Vacuum heat-insulated pipe regulation and control method based on adaptive heat flow prediction algorithm

Abstract

The present application relates to the technical field of vacuum heat insulation pipe, and particularly relates to a vacuum heat insulation pipe regulation and control method based on an adaptive heat flow prediction algorithm, and proposes a vacuum heat insulation pipe regulation and control method based on an adaptive heat flow prediction algorithm: S1, synchronous data such as partition temperature, vacuum degree and flow rate are collected; S2, physical quantities such as partition heat flow density are calculated; S3, a feature vector is constructed and parameters are recursively updated; S4, a heat flow prediction model and a vacuum conduction prediction model are established; S5, heat modal residual and vacuum modal residual are jointly normalized; S6, a target partition is determined through sparse inversion, and partition regulation and control instructions are generated; S7, under constraints, air extraction, air supplement, bypass and heating are cooperatively set; and S8, model parameters are updated online through recursive least squares combined with a forgetting factor. The present application realizes real-time positioning and inhibition of partition abnormalities under non-stop transportation conditions, so that the vacuum degree of the partition interlayer cavity and the temperature of the outer wall of the inner tube are stably maintained in a preset working condition window.

Description

Technical Field

[0001] This invention relates to the field of vacuum insulation pipe technology, and in particular to a method for controlling vacuum insulation pipes based on an adaptive heat flow prediction algorithm. Background Technology

[0002] Vacuum-insulated pipes are widely used for cryogenic media transportation, long-distance heat tracing and insulation in chemical industries, and heat loss control in precision process pipelines. A typical structure consists of an inner and outer pipe arranged coaxially, forming a vacuum insulation layer between them. In engineering, for ease of construction and maintenance, they are often segmented axially to form several maintainable units. However, in actual operation, the vacuum level of the partitioned interlayer cavity slowly deteriorates due to factors such as venting, micro-leakage, and valve aging. The temperature of the outer wall of the inner pipe also slowly drifts with environmental disturbances and changes in operating conditions. The coupling of these two factors leads to increased heat loss and energy consumption. Existing monitoring and maintenance methods commonly used in engineering projects mainly include: static vacuum testing based on "pressure rise rate," intermittent evacuation and gas replenishment, power step tests with heating bands on certain pipe sections, and helium mass spectrometry leak detection under shutdown conditions. The above methods share common limitations: First, measurement methods based on a single mode (vacuum only or temperature / heat flow only) are highly sensitive to environmental and load fluctuations, making them prone to misdiagnosis or missed diagnosis. Second, they lack identifiable active excitation and system identification links, making it difficult to effectively separate cross-zone coupling from local zone anomalies in an online state without interruption, resulting in limited positioning resolution. Third, the execution layer typically drives independent zone extraction valves, zone replenishment valves, zone bypass microfluidic valves, and heating components based on empirical thresholds, lacking a collaborative setting mechanism oriented towards constraints (valve network binary switching, energy budget, rate limit), resulting in delayed and mutually interfering control actions. Fourth, the model parameters drift over time and lack online updates, causing subsequent judgments to rely on fixed thresholds and static experience, resulting in poor cross-seasonal and cross-operating condition transferability.

[0003] In long-distance devices, the above problems are amplified: cross-zone coupling caused by axial heat conduction and radiation normalizes the "single-point anomaly - multi-point response" scenario; the slow changes in vacuum level of the zoned interlayer cavity and temperature of the outer wall of the inner tube of the zone, coupled with short-term disturbances, make it difficult for traditional threshold discrimination and static comparison to guarantee a stable false alarm rate and false negative rate; at the same time, the topology switching of the switchable valve network and the power step of the heating component, without a unified timing and coding strategy, will form a strong correlation background at the measurement end, further deteriorating the identifiability of zone-level positioning. Even if anomalies can be identified in certain periods, there is a lack of systematic methods for generating consistent, feasible, and rapidly converging collaborative settings under the premise of valve position dispersion, Hamming distance-limited switching, and energy and temperature rise constraints.

[0004] In summary, existing technologies generally lack an online methodology chain that connects "identifiable joint excitation - bimodal predictive modeling - joint residual standardization - sparse localization - constrained collaborative setting - online parameter update". This results in fragmented partition-level diagnosis and control, uncontrollable time-varying parameters, and a lack of global consistency and compliance with engineering constraints in the execution actions. It is difficult to maintain the vacuum level of the partitioned interlayer cavity and the temperature of the outer wall of the inner tube of the partitioned section within the preset operating condition window for a long period of time without interrupting the supply.

[0005] Based on this, the core technical problem to be solved in this application is: how to construct an online integrated method for vacuum insulation pipes under conditions of uninterrupted supply, strong coupling, and time-varying operating conditions, so that a zoned heat flow prediction model and a zoned vacuum conduction prediction model can be jointly established through synchronous excitation of a switchable valve network with identifiable characteristics and heating components, and a unified residual standardization can be implemented for dual-modal measurements. Anomalies can be located at the zoned scale in a sparse manner. Then, under the constraints of discrete valve network switching, energy and rate, the zoned evacuation valve, zoned replenishment valve, zoned bypass microfluidic valve and heating components can be collaboratively set. At the same time, the model parameters can be updated online and stability projected. Thus, the vacuum degree of the zoned interlayer cavity and the temperature of the outer wall of the inner tube in the zone can be stably maintained within the preset operating condition window throughout the entire service life, with traceable numerical stability and engineering feasibility. Summary of the Invention

[0006] To overcome the aforementioned technical deficiencies, the present invention aims to provide a vacuum insulation pipe control method based on an adaptive heat flow prediction algorithm. This method involves axially partitioning the vacuum insulation pipe, using a switchable valve network combined with orthogonal binary sequences and synchronous power disturbances for joint excitation, establishing partitioned heat flow prediction models and partitioned vacuum conduction prediction models, and generating joint normalized residuals. After locking the target partition based on sparse inversion, the pumping, gas replenishment, bypass, and heating actuators are collaboratively set under valve network discrete switching, energy, and rate constraints. Online parameter updates are performed using recursive least squares and stability projection, thereby enabling real-time location and suppression of partition anomalies without interrupting operation, and maintaining the vacuum level of the partitioned interlayer cavity and the temperature of the inner tube's outer wall stably within a preset operating window.

[0007] This invention discloses a method for controlling vacuum insulation pipes based on an adaptive heat flow prediction algorithm, comprising the following steps:

[0008] Step S1: Divide the vacuum insulation pipe into N independent control zones along the axial direction. A zoned sandwich cavity is formed between the inner tube and the outer tube of each independent control zone. A monitoring unit and an execution unit are set in each independent control zone. The monitoring unit includes a heat flow sensor, a temperature sensor and a vacuum sensor. The execution unit includes a heating component, a zoned air extraction valve, a zoned air replenishment valve and a zoned bypass microfluidic valve.

[0009] Step S2: Construct a switchable valve network. The switchable valve network is formed by valve groups located at both ends of each independent control zone. During the sampling period, the connectivity of the valve network and the power switch of the heating component are switched according to a predetermined binary coding sequence to form a joint excitation that can be identified by the zone.

[0010] Step S3: Synchronously collect zoned data. The zoned data includes zoned pipe wall heat flow, zoned inner pipe outer wall temperature, zoned outer sleeve inner wall temperature, and zoned interlayer cavity vacuum degree, and generate thermal mode measurement vectors and vacuum mode measurement vectors respectively.

[0011] Step S4: Establish a zoned heat flow prediction model and a zoned vacuum conduction prediction model. The zoned heat flow prediction model uses historical data of joint excitation and thermal mode measurement vectors as input and output mapping, and the zoned vacuum conduction prediction model uses historical data of joint excitation and vacuum mode measurement vectors as input and output mapping.

[0012] Step S5: Generate thermal residual vector and vacuum residual vector based on the predicted output and measured output of the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model, and perform joint normalization processing on the thermal residual vector and vacuum residual vector to obtain joint normalized residual.

[0013] Step S6: Using the joint normalized residual as the observation, solve the sparse leakage vector to obtain the target partition set, and generate the partition control command corresponding to the target partition set;

[0014] Step S7: Send a partition control command to the execution unit of the target partition to coordinate the setting of the partition air extraction valve, partition air replenishment valve, partition bypass microfluidic valve and heating component so that the vacuum degree of the partition interlayer cavity and the temperature of the outer wall of the partition inner tube meet the preset working condition window.

[0015] Step S8: Update the parameters of the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model online, apply stability projection constraints to the updated parameters, and then return to step S2 for repeated execution.

[0016] Preferably, the predetermined binary encoding sequence in step S2 is a Walsh encoding sequence or an equivalent orthogonal binary sequence with a length of 16 to 64. Each bit of the encoding sequence drives the connection state of the switchable valve network and the power switch of the heating component, respectively.

[0017] Preferably, the joint excitation in step S2 further includes applying a linear sweep frequency microfluidic signal to the partitioned bypass microfluidic valve, with a sweep frequency range of 0.05 Hz to 2 Hz, and synchronized with the binary power sequence of the heating element, so as to obtain mutually independent thermal mode response and vacuum mode response in step S3.

[0018] Preferably, the solution for the sparse leakage vector in step S6 satisfies the following equation:

[0019]

[0020] in, A sparse leakage vector of dimension N; The response matrix is ​​obtained from the partitioned heat flow prediction model; The response matrix is ​​obtained by the partitioned vacuum conduction prediction model; The thermal residual vector; The vacuum residual vector; and Positive scalar weights; and Let them represent the 2-norm and the 1-norm, respectively.

