Energy-saving control system of magnetic suspension flash-type low-temperature refrigeration system

Abstract

The application discloses a kind of magnetic suspension flash evaporation low-temperature refrigeration system energy-saving control systems, it is related to industrial low-temperature refrigeration and energy-saving control technical field, the constrained model predictive control is combined with economizer liquid level scheduling, disturbance observation feedforward, multi-time scale stratification in the present application, the collaborative optimal regulation of key execution quantity is realized under nonlinear, time-varying and multi-disturbance concurrent scene: fast layer level is carried out real-time response around compressor speed and liquid level target, slow layer level is carried out steady-state optimization to high-pressure side target pressure and inlet guide vane angle, so that temperature and pressure are kept smooth transition when load step and environmental mutation appear simultaneously;Economizer management module is coupled with permission logic with intermediate air supply, so that energy efficiency is improved and safety boundary is uniformly constrained;Disturbance observer estimates equivalent disturbance, and modifies prediction model in feedforward mode, reduces the command lag and overshoot caused by model mismatch.

Description

Technical Field

[0001] This invention relates to the field of industrial cryogenic refrigeration and energy-saving control technology, and in particular to an energy-saving control system for a magnetic levitation flash cryogenic refrigeration system. Background Technology

[0002] In the field of industrial cryogenic refrigeration, magnetic levitation flash refrigeration systems are widely used in chemical, pharmaceutical and food processing and other scenarios that require stable low-temperature environments. Such systems face complex external disturbances, such as sudden changes in ambient temperature due to seasonal changes or production scheduling, as well as step fluctuations in internal process loads. These disturbances cause the system to operate under nonlinear and time-varying conditions for a long time, requiring high temperature control accuracy and energy efficiency stability.

[0003] To improve system robustness, some existing technologies employ model predictive control and adaptive strategies for multivariate coordination. For example, by identifying the dynamic characteristics of the system online, the compressor speed and flash economizer level setpoints are adjusted in real time. Alternatively, a feedforward mechanism is introduced to pre-correct high-pressure control and guide vane opening based on predicted signals of ambient temperature and load changes. Some other solutions combine multivariate decoupled control with disturbance observers to suppress oscillations caused by parameter coupling.

[0004] However, when dealing with multiple disturbance coupling, the performance of model predictive control depends on an accurate system model, and model mismatch is prone to occur under strongly nonlinear conditions, resulting in control command lag or overshoot. Although the adaptive strategy can track slowly changing parameters, it is insufficient in response to sudden disturbances and is difficult to maintain stability when sudden changes in ambient temperature and load steps occur simultaneously. In addition, the multivariable control does not fully consider the change in the dynamic coupling strength between the flash system and the compressor with the operating conditions, resulting in a decrease in the adaptability of control parameters under partial load. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] This invention provides an energy-saving control system for a magnetic levitation flash cryogenic refrigeration system, which solves the problems of existing control systems being unable to resist multiple disturbance couplings and having insufficient energy efficiency and operational stability under nonlinear time-varying conditions.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This invention provides an energy-saving control system for a magnetic levitation flash cryogenic refrigeration system, comprising:

[0009] The magnetic levitation centrifugal compressor features variable frequency drive and an intermediate air inlet;

[0010] The flash evaporation economizer is equipped with a liquid level sensor and a liquid level control valve, and is connected to the intermediate air supply port.

[0011] The heat exchange circuit includes the evaporator and condenser side equipment;

[0012] Multiple sensors are used to collect at least one or more of the following data: ambient temperature, evaporator / condenser side pressure and temperature, refrigerant or secondary side flow rate, flash economizer level, vibration and electrical power;

[0013] The controller includes a processor and a memory. When the instructions stored in the memory are executed by the processor, the controller causes the controller to:

[0014] a) Generate control sequences based on sensor data and constrained model predictive control;

[0015] b) Apply the control sequence to at least one actuator: compressor speed, the level control valve, and a condenser-side actuator for achieving the target pressure setpoint on the high-pressure side;

[0016] c) Apply safety constraints to the control sequence, which include at least one of the following: upper and lower limits of liquid level, backflow margin, exhaust temperature, or upper limit of motor current;

[0017] When the compressor is equipped with inlet guide vanes, the controller applies the control sequence to the angle of the inlet guide vanes.

[0018] As a preferred embodiment of the energy-saving control system of the magnetic levitation flash cryogenic refrigeration system described in this invention, the controller adopts a layered multi-time-scale strategy: the first time scale is used to quickly adjust the compressor speed and the target set value of the flash economizer liquid level, and the second time scale is used to adjust the target set value of the high-pressure side and the angle of the inlet guide vanes.

