layers with measured temperatures in the effective cooling temperature range of 4.2℃ to 11℃, so that the calculated real-time state of charge always monotonically converges to the range of 0% to 100% under the conditions of full cooling and high-temperature return water disturbance, representing the remaining usable cooling energy of the water. At 06:40 in the morning, the temperature sensor at the bottom of the tank detected that the water temperature dropped to 4.4℃ and the centrifugal chiller unit load rate adaptively dropped to 18.5%, and the system calculated the real-time state of charge. When the energy storage capacity reaches 96.2%, it is determined that the energy storage space is fully filled. The system automatically and smoothly shuts down the centrifugal chiller unit and cuts off the cold storage circuit 20 minutes before the end of off-peak electricity to avoid ineffective cold energy loss and overcharging energy consumption.

[0033] During the peak electricity price and cooling demand period from 09:00 to 11:00 during the daytime, the prediction model shows that the total end-point cooling load climbs to 4200kW to 4850kW, exceeding the rated cooling capacity of 3516kW for a single medium-temperature centrifugal chiller unit. The system initiates predictive control rolling optimization with a control step size of 15 minutes and a look-ahead prediction time domain of 4 hours. Based on the measured load deviation and the temperature stratification status of the water tank, the measured cooling load at the current moment, the real-time state of charge of the water storage tank, the measured thickness of the thermocline layer, and changes in meteorological parameters are used as state feedback inputs to the optimization solver. When the predicted state of charge evolution trend in the time domain indicates that the effective cooling capacity of the water-cooled storage tank is insufficient to cover the cooling load demand during peak power periods, the controller issues a feedforward scheduling command 30 to 60 minutes before the expected time of insufficient cooling capacity. This command either smoothly switches to the combined cooling mode or starts the chiller unit in advance to increase its operating frequency. In this embodiment, the former execution method is adopted, smoothly switching to the M4 combined cooling mode. During the rolling optimization process, the optimization solver performs deductions based on the discrete state transition model of the cold source system, setting the real-time state of charge of the water-cooled storage tank, the thickness of the thermocline layer, and the end-of-process water supply temperature as states. The variables are set as follows: the operating frequency of the variable frequency storage and dissipation chiller pump, the setpoint of the centrifugal chiller outlet water temperature, and the opening degree of the regulating valves in each loop are set as control variables; the hourly cooling load output by the time-series prediction network and the outdoor wet-bulb temperature are set as measurable disturbance variables. Within the 4-hour look-ahead prediction time domain, the optimization solver constructs a comprehensive optimization objective functional that minimizes the sum of the sum of the daily operating electricity cost and the squared penalty terms for the deviation of the process water supply temperature from the target value of 7°C. A weighted smoothing penalty is applied to the variation amplitude of the control variables between adjacent control steps. Simultaneously, the boundary conditions of the chiller unit's 30% to 100% rated load rate and the pump 2 are embedded. The frequency boundary from 0Hz to 50Hz and the valve opening rate of change are constrained; the solver calls the sequential quadratic programming algorithm to perform online iterative optimization of the objective functional within each 15-minute control step, calculates the optimal control variable time series for the next 4 hours, and only issues the control command for the first control step of the time series to the field frequency converter and electric valve actuator. In the next control step, the state is updated and the solution is rolled over again based on the feedback of measured sensor data; the variable frequency storage and discharging cold pump starts and controls the bottom water distributor to draw low-temperature cold water from the bottom of the pool at a flow rate of 0.11 m / s, and then pumps it at a flow rate of 620 m. 3A flow rate of / h is delivered to the primary side of the plate heat exchanger, while the secondary side stably outputs 10℃ chilled water, releasing 2150kW of stored cooling capacity to the terminal pipe network. Simultaneously, a single medium-temperature centrifugal chiller unit is activated to handle the remaining 2700kW cooling load, ensuring that the unit operates stably within its optimal energy efficiency range of 76.8%, replacing the operation mode of activating a second high-power centrifugal chiller unit. Throughout the cooling cycle, the system follows a smooth switching operation, first activating and then deactivating the chilled water. After the cooled water outlet establishes a stable temperature difference through the plate heat exchanger and runs continuously for 60 seconds, the cooling capacity output ratio is locked. Model predictive control monitors the evolution trend of the state of charge in real time through look-ahead prediction, maintaining the chilled water supply temperature at the end of the condensate bath process within the range of 7℃±0.2℃ during peak cooling periods, ensuring the continuous stability of constant temperature parameters for precision production processes.

[0034] Throughout the operation and cold storage process of the chiller unit, the four variable frequency cooling towers adopt dynamic approximation control based on outdoor wet-bulb temperature and real-time system load rate instead of constant outlet water temperature control, optimizing the controller based on outdoor wet-bulb temperature. The polynomial fitting curve is used to calculate the baseline approximation, and combined with the real-time system load rate. Interpolation determines the dynamic approximation target The relationship is as follows: ,in: ,in, This is the baseline approximation obtained by polynomial fitting based on the outdoor wet-bulb temperature when the system load rate is 1. The baseline approximation is obtained by polynomial fitting of the outdoor wet-bulb temperature when the system load rate is 0.5. For the system's real-time load rate, Given the outdoor wet-bulb temperature, the calculated target outlet temperature of the cooling water is: The controller imposes physical limit constraints ranging from 18°C ​​to 32°C, and further maps the fan frequency to segmented frequency-limiting ranges. When the target cooling water outlet temperature is between 25°C and 28°C, the upper limit frequency of the fan is locked at 45Hz; when it is not lower than 28°C, it is released to 50Hz; when the outlet temperature briefly exceeds 32°C, all fans are forcibly triggered to operate at full frequency of 50Hz; and when the temperature drops below 30°C, adaptive optimization resumes. The polynomial fitting coefficients mentioned above are based on a rated circulating water volume of 800m³ / h. 3On a counter-flow variable frequency cooling tower thermal performance calibration test bench, data was obtained through a test condition matrix covering outdoor wet-bulb temperatures from 0℃ to 30℃ and load rates from 50% to 100%. Under constant air-to-water ratio and packing heat transfer characteristics, the optimal heat dissipation approximation discrete data points were measured at 50% and 100% load rates when the sum of the power consumption of the centrifugal chiller compressor and the cooling tower fan reached its minimum value. A quadratic polynomial regression fitting was performed using the least squares method, achieving a goodness-of-fit coefficient of 0.985 and a maximum fitting residual of less than 0.15℃. Through the aforementioned gravity-buoyancy stratified flow velocity suppression, long-short-term memory time-series load look-ahead prediction, day-ahead particle swarm multi-objective scheduling, real-time model predictive control rolling correction, and cooling tower variable frequency adaptive approximation optimization, the system, while fully retaining 900m... 3 Under the premise of dead water level in fire emergency, the irregular rectangular civil engineering water body is kept stable for cold storage, and the cold capacity is allocated according to the day-ahead operation mode scheduling sequence and real-time rolling results during extreme peak load and peak electricity price periods, so as to improve the energy efficiency of the cold source system and maintain a stable cooling state.