[0021] Preferably, the response matrix and The online parameter update employs a recursive least squares method with an adaptive forgetting factor. The forgetting factor is updated in each sampling period according to the following formula and projected onto the closed interval 0.97, 0.995:

[0022]

[0023] in, The current forgetting factor; For a fixed weight constant, satisfying ; It is a scalar function obtained based on the joint normalized residuals; A projection operator that limits the input value to the interval 0.97, 0.995.

[0024] Preferably, the cooperative setting in step S7 is obtained by solving the constrained actuator allocation problem, which satisfies the following equation:

[0025]

[0026] in, It is a control vector that includes the setpoint of the zone vacuum pump speed, the opening degree of the zone extraction valve, the opening degree of the zone replenishment valve, the opening degree of the zone bypass microfluidic valve, and the setpoint of the heating component power. This is the linear mapping matrix from the actuator to the state; The target vector is determined by the sparse leakage vector and the preset operating condition window; It is a positive scalar; and Define the upper and lower bounds of the vector; Let m be the binary selection vector of the switchable valve network.

[0027] Preferably, before executing step S1, an initialization step S0 is included. The initialization step S0 connects a reference capillary segment in parallel to each independent control zone. The equivalent conductivity of the reference capillary is... And under the encoding switch in step S2, the on / off calibration of the reference capillary is performed. Used for the response matrix The column vectors are used for magnitude scaling.

[0028] Preferably, before step S3, time alignment is performed on the partitioned data acquisition data, and the partition delay parameter for time alignment is... Determine using the following formula:

[0029]

[0030] in, In order to delay Next Multi-channel predicted sequences for each partition The corresponding multi-channel measured sequence; Represents the set of real numbers; This indicates the vector in the channel dimension. and The sum of the products of each term.

[0031] Preferably, the joint normalization process in step S5 includes applying two-dimensional adaptive filtering along the time index and partition index to the thermal residual vector and the vacuum residual vector respectively, and normalizing the amplitude using their respective calibration standard deviations. The time window length of the two-dimensional adaptive filtering is 32 to 128, and the partition window length is 3 to 7.

[0032] Preferably, the constrained actuator allocation problem allows only one switching of the binary selection vector s of the switchable valve network with a Hamming distance of 1 in each sampling period, and the opening degree of the partitioned bypass microfluidic valve is discretized into a four-level set {0, 0.25, 0.5, 0.75}, and the power setting value of the heating component is discretized into a four-level set {0, 0.2, 0.5, 0.8} (unit is relative rated value).

[0033] Compared with existing technologies, the above technical solution has the following advantages:

[0034] 1. Improve the accuracy of zoned heat flux and vacuum level prediction. Existing technologies for heat flux prediction and vacuum level monitoring typically rely on a single mode (e.g., pressure or temperature monitoring alone), which cannot effectively address system disturbances or anomalies and is prone to misdiagnosis or missed diagnosis. The lack of a joint prediction model for system coupling results in low prediction accuracy for zoned states. This invention establishes a joint zoned heat flux prediction model and a zoned vacuum conduction prediction model, respectively, and performs joint normalization and residual standardization based on real-time data from the thermal and vacuum modes. This technical solution effectively improves the system's prediction accuracy by adaptively adjusting parameters and optimizing prediction results in real time. It significantly improves the accuracy of heat flux and vacuum level predictions for zoned pipelines, reduces errors and uncertainties caused by single-mode predictions, and enhances the system's monitoring accuracy.

[0035] 2. Enhanced ability to locate and diagnose zonal anomalies. Existing technologies mostly rely on static thresholds or independent sensor data to locate pipeline problems, lacking a comprehensive analysis and localization mechanism, making it difficult to effectively distinguish which zones have anomalies. This invention employs a sparse inversion method based on joint normalized residuals. Based on zone heat flow and vacuum level data, it uses a sparse decoupling algorithm to accurately locate anomalies in the target zone. This method utilizes residual analysis and cross-modal joint diagnosis to identify and locate anomalous zones in real time and accurately, effectively improving the diagnostic capability for zonal anomalies, avoiding misdiagnosis or missed diagnosis, and improving system stability and reliability.

[0036] 3. Optimize actuator coordination settings to reduce energy waste. In existing technologies, multiple actuators (such as zoned extraction valves and make-up valves) typically operate independently and are set based on experience, without considering the synergistic effect between valve positions and heating power, often leading to energy waste or over-adjustment. This invention, by jointly setting the control commands of the actuators and introducing rate limits, energy budgets, and actuator action synchronization constraints during valve position switching and heating power setting, ensures that all actuators work collaboratively within a reasonable range. Through the coordinated control of actuators, unnecessary energy waste is reduced, the system's energy efficiency ratio is improved, and operating costs are lowered.

[0037] 4. Online model updates improve system adaptability. Existing heat flux and vacuum level models are often static. Once the system changes or ages, the original model becomes difficult to adapt to new operating conditions, leading to a decline in system performance and requiring manual intervention for updates. This invention employs an online parameter update mechanism, using adaptive heat flux prediction algorithms and vacuum conduction prediction algorithms, combined with recursive least squares and stability projection strategies, to achieve online model updates and adaptive adjustments. This improves the system's adaptability during long-term operation, enabling it to automatically adapt to system changes in practical applications, reducing the need for manual intervention, and enhancing the system's operational stability and reliability.

[0038] 5. Reduce false alarms and missed alarms, and improve system security. Traditional monitoring systems often rely on fixed thresholds, which are prone to false alarms and missed alarms, leading to delayed system response and potentially endangering equipment and operator safety. This invention, through combined normalized residual analysis of thermal and vacuum modes, along with zonal evidence scores and a multi-criteria screening mechanism, further improves the accuracy of the target zonal set, ensuring accurate alarms and responses. This significantly reduces false alarms and missed alarms, enhances system security, and ensures stable equipment operation and personnel safety.

[0039] 6. Supports multi-zone, large-scale applications. Traditional methods typically can only handle a small number of zones or require manual settings and frequent adjustments, making it difficult to achieve automated adjustment and maintenance of large-scale systems. This invention divides the zones into multiple independent control units, models and controls each zone separately, and optimizes the overall system through collaborative settings and large-scale data processing. This allows the invention to support intelligent control of multi-zone, large-scale vacuum insulation pipe systems, expanding the application scope and improving the system's automation level.

[0040] In summary, this invention solves several problems in the prior art by introducing bimodal prediction models, sparse inversion localization, cooperative actuator setting, online parameter updating, and multi-criteria decision-making, thereby improving the accuracy, efficiency, security, and adaptability of the system. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the steps of a vacuum insulation pipe control method based on an adaptive heat flow prediction algorithm according to the present invention.

[0042] Figure 2 This is a schematic diagram of the time series of binary encoding and synchronization marking of the valve network;

[0043] Figure 3 A graph showing the comparison between measured and predicted values ​​of heat flux density in the pipe wall of each zone;

[0044] Figure 4 A curve comparing the measured and predicted values ​​of the vacuum degree of the partitioned sandwich cavity;

[0045] Figure 5 For the joint normalized residual time series curve;

[0046] Figure 6 A curve showing the comparison between the evidence score for each partition and the target partition threshold;

[0047] Figure 7 Set a time curve for actuator coordination;

[0048] Figure 8 The tracking error curve is displayed when entering the preset operating condition window.

[0049] Figure 9 This is a convergence curve of the forgetting factor and model parameters during the online update process. Detailed Implementation

[0050] The advantages of the present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments.

[0051] See Figure 1 As shown, this embodiment provides a method for controlling a vacuum insulation pipe based on an adaptive heat flow prediction algorithm, including the following steps: Step S1: Divide the vacuum insulation pipe into N independent control zones along the axial direction. A zoned interlayer cavity is formed between the inner tube and the outer tube of each independent control zone. A monitoring unit and an execution unit are set in each independent control zone. The monitoring unit includes a heat flow sensor, a temperature sensor, and a vacuum sensor. The execution unit includes a heating component, a zoned evacuation valve, a zoned replenishment valve, and a zoned bypass microfluidic valve. Step S2: Construct a switchable valve network. The switchable valve network is formed by valve groups located at both ends of each independent control zone. During the sampling period, the connectivity of the valve network and the power switch of the heating component are switched according to a predetermined binary encoding sequence to form a zone-identifiable joint excitation. Step S3: Synchronously collect zone acquisition data. The zone acquisition data includes zoned pipe wall heat flow, zoned inner tube outer wall temperature, zoned outer tube inner wall temperature, and zoned interlayer cavity vacuum degree. Generate thermal mode measurement vectors and vacuum mode measurement vectors respectively. Step S4: Establish a zoned heat flow prediction model and a zoned vacuum prediction model. The conduction prediction model and the partitioned heat flow prediction model use historical data of joint excitation and thermal mode measurement vectors as input-output mappings. The partitioned vacuum conduction prediction model also uses historical data of joint excitation and vacuum mode measurement vectors as input-output mappings. Step S5: Based on the predicted and measured outputs of the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model, generate thermal residual vectors and vacuum residual vectors, and perform joint normalization processing on the thermal residual vectors and vacuum residual vectors to obtain joint normalized residuals. Step S6: Using the joint normalized residuals as observations, solve the rarefaction problem. The leakage vector is used to obtain the target partition set, and the partition control command corresponding to the target partition set is generated; Step S7: The partition control command is issued to the execution unit of the target partition to coordinate the setting of the partition air extraction valve, partition air replenishment valve, partition bypass microfluidic valve and heating component, so that the vacuum degree of the partition interlayer cavity and the temperature of the outer wall of the partition inner tube meet the preset working condition window; Step S8: The parameters of the partition heat flow prediction model and the partition vacuum conduction prediction model are updated online, and the stability projection constraint is applied to the updated parameters before returning to step S2 for repeated execution.