[0019] As a preferred embodiment of the energy-saving control system for the magnetic levitation flash cryogenic refrigeration system described in this invention, the controller is equipped with an economizer management module.

[0020] The intermediate gas replenishment permit shall be issued when the target liquid level setting value and the upper and lower limits of the liquid level are met.

[0021] The target liquid level setting is adjusted based on the load, condenser side conditions, and evaporator side superheat.

[0022] As a preferred embodiment of the energy-saving control system of the magnetic levitation flash cryogenic refrigeration system described in this invention, the controller includes a disturbance observer for estimating the equivalent disturbance caused by sudden changes in ambient temperature and step changes in process load, and superimposing the estimation result onto the state prediction and constraint correction of the model predictive control using feedforward terms.

[0023] As a preferred embodiment of the energy-saving control system of the magnetic levitation flash cryogenic refrigeration system described in this invention, the controller includes an online parameter adjustment module, which is used to update the weights, prediction time domain or constraint margin of the model predictive control online based on the operating performance index under the condition of satisfying the preset safety boundary and convergence criterion, and to fall back to the conservative parameter set when communication is abnormal or out of bounds.

[0024] As a preferred embodiment of the energy-saving control system for the magnetic levitation flash cryogenic refrigeration system described in this invention, the system includes multiple magnetic levitation centrifugal compressors, and the controller employs distributed consensus optimization.

[0025] Each unit acts as an intelligent agent, exchanging load and marginal energy consumption information through a communication network, collaboratively determining its own rotational speed and inlet guide vane angle, and dynamically allocating the injection steam mass flow rate or valve opening of the flash economizer.

[0026] As a preferred embodiment of the energy-saving control system of the magnetic levitation flash cryogenic refrigeration system described in this invention, the controller performs trajectory planning and tracking control during system startup or shutdown. The trajectory planning generates a smooth reference sequence of pressure, temperature and rotational speed, and the tracking control enables the actual variables to track the reference sequence under limited slope and constraint conditions.

[0027] As a preferred embodiment of the energy-saving control system of the magnetic levitation flash cryogenic refrigeration system described in this invention, the controller includes a multivariable decoupling and gain scheduling module, which is used to estimate the coupling strength between the compressor speed, intermediate pressure and liquid level online based on the condenser side and load conditions, and adjust the controller gain or local model parameters accordingly.

[0028] As a preferred embodiment of the energy-saving control system of the magnetic levitation flash cryogenic refrigeration system described in this invention, the controller adjusts the control sequence based on the real-time diagnostic results of the vibration sensor and the magnetic bearing status, and executes a load reduction and intermediate gas replenishment switching strategy when the diagnostic results trigger a preset threshold.

[0029] As a preferred embodiment of the energy-saving control system of the magnetic levitation flash cryogenic refrigeration system described in this invention, the controller is deployed on an edge computing device, which communicates with the cloud platform to receive model or parameter updates within a non-real-time window, and the updates take effect after the stability verification is satisfied; the edge side maintains local control and recording when communication is interrupted or parameter verification fails.

[0030] The beneficial effects of this invention are as follows: By combining constrained model predictive control with economizer level scheduling, disturbance observation feedforward, and multi-timescale hierarchical management, this invention achieves coordinated optimal adjustment of key execution variables in nonlinear, time-varying, and multi-disturbance concurrent scenarios. At the fast level, it provides real-time responses around compressor speed and level targets; at the slow level, it performs steady-state optimization of the high-pressure side target pressure and inlet guide vane angle, thus maintaining a smooth transition in temperature and pressure even when load jumps and environmental abrupt changes occur simultaneously. The economizer management module couples intermediate gas injection with level targets and permissive logic, unifying the constraints on energy efficiency improvement and safety boundaries (anti-surge margin, upper and lower limits of level, exhaust temperature, and upper limit of motor current). The disturbance observer estimates equivalent disturbances and corrects the prediction model using a feedforward approach, reducing command lag and overshoot caused by model mismatch. The online parameter adjustment module updates weights and the prediction time domain under safety boundaries and convergence criteria, taking into account both steady-state energy consumption and dynamic quality. In multi-machine scenarios, distributed consensus optimization is based on marginal energy consumption and disturbance estimation for steam injection and load allocation, reducing internal friction in group control and preventing individual machines from approaching unsafe zones. Start-up / shutdown trajectory planning and vibration diagnosis further improve equipment lifespan and operational reliability; the edge-cloud architecture enables models and parameters to be updated after verification, ensuring that control strategies remain adaptable and energy-efficient over the long term. Attached Figure Description

[0031] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the scope of this application.

[0032] Figure 1 This is a schematic diagram of the energy-saving control system of the magnetic levitation flash cryogenic refrigeration system in an embodiment of the present invention.