[0035] Example 2: The current industrial park cooling energy station test platform is equipped with two medium-temperature centrifugal chillers, each with a rated cooling capacity of 3516kW; one low-temperature magnetic levitation chiller, with a rated cooling capacity of 1407kW; one plate heat exchanger, with a rated heat exchange capacity of 2800kW; and four variable frequency cooling towers, each with a rated circulating water volume of 800m³. 3 / h, the total volume of the water-cooled storage tank is 2700m³. 3 An effective cold storage water depth of 4m corresponds to an effective cold storage volume of 1800m³. 3 A water depth of 2 meters at the bottom of the pool corresponds to a reserved volume of 900 cubic meters. 3 As the minimum fire-fighting reserve water volume; a layer of PT1000 temperature sensors is installed every 0.5m along the water depth direction, with a temperature measurement accuracy of ±0.05℃ and a range covering -20℃ to 60℃; the pipeline is equipped with an electromagnetic flowmeter with a measurement accuracy of ±0.2% and a sampling period set to 1s; the power acquisition unit adopts a 0.5S-class multi-functional microelectromechanical meter. During the test, Gaussian distributed noise with an amplitude of 5% of the standard deviation of the measurement signal is superimposed in the data acquisition channel, and a 50Hz power frequency induced voltage ripple is injected into the temperature measurement cable circuit to simulate electromagnetic disturbances and turbulent fluctuations of pipeline fluid in the industrial environment.

[0036] System overall energy efficiency ratio The calculation is expressed as: ,in, This represents the total cooling load at the terminal. Input power to the chiller unit. This represents the total power of the cooling tower fan. This refers to the total power of the chilled water pumps. This refers to the total power of the cooling water pump. The total power of the variable frequency chiller pump and the outdoor wet-bulb temperature threshold for switching between free cooling modes. The calculation is expressed as: ,in, The baseline value for the end-of-pipe chilled water return temperature is set at 12.5℃. The logarithmic mean design temperature difference for the plate heat exchanger is set to 1.8℃. The aforementioned dynamic approximation target represents the same cooling tower approximation physical quantity. In the calculation of the free cold cut-in threshold, a calibration benchmark value of 2.7℃ is used, resulting in a free cold cut-in threshold benchmark of 8℃ and an effective cold storage capacity of the water-cooled storage tank. The calculation is expressed as: ,in, The density of the water is taken as 999.8 kg / m³. 3 , The isobaric specific heat capacity of water is taken as 4.186 kJ / (kg·℃). For effective cold storage capacity, a value of 1800m³ is chosen. 3 , The final cooling water temperature is set at 11℃. The final water temperature for cold storage is set at 4.2℃.

[0037] Five test conditions were set up under a continuous 48-hour meteorological and terminal cooling load disturbance sequence. The test group used a time-series prediction network and model predictive control for rolling optimization, and arranged the storage and release of cold water in the water storage tank and the variable frequency load distribution of the chiller units according to the day-ahead operation mode scheduling sequence. When the outdoor wet-bulb temperature was not higher than 8°C and it was during the off-peak or peak power period of the power grid, the natural cold source storage mode was activated, and the chilled water produced by the natural cold source was stored in the water storage tank. During the peak power period, the water storage release mode or the combined cooling mode was activated according to the real-time charge status of the water storage tank. In the control group 1, the time-series prediction network was removed, and the storage and release of cold water in the water storage tank were activated according to the fixed tiered electricity price schedule. The unit and plate heat exchanger are stopped, and proportional-integral-derivative (PID) regulation is performed based on the real-time terminal return water temperature. Control group 2 disconnects the water-cooled storage and release circuits, with only the chiller, cooling tower, and plate heat exchanger bearing the terminal load. Control group 3 increases the frequency of the variable frequency storage and release pump, making the flow velocity at the distributor outlet orifice reach 0.22 m / s, serving as a control condition outside the 0.05 m / s to 0.15 m / s flow velocity control range. The corresponding Reynolds number at the distributor outlet orifice is determined based on the orifice velocity, orifice characteristic size, and water kinematic viscosity, deviating it from the laminar flow disturbance suppression range. Control group 4 switches to the free cooling mode and enters the wet-bulb temperature threshold. The temperature is set to 5℃, which is lower than the 8℃ benchmark obtained from the heat transfer temperature difference calculation.

[0038] When the five test groups adopted the same meteorological and load input conditions at the beginning of the sequence, the dry-bulb temperature was 31.5℃, the wet-bulb temperature was 7.6℃, and the hourly average cooling load was 2130kW. The test group processed the continuous meteorological and terminal load data through a time-series prediction network, and the root mean square error of the prediction was 48.2kW. During the storage and release of cold energy, the variable frequency storage and release pump was controlled to keep the Reynolds number at the outlet of the water distributor within the laminar flow disturbance suppression range and the thickness of the thermocline layer was kept at 0.38m. When the natural cold source met the triggering conditions of the natural cold source storage mode, it first stored cold energy in the water storage tank, and then released the corresponding cold energy from the water storage tank during the peak period. The free cold energy bearing ratio was calculated to be 58.4%. After combining the model prediction control rolling optimization allocation of the chiller unit and the water storage tank, the peak power grid power was 142.6kW, and the water supply temperature was kept at 7℃±0.18℃. The system comprehensive energy efficiency ratio was calculated to be 5.82 based on the measured power of each cold source equipment, water pump and cooling tower fan.