[0052] This embodiment will describe step S1 in detail. Step S1 is used to establish the basic conditions for hardware and measurement and control topology. Its core lies in dividing the vacuum insulation tube into multiple independent control zones along the axial direction, so that the interlayer cavity of each independent control zone can be independently monitored and controlled, and at the structural and interface level, it provides sufficient identifiability for subsequent switchable valve networks, coded excitation, dual-mode measurement and model identification. Specifically, step S1 includes the following steps: zone geometry construction, sealing and isolation of the interlayer cavity, arrangement of monitoring units, integration of execution units, vacuum penetration of measurement and execution wiring, constraint design of thermal parasitic coupling, and traceable identification and coding system. Preferably, in the implementation of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm", the vacuum insulation pipe is configured as a coaxial structure of an inner pipe and an outer pipe. The inner pipe is used for medium transport, and a vacuum insulation layer is formed between the outer pipe and the inner pipe. The vacuum insulation layer is axially divided into multiple partitioned sandwich cavities by a circumferential partition member. Each partitioned sandwich cavity is in a vacuum-connected and isolated state with adjacent partitioned sandwich cavities under static conditions, and can be controllably connected through a valve group under dynamic conditions. To quantitatively describe the key quantities involved in step S1, different symbols are introduced within the formula range below: the number of independently controllable partitions is defined as... , will the The axial length of each independent control zone is defined as The nominal outer diameter of the inner tube is defined as The nominal inner diameter of the outer tube is defined as The equivalent leakage flux of each partitioned cavity under molecular flow conditions is defined as... .

[0053] Regarding the partitioned geometry construction, step S1 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" preferably sets the number of independent control partitions to no less than four to support subsequent joint excitation and cross-partition coupling identification based on coding. Prototype testing showed that with eight independent control partitions, an axial length of 3.0 m for each partition, a nominal outer diameter of 38 mm for the inner tube, and a nominal inner diameter of 60 mm for the outer tube, stable partition-level heat flow and vacuum conduction responses could be obtained in subsequent steps on a device with a total length of approximately 24 m. Each partitioned interlayer cavity is sealed by circumferential welding of a stainless steel circumferential partition member. The circumferential partition member forms a double-sided lip seal transition between the outer tube and the inner tube, preferably with a lip thickness of 0.8 mm to 1.2 mm and a circumferential weld leg width of no less than 1.5 mm, to ensure sealing strength under vacuum and heating disturbance conditions. To reduce unwanted heat conduction across zones, step S1 of the "vacuum insulation pipe control method based on adaptive heat flow prediction algorithm" involves setting a circumferential thermal isolation groove between the circumferential partition member and the outer jacket. The thermal isolation groove preferably has a depth of 0.6 mm to 0.9 mm and a width of 1.5 mm to 2.0 mm. A temperature equalization ring is also set on the outside of the circumferential partition member to suppress local temperature spikes. To reduce radiation coupling uncertainty, a low-emissivity coating patch is applied to the inner surface of each zone's interlayer cavity, and a sensor window is provided. The sensor window preferably has a diameter of 18 mm to 22 mm to meet the requirements for heat flow sensor attachment.

[0054] Regarding the sealing and isolation of the partitioned interlayer cavities, step S1 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" achieves static isolation and dynamic controllable connectivity through a three-layer mechanism: structural isolation of the circumferential partition components, static backflow prevention by micro-check valves, and dynamic topology switching of the switchable valve network. Preferably, a micro-check valve is arranged between every two adjacent partitioned interlayer cavities, with the opening pressure differential of the micro-check valve limited to the range of 40Pa to 80Pa, to suppress reverse crosstalk during high-dynamic pumping or inflation transients. Regarding the equivalent leakage flux, leakage bench testing revealed the median value of the equivalent leakage flux for the eight independently controlled partitions when a capillary bypass with an equivalent diameter of no more than 0.3mm and an equivalent length of no less than 15mm was introduced on the outside of each partitioned interlayer cavity. Can be controlled in Within the range of levels, subsequent steps employing coded excitation and joint sparse inversion can clearly distinguish the sources of disturbances at the single-partition level.

[0055] Regarding the arrangement of monitoring units, step S1 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" sets up a set of heat flow sensors, a set of temperature sensors, and a set of vacuum sensors in each independent control zone. Preferably, the heat flow sensors are thin-film heat flow plates, the temperature sensors are platinum resistance thermometers, and the vacuum sensors are a combination of a miniature thermal conductivity vacuum gauge and a cold cathode vacuum gauge to cover 0.1 Pa to... The range is Pa. In terms of spatial arrangement, the heat flow sensor is attached to the upstream quarter of the outer wall of the inner tube, the temperature sensors are respectively set at the axial midpoint mirror positions of the outer wall of the inner tube and the inner wall of the outer sleeve, and the vacuum sensor is set at the downstream quarter of the partitioned interlayer cavity, forming a resolvable measurement channel under the three mechanisms of convection, conduction, and radiation. To facilitate subsequent topology switching and coded excitation of the switchable valve network, step S1 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" leads out the monitoring unit of each independent control zone using an independent vacuum-insulated electrical connector. The vacuum-insulated electrical connector uses a glass-metal sealing component to achieve both electrical insulation and vacuum sealing requirements.

[0056] Regarding the integration of execution units, step S1 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" sets up a partitioned extraction valve, a partitioned replenishment valve, a partitioned bypass microfluidic valve, and a heating component in each independent control zone. Preferably, the partitioned extraction valve and the partitioned replenishment valve are high-vacuum electric valves with a minimum resolvable opening of no more than 0.5%; the partitioned bypass microfluidic valve adopts a mass flow adjustable structure for subsequent frequency sweep microfluidic excitation; the heating component adopts a wound constant resistance heating belt with a calibration port for lead-out power. To enhance the identifiability of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" at the structural level, step S1 arranges the valve seats of the partitioned extraction valve and the partitioned replenishment valve of each independent control zone at a relative axial quarter position, and arranges the partitioned bypass microfluidic valve and the heating component at symmetrical positions on both sides of the axial center, thereby forming a superposition of responses of different spatial modes in subsequent encoded excitation. This arrangement has been verified by prototype testing to reduce the peak ratio of coupling residuals between adjacent zones to below 0.25. To facilitate a quantitative description of the isolation improvement brought about by this arrangement, the cross-partition coupling ratio is defined as... The peak residual value of the target partition is defined as The peak residual of adjacent partitions is defined as Under standard experimental conditions, the following measurement relationship can be obtained: Among them, cross-partition coupling ratio The experimental evaluation parameters are used only in the formula description of this section; the full technical names will continue to be used in the main text. The results are derived from statistical measurements of a prototype with a total length of 24 m and eight independently controlled zones under conditions of a sandwich vacuum of approximately 2 Pa and a linear power of 0.5 W / m step disturbance.

[0057] Regarding the vacuum penetration of measurement and execution wiring, step S1 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" connects all monitoring units and execution units to a bus coupler via a partitioned junction box. The partitioned junction box is located outside the outer casing and is epoxy-encapsulated to form a secondary seal. The bus coupler uses a differential bus structure to support the timing requirements of subsequent synchronous sampling and synchronous excitation. Preferably, the metal shell of the partitioned junction box and the outer casing are equipotentially connected via flexible copper braided tape to reduce the impact of electromagnetic interference on the thin-film heat flow sensor. To support the synchronous driving of subsequent coded sequences on the valve network and heating components, step S1 predefines the topology mapping of the execution ports and sets machine-readable identifiers on the junction boxes of each independent control partition. The machine-readable identifiers use both one-dimensional barcodes and two-dimensional codes for online verification of hardware wiring correctness. According to assembly line data statistics, compared with the control tooling without machine-readable identifiers, the probability of incorrect wiring decreased from 3.1% to below 0.2%.

[0058] Regarding the traceable identification and coding system, step S1 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" assigns a unique zone identification number to each independent control zone and sprays it onto the outer surface of the outer jacket. Simultaneously, a passive high-frequency RFID tag is embedded within the zone's interlayer cavity. The RFID tag is coupled to an external reader / writer through a glass-metal sealing window, achieving zone identification without disrupting the vacuum seal. Preferably, the zone identification number is at least six digits long and includes the assembly date and assembly station information, allowing for traceability back to batch process parameters during subsequent maintenance and calibration.

[0059] To further support the identifiability of subsequent dual-mode measurements and joint inversion, step S1 of the "vacuum insulation pipe control method based on adaptive heat flux prediction algorithm" also reserves a reference capillary interface in the partition cavity of each independent control zone. The reference capillary interface is used to connect a reference capillary during the initialization phase for amplitude calibration and time constant calibration. In the prototype test, the equivalent flux of the reference capillary is... to The measurement can be repeated within the interval. After calibration using the reference capillary, the amplitude estimation deviation of the partitioned vacuum conduction prediction model converges from ±18% before calibration to within ±5%. To ensure the reusability of the reference capillary at the structural level, step S1 sets the reference capillary interface to a VCR metal seal and installs a torque limiting nut on the outside. The limiting value is preferably 10 N·m to avoid stress cracking of the glass-metal sealing window due to excessive tightness.