[0033] The components include: 1. Magnetic levitation centrifugal compressor; 2. Gas-liquid separator; 3. Evaporator; 4. Flash economizer; 5. Condensation side equipment; 6. Controller; 7. Oil separator; 101. Intermediate air supply port; 401. Liquid level sensor; 402. Liquid level control valve. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0035] All terms used in this application (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0036] For example, the terms “first” and “second” used in this application are only used to distinguish and describe similar objects, to differentiate the first object from another object, and are not used to describe a specific order or sequence, nor should they be interpreted as indicating or implying relative importance.

[0037] This application proposes an energy-saving control system for a magnetic levitation flash cryogenic refrigeration system, combined with... Figure 1 As shown, the system includes a magnetic levitation centrifugal compressor 1, a gas-liquid separator 2, a heat exchange circuit, and a controller 6. The gas-liquid separator 2 is connected between the magnetic levitation centrifugal compressor 1 and the heat exchange circuit. The heat exchange circuit includes an evaporator 3 and a condenser-side device 5. The magnetic levitation centrifugal compressor 1 has a variable frequency drive and an intermediate air supply port 101. The flash economizer 4 is equipped with a liquid level sensor 401 and a liquid level control valve 402, and is connected to the intermediate air supply port 101.

[0038] It also includes multiple sensors for collecting at least one or more of the following data: ambient temperature, evaporator / condenser side pressure and temperature, refrigerant or secondary side flow rate, flash economizer 4 liquid level, vibration, and electrical power; the sensor for collecting the liquid level of flash economizer 4 is the liquid level sensor 401. The controller 6 includes a processor and a memory. When the instructions stored in the memory are executed by the processor, the controller 6:

[0039] a) Generate control sequences based on sensor data and constrained model predictive control;

[0040] b) Apply the control sequence to at least one actuator: compressor speed, level control valve 402, and high-pressure side target pressure setpoint;

[0041] c) Apply safety constraints to the control sequence, including at least one of the following: upper and lower limits of liquid level, backflow margin, exhaust temperature, or upper limit of motor current.

[0042] When the compressor is equipped with inlet guide vanes, the controller 6 further applies the control sequence to the angle of the inlet guide vanes.

[0043] In this embodiment, the engineering specifications for environmental and process measurements are as follows: ambient temperature is obtained from the cooling tower return air inlet or the fresh air sampling point outside the machine room; evaporation and condensation side pressures and temperatures are measured at points near the heat exchanger and calibrated at a single point; secondary side flow is provided by pipeline electromagnetic or ultrasonic meters; liquid level is provided by flash tank differential pressure or guided wave radar; vibration is measured from the compressor housing or the built-in monitoring channel of the magnetic bearing; and electrical power is provided by the unit or compressor-side power metering unit. All sensor signals are entered into controller 6 after being aligned with a unified timestamp and deburred. The default control sampling period is set to 100 milliseconds, which can be adjusted within 50 to 200 milliseconds, determined based on communication latency and actuator response testing. The execution side includes the control words corresponding to the variable frequency speed command, liquid level valve opening target, and condensation side target pressure. Command issuance adopts a two-level handshake of pre-write limiting and post-write confirmation. The confirmation timeout threshold is 500 milliseconds by default and adjustable from 200 to 1000 milliseconds, set based on the manufacturer's interface specifications and the measured values ​​of the field loop latency. Optionally, missing or out-of-bounds single-point measurements should be kept for no more than 2 seconds based on the most recent valid value and marked in the record for subsequent auditing.

[0044] In one embodiment, the magnetic levitation centrifugal compressor 1, evaporator 3, and condenser-side equipment 5 form a refrigeration heat exchange loop through pipelines. An oil separator 7 is provided between the magnetic levitation centrifugal compressor 1 and the condenser-side equipment 5. The refrigerant output from the magnetic levitation centrifugal compressor 1 is first separated by the oil separator 7, then enters the condenser-side equipment 5, and the oil returns to the magnetic levitation centrifugal compressor 1. A flash economizer 4 is installed on the liquid pipe, with its liquid phase outlet connected to the inlet of the evaporator 3 and its gas phase outlet connected to the intermediate gas injection port 101 of the magnetic levitation centrifugal compressor 1. A liquid level sensor 401 and a liquid level control valve 402 are installed on the economizer. The controller 6 is electrically connected via signal lines to sensors such as pressure, temperature, flow rate, liquid level, vibration, and electrical power on the magnetic levitation centrifugal compressor 1, the liquid level control valve 402, and the heat exchange loop, and is used to execute the aforementioned model predictive control and safety constraint strategies. The gas-liquid separator 2 is a conventional component, and its structure and operation are well known in the art. It does not limit the control method and energy-saving effect of the present invention. Other valves, pipe fittings and auxiliary equipment not shown in the figure can also be configured in a conventional manner according to specific engineering requirements.