[0039] The control group 1 used the same initial meteorological and load inputs, but did not perform time-series load forecasting. The water-cooled storage tank only stored and released cold according to a fixed schedule. During the morning load ramp-up phase, the lack of feedforward load information caused the release process to consume the available cold energy in the water-cooled storage tank in advance. The slope layer thickness was 0.42m, and the free cold energy share was 32.1%. During the midday peak electricity price period, the centrifugal chiller unit increased its operating load to supplement the cold energy. The peak grid power consumption was 418.5kW, the water supply temperature was 7℃±0.65℃, and the system's comprehensive energy efficiency ratio was 4.21.

[0040] The control group 2 retained the time-series prediction process, with a root mean square error of 48.2 kW. However, the cold storage and release loops of the water-cooled storage pool were removed, and the water-cooled storage pool was no longer used for cross-period cold transfer. During continuous meteorological disturbances, the proportion of free cooling was 55.2%, and the peak cooling load was mainly borne by electric compressor chillers. The peak grid power consumption was 785.4 kW, the water supply temperature was 7℃±0.22℃, and the system's comprehensive energy efficiency ratio was 3.95.

[0041] The control group 3 also retained the time-series prediction process, with a root mean square error of 48.2 kW. However, the flow velocity at the outlet of the water distributor was increased to 0.22 m / s. Under this condition, the increased flow velocity caused the Reynolds number at the outlet of the water distributor to deviate from the laminar flow disturbance suppression range, enhancing the shear disturbance inside the water tank. The thickness of the inclined temperature layer increased from 0.38 m in the test group to 1.85 m, and the temperature of the cooled water outlet rose to above 8.6℃. Although the free cooling load was 58%, the chiller unit needed to be prepared in advance. With supplemental cooling, the peak grid power consumption is 365.8kW, the water supply temperature is 7℃±0.82℃, and the system's comprehensive energy efficiency ratio is 4.63. The changes in flow velocity, thermocline thickness, and outlet water temperature corresponding to this set of data indicate that, under the test water tank and water distribution structure, when the outflow velocity of the water distributor is controlled between 0.05m / s and 0.15m / s, and the Reynolds number at the corresponding water distributor outlet orifice is within the laminar flow disturbance suppression range, the thermocline thickness can be controlled within 0.5m.

[0042] The time-series prediction process was retained for control group 4, with a root mean square error of 48.2 kW and a slope layer thickness of 0.39 m. However, the wet-bulb temperature threshold for the free cooling mode was adjusted to 5 ℃. Under the input condition of a wet-bulb temperature of 7.6 ℃, this threshold could not meet the entry conditions for the natural cold source related mode, and the free cooling share was 0%. The peak grid power was 612.3 kW, the water supply temperature was 7 ℃ ± 0.2 ℃, and the system's comprehensive energy efficiency ratio was 4.38. This result corresponds to the 8 ℃ free cooling entry threshold benchmark calculated based on the terminal chilled water return temperature, the logarithmic mean design temperature difference of the plate heat exchanger, and the cooling tower approximation degree.

[0043] During the continuous 48-hour test cycle, the total chilled water supply temperature of the test group was maintained within the range of 7℃±0.18℃, the switching transition time between each operating mode was less than 180s, and the fluctuation of the pipeline hydraulic pressure was limited to within ±0.02MPa. In the above tests, the stratification state of the water storage tank, the proportion of free cooling, the peak power grid power consumption, the water supply temperature, and the overall energy efficiency ratio of the system all corresponded to the changes in the control conditions of each group. This can be used to compare the cold source scheduling status when the fire water tank is reused for cold storage, natural cold source is utilized, and time-series load prediction and model predictive control are used for rolling optimization.

[0044] Example 3: Currently, the outdoor dry-bulb temperature of a cooling energy station in an industrial park in East China has dropped to -3.5℃, and the outdoor wet-bulb temperature... The temperature remains consistently between -1.2℃ and 1.8℃. The energy station is equipped with two medium-temperature centrifugal chillers with a rated cooling capacity of 3516kW each, one low-temperature magnetic levitation chiller with a rated cooling capacity of 1407kW, one water-to-water plate heat exchanger with a rated heat exchange capacity of 2800kW, and a water-cooled storage tank converted from an existing rectangular reinforced concrete fire-fighting water tank. The effective dimensions of the tank are 30m long, 15m wide, and 6m deep, with a 2m water depth at the bottom corresponding to a depth of 900m. 3 As the minimum fire-fighting water reserve, the upper 4m effective cold water storage depth corresponds to 1800m. 3 With an effective cold storage capacity, the condensation bath process workshop is in a low-load production state, and the process terminal continuously supplies 7°C cold water. The process cooling load is 780kW, which is 35% lower than the rated heat exchange capacity of the plate heat exchanger. When switching from the direct supply mode of natural cold source to the topological coupling mode of natural cold source and low temperature system, there is a difference in the static pressure reference of the two sets of pipe networks. At the same time, it is necessary to deal with the low-temperature frost on the primary side of the plate heat exchanger under extremely cold and large temperature difference heat exchange conditions, as well as the degradation of the slope temperature layer boundary caused by the variable flow velocity disturbance of the rectangular water pool.