[0060] In terms of the constraint design for thermal parasitic coupling, step S1 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" optimizes the local thermal resistance of the inner tube support structure. It preferably uses a three-point low thermal conductivity support for each independent control zone. By creating radial thinning grooves and installing ceramic gaskets on the support feet, the equivalent thermal conductivity of a single-point support is controlled to less than 0.35 W / K. To illustrate the quantification effect of this constraint, the equivalent thermal conductivity of a single-point support is defined as... The total equivalent thermal conductivity at three points is defined as follows: The results were obtained through prototype measurement. This can significantly suppress the systematic error drift caused by the supporting thermal bridge in subsequent heat flow prediction.

[0061] To ensure compatibility with different media and on-site constraints, step S1 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" allows for the replacement of the inner and outer tube materials without altering the functional boundaries of the independent control zones. For example, the inner tube can be replaced from stainless steel to a low-temperature austenitic alloy to reduce low-temperature shrinkage stress, and the outer tube can be replaced from stainless steel to aluminum alloy to reduce the overall mass. Under this replacement, the sealing boundary of the zoned interlayer cavity and the connection method of the monitoring and execution units remain unchanged, thus not affecting the switchable valve network and coded excitation in subsequent steps. As a comparative verification, a unified step heating disturbance test was conducted on prototypes with two different material combinations. The difference in temperature response rise time (10% to 90%) at the zone level was less than 6%, indicating that the measurement and control topology established in step S1 has material independence and is beneficial for promotion in different engineering environments.

[0062] In summary, step S1 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" provides structured identifiable conditions and stable boundary conditions for subsequent steps, including switchable valve network, joint excitation of binary coding and frequency sweeping microfluidics, establishment of partitioned heat flow prediction model and partitioned vacuum conduction prediction model, construction of joint normalized residuals, and solution of sparse leakage vectors. This is achieved through multi-level partitioned geometry construction, partitioned interlayer cavity sealing and isolation, monitoring unit arrangement, execution unit integration, vacuum penetration of measurement and execution wiring, constraint design of thermal parasitic coupling, and traceable identification coding system.

[0063] In this embodiment, step S2 will be described in detail. In the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm", step S2 is used to construct a switchable valve network and synchronously switch the connectivity of the switchable valve network and the power switch of the heating component according to a predetermined binary encoding sequence within the sampling period. This forms a joint excitation that is mutually orthogonal, statistically independent, and energy-controlled in each independent control zone, providing a high signal-to-noise ratio and low coupling identification input for the zone heat flow prediction model and zone vacuum conduction prediction model in subsequent steps. The basic topology of the switchable valve network consists of valve groups arranged at both ends of each independent control zone. The valve groups are connected to the zone bypass microfluidic valve, the zone extraction valve, and the zone replenishment valve through an integrated manifold, so that any independent control zone can be controlled to switch between states such as "isolation", "connection to the main extraction branch", "connection to the reference capillary calibration branch", and "controlled connection to the adjacent independent control zone via a flow-limiting branch". To ensure repeatability and durability under vacuum conditions, the switchable valve network preferably adopts a structure with a metal-sealed valve core and magnetic coupling drive, and buffer holes are provided on both sides of the valve core to suppress the impact of transient pressure differences on the interlayer cavity. Prototype tests conducted on eight independently controlled zones in the experimental setup showed that, within the interlayer cavity pressure range of 0.5 Pa to 5 Pa, no jamming or leakage level drift occurred after 10,000 repeated switching cycles, verifying that the switchable valve network constructed in step S2 meets the durability requirements for subsequent long-term online identification.

[0064] To quantify the timing constraints of encoding drive and topology switching in step S2, the sampling period of the "vacuum insulation pipe control method based on adaptive heat flow prediction algorithm" is defined in the formula as […]. The symbol duration of a binary coded sequence is defined in the formula as . The state vector of the switchable valve network is defined in the formula as follows: The binary sequence of power switches for the heating element is defined in the formula as follows: The connectivity matrix of the switchable valve network is defined in the formula as follows: To balance the mechanical response of the valve core with the synchronization of measurement, it is preferable to make the duration of the code elements in the binary encoded sequence an integer multiple of the sampling period, i.e., using a standard mathematical expression. Under this constraint, the sum of the valve core's mechanical switching time and signal establishment time should not exceed one-fifth of the symbol duration to avoid fuzzy excitation caused by opening and closing transitions. Prototype testing results show that when a sampling period is used... ms, symbol duration At ms, the average valve core switching time for the eight independent control zones across the entire temperature range is approximately 42 ms, and the signal setup time is approximately 23 ms, which is compatible with the above constraints.

[0065] To ensure the separability of responses in each independent control zone under joint excitation, the predetermined binary coding sequence preferably employs a family of orthogonal binary sequences. A bipolar or unipolar "zero-mean relative perturbation" strategy with a strict 50% duty cycle is used for the power switching of the heating element. This involves performing symmetrical "on / off" perturbations of equal duration near the steady-state baseline power of the heating element, ensuring that the algebraic sum of the heat input within each frame of coding is close to zero, thereby suppressing steady-state temperature drift. To quantify its orthogonality and separability, a sequence is defined to be assigned to the ... The binary coded sequence of each independent control zone is denoted in the formula as follows: The normalized correlation metric for a binary encoded sequence is defined in the formula as: ;in, The number of code elements encoded in one frame. By employing orthogonal binary coding and performing discrete searches for polarity and phase before loading, the experimental setup... and Both frame lengths can achieve (when )or The approximate orthogonality condition. With eight independent control partitions and frame length... Taking the prototype test as an example, in the coded correlation spectrum constructed based on the partitioned heat flow measurement, the ratio of the main diagonal peak to the maximum sidelobe (in terms of linear ratio) is better than 20:1, and the corresponding cross-partition coupling residual ratio drops to below 0.2, indicating that the coding drive in step S2 enables the identification input of the "vacuum insulation pipe control method based on adaptive heat flow prediction algorithm" to have good mutual uncorrelation.

[0066] The connectivity of the switchable valve network is switched discretely in step S2 according to a binary encoded sequence. Its structure can be represented by the connectivity matrix in the formula. ,in To map the state vector of a switchable valve network to a deterministic operator of topological connectivity, considering the shock sensitivity of the vacuum system, changes to the state vector of the switchable valve network between adjacent symbols are restricted to only one connectivity change per symbol, thus ensuring that the total number of connectivity changes within a frame does not exceed the number of symbols. To formalize this constraint, the Hamming distance between adjacent symbols is defined in the formula as... and apply The switching constraint, as recorded by the prototype, can control the peak transient pressure difference of the partitioned interlayer cavity to within 20 Pa, significantly reducing vacuum disturbances caused by rapid valve switching. To ensure strict synchronization between the power switch of the heating element and the switching of the switchable valve network, time synchronization is performed between the controller of the switchable valve network and the power controller of the heating element via hardware synchronization pulses. Clock drift is corrected by inserting a synchronization marker into the first symbol of each frame, thereby maintaining the phase consistency of the binary coded sequence during long-term operation.

[0067] Regarding energy and safety constraints, step S2 sets an intra-frame energy limit for the power switch of the heating component and limits it using a standard mathematical expression. ,in The relative step power amplitude of the heating element. This represents the upper limit of the energy allowed within the operating window. The prototype is in... The intra-frame energy under the condition is approximately Even with a background of slow temperature changes on the outer wall of the tube within the partition, a separable thermal modal response can still be obtained without reaching the upper temperature limit. Compared with direct square wave drive without energy constraints, the peak-to-peak temperature within the frame is reduced by about 37%, which is beneficial for keeping the outer wall temperature of the tube within the partition within the preset operating window in subsequent steps.

[0068] To improve robustness and engineering usability, step S2 sets up coded self-test frames and calibration frames. Specifically, after every few frames of the regular binary coded sequence, a self-test frame is inserted stating "all switchable valves are in isolation, and the heating component power switch is constantly at zero" to determine the measurement noise baseline. Then, a calibration frame is inserted stating "all switchable valves are connected to the main exhaust branch, and the heating component power switch flashes at a fixed duty cycle" to verify the actuator state consistency. During a two-hour endurance run, the prototype operated in a rotating pattern of "binary coded sequence × 15 frames + self-test frame × 1 + calibration frame × 1," and the noise standard deviation of the zoned heat flow measurement stabilized at [value missing]. At the level, the actuator consistency pass rate through the calibration frame criterion reaches 99.6%. When a local failure of the switchable valve network is detected (e.g., sticking or insufficient stroke at a certain valve position), step S2 triggers code reconstruction, that is, removing the affected column vectors from the predetermined binary code sequence family and performing a secondary search for polarity and phase on the remaining column vectors to ensure that the remaining column vectors still satisfy the requirements. The approximate orthogonality condition is obtained while maintaining the Hamming distance constraint. This encoding reconstruction can be completed within one frame in the simulated fault scenario of the prototype, ensuring that the identification of subsequent steps is not interrupted.

[0069] It should be noted that step S2 supports hierarchical coding and staggered multiplexing. Hierarchical coding refers to dividing independent control zones into several clusters based on physical adjacency and assigning a set of low-speed binary coding sequences to each cluster. Then, within each cluster, high-speed binary coding sequences are assigned to each independent control zone, allowing the overall code rate and energy distribution to scale linearly with length expansion. Staggered multiplexing refers to introducing a time misalignment of half-symbols or whole-symbols into the binary coding sequences of adjacent clusters to reduce cross-cluster correlation caused by radial heat transfer through the outer casing. Long-term tests of the prototype (total length 60 m, twenty independent control zones) show that after adopting hierarchical coding and staggered multiplexing, the average cross-correlation sidelobes obtained from zone heat flow measurements decreased by approximately 44%, and the false alarm rate obtained from zone vacuum conduction measurements decreased from 0.9% to 0.3%, significantly improving the separability of zone-level responses.