[0045] In one embodiment, the controller 6 employs a layered multi-timescale strategy: the first timescale is used to rapidly adjust the compressor speed and the target liquid level setpoint of the flash economizer 4, and the second timescale is used to adjust the target pressure setpoint on the high-pressure side and (when configured) the inlet guide vane angle.

[0046] Specifically, the time scales of the fast and slow layers are decoupled: the fast layer's update cycle is 100 milliseconds by default and adjustable from 50 to 200 milliseconds, used for speed and liquid level targets; the slow layer's update cycle is 10 seconds by default and adjustable from 5 to 30 seconds, used for high-pressure side target pressure and inlet guide vane angle. Inter-layer linkage is triggered by a threshold: the slow layer is only allowed to issue a new steady-state target when there are no over-limit events in the fast layer within the past 3 seconds and the critical constraint deviation is continuously less than the set threshold; the threshold is 2% of the rated range by default and can be tuned from 1% to 5%, determined based on the steady-state deviation statistics of the start-up and step experiments. Optionally, to avoid frequent switching, the minimum interval between two adjacent target changes in the slow layer is no less than 30 seconds.

[0047] In one embodiment, controller 6 includes an economizer management module: it issues intermediate gas replenishment permission when the target liquid level setting and upper / lower liquid level limits are met; and it schedules the target liquid level setting based on load, condenser-side operating conditions, and evaporator-side superheat to reduce overall power consumption. For example, the load is estimated from the secondary-side flow rate and inlet / outlet water temperature, the evaporator-side superheat is calculated from the evaporator outlet dryness criterion, and the condenser-side target pressure comes from the upper-layer strategy or an empirical curve corrected according to ambient temperature; the three quantities are filtered by a first-order low-pass filter and a median filter to form a smooth input. The filter time constant is 1 second by default and adjustable from 0.5 to 3 seconds, set according to the measurement point noise and actuator inertia test results. To ensure consistency, the update frequency of the liquid level target is consistent with that of the fast layer, and a maximum change rate limit is applied to adjacent updates, which by default does not exceed 5% of the range each time and can be tuned from 2% to 10%.

[0048] The scheduling method for the target liquid level setpoint is as follows:

[0049] Step C1 uses three types of operating conditions to drive scheduling: target pressure on the condenser side, superheat on the evaporator side, and load flow on the secondary side. The goal is to moderately raise the liquid level target to reduce overall power consumption when the load and target pressure on the condenser side increase, and to lower the liquid level target to maintain permissible safety when the superheat is low.

[0050] Step C2: Within the control cycle, the target liquid level is obtained by linear superposition and interval saturation.

[0051] ,

[0052] in, For the target liquid level, the range ratio is set to 0-1. , This refers to the lower / upper limit of the liquid level, the range ratio. Reference liquid level, range ratio The weights for condenser side, load, and superheat are dimensionless. The three operating condition normalized indicators are dimensionless and range from 0 to 1. Similarly, the weights are assigned according to the principle of safety first, followed by energy saving: it is recommended to set the weights of the condenser side and load to a medium level initially, and the superheat weight to a higher level to protect the safety of the evaporator side. The initial settings can be determined based on the operating condition range recommended by the manufacturer and the variance contribution rate of the operation statistics over the past week. The upper and lower bounds of the normalized indicators are derived from the available range recorded during on-site commissioning and can be reviewed periodically by season. To avoid bias caused by differences in the range of different models, all inputs involved in scheduling must be calibrated at one or two points before being put into operation. The calibration error is recommended not to exceed 1% of the full scale.

[0053] In the formula, the normalization index is defined as:

[0054] ,

[0055] ,

[0056] ,

[0057] in, The target pressure on the condenser side. Set its lower / upper bound for scheduling. Secondary load flow Set its lower / upper bound for scheduling. This refers to the superheat on the evaporator side. In order to achieve excessive hype, This is the lower / upper limit of the superheat safety level; marked with a symbol. For interval saturation operators, This is a amplitude limiting operator.

[0058] Optionally, the interval saturation and amplitude limiting are implemented using a numerical safety strategy: when the input exceeds the defined boundary for more than one update cycle, the target is first fixed at the boundary value, and an alarm count is initiated; the quantization accuracy of the target is not less than one-thousandth of the range, and a rate of change limit is applied before final issuance to suppress jitter. To match the valve resolution, the liquid level target can be discretized according to the valve's minimum recognizable step distance before issuance, and the step distance parameter is determined by the valve actuator's factory data or on-site step test.