[0045] A layer of PT1000 temperature sensors is installed every 0.5m along the vertical direction of the water-cooled storage tank to obtain discrete temperature measurement sequences. The edge controller reads the temperature values ​​of each layer with a discrete sampling period of 10 seconds and calculates the vertical temperature gradient between adjacent temperature measurement nodes. ,in, For the first Vertical temperature change rate of the thermometer layer and These are the measured temperature values ​​of adjacent upper and lower temperature measuring layers, respectively. and The vertical installation height coordinates of adjacent upper and lower temperature measurement layers are used as examples. The controller traverses the gradient sequence of each layer, determines the temperature measurement node where the absolute value of the rate of change reaches its maximum value as the center position of the inclined temperature layer, and determines the thickness of the inclined temperature layer by the height difference between the uppermost and lowermost layers where the absolute value of the rate of change of temperature is greater than the critical temperature gradient threshold of 0.8℃ / m. The critical temperature gradient threshold of 0.8℃ / m was determined by the Richardson number dynamic criterion of buoyancy damping and shear mixing in water body, combined with the calibration measurement in the pool. When the vertical temperature change rate is less than 0.8℃ / m, the buoyancy suppression effect generated by the density difference between water layers is insufficient to offset the inflow and outflow disturbances, the Richardson number is lower than the critical stability value of 0.25, and local micro-vortex heat dissipation is generated. When the vertical temperature change rate reaches or exceeds 0.8℃ / m, gravity buoyancy damping dominates, and the shear mixing of fluid micro-clusters is suppressed. During the calibration process, the diffusion decay rate of the oblique temperature layer under different temperature gradients was measured in the range of 0.2℃ / m to 1.5℃ / m with a step size of 0.1℃ / m. Based on this, 0.8℃ / m was taken as the critical boundary threshold between the stable gravity stratification core area and the upper and lower isothermal mixing area.

[0046] In the cold storage and release cycle, the edge controller adjusts the output frequency of the variable frequency cold storage and release pump according to the measured flow rate, limiting the outflow velocity of the top and bottom low-speed radial distributors to between 0.05 m / s and 0.15 m / s. Simultaneously, the Reynolds number at the distributor outflow orifice is calculated based on the actual flow velocity, orifice characteristic dimensions, and water kinematic viscosity. This Reynolds number is used to determine if the distribution falls within the laminar flow disturbance suppression range. At this time, the edge controller triggers the flow rate suppression protection action, causing the operating frequency of the variable frequency storage and dissipation chiller pump to be reduced by 20% in a stepwise manner to reduce inflow shear disturbance; until the thickness of the inclined thermosphere obtained by fitting the vertical temperature measurement sequence meets the requirements. Furthermore, after the extreme value of the temperature change rate recovers steeply, the original set flow rate is restored. In the water storage tank with an effective storage depth of 4m, the volume of the inclined temperature layer transition zone corresponding to the upper limit of 0.5m thickness accounts for 12.5% ​​of the total effective storage volume. After the water pump operating frequency is reduced by 20%, the pipeline circulation flow rate is reduced by 20% simultaneously. The outflow velocity of the water distributor drops from 0.14m / s to 0.11m / s, and the outflow pressure decreases by 36%. The 0.35m recovery threshold corresponds to the inclined temperature layer thickness converging to below 8.8% of the effective water depth. At this time, the measured vertical temperature gradient recovers to a steep distribution, and the rated cooling capacity output requirement is still met after the original set flow rate is restored.

[0047] When the outdoor wet-bulb temperature consistently meets When the load at the end of the low-temperature process is less than 35% of the rated heat exchange capacity of the plate heat exchanger, the system initiates a topology coupling mode conversion between the natural cold source and the low-temperature system. Before the switch, the medium-temperature loop of the natural cold source is at an initial static pressure reference of 10m head, and the low-temperature chilled water loop is at an initial static pressure reference of 40m head. The main connecting valves V1 and V2 remain closed, and the initial pressure difference at the interface between the two sides is 32.4kPa. The edge controller first adjusts the bypass balancing valve V3, and the differential pressure sensor continuously detects the pressure difference between the two sides. After the absolute value of the pressure difference converges to 3.1kPa and stably meets the condition of not exceeding 5kPa, the main connecting valves V1 and V2 are opened, and the constant pressure point switching device is driven to switch the system makeup water constant pressure reference to a 40m high head. The medium-temperature loop of the natural cold source includes a variable frequency cooling tower, a medium-temperature circulating water pump, and the primary side flow path of the plate heat exchanger. The low-temperature chilled water loop includes the end of the low-temperature process, the chilled water circulating pump, and the secondary side flow path of the plate heat exchanger. Two sets of return The main water supply and return water connections are connected in parallel. The main water supply connection is equipped with a main connecting valve V1, and the main return water connection is equipped with a main connecting valve V2. A bypass balancing branch with a bypass balancing valve V3 is connected across both ends of V1 and V2. Differential pressure transmitters are installed on both sides of the connection section. The constant pressure switching device includes a bidirectional constant pressure electromagnetic shut-off valve group, a diaphragm-type air pressure stabilizing tank, and a variable frequency water supply pump group, which are respectively connected in parallel on the medium temperature circuit return water main and the low temperature circuit return water main. After receiving the constant pressure switching command, the controller hardwires to shut off the electromagnetic shut-off valve of the 10m low constant pressure branch and simultaneously opens the electromagnetic shut-off valve of the 40m high constant pressure branch. The variable frequency water supply pump adjusts its speed according to the feedback signal of the high constant pressure point pressure sensor to raise the water supply pressure benchmark to 40m head. After the pipeline pressure stabilizes, the controller adjusts the compressor speed of the low temperature magnetic levitation chiller in stages until it stops and the power is cut off. The plate heat exchanger then takes over the 780kW process cooling load.

[0048] During operation in the natural cold source and low-temperature system topological coupling mode, the bypass balancing valve V3 continuously adjusts the return water mixing flow rate based on the feedback signal from the plate heat exchanger primary side inlet temperature sensor, maintaining the primary side inlet water temperature between 16.5℃ and 18.2℃, and keeping it no lower than 15℃. The control target is set at 18℃. The return water mixing branch used for antifreeze regulation is connected to the antifreeze mixing node before the primary side inlet, and the temperature sensor is set at the inlet control position corresponding to this node. The high-temperature return water at the end of the process participates in the inlet antifreeze regulation through this branch. The secondary side process chilled water return water of 12℃ to 14℃ still enters the secondary side of the plate heat exchanger along the low-temperature chilled water loop. In the natural cold source related mode, the target outlet water temperature of the variable frequency cooling tower is determined according to the outdoor wet-bulb temperature plus 1.5℃, and limited to the range of 4℃ to 12℃. Under extremely cold conditions ranging from -1.2℃ to 1.8℃, the target value is constrained by a lower limit of 4℃. The actual low-temperature cooling water produced by the cooling tower can be between 2.5℃ and 4℃ due to environmental disturbances. This low-temperature cooling water is sent to the primary side heat exchange area of ​​the plate heat exchanger as a cold source medium. The bypass balance valve V3 adjusts the temperature of the antifreeze control position by mixing the inlet return water. The secondary side of the plate heat exchanger is supplied with process chilled water return water at 12℃ to 14℃ and outputs process supply water at 7℃. During the heat exchange process, the high-temperature return water at the end of the process provides an antifreeze buffer at the inlet of the primary side. The low-temperature cooling water at 2.5℃ to 4℃ from the cooling tower enters the heat exchange area to participate in heat exchange. The thermodynamically driven temperature difference is maintained between 3℃ and 5℃.