[0070] Considering potential strong disturbances in the field (such as sudden charging / discharging or a step change in ambient temperature), step S2 incorporates a gating matrix and a pause condition into the drive logic of the switchable valve network and the heating component power switch. The gating matrix is ​​used to forcibly maintain the switchable valve network in an isolated state and keep the heating component power switch at zero value when the vacuum level of the partitioned interlayer cavity exceeds the safety threshold, the temperature of the outer wall of the inner tube in the partition approaches the upper limit, or a pause command is issued by the upper-level monitoring system. This ensures the valve network remains within the safe zone and the power switch returns to zero until it is restored to the safe zone, then smoothly transitions back to the binary encoding sequence. This gating strategy reduced the false trigger rate to below 0.1% in the prototype's disturbance injection experiment and avoided model updates errors caused by strong disturbances. Furthermore, to reduce mechanical wear, step S2 uses a dual-edge compensated pulse for driving the switchable valve network. This means that after each state transition, a holding pulse with attenuated amplitude is applied to the drive coil to ensure the valve core is in position, while limiting the holding current to reduce valve body heating, thereby extending its lifespan.

[0071] At the implementation level, step S2 outputs a binary encoded sequence through an integrated timing generator and provides independent channels for the switchable valve network and the power switch of the heating component, ensuring that the two types of actions are hardware-independent but strictly aligned on the time axis. The timing generator provides three types of state register information: frame count, symbol count, and synchronization mark, which can be read and recorded by the host state machine at any time for subsequent time alignment during online parameter updates of the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model. To further improve clock stability, step S2 uses a temperature-compensated crystal oscillator as the master clock source, with clock drift maintained within ±2 ppm over 24 hours. Combined with the frame synchronization mark mechanism, the phase error between the binary encoded sequence and the sampling system can be controlled within 5% of the symbol duration.

[0072] In summary, step S2, by constructing a switchable valve network joint excitation scheme with mechanisms such as Hamming distance-constrained switching, orthogonal binary coding drive, zero-mean power perturbation, intra-frame energy constraint, synchronization marking and coding self-check / calibration frames, hierarchical coding and misalignment multiplexing, and gating matrix and dual-edge compensation pulses, provides highly identifiable, energy-constrained, and robust input conditions for the subsequent identification and control of the "vacuum insulation pipe control method based on adaptive heat flow prediction algorithm". Taking prototype tests with eight independent control zones and twenty independent control zones as examples, based on the joint excitation of step S2, the goodness of fit of the zone heat flow prediction model and the zone vacuum conduction prediction model obtained in subsequent steps can reach above 0.93 and 0.90, respectively. The positioning accuracy of sparse inversion is better than that of a single independent control zone, and it can still maintain stable convergence after coding reconstruction.

[0073] In this embodiment, step S3 will be described in detail. The goal of step S3 is to perform strict synchronous acquisition and encoding association synchronous demodulation of the partition acquisition data of each independent control zone. The partition acquisition data includes the heat flow of the partition tube wall, the temperature of the outer wall of the inner tube of the partition, the temperature of the inner wall of the outer tube of the partition, and the vacuum degree of the partition interlayer cavity. Then, according to the decoding criteria that match the binary encoding and frequency sweep microfluidics in step S2, the thermal mode measurement vector and the vacuum mode measurement vector are constructed. To ensure the discriminability and robustness of the subsequent partitioned heat flow prediction model and partitioned vacuum conduction prediction model, step S3 employs a series of measures at the hardware level: simultaneous sampling analog-to-digital conversion; at the timing level, alignment with hardware synchronization markers of the switchable valve network and heating component power switches; and at the signal processing level, correlation demodulation matching binary encoding and phase-locked extraction matching swept-frequency microfluidics. Furthermore, it incorporates anti-aliasing filtering, quantization calibration, temperature cross-sensitivity compensation, outlier suppression, and data integrity verification. These measures ensure that the thermal mode measurement vector and vacuum mode measurement vector exhibit zero mean, low correlation, and high signal-to-noise ratio characteristics at the frame-level time scale. To quantify the timing and data structure of step S3, the sampling interval is defined within the formula range as follows: (Corresponding to the hardware sampling period of "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm"), the sampling frequency is defined as... The duration of a symbol in a binary code is defined as follows: (corresponding to the symbol duration in step S2), the number of symbols encoded in one frame is defined as follows: The number of sampling points contained in each symbol is defined as follows: and order With frame duration .

[0074] Regarding synchronous acquisition, step S3 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" integrates the heat flow sensors, temperature sensors, and vacuum sensors of each independent control zone into a single multi-channel analog-to-digital converter for simultaneous sampling, and locks the start and end of sampling with a hardware synchronization marker from the timing generator in step S2; in the prototype, the sampling interval is taken as... (corresponding sampling frequency) Hz), symbol duration =0.4s, number of sampling points per symbol =80, frame code element number is taken This ensures that a frame of data contains 256025602560 synchronous sampling points and is strictly in phase with the encoding and driving in step S2.

[0075] Regarding the synchronous demodulation associated with the encoding, step S3 constructs a thermal mode measurement vector for the heat flow of the partitioned pipe wall based on the power switch of the heating component driven by binary encoding, and constructs a vacuum mode measurement vector for the vacuum degree of the partitioned interlayer cavity based on the binary encoding and frequency sweep microfluidic excitation. To formally describe the synchronous demodulation operation, the first step is defined within the formula range. The heat flow sampling sequence of the pipe wall in each independent control zone is as follows: Define the first step consistent with step S2. The binary encoding sequence of the heating element power for each independent control zone is as follows: , where index Spanning a frame sampling point, index Spanning one frame of symbols; then the first of the thermal mode measurement vectors Each component can be obtained using the following formula: ,in In the formula, the corresponding components of the thermal modal measurement vector are represented. To construct the vacuum modal measurement vector, the first component is defined within the formula's scope. The vacuum degree sampling sequence of the partition interlayer cavity of each independent control zone is as follows: , define the first The binary encoding sequence of the valve network for each independent control zone is as follows: Furthermore, orthogonal reference sequences are constructed for the frequency-sweeping microfluidic excitation within each frame. and (Generated from a known swept frequency trajectory); then the vacuum mode measurement vector can be obtained by superimposing correlation demodulation and phase-locked loop extraction:

[0076]

[0077] ,

[0078] in This indicates the vacuum modal measurement components based on binary encoding correlation. and This represents the in-phase and quadrature components of the swept-frequency microfluidic. The above measurement components, after amplitude normalization, are concatenated to form the first component of the vacuum mode measurement vector. Each partition element. Prototype results show that, at frame length =32, sampling points per symbol With a configuration of 80, the sidelobe ratio of the thermal mode measurement vector and the vacuum mode measurement vector relative to the binary coded reference is better than 20:1. The residual signal-to-noise ratio based on the partitioned acquisition data is improved to approximately [amount missing] of the undemodulated state after synchronous demodulation. The magnitude is several times that of the previous model, supporting the stable identification of subsequent zoned heat flow prediction models and zoned vacuum conduction prediction models.

[0079] Regarding anti-aliasing and quantization calibration, step S3 configures a fourth-order active low-pass anti-aliasing filter before analog-to-digital conversion to limit the bandwidth, with the cutoff frequency preferably set to one-quarter to one-fifth of the sampling frequency; in the prototype, the cutoff frequency is set to 40 Hz to match the sampling frequency of 200 Hz. After analog-to-digital conversion, step S3 applies a step suppression window function at the symbol boundary to the partitioned acquisition data to reduce leakage of the valve network switching transient in the demodulation domain, and uses a reference zero input frame to estimate and eliminate the quantization offset. To reduce the impact of temperature cross-sensitivity on the vacuum level of the partitioned interlayer cavity, step S3 introduces temperature compensation for the vacuum level of the partitioned interlayer cavity, defining the temperature sampling sequence of the inner wall of the partitioned outer tube within the formula range as follows: The temperature compensation coefficient is defined as With baseline temperature The vacuum degree of the partitioned interlayer cavity after temperature compensation is: ,in The results were obtained from multi-temperature point calibration during the initialization phase. After applying this compensation, the temperature-vacuum cross-sensitivity residual of the prototype in the range of 0.5–5 Pa was reduced from ±8% to within ±2.5%.

[0080] Regarding outlier suppression and data integrity, step S3 calculates the median absolute deviation for each frame of partitioned data acquisition and implements dual-threshold Hempel filtering. Sampling points exceeding the adaptive threshold are marked as outliers and replaced with local linear interpolation. Simultaneously, cyclic redundancy check is introduced at the frame level to ensure end-to-end data consistency. During two hours of endurance operation, the outlier rates for partitioned pipe wall heat flux, partitioned inner pipe outer wall temperature, partitioned outer pipe inner wall temperature, and partitioned interlayer cavity vacuum degree were 0.18%, 0.05%, 0.06%, and 0.22%, respectively, all below the engineering threshold of 1%. To complement the coded self-test frame and calibration frame in step S2, step S3 calculates the standard deviation of noise for each channel within the self-test frame and triggers a health indicator using the standard deviation threshold. In the calibration frame, the consistency of the response to the switchable valve network and heating component power switch is checked. Statistical analysis of eight independently controlled partitioned prototypes shows that the pass rate for health indicators reaches 99.6%. When local degradation occurs, step S3 writes this information into the subsequent parameter update stage as a basis for weight deduction.