[0059] Step C3, assuming the liquid level measurement is... The hysteresis bandwidth is Range ratio; when lie in and At that time, intermediate air replenishment is permitted and can proceed. or When the valve is in an intermediate range, the permission is revoked. After returning to the middle range and meeting the aforementioned conditions, the permission is restored. Furthermore, the permission logic employs a combination of hysteresis and minimum duration: the minimum duration for permission establishment is 3 seconds by default and adjustable from 1 to 5 seconds; the minimum waiting time for re-establishing a permission after revocation is 5 seconds by default and adjustable from 3 to 10 seconds, to avoid frequent start-stop cycles that could reduce energy efficiency and stability. The level valve closed-loop uses a proportional-integral (PI) structure. The proportional gain and integral time are tuned during the commissioning phase through small-amplitude step tests. It is recommended that the steady-state error of a single step be less than 2% of the range, and the integral time be within the range of 1 to 3 seconds. When the valve position approaches the 5% neighborhood of the upper or lower limit, the closed-loop gain is automatically reduced to prevent saturation and stagnation.

[0060] Reference range of values:

[0061] Liquid level and hysteresis: , , , Weight: ,and ;boundary: The pressure can be 0.7-1.1 times the rated condensing pressure. Take the 5%-95th percentile of the operational statistics. , , .

[0062] Specifically, the dispatch law uses the above formula to give the functional relationship between the liquid level target and the operating conditions. It linearly combines three types of normalized indices and subjectes them to interval saturation constraints to form a directly issued liquid level target. When the target pressure and load flow on the condenser side increase, the liquid level target increases accordingly, enhancing the contribution of the economizer and intermediate gas injection to energy efficiency. When the superheat on the evaporator side is lower than expected, the negative term suppresses the liquid level rise, maintaining a safety margin. The permissive rule uses liquid level measurement and hysteresis band to form three intervals, reducing switching jitter and limiting boundary operations. The parameter range is described using range ratio and quantile statistics, facilitating cross-model migration and online tuning. Normalization uses a limiting form to avoid excessive amplification effects of extreme values ​​on dispatch. Dispatch, permissive, and boundary... By coordinating within the same parameter system, the fast layer can respond to operational disturbances in real time, while the slow layer only needs to adjust the target pressure on the high-pressure side and other execution quantities over a wider time scale, resulting in lower overall power consumption and a more stable operating state. In this embodiment, the parameter approval of the liquid level scheduling and permitting strategy is completed through phased trial operation: first, the target change rate and hysteresis width are verified under low load and stable cooling conditions; second, the frequency of perturbation of the permitting switch is verified under the condition of synchronous step change of load and ambient temperature. The statistical caliber adopts a continuous 15-minute window, and the number of permitting switches within the window is defaulted to no more than 3 times as the qualified standard. After approval, the parameters are solidified, and the boundary values ​​are routinely reviewed seasonally within the maintenance window.

[0063] In one embodiment, controller 6 includes a disturbance observer for estimating the equivalent disturbance caused by sudden changes in ambient temperature and a step change in process load, and superimposing the estimation result onto the state prediction and constraint correction of model predictive control using feedforward terms. Specifically, the disturbance observation takes the deviation between the measured and predicted values ​​as input, and uses a low-pass filter to form the equivalent disturbance estimate. The filter time constant is 2 seconds by default and can be adjusted from 1 to 5 seconds. The feedforward superposition has an amplitude limit, with the limit threshold defaulting to 10% of the rated range of the corresponding state variable, and can be tuned from 5% to 15%. When three consecutive unexplained sudden increases in deviation occur within a short period of time, the feedforward channel is paused and recording is triggered, while the main loop constraints remain unchanged to ensure safety.

[0064] In one embodiment, the controller 6 includes an online parameter adjustment module, which is used to update the weights, prediction time domain or constraint margin of the model predictive control online based on the operating performance index, provided that the preset safety boundary and convergence criterion are met, and to fall back to the conservative parameter set when communication is abnormal or out of bounds.

[0065] For example, online parameter adjustments are triggered by the energy efficiency index and constraint violation count within a rolling window. The rolling window is set to 10 minutes by default and can be adjusted from 5 to 30 minutes. The step size for each parameter update is set to be small and gradual, with weight changes not exceeding 20% ​​of the original value, prediction time-domain changes not exceeding 20% ​​of the original value, and constraint margin tightening not exceeding 20% ​​of the original value. The upper limit of the step size is determined based on sensitivity testing during the commissioning period. In the event of communication interruption, convergence failure, or the cumulative number of constraint violations exceeding the threshold, the system immediately switches to preset conservative parameters and archives the context of this adjustment in the log.