[0049] When the AI ​​optimization decision server extracts the historical time-series feature matrix of the past 72 hours at 00:00 every day, it includes outdoor dry-bulb temperature, outdoor wet-bulb temperature, relative humidity, solar radiation intensity, wind speed, calendar time type label, end-point chilled water supply and return temperature difference, circulation flow rate, and the vertical temperature matrix of the water storage tank. The calendar time type label carries the original calendar time label and weekday identification information. The vertical temperature matrix of the water storage tank retains the temperature data of each layer according to a fixed temperature measurement layer order. Each historical time step uses the same temperature measurement layer order and feature arrangement order to complete time alignment. After zero-mean standardization preprocessing, the historical time-series feature matrix is ​​input into a time-series prediction network composed of multiple long short-term memory units and attention layers. The long short-term memory units extract time-series correlations through a gating mechanism. The attention layer calculates the weight coefficients of each historical time step and generates a context vector. Then, after mapping through a fully connected layer, it outputs the hourly cooling load prediction sequence for the next 24 hours. The time-series prediction network loads the most recent 7 days of measured operating samples at a fixed 7-day cycle, and performs online incremental fine-tuning of the network connection weights. The average absolute percentage error of the prediction is maintained within a range of no more than 8%. During real-time scheduling, the controller performs model predictive control rolling optimization with a control step size of 15 minutes and a look-ahead prediction time domain of 4 hours, incorporating the measured cooling load and the real-time state of charge of the water storage tank. and the thickness of the thermocline As a state feedback input to the solver, the device frequency conversion command and valve opening are corrected based on the rolling solution results for the current step size.

[0050] Example 4: This example combines Figures 1 to 2 This paper describes a multi-mode cold source coordinated control method and system that integrates water storage cooling, free cooling, AI prediction, and fire water tank reuse. Figure 1As shown, in step S101, the system collects vertical temperature data along the height direction inside the water-cooled storage tank reused from the rectangular fire water tank and fits the thickness of the inclined temperature layer. The variable frequency cold storage pump is adjusted to limit the outflow velocity of the water distributor to between 0.05 m / s and 0.15 m / s. When the inclined temperature layer thickness is greater than 0.5 m, the operating frequency of the variable frequency cold storage pump is reduced by 20% until the inclined temperature layer thickness converges to below 0.35 m, after which the original set flow rate is restored. In step S102, the system collects meteorological parameters, the temperature difference between the supply and return chilled water at the terminal, and the time-series data of the circulating flow rate. A time-series prediction network incorporating an attention mechanism is used to calculate the hourly cooling load prediction sequence for the next 24 hours. Combined with the real-time state of charge of the water-cooled storage tank, the day-ahead operation mode scheduling sequence is solved. In step S103, the system performs model predictive control rolling optimization with a control step size of 15 minutes, and performs smooth switching between the chiller direct supply mode, water storage mode, water storage and release mode, combined cooling mode, natural cold source direct supply mode, natural cold source storage mode, and natural cold source and low temperature system topology coupling mode. In step S104, when the outdoor wet-bulb temperature is not higher than 2℃ and the system switches to the natural cold source and low temperature system topology coupling mode, the system adjusts the bypass balancing valve to bring the pressure difference on both sides of the connecting pipeline to within 5kPa before opening the main connecting valve. At the same time, the pipeline water makeup pressure benchmark is switched to a 40m high head, and the opening of the bypass balancing valve is adjusted to maintain the primary side inlet water temperature of the plate heat exchanger not lower than 15℃.

[0051] like Figure 2 As shown, the multi-mode cold source coordinated control system includes a water state perception and flow velocity control unit, a time-series load prediction and day-ahead scheduling unit, a multi-mode rolling optimization control unit, and a pipeline topology and constant pressure antifreeze control unit. The water state perception and flow velocity control unit, based on a rectangular fire water tank reuse structure, collects vertical temperature data and fits the thickness of the inclined thermocline, controlling the outflow state to be within the laminar flow disturbance suppression range. The time-series load prediction and day-ahead scheduling unit inputs a historical time-series feature matrix, extracts long-term and short-term time correlations, adaptively allocates weight distribution, generates and outputs hourly cold load prediction sequences for future periods. The multi-mode rolling optimization control unit performs model predictive control rolling optimization, issues feedforward scheduling commands for forward control, drives the system to perform smooth switching between different operating modes (first-in, then-out), and performs grid demand response peak shaving and valley filling. The pipeline topology and constant pressure antifreeze control unit opens the main connecting valve after the pressure difference in the connected loop converges, performs coordinated switching of the water replenishment constant pressure benchmark, and fine-tunes the opening of the bypass balancing valve during operation to maintain the inlet water temperature not lower than the set temperature value.