[0081] Regarding multi-channel time alignment, in order to further suppress minor intra-channel delay differences beyond the synchronization marker in step S2, step S3 performs fine-grained alignment within the frame using the maximum cross-correlation criterion for the heat flow of the partitioned pipe wall, the temperature of the outer wall of the inner pipe of the partitioned pipe, and the temperature of the inner wall of the outer casing of the partitioned pipe, and uses the heat flow of the partitioned pipe wall as the reference channel; the reference channel is defined within the formula range as... Define the channel to be aligned as to make The obtained channel delay correction amount In prototype statistics, the time falls within ±6 ms, thus ensuring intra-frame phase consistency between the thermal mode measurement vector and the vacuum mode measurement vector.

[0082] Regarding measurement vector splicing and normalization, step S3 writes the thermal mode measurement vector and vacuum mode measurement vector of each independent control zone into the buffer at the end of the frame in a fixed order, and normalizes the amplitude using the standard deviation of their respective reference frames; the thermal mode normalization coefficient is defined within the formula range as follows: The vacuum mode normalization coefficient is The normalized results for the thermal mode and the vacuum mode are respectively , , , The normalized components are then concatenated according to the partition index to form a frame-level data structure for subsequent identification and control.

[0083] In terms of engineering implementation, step S3 provides hardware interlocking between the front-end actuator and the sampling end for partitioned data acquisition. Specifically, when the vacuum level of the partitioned interlayer cavity exceeds the limit or the temperature of the outer wall of the inner tube in the partition approaches the upper limit, the sampling end sends a pause instruction to step S2 through gating logic, ensuring that binary encoding and frequency sweeping microfluidics operate within a safe zone. Synchronous demodulation is then restarted smoothly after recovery. On eight independently controlled partition prototypes, this interlocking strategy reduces the probability of spurious signal injection under strong disturbances to below 0.1% and avoids erroneous convergence in subsequent parameter updates. Furthermore, to accommodate devices with longer distances, step S3 supports a hierarchical buffering and staggered multiplexing data aggregation strategy. Combined with the hierarchical encoding in step S2, in an experiment with a total length of 60 m and twenty independently controlled partitions, the average sidelobe related to cross-cluster correlation is reduced by approximately 44%, and the goodness of fit of the corresponding partitioned vacuum conduction prediction model improves from 0.86 to over 0.90.

[0084] In summary, step S3 of the "Vacuum Insulation Tube Control Method Based on Adaptive Heat Flow Prediction Algorithm" constructs thermal mode measurement vectors and vacuum mode measurement vectors that meet the requirements of high identifiability and high signal-to-noise ratio through strict time alignment of simultaneous sampling and hardware synchronization marking, correlation demodulation matching binary encoding, phase-locked extraction matching frequency sweep microfluidics, anti-aliasing and quantization calibration, temperature cross-sensitivity compensation, outlier suppression and data integrity verification, fine-grained multi-channel time alignment, and normalization splicing.

[0085] In this embodiment, step S4 will be described in detail. Step S4 organizes data and performs modeling using frames as the time unit. For ease of formula expression, the frame index is denoted as […]. The independent control partition index is denoted as The thermal mode measurement vector is in the first... The components of each independent control zone are denoted as The magnitude or orthogonal component of the vacuum modal measurement vector is measured in the 1st... The components of each independent control zone are denoted as The frame-level coding statistics of the heating component power are denoted as: (Determined by steps S2 and S3), the frame-level topology statistics of the switchable valve network are denoted as... (Determined by steps S2 and S3).

[0086] The zoned heat flux prediction model is used to predict the thermal mode measurement vector for the next frame from frame-level statistics of heating component power, frame-level topological statistics of switchable valve networks, and historical values ​​of thermal mode measurement vectors. In a preferred embodiment, this model employs a multi-input autoregressive-exogenous form with causal masking and cross-zone coupling terms, expressed as follows: ,in The predicted values ​​of the thermal mode measurement vectors by the zoned heat flow prediction model; For bias terms; , , , These are the autoregressive coefficient, the input coefficient of this zone corresponding to the power of the heating component, the coupling coefficient of the input of adjacent independent control zones, and the input coefficient of the switchable valve network, respectively. , , , For order; Indicates the relationship with the first Each independent control zone has a thermally coupled set of neighboring zones. In implementation, the zone heat flow prediction model ensures low and identifiable input cross-correlation through Hamming distance-limited switching and binary encoding generated in step S2. The coupling terms of adjacent independent control zones in the model are restricted by a causal mask to "only allow finite hysteresis terms along the axial nearest neighbor," thus satisfying the physical reachability of heat diffusion. The zone heat flow prediction model takes frame-level statistics of heating component power, frame-level topological statistics of switchable valve networks, and historical values ​​of thermal mode measurement vectors as inputs, and uses a multi-input autoregressive-exogenous form to predict the thermal mode measurement vector of the next frame. Simultaneously, the model explicitly includes coupling terms of adjacent independent control zones to reflect the axial thermal coupling relationship. The predicted heat flow density output by this zone heat flow prediction model can be compared frame-by-frame with the actual heat flow density obtained in step S2 to verify prediction accuracy and evaluate modeling quality. Figure 3 As shown.

[0087] The partitioned vacuum conduction prediction model is used to predict the vacuum mode measurement vector of the next frame from the frame-level topology statistics of the switchable valve network, the in-phase / orthogonal extraction results of the swept-frequency microfluidic, and the historical values ​​of the vacuum mode measurement vector. In a preferred embodiment, the partitioned vacuum conduction prediction model adopts a multi-input autoregressive-exogenous form and superimposes a phase-locked term with the swept-frequency microfluidic as a reference, as expressed in the formula: ,in, The predicted value of the vacuum mode measurement vector by the partitioned vacuum conduction prediction model; For bias terms; , , , These are the autoregressive coefficient, the switchable valve network input coefficient, and the sweep frequency microfluidic in-phase / orthogonal input coefficient, respectively. , , For order; and Phase-locked extraction of the swept-frequency microfluidic from step S3. In an improved embodiment, the partitioned vacuum conduction prediction model can introduce flow conductance scheduling, that is, using the intra-frame mean of the vacuum degree of the partitioned interlayer cavity as the scheduling variable, and using piecewise affine interpolation of the coefficient vector in several discrete vacuum degree segments to cover the operating condition changes of molecular flow-transition flow.

[0088] The partitioned vacuum conduction prediction model takes frame-level topological statistics of a switchable valve network, in-phase / orthogonal extraction results of swept-frequency microfluidics, and historical vacuum mode measurement vectors as inputs. It employs a multi-input autoregressive-exogenous structure with superimposed phase-locked terms to predict the vacuum mode measurement vector for the next frame. This model can introduce segmented scheduling to cover changes in operating conditions under different partitioned cavity vacuum levels. Comparing the predicted partitioned cavity vacuum level with the actual measured partitioned cavity vacuum level can be used to verify the fitting ability and stability of the partitioned vacuum conduction prediction model. Figure 4 As shown.

[0089] To uniformly manage model errors and generate residuals for subsequent inversion and control, step S4 defines the thermal mode residuals and vacuum mode residuals, and expresses them using standard mathematical expressions. , ,in, and These are the thermal mode residual and the vacuum mode residual, respectively. Based on the reference capillary calibration in the initialization phase (see step S0) and the encoding orthogonality in step S2, step S4 aggregates the residuals within the frame and normalizes their magnitudes to obtain the joint residual structure used in subsequent steps.

[0090] To ensure the clarity and traceability of the column vector meanings in the subsequent sparse leakage vector solution, step S4 provides the method for constructing the response matrix. First, a reference capillary is connected to the... Using the standard disturbances formed by each independent control zone as a basis, and while maintaining the same coding for the switchable valve network and heating component power switches as in normal operation, the frame-level sequences of the thermal modal residuals and vacuum modal residuals of all independent control zones are recorded. The thermal modal response matrix and vacuum modal response matrix are then defined using standard mathematical expressions. , ,in, This is the thermal modal response matrix. The vacuum modal response matrix is... To independently control the number of partitions; the i-th column vector and The steady-state residual signatures generated by the on / off switching of the reference capillary in the i-th independent control zone are normalized, and a scaling factor is introduced in different vacuum degree scheduling segments for amplitude correction. This construction ensures that in subsequent inversion, the column vectors of the response matrix correspond one-to-one with the physical "leakage location", and are consistent with the scheduling segments of the zoned vacuum conduction prediction model.

[0091] Regarding parameter estimation and online updates, step S4 applies a combination of constrained recursive least squares and robust loss to both types of models to suppress the impact of outliers on the coefficients. For consistency, the parameter vectors of the two types of models are denoted as follows within the formula range. and Let the regression vector be denoted as and In single-frame updates, parameter recursion satisfies... , The above formula uses unified notation to represent the recursion of the two types of models. and Given the current parameters and gain matrix; The step size is adaptively adjusted based on the health status indicator in step S3; The residuals after suppressing outliers using Hampshire weights; The projection operator projects the parameters onto a feasible region C defined by a causal mask, physical symbol priors, and the scheduling segment boundary. This recursion is equivalent to a regularized least squares one-step approximation with adaptive forgetting and robust weights, which can suppress the abrupt impact of abnormal frames on the parameters while maintaining linearity.

[0092] To improve the synergy between the two models at the physical consistency level, step S4 introduces a column vector registration regularization term for the two types of response matrices. Specifically, within the formula range, the scaling factor vector is defined as... And minimized using standard mathematical expressions during offline tuning or online small-step correction. ,in, For the first The registration term aligns the residual signatures of the same physical leakage in the thermal and vacuum modes, thereby obtaining a more stable column space when solving for the sparse leakage vector.