[0066] In one embodiment, when the system includes multiple magnetically levitated centrifugal compressors 1, the controller 6 employs distributed consensus optimization:

[0067] Each unit acts as an intelligent agent, exchanging load and marginal energy consumption information through a communication network, collaboratively determining its own speed and (when configured) inlet guide vane angle, and dynamically allocating the injection steam mass flow rate or valve opening of the flash economizer 4 to minimize the overall power consumption of the system.

[0068] Optionally, multi-machine communication adopts a fixed-period time synchronization and timeout rejection strategy: the time synchronization error is not more than 200 milliseconds by default and can be adjusted to 500 milliseconds; the uplink and neighbor communication timeout threshold is 1 second by default and can be adjusted from 0.5 to 2 seconds; when the communication timeout of a single unit exceeds two cycles, the unit temporarily exits the consensus iteration and runs according to the local minimum power consumption strategy, while retaining the boundary check of system-level constraints to avoid adverse effects on other units.

[0069] Distributed consensus optimization steps include:

[0070] Step D1: Treat each magnetic levitation centrifugal compressor 1 as an intelligent agent and conduct neighborhood communication through a directed network. In each round of communication, exchange two types of quantities: marginal power consumption estimate and disturbance estimate. These are used to coordinate the allocation of rotational speed, inlet guide vane angle and steam injection mass flow rate, so as to reduce the overall power consumption of the system and maintain the anti-surging margin and liquid level safety boundary.

[0071] Similarly, to ensure the feasibility of consensus convergence, the neighborhood set is given by the network topology configuration file. It is recommended to use sparse connections to reduce the load. The number of neighborhoods for a single unit should be no less than one and no more than half of the total number. When neighborhood information is missing, local optimization still uses the most recent valid consensus result as a reference and automatically restores to the overall consensus goal after the next round of communication is successful.

[0072] Step D2, using total power consumption combined with soft constraint penalties as the objective:

[0073]

[0074] in, For the overall goal, For unit indexing, For the number of units, For the first Taiwanese generator unit power model For rotational speed, The angle of the inlet guide vane (which is fixed as a constant in the constraints when this component is not present). For the first Steam injection mass flow rate Set the target pressure value for the high-pressure side. As weight, For the liquid level target, This is the measured liquid level value. For the first Estimated local equivalent disturbances in Taiwan (derived from disturbance observers). The total steam injection reference given by the upper-level dispatcher. For the first Taiwan cooling capacity, For load demand.

[0075] In this embodiment, the power model of a single unit is obtained by fitting the performance chart and commissioning data provided by the manufacturer. The fitted sample covers at least three load ranges: low, medium, and high, with a sample size of no less than one hundred valid points. The cooling capacity is estimated based on the secondary side flow rate and the temperature difference between the outlet and return water, taking into account sensor correction factors. The load demand is given by the upper-level system or the data center group control; when not connected to the upper-level system, it is converted according to the secondary side temperature control deviation. To ensure the robustness of the target calculation, the fitted model is periodically re-evaluated within the maintenance window.

[0076] Step D3: Define the feasible region for each unit and add an upper limit to the inter-unit coupling:

[0077]

[0078] in, For the first Taiwan feasible domain, Lower / upper speed, The lower / upper angle of the guide vane. This is the upper limit for steam injection. For the sake of apnea margin, As the lower limit of the margin, This is the motor current. Its upper limit, The exhaust temperature, Its upper limit, Lower / upper limit of liquid level This represents the upper limit of the total steam injection capacity; without inlet guide vanes, To achieve fixation within constraints, The angle value is fixed. Furthermore, the anti-surge margin is calculated based on the relative distance between the compressor's stable operating domain and the real-time operating point collected during commissioning. The distance is expressed using the manufacturer's recommended dimensionless method and verified on-site. The upper limits for exhaust temperature, motor current, and liquid level are derived from equipment technical conditions and safety regulations, with safety regulations taking precedence over energy-saving targets. The constraint assessment update cycle is consistent with the rapid response layer. When a trend approaching the upper limit is detected, soft limiting is triggered in advance. The minimum duration for entering and exiting soft limiting is no less than 3 seconds.

[0079] Step D4: Update local solutions and consistency variables using consensus-ADMM:

[0080]

[0081] in, For the first The decision vector is k, where k is the iteration step. To augment the parameters, Marginal power consumption is defined as follows: , To adjust the weights for perturbation, Z k As a consensus variable, As dual variables, and These are the original and dual residuals, respectively; the convergence criterion is: when Less than the threshold and Less than the threshold or the number of iterations reaches When time stops, the mean of Z can be achieved by neighborhood averaging;

[0082] Optionally, the consensus iteration is initialized using the previous converged solution as the starting point. During the initial run, the local optimal solution of each unit is used, and the consensus variable is set to its simple average. The initial values ​​of the augmented parameters are determined offline based on the target scaling and constraint sensitivity, and are allowed to be adaptively fine-tuned during operation according to the residual size, with each fine-tuning not exceeding 20% ​​of the original value. When the iteration reaches the upper limit but still fails to meet the residual threshold, the results of this round are frozen at the current value and enter the next cycle to avoid computational blockage.