[0052] Example 5: During the engineering commissioning and initial calibration of the rectangular fire water tank reuse and renovation cold storage system, the fire water storage capacity and the operating range of the water distributor were first calibrated. The liquid level sensing unit was used to measure the 900m depth corresponding to a 2m water depth at the bottom of the tank. 3The minimum fire-fighting water reserve was calibrated. A drainage test confirmed that when the water level in the tank reached the minimum fire-fighting water reserve warning line corresponding to a 2m water depth, the controller hardwired off the electric valve circuits for both cold storage and cold release, while maintaining the dedicated fire-fighting direct supply pipeline fully open. Subsequently, the outflow calibration of the water distributor was performed in the cold storage working area 4m above the tank. The edge controller drove the variable frequency cold storage pump to operate in 5Hz increments within the 15Hz to 50Hz range. Ultrasonic flow meters and orifice velocity meters detected the outflow parameters of the top and bottom low-speed radial water distributors, with the circulation flow rate at 200m³ / h. 3 / h to 650m 3 When the flow rate varies within the range of / h, the measured flow velocity at the outlet of the distributor increases linearly from 0.05 m / s to 0.14 m / s. The controller calculates the Reynolds number at the outlet of the distributor based on the measured flow velocity, characteristic dimensions of the distributor orifice, and kinematic viscosity of the water. The Reynolds number at the outlet of the distributor is used to determine the flow regime at the orifice. Within the range of the above-mentioned circulating flow rate and outlet velocity, the Reynolds number at the outlet of the distributor is in the laminar flow disturbance suppression range, and the corresponding outlet velocity of the distributor remains within the range of 0.05 m / s to 0.15 m / s. The thickness of the oblique temperature layer corresponding to the extreme value layer of the vertical temperature gradient change rate of the water body is... Maintain below 0.35m.

[0053] During the parameter calibration and operating condition boundary tuning process of the adaptive control algorithm for the variable frequency cooling tower group, the controller is optimized for outdoor wet-bulb temperature. With system real-time load rate A mapping relationship between the approximation degree polynomial function and the wind turbine frequency range is established. Within a test matrix covering outdoor wet-bulb temperatures from -1.2℃ to 28.5℃ and real-time system load rates from 0.5 to 1, the controller calculates the baseline approximation parameters using the established polynomial relationship. and Then, the dynamic approximation target is obtained by interpolation based on the real-time system load rate. and the target outlet temperature of cooling water When the target cooling water outlet temperature is between 18℃ and 22℃, the upper limit frequency of the cooling tower fan is set to 30Hz; when it is between 22℃ and 25℃, the upper limit frequency is set to 40Hz; when it is between 25℃ and 28℃, the upper limit frequency is set to 45Hz; and when it is not lower than 28℃, the upper limit frequency is released to 50Hz. This stepped upper limit frequency is determined based on the cubic relationship between fan air volume and shaft power, as well as the variation law of the chiller unit's energy efficiency ratio with condensing temperature. When the target cooling water outlet temperature is between 18℃ and 22℃, the unit's condensing pressure is in the high-efficiency operating range, and the increase in fan power consumption from increasing the fan frequency from 30Hz to 50Hz exceeds the reduction in power consumption caused by further decreasing the unit's condensing temperature. To reduce compressor power consumption, the upper limit frequency is maintained at 30Hz. When the target temperature enters the ranges of 22℃ to 25℃ and 25℃ to 28℃, the unit's condensing pressure gradually increases. The reduction in compressor power consumption due to increasing fan speed is greater than the corresponding increase in fan power consumption, so the upper limit frequency is increased to 40Hz and 45Hz, respectively. When the target temperature reaches or exceeds 28℃, the upper limit frequency is released to the rated full frequency of 50Hz to meet the heat dissipation requirements of the refrigeration unit under high load and to avoid high-pressure overload tripping. When the measured cooling water temperature exceeds 32℃, the controller forces all cooling tower fans to run at 50Hz. After the cooling water temperature drops below 30℃, the adaptive segmented frequency limiting adjustment is restored.

[0054] Example 6: After the multi-mode cold source coordinated control system is connected to the power grid demand response platform and the virtual power plant dispatch platform, the system establishes a communication connection with the platform and realizes bidirectional adjustment of load power. The edge controller collects the current operating mode and the operating power of each device through the communication gateway with a data refresh cycle of 1 second. Based on this, it calculates the maximum peak-shaving load capacity that can be adjusted downward and the maximum valley-filling load capacity that can be absorbed upward in the current period, and reports the system's adjustable load capacity and the operating status of each mode to the demand response platform in real time. When receiving the power grid peak-shaving command, the control system prioritizes determining the real-time charge status of the water storage tank. Compared with outdoor wet-bulb temperature ;when When the M5 natural cold source direct supply mode is first switched on, the operating load of the medium-temperature centrifugal chiller unit compressor is unloaded in stages; when the real-time charge status of the water storage tank meets the requirements... During peak daytime electricity price periods, dispatch commands drive the switching valve group to switch to M3 water storage cooling mode. When chiller units are needed to share the remaining load, the switching is to M4 combined cooling mode. The mode switching adopts a first-in, then-out sequence, first activating the associated valves and equipment of the target flow path; when the target loop is activated, the cooling tower electric water valve is opened first, the variable frequency cooling tower fan is started after a 15-second delay, and the variable frequency cooling water pump is started after a 30-second delay. Subsequently, the electric water valves on the plate heat exchanger or chiller unit evaporator side are opened, and the variable frequency chilled water pump is started after a 30-second delay; after the chilled water loop circulation flow is established and the supply and return water temperature difference is stable for 60 seconds, the target mode is activated. When a chiller unit is required to participate, the chiller unit compressor is started and loaded. After the target flow path is established for cooling, the target outlet temperature of the chilled water corresponding to the target operating mode is used as the set cooling threshold. The outlet temperature of the secondary side of the plate heat exchanger is monitored. After the set cooling threshold is reached and stabilized for 60 seconds, the load of the original operating unit is gradually reduced to the shutdown boundary and then shut down. Finally, the isolation valve of the original flow path is closed. When a valley filling command is received from the power grid, the control system prioritizes switching to the M2 water storage mode, or switches to the M6 ​​natural cold source storage mode under low wet-bulb temperature conditions at night to increase the amount of electricity consumed during the valley period and transmits the adjustment response status back to the virtual power plant dispatch platform in real time.