[0093] To further control model complexity and avoid overfitting, step S4 uses an information criterion to automatically select the order. Within the formula range, the set of candidate orders is denoted as... Calculate for each candidate order ,in Information criterion value; The number of sample points; This is the current candidate order; This is a complexity penalty coefficient. The partitioned heat flux prediction model and the partitioned vacuum conduction prediction model are independently selected to... The minimum order is used, allowing for minor order fine-tuning within a small range during the online phase. Prototype statistics show that, by adopting this criterion, the average number of parameters for both types of models is reduced by approximately 28% compared to the fixed-order scheme, while maintaining a essentially unchanged goodness of fit.

[0094] At the coupling constraint level with steps S2 and S3, step S4 adds a causal mask derived from the connectivity matrix of the switchable valve network to the regression vector. This means that only inputs satisfying the condition "reachable from the current frame and several lag frames" are allowed to enter the regression. The mask generation is jointly determined by the state vector sequence from step S2 and the Hamming distance constraint. This approach reduces the number of redundant coefficients in cross-regional coupling terms to approximately 45% on eight independent control partition prototypes and reduces column space correlation when subsequently solving for sparse leakage vectors. As a complementary robustness enhancement measure, step S4 weights the health indicators from step S3, applying small weights to unqualified channels when constructing the residual sum of squares, thereby ensuring stable updates for both types of models even in local degradation scenarios.

[0095] In terms of engineering implementation, step S4 stores the parameters, information criteria, health weights, causal masks, scaling factors, and response matrices of the two types of models as a frame-level data block, with each frame having a timestamp, encoded frame number, and synchronization marker, facilitating playback and comparison during on-site diagnosis or remote maintenance. To accommodate long-distance devices and hierarchical coding, step S4 supports a hybrid structure of cluster-level shared coupling terms and partition-level independent local terms, i.e., maintaining a shared adjacent coupling kernel at the cluster level, while estimating only local autoregressive and local input terms at the partition level.

[0096] Through the aforementioned structured partitioned heat flow prediction model and partitioned vacuum conduction prediction model, step S4 provides a stable and usable prediction and sensitivity foundation for the subsequent joint residual normalization, sparse leakage vector solution, and constrained actuator allocation of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm". Experimental comparisons show that, with the use of column vector registration regularization terms and causal masking, the positioning accuracy of sparse inversion is improved by about 14 percentage points compared to the baseline scheme without registration and masking, and the false alarm rate is reduced by about two-thirds; with the use of conductance scheduling, the parameter drift of both models is significantly suppressed across the vacuum range of 2 Pa to 10 Pa.

[0097] In this embodiment, step S5 will be described in detail. The purpose of step S5 is to compare the predicted outputs of the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model with the corresponding measured outputs to obtain the thermal mode residuals and vacuum mode residuals, respectively. At the frame level, the above residuals are subjected to joint normalization processing to form a standardized input for subsequent solving of sparse leakage vectors and actuator co-setting. The joint normalization processing covers baseline detrending, outlier suppression, scale uniformity, cross-partition and cross-modal covariance shaping, and safety gating masking, so that the thermal mode residuals and vacuum mode residuals are comparable across different independent control partitions and different time frames and meet the numerical conditions for subsequent inversion.

[0098] Step S5 takes the thermal mode residuals and vacuum mode residuals output from step S4 as inputs, and performs baseline detrending, outlier suppression, scale uniformity, covariance shaping, and energy balancing on them to obtain a joint normalized residual vector that is comparable across different independent control zones and has a consistent scale across thermal and vacuum modes. Step S5 further introduces a safety gating mask. When the vacuum level of the zoned interlayer cavity or the temperature of the outer wall of the inner tube in the zone is detected to be close to the safety boundary, the corresponding component is masked to prevent erroneous information from entering subsequent decisions. The behavior of the thermal mode normalized residuals and vacuum mode normalized residuals after processing in step S5 over time can be displayed using two aligned residual curves on a time axis, such as... Figure 5 As shown.

[0099] To formally represent the input and output of step S5, within the scope of the formula, the first... The independent control zone in the first The thermal modal measurement vector of the frame is denoted as The partitioned heat flow prediction model will be applied to the first... The independent control zone in the first The predicted value of the frame is denoted as Based on this, the thermal modal residual is defined as... (Equation 1), the first The independent control zone in the first The vacuum modal measurement vector of a frame (including the splicing amount of amplitude or in-phase / quadrature components) is denoted as... The partitioned vacuum conduction prediction model will be applied to the first... The independent control zone in the first The predicted value of the frame is denoted as Therefore, the vacuum modal residual is defined as follows: (Equation 2), where the four quantities appearing in equations (1) and (2) have been determined in steps S3 and S4. To improve the robustness of the residuals, step S5 first performs baseline detrending processing on the thermal mode residuals and vacuum mode residuals in each frame, using a length of The baseline sequence is obtained by moving average or first-order polynomial fitting, and then expressed using standard mathematical expressions. , (Equation 3) yields the detrended thermal mode residuals and vacuum mode residuals, where and This is the corresponding intra-frame baseline estimate.

[0100] Regarding outlier suppression, step S5 calculates the median absolute deviation for each independent control partition and each time frame, and constructs an adaptive threshold. If a sampling point exceeds the threshold, it is replaced by linear interpolation of the adjacent time frame. To express this operation in the formula, the window length is denoted as... The median value and median absolute deviation of the thermal modes are denoted as follows: and The median value and median absolute deviation of the vacuum mode are denoted as follows: and Then the anomaly discrimination conditions for thermal modes and vacuum modes are uniformly written as: , (Equation 4), where and This is the normalization coefficient. Prototype statistics show that taking... =8 frames =9 frames =3 frames When the coefficient is 3, the abnormal replacement rates of the thermal mode and the vacuum mode are approximately 0.20% and 0.25%, respectively, which are lower than the engineering threshold of 1%.

[0101] Regarding scale uniformity, step S5 constructs a scale factor for each independent control partition based on the reference capillary calibration results from the initialization phase and the noise estimation of the self-test frame, making the thermal mode residual and the vacuum mode residual comparable in amplitude. To this end, the standard deviation estimate obtained from the self-test frame is introduced into the formula, denoted as follows: and And define the scaled quantity as , Equation (5) eliminates the unit differences between different independent control zones and different channels. To further eliminate statistical correlations across zones and modes, step S5 introduces covariance shaping at the frame level, i.e., in the most recent The covariance matrices of the thermal and vacuum modes are estimated intra-frame and decomposed into lower triangular factors, followed by pre-whitening using the factor inverse matrix. For consistency in the formulas, the normalized residual vector of the thermal modes is denoted as the partitioned concatenation within a frame. Let the corresponding covariance matrix be denoted as ,make To meet The lower triangular factor; similarly, the normalized residual vector of the vacuum mode is denoted as... The corresponding covariance matrix and lower triangular factor are denoted as follows: and The pre-whitening result is: , (Equation 6), where the matrix inverse is implemented using a linear solver with positive definite correction to ensure stability. Prototype comparison shows that, compared with amplitude normalization using only Equation (5), after superimposing covariance shaping using Equation (6), the median value of the cross-regional correlation coefficient decreases by about 45%, and the cross-correlation sidelobes between the vacuum mode and the thermal mode decrease by about 38%, providing a residual input that is closer to independent and identically distributed for subsequent solution of sparse leakage vectors.

[0102] To form a unified joint normalized residual structure and provide weight parameters for subsequent steps, step S5 defines the frame-level joint normalized residual splicing vector and modal balance coefficients. First, the energy balance coefficients for the thermal and vacuum modes are set using standard mathematical expressions. (Equation 7), where To prevent extremely small positive numbers with a denominator of zero, the two types of results from equation (6) are then output as two separate components, i.e. , (Equation 8), in equation (8) and This refers to the two types of jointly normalized residual vectors output in step S5 of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm" and used in subsequent steps. The thermal mode residuals maintain their original scale, while the vacuum mode residuals are adjusted by a balance coefficient to have comparable energy in the objective function. Comparative experiments show that the energy self-balancing strategy obtained by using equations (7) to (8) reduces the column space condition number of the subsequent sparse inversion by about 31% and improves the positioning accuracy by about 12 percentage points.

[0103] To ensure safety and stability, step S5 incorporates the gating indication and health status identifier generated in steps S2 and S3 into a frame-level mask, and defines the frame-level mask vector using a standard mathematical expression. The mask component values ​​are in Then, element-wise multiplication is performed on the output of equation (8) to obtain... , (Equation 9), where This indicates element-wise multiplication. The mask is used to temporarily withhold the corresponding components from subsequent steps when limiting conditions occur (such as the vacuum level of the partitioned cavity exceeding the limit, the temperature of the outer wall of the inner tube of the partition approaching the upper limit, or a self-test frame), to avoid erroneous information affecting the inversion and control.

[0104] It should be noted that step S5 writes the thermal mode residual, vacuum mode residual, scale factor, covariance matrix factor, energy balance coefficient, and mask vector of each frame into a buffer in a versioned manner, and records the timestamp, encoded frame number, and synchronization mark for subsequent playback or offline diagnosis. Simultaneously, step S5 supports constructing a two-dimensional buffer array of frame-level joint normalized residuals in the time and partition dimensions and deriving statistical summaries. For example, it defines the most recent... Intra-frame evidence score (Equation 10), where and These are the components corresponding to Equation (6). The evidence score of Equation (10) can be used for visualization and threshold early warning in the upper-level monitoring system, and can also be written into subsequent documents as an optional technical feature to enhance operability in the monitoring and diagnosis process when needed.