[0083] The weight and threshold ranges, communication load, and update frequency settings are as follows:

[0084] , , , , , The communication load is broadcast in rounds. The control layer updates with necessary neighborhood duals every 100-500ms. Consensus iteration is performed every 1-5 seconds, with a total injection reference. Issued by the upper-level economic management module;

[0085] Specifically, this distributed scheme is based on a single-machine power consumption model. By superimposing liquid level deviation, disturbance robustness, and steam injection and load matching terms into the total cost, collaborative optimization directly affects energy consumption and safety variables. The constraint set covers anti-surge margin, liquid level boundary, and actuator saturation, and limits the total steam injection capacity through inter-group coupling. The consensus-ADMM uses marginal power consumption as the alignment object, combines disturbance correction weights to suppress deviations caused by environmental and load abrupt changes, and the residual threshold and step limit constitute the convergence shutdown criterion. Here, the direction of collaborative optimization is given, and the stopping rule is defined. The communication load only includes two categories: marginal power consumption and disturbance estimation. Key-based data structures are simple and easy to average among edge devices, enabling multi-machine collaboration under limited bandwidth. For example, to reduce communication load, edge power consumption and disturbance estimates can be quantized at fixed points and sent only when the changes exceed a set dead zone. The dead zone width is 1% of the range by default and can be tuned from 0.5% to 2%. At the same time, to prevent a unit from being in extreme conditions for a long time, an equalization counter is set to count the relative load percentage of each unit. The counting window is 30 minutes by default. When the percentage of any unit continues to exceed 70%, its allocation coefficient will automatically decrease by one level in the next round of consensus until the percentage falls back to the target range.

[0086] In one embodiment, controller 6 performs trajectory planning and tracking control during system startup or shutdown. Trajectory planning generates a smooth reference sequence of pressure, temperature, and speed, while tracking control ensures that the actual variables track the reference sequence under constrained slope and conditions. In this embodiment, the reference sequence during startup employs a combination of segmented ramps and platforms. The upper limits of the pressure and temperature slopes are set based on the allowable thermal stress of the equipment and valve capacity. By default, pressure does not exceed 2% of the rated range per second, temperature does not exceed 1 Kelvin per second, and speed does not exceed 5% of the rated value per second. During shutdown, the process is executed in reverse order, and the circulating pump is only allowed to stop when the evaporator-side temperature approaches ambient temperature. If any critical constraint approaches within 5% of the upper limit during tracking, the reference sequence is frozen and a conservative mode is entered until the deviation is resolved.

[0087] In one embodiment, controller 6 includes a multivariable decoupling and gain scheduling module, used to estimate the coupling strength between compressor speed, intermediate pressure, and liquid level online based on the condenser side and load conditions, and adjust the gain or local model parameters of controller 6 accordingly. Similarly, the online identification of multivariable decoupling and gain scheduling adopts a combination of windowed small disturbance probing and natural condition observation. The window length is 300 seconds by default and adjustable from 120 to 900 seconds. Gain update is triggered only when the estimated change in coupling strength exceeds a preset threshold. The threshold is recommended to be set to 20% of the average of the last three windows. To prevent frequent switching, the minimum interval for gain update is not less than 60 seconds, and the adjustment range is limited to ±15% of the original value each time.

[0088] In one embodiment, controller 6 adjusts the control sequence based on real-time diagnostic results from vibration sensors and magnetic bearing status to suppress equipment oscillations and maintain anti-surge margin. When the diagnostic results trigger a preset threshold, a load reduction and intermediate air replenishment switching strategy is executed. Furthermore, the vibration and magnetic bearing diagnostics employ multi-channel comprehensive criteria, including root mean square, peak factor, and frequency band energy ratio. The criterion threshold is determined by baseline testing during commissioning; it is recommended to set the alarm threshold to the baseline mean plus three standard deviations. When the alarm persists for more than 3 seconds, load reduction is executed; when it exceeds 10 seconds, intermediate air replenishment is canceled and the upper speed limit is lowered. After the criteria recover to below the baseline mean plus one standard deviation and remain below this level for 30 seconds, the original settings are gradually restored. In the event of a missing measurement or sensor self-test failure, the load reduction strategy takes effect immediately and the event is recorded.