[0055] Under extreme cold weather and high-impurity water quality disturbance conditions, the system implements anti-freezing control for the plate heat exchanger and cooling tower packing, as well as abnormal sewage discharge control for the cooling water circulation. When the system operates in winter natural cold source direct supply mode and the outdoor wet-bulb temperature continues to drop, temperature sensors monitor the cooling water supply and return water temperature difference, return water temperature, and inlet water temperature in real time. When the cooling water supply and return water temperature difference narrows to within 3℃ and the cooling water inlet water temperature is below 3℃, the edge controller triggers anti-freezing suppression protection, gradually increasing the target outlet water temperature setpoint of the chilled water to 12℃ in 0.5℃ increments, while simultaneously locking the variable frequency cooling tower fan operating frequency to the minimum operating frequency of 10Hz. z. By reducing the cooling tower airflow to limit evaporative heat dissipation, the secondary side flow channel of the plate heat exchanger and the cooling tower water collection pan are kept in an anti-freezing operating state. The conductivity meter installed on the main cooling water pipeline continuously monitors the salt content of the circulating water at a cycle of 10 seconds. When the measured conductivity of the circulating water reaches the preset discharge threshold of 2000 μS / cm, the controller outputs an analog control signal to open the electric discharge regulating valve and link it with the electric water supply valve to add softened water. When the measured conductivity of the circulating water drops to below 1600 μS / cm, the discharge regulating valve is closed, so that the salt content of the circulating water falls back to the discharge control range and the heat transfer coefficient of the heat exchange surface is maintained within the design reference range.

[0056] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A multi-mode cold source coordinated control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse, characterized in that, Includes the following steps: Step S101: Collect vertical temperature data along the height direction inside the water storage tank reused from the rectangular fire water tank and fit the thickness of the inclined temperature layer. Adjust the variable frequency storage and release pump to limit the outflow velocity of the water distributor to between 0.05m / s and 0.15m / s. When the thickness of the inclined temperature layer is greater than 0.5m, reduce the operating frequency of the variable frequency storage and release pump by 20% until the thickness of the inclined temperature layer converges to below 0.35m and then restore the original set flow rate. Step S102: Collect meteorological parameters, end-of-pipe chilled water supply and return temperature difference and circulation flow time series data, calculate the hourly cooling load prediction sequence for the next 24 hours by introducing a time series prediction network with attention mechanism, and solve the day-ahead operation mode scheduling sequence by combining the real-time charge state of the water storage tank. Step S103: Model predictive control is rolled to optimize with a control step size of 15 minutes, and a smooth switching is performed between the chiller direct supply mode, water storage mode, water storage release mode, combined cooling mode, natural cold source direct supply mode, natural cold source storage mode, and natural cold source and low temperature system topology coupling mode. Step S104: When the outdoor wet-bulb temperature is not higher than 2℃ and the natural cold source and low-temperature system topological coupling mode is switched, the bypass balancing valve is adjusted to reduce the pressure difference on both sides of the connecting pipeline to within 5kPa, and then the main connecting valve is opened. At the same time, the pipeline water replenishment pressure benchmark is switched to 40m high head, and the opening of the bypass balancing valve is adjusted to maintain the primary side inlet water temperature of the plate heat exchanger not lower than 15℃.

2. The multi-mode cold source coordinated control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse as described in claim 1, characterized in that, In step S101, the fire-fighting water reserve at the bottom of the rectangular fire-fighting water tank is not less than 1 / 3 of the total volume of the tank; PT1000 temperature sensors are arranged in a vertical temperature array every 0.5m to 1.0m along the water depth direction inside the water storage tank; the Reynolds number at the outlet of the water distributor is in the laminar flow disturbance suppression range; when the liquid level sensor detects that the liquid level has reached the fire-fighting minimum water storage warning line, the valves of the cold storage and cold release circuits are cut off; when a fire alarm signal is received, the refrigeration circuit is closed and the fire-fighting dedicated direct supply pipeline is fully opened.

3. The multi-mode cold source coordinated control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse as described in claim 1, characterized in that, In step S102, the time-series prediction network takes the historical time-series feature matrix of the past 72 hours as input. The feature matrix includes outdoor dry-bulb temperature, outdoor wet-bulb temperature, relative humidity, solar radiation intensity, wind speed, calendar time type label, end-point chilled water supply and return temperature difference, circulation flow rate, and the vertical temperature matrix of the water storage tank. The network extracts long-short-term time correlation features through multi-layer long short-term memory units, adaptively allocates the weight distribution of each historical time step through an attention layer, and outputs the hourly cooling load prediction sequence for the next 24 hours through dimensionality reduction by a fully connected layer. The time-series prediction network performs online incremental fine-tuning every 7 days using newly added running samples to maintain the average absolute percentage error of the prediction at no more than 8%.

4. The multi-mode cold source coordinated control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse as described in claim 1, characterized in that, In step S103, the model predictive control performs rolling calculations in the forward prediction time domain of 4 hours; the measured cooling load at the current moment, the real-time charge state of the water storage tank, the measured thickness of the thermocline layer, and the changes in meteorological parameters are used as state feedback inputs to the optimization solver; when the evolution trend of the charge state in the forward prediction time domain indicates that the effective cooling capacity of the water storage tank is insufficient to cover the cooling load demand during peak power periods, a feedforward scheduling command is issued 30 to 60 minutes in advance to smoothly switch to the combined cooling mode or start the chiller unit in advance to increase the operating frequency.

5. The multi-mode cold source coordinated control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse as described in claim 1, characterized in that, The switching priority between modes is as follows: natural cold source related modes are called first, followed by water storage and cooling mode, and finally the chiller unit operation mode is started. When the outdoor wet-bulb temperature is not higher than 5℃, the natural cold source direct supply mode is triggered, and the variable frequency cooling tower group supplies cooling to the terminal through the plate heat exchanger while the chiller unit is shut down. When the outdoor wet-bulb temperature is not higher than 8℃ and it is during the off-peak or normal power period of the power grid, the natural cold source storage mode is triggered, and the variable frequency cooling tower group produces chilled water of 4℃ to 12℃ and injects it into the water storage tank for storage. When it is during the peak power period of the power grid and the charge status of the water storage tank meets the cooling demand, the water storage and cooling mode is triggered, and the cooling pump outputs chilled water of 10.0℃ plus or minus 0.5℃ through the plate heat exchanger for cooling. When the cooling outlet water temperature is higher than 11.0℃, the chiller unit is started first to establish a stable operating condition before the cooling circuit is cut off.