[0089] In one embodiment, controller 6 is deployed on an edge computing device. The edge computing device communicates with the cloud platform to receive model or parameter updates within a non-real-time window. The updates take effect after stability verification is satisfied. The edge side maintains local control and recording when communication is interrupted or parameter verification fails. Optionally, model or parameter updates between the edge and the cloud are performed during a preset maintenance period. The update package includes a version number and verification information. After transmission, a 10-minute dry run verification is performed in the simulation sandbox. After passing the verification, the system switches to online use. If verification fails or an anomaly occurs after switching, the system rolls back to the previous stable version within 1 second and prohibits automatic retry until manual confirmation. The minimum retention period for records is 30 days by default and can be adjusted to 180 days for tracing control behaviors and events.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0091] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of this application and form different embodiments. For example, all the embodiments above can be used in any combination. The information disclosed in this background section is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

Claims

1. An energy-saving control system for a magnetic levitation flash cryogenic refrigeration system, characterized in that, include: The magnetic levitation centrifugal compressor features variable frequency drive and an intermediate air inlet; The flash evaporation economizer is equipped with a liquid level sensor and a liquid level control valve, and is connected to the intermediate air supply port. The heat exchange circuit includes the evaporator and condenser side equipment; Multiple sensors are used to collect at least one or more of the following data: ambient temperature, evaporator / condenser side pressure and temperature, refrigerant or secondary side flow rate, flash economizer level, vibration and electrical power; The controller includes a processor and a memory. When the instructions stored in the memory are executed by the processor, the controller causes the controller to: a) Generate control sequences based on sensor data and constrained model predictive control; b) Apply the control sequence to at least one actuator: compressor speed, the level control valve, and a condenser-side actuator for achieving the target pressure setpoint on the high-pressure side; c) Apply safety constraints to the control sequence, which include at least one of the following: upper and lower limits of liquid level, backflow margin, exhaust temperature, or upper limit of motor current; When the compressor is equipped with inlet guide vanes, the controller applies the control sequence to the angle of the inlet guide vanes. The controller adopts a hierarchical multi-timescale strategy: the first timescale is used to quickly adjust the compressor speed and the target set value of the flash economizer liquid level, and the second timescale is used to adjust the target pressure set value on the high-pressure side and the inlet guide vane angle. The controller includes a disturbance observer for estimating the equivalent disturbances caused by sudden changes in ambient temperature and step changes in process load, and superimposing the estimation results onto the state prediction and constraint correction of the model predictive control using feedforward terms. The controller includes an online parameter adjustment module, which is used to update the weights, prediction time domain or constraint margin of the model predictive control online based on the operating performance index, under the condition of satisfying the preset safety boundary and convergence criterion, and to fall back to the conservative parameter set when communication is abnormal or out of bounds.

2. The energy-saving control system for a magnetic levitation flash cryogenic refrigeration system as described in claim 1, characterized in that, The controller is equipped with an economizer management module: The intermediate gas replenishment permit shall be issued when the target liquid level setting value and the upper and lower limits of the liquid level are met. The target liquid level setting is adjusted based on the load, condenser side conditions, and evaporator side superheat.

3. The energy-saving control system for a magnetic levitation flash cryogenic refrigeration system as described in claim 1, characterized in that, The system includes multiple magnetically levitated centrifugal compressors, and the controller employs distributed consensus optimization. Each unit acts as an intelligent agent, exchanging load and marginal energy consumption information through a communication network, collaboratively determining its own rotational speed and inlet guide vane angle, and dynamically allocating the injection steam mass flow rate or valve opening of the flash economizer.

4. The energy-saving control system for a magnetic levitation flash cryogenic refrigeration system as described in claim 1, characterized in that, The controller performs trajectory planning and tracking control during system startup or shutdown. The trajectory planning generates a smooth reference sequence of pressure, temperature, and rotational speed, and the tracking control enables the actual variables to track the reference sequence under limited slope and constraints.

5. The energy-saving control system for a magnetic levitation flash cryogenic refrigeration system as described in claim 1, characterized in that, The controller includes a multivariable decoupling and gain scheduling module, which is used to estimate the coupling strength between compressor speed, intermediate pressure and liquid level online based on the condenser side and load conditions, and adjust the controller gain or local model parameters accordingly.

6. The energy-saving control system for a magnetic levitation flash cryogenic refrigeration system as described in claim 1 or 5, characterized in that, The controller adjusts the control sequence based on the real-time diagnostic results of the vibration sensor and the magnetic bearing status. When the diagnostic results trigger a preset threshold, it executes a strategy of switching between load reduction and intermediate air replenishment.

7. The energy-saving control system for a magnetic levitation flash cryogenic refrigeration system as described in claim 1, characterized in that, The controller is deployed on an edge computing device, which communicates with the cloud platform to receive model or parameter updates within a non-real-time window. The updates take effect after stability verification is satisfied. The edge side maintains local control and recording when communication is interrupted or parameter verification fails.

Images