6. The multi-mode cold source coordinated control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse as described in claim 1, characterized in that, In step S104, the initial constant pressure reference head of the free cold medium temperature loop is 10m, and the initial constant pressure reference head of the low temperature chilled water loop is 40m. When the outdoor wet-bulb temperature is not higher than 2℃ and the load at the end of the low temperature process is lower than 35% of the rated heat exchange capacity of the plate heat exchanger, the topological coupling mode between the natural cold source and the low temperature system is triggered. The opening of the bypass balancing valve is adjusted so that the absolute value of the pressure difference on both sides of the junction of the two loops drops to within 5kPa. The main connecting valve is opened and the constant pressure point switching device is triggered simultaneously to lock the constant pressure reference of the entire system's water supply to a high head of 40m. In the topological coupling mode between the natural cold source and the low temperature system, the bypass balancing valve performs closed-loop fine adjustment according to the primary side inlet temperature of the plate heat exchanger to maintain the primary side inlet water temperature of the plate heat exchanger in the range of 16.5℃ to 18.2℃, and completely disconnects the power supply of the low temperature magnetic levitation chiller unit.

7. The multi-mode cold source coordinated control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse as described in claim 1, characterized in that, The frequency control of the variable frequency cooling tower fan includes: in the direct supply mode of the chiller unit, the water storage mode, and the combined cooling mode, the baseline approximation is calculated based on the polynomial fitting curve of the outdoor wet-bulb temperature, and the dynamic approximation target is determined by interpolation of the real-time load rate of the system, and the target cooling water outlet temperature is generated; a stepped upper limit frequency constraint is applied to the fan according to the range of the target cooling water outlet temperature: the upper limit frequency is 30Hz when the target temperature is between 18℃ and 22℃, 40Hz when it is between 22℃ and 25℃, 45Hz when it is between 25℃ and 28℃, and 50Hz when it is not lower than 28℃; when the measured cooling water outlet temperature is higher than 32℃, the fan is forced to operate at full frequency of 50Hz, and when it drops below 30℃, adaptive control is restored; in the natural cold source related mode, the target cooling water outlet temperature is set to the outdoor wet-bulb temperature plus 1.5℃ and limited to the range of 4℃ to 12℃, and the fan frequency is continuously adjusted in the range of 10Hz to 50Hz.

8. The multi-mode cold source coordinated control method integrating water storage cooling, free cooling, AI prediction and fire water tank reuse as described in claim 1, characterized in that, During the transition between different operating modes, the following sequence is followed: first, open the associated valves and equipment on the target flow path; after monitoring the secondary side outlet water temperature of the plate heat exchanger to reach the set cooling threshold and remain stable for 60 seconds, gradually reduce the load of the original operating unit to the shutdown boundary and shut it down; finally, close the isolation valve of the original flow path. Under the winter natural cold source operating condition, when the temperature difference between the cooling water supply and return water is reduced to within 3℃ and the water temperature is below 3℃, the target outlet water temperature of the chilled water is increased in stages and the minimum fan operating frequency is locked.

9. The multi-mode cold source coordinated control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse as described in claim 1, characterized in that, It also includes grid demand response control steps: when responding to grid peak shaving instructions, priority is given to dispatching water storage cooling mode or natural cold source direct supply mode to reduce the operating load of chiller units; when responding to grid valley filling instructions, priority is given to dispatching water storage cooling mode or natural cold source cooling mode to improve the absorption of off-peak electricity. The system can report its adjustable load capacity and operating status in various modes to the demand response platform in real time.

10. A multi-mode cold source collaborative control system integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse, used to implement the multi-mode cold source collaborative control method integrating water storage cooling, free cooling, AI prediction, and fire water tank reuse as described in claim 1, characterized in that, include: The system includes a water condition sensing and flow rate control unit, a time-series load prediction and day-ahead scheduling unit, a multi-mode rolling optimization control unit, a pipeline topology and constant pressure antifreeze control unit, and a cold source heat exchange unit. The water condition sensing and flow rate control unit is used to collect vertical temperature data along the height direction inside the water storage tank reused by the rectangular fire water tank and fit the thickness of the inclined temperature layer. It controls the variable frequency storage and release pump to limit the water outflow velocity between 0.05m / s and 0.15m / s. When the thickness of the inclined temperature layer is greater than 0.5m, the operating frequency of the variable frequency storage and release pump is reduced by 20% until the thickness of the inclined temperature layer converges to below 0.35m, and then the original set flow rate is restored. The time-series load forecasting and day-ahead scheduling unit is used to collect meteorological parameters, end-point chilled water supply and return temperature difference and circulation flow time-series data. It calculates the hourly cooling load forecast sequence for the next 24 hours by introducing a time-series forecasting network with an attention mechanism, and solves the day-ahead operation mode scheduling sequence by combining the real-time charge state of the water storage tank. The multi-mode rolling optimization control unit is used to perform model predictive control rolling optimization with a control step size of 15 minutes, and to implement smooth switching between the chiller direct supply mode, water storage cold storage mode, water storage cold release mode, combined cooling mode, natural cold source direct supply mode, natural cold source cold storage mode, and natural cold source and low temperature system topology coupling mode. The pipeline topology and constant pressure antifreeze control unit is used to adjust the bypass balancing valve to reduce the pressure difference on both sides of the connecting pipeline to within 5 kPa when the outdoor wet bulb temperature is not higher than 2℃ and the natural cold source and low temperature system topology coupling mode is switched. Then, the main connecting valve is opened, the pipeline water makeup constant pressure reference is switched to 40m high water head, and the opening of the bypass balancing valve is adjusted to maintain the primary side inlet water temperature of the plate heat exchanger not lower than 15℃. The cold source heat exchange unit includes a medium-temperature refrigeration unit, a low-temperature magnetic levitation refrigeration unit, a plate heat exchanger, and a variable frequency cooling tower, which are used to generate cold energy and supply heat in response to control commands issued by the multi-mode rolling optimization control unit.

Images