[0105] It should be noted that on prototypes with eight independent control zones and twenty independent control zones, with a frame length of 32, a sampling frequency of 200 Hz, and a covariance window... With energy window After running for two hours, the joint normalization process completed in step S5 reduced the median value of the cross-regional correlation coefficient from 0.34 in the unshaped state to 0.19, and reduced the cross-correlation sidelobe of the thermal mode and the vacuum mode from 0.28 to 0.17. Under the same conditions, the average column space condition number of the subsequent sparse leakage vector solution was reduced from 18.7 to 12.9, the positioning accuracy was improved from 85% to around 97%, and the false alarm rate was reduced from 0.9% to 0.3%.

[0106] Step S5 integrates baseline detrending, anomaly suppression, scale uniformity, covariance shaping, energy balance, and safety masking to output a joint normalized residual with statistical independence and scale comparability. This directly improves the numerical stability and positioning accuracy of the subsequent inversion and control stages of the "Vacuum Insulation Pipe Control Method Based on Adaptive Heat Flow Prediction Algorithm".

[0107] In this embodiment, step S6 will be described in detail. The goal of step S6 is to use the joint normalized residual obtained in step S5 as the observation, and combine the thermal modal response matrix of the partitioned heat flow prediction model constructed in step S4 with the vacuum modal response matrix of the partitioned vacuum conduction prediction model to solve for the sparse leakage vector for the independently controlled partition, thereby determining the target partition set, and generating a partitioned control instruction list for subsequent collaborative settings. Specifically, step S6 completes three levels of work: First, constructing a cross-modal linear forward mapping and sparse inversion target to estimate the sparse leakage vector; Second, determining the target partition set and giving the partition-level action intention based on the estimation results and confidence criteria; Third, generating a partitioned control instruction list corresponding to the target partition set, including suggested setpoints or discrete levels for evacuation, gas replenishment, bypass, and heating, providing input for subsequent collaborative settings steps.

[0108] Step S6 correlates the joint normalized residuals output in step S5 with the thermal modal response matrix and vacuum modal response matrix constructed in step S4, establishing an inversion objective function with non-negativity constraints and composite sparse regularization, and solving for the sparse leakage vector, thereby locking down the target partition set of anomaly sources at the partition scale. Step S6 then calculates the partition evidence score, matching statistic, and persistence statistic, and compares them with preset thresholds to select the target partition set that meets the amplitude criterion, cross-modal consistency criterion, and persistence criterion; this selection process can be represented as "each partition corresponds to an evidence score curve and is compared with a target partition threshold baseline," such as... Figure 6 As shown.

[0109] First, a unified definition is established at the symbol level. The thermal modal response matrix identified by the partitioned heat flux prediction model in step S4 is denoted in the formula as follows: The vacuum modal response matrix obtained from the partitioned vacuum conduction prediction model in step S4 is denoted in the formula as follows: The joint normalized residual vector of the thermal modes output in step S5 is denoted in the formula as follows: The vacuum mode joint normalized residual vector output in step S5 is denoted in the formula as follows: The sparse leakage vector to be estimated is denoted as l in the formula (its dimension is equal to the number of independent control partitions and each component is non-negative), and the thermal mode weighting coefficient is denoted as... The vacuum mode weighting coefficient is denoted as in the formula. The norm regularization coefficient is denoted as in the formula. The first-order difference regularization coefficient of the time continuity constraint is denoted in the formula as... The sparse regularization coefficient of the partitioned cluster level group is denoted in the formula as follows: The feasible region of the nonnegativity constraint is denoted in the formula as .

[0110] To fuse information from thermal and vacuum modes within the same inversion framework, step S6 establishes a cross-modal forward mapping: the joint normalized residual vector for thermal modes is approximately generated by the product of the thermal mode response matrix and the sparse leakage vector, and the joint normalized residual vector for vacuum modes is approximately generated by the product of the vacuum mode response matrix and the sparse leakage vector. On a single frame of data, the basic form of sparse inversion employs a least-squares objective with non-negativity constraints and a composite regularization term, with the standard mathematical expression as follows: (Equation 11), where This is a first-order difference operator constructed based on the partition index, used to suppress unnecessary oscillations between adjacent partitions; It is a set of partitioned clusters defined by physical adjacency or hierarchical coding. The term represents a subset of sparse leakage vectors belonging to the same partition cluster, used to introduce group sparsity at the partition cluster scale to enhance robustness. The five terms in Equation (11) correspond to the thermal mode fitting term, vacuum mode fitting term, norm 1 sparsity term, temporal continuity sparsity term, and partition cluster-level group sparsity term, respectively. The introduction of this composite objective is to simultaneously satisfy the prior that "local leakage is usually sparsely distributed, has finite spatial extension, and is correlated at the partition cluster scale".

[0111] To ensure the traceability of parameter selection, step S6 employs a two-stage strategy in engineering implementation to determine the thermal mode weighting coefficient, vacuum mode weighting coefficient, and various regularization coefficients. The first stage, during the initialization phase, calculates the mode ratio based on the reference capillary calibration and self-test frame noise statistics, and provides... , (It is 12), of which and The first part estimates the noise variance for the thermal mode and the vacuum mode, respectively. The second part, in the online phase, uses both the piecewise linear criterion and the residual consistency criterion to jointly tune the regularization coefficients. That is, while ensuring that the weighted residuals of the two modes have similar energies, the search proceeds along the vicinity of the inflection point of the piecewise linear curve. , and The combination of these terms ensures that the fitted term and the regularization term of the objective function are in a balanced range.

[0112] To improve numerical efficiency and portability, step S6 employs a split-solution approach at the algorithm level, preferably using the alternating direction multiplier method: the first-norm and group-norm terms in equation (11) are separated using auxiliary variables and updated using closed-form expressions of soft thresholding and group thresholding, while the second-norm fitting problem is solved quickly using conjugate gradient or preconditioned minimum residual methods. Two typical threshold operators are given in standard mathematical expressions: the soft thresholding operator for the first-norm term is... (Equation 13) The group threshold operator for the group sparse terms is: (Equation 14), where For threshold parameters, This is a group vector. The nonnegativity constraint is achieved through component-wise projection, i.e., l In the prototype implementation, using a normalized scale for the thermal mode weighting coefficient and the vacuum mode weighting coefficient, it converges to the residual tolerance within ten iterations. The average frame computation time is less than 20 milliseconds, meeting the requirements of online applications.

[0113] After obtaining the sparse leakage vector, step S6 needs to convert it into a target partition set. To avoid false triggers caused by occasional noise, step S6 introduces a multi-criteria confidence screening mechanism: first, the amplitude criterion, requiring that the component of the sparse leakage vector is greater than the amplitude threshold; second, the correlation criterion, requiring that the matching between the thermal mode and the vacuum mode for this component is statistically significant; and third, the persistence criterion, requiring that the component is hit multiple times within the sliding window. To formalize the above process, the first... The amplitude statistics for each independent control zone are as follows: Define the matching statistics as (Equation 15), where and Representing the first and second classes of response matrices respectively List; Let be the modal equilibrium coefficient. The persistence statistic is defined as... (Equation 16), where For indicator functions, For amplitude threshold, Let be the length of the sliding window. Then the target partition set is defined as... (Equation 17) During the two-hour operation of the prototype in eight independently controlled zones, the following was taken: Four times the median of the sparse leakage vector To match the quantile threshold (90th percentile) of the statistic. =0.5 and At 5 frames, the localization accuracy of the target partition set reached 97 percentage points, and the false alarm rate was about 3,000. This represents an improvement of 12 percentage points and a reduction of 2 / 3 compared to the baseline strategy that did not use matching statistics and persistence statistics.

[0114] After forming the target partition set, step S6 further generates a partition control instruction list. Considering that subsequent collaborative setting steps need to be solved within a constrained actuator space, the partition control instruction list given in step S6 is output in the form of "action intention + suggested setting", including partition pumping action intention, partition gas replenishment action intention, partition bypass microfluidic action intention, and partition heating action intention, and provides amplitude suggestions or discrete levels. To facilitate the direct derivation of action intentions from observations, step S6 calculates symbolic indicators for two types of modes: the indicator for the vacuum mode is defined within the formula range as... (Equation 18) defines the indicator quantity of the thermal mode as follows: (Equation 19) In a vacuum sandwich cavity environment, a positive vacuum mode indicator typically corresponds to the intention to pump air from the independent control zone, while a negative vacuum mode indicator corresponds to the intention to replenish air. The thermal mode indicator is used to determine whether to superimpose a compensation action of the heating element or an action to reduce the power of the heating element. Recommended settings for subsequent collaborative settings are then given: the target vector is denoted in the formula as... , of which Each component is composed of the following formula. (Equation 20), the three coefficients in equation (20) The proportional coefficients corresponding to the zone extraction, zone replenishment, and zone heating recommendations are obtained through the initialization phase and on-site commissioning. Considering the binary selection of the switchable valve network in subsequent steps, step S6 simultaneously outputs the initial binary selection vector recommendation, denoted in the formula as... The rules are as follows: for each independent control zone belonging to the target zone set, if the vacuum mode indicator is positive, the switchable valve network is connected to the main pumping branch; if the vacuum mode indicator is negative, the switchable valve network is connected to the reference capillary calibration branch or the buffer branch; the remaining independent control zones are isolated. Therefore, the output of step S6 includes the target zone set, the zone control instruction list, the target vector, and the initial binary selection vector suggestion, all of which are passed to the subsequent collaborative setting steps in the form of frame-level data blocks.

[0115] It should be noted that, in order to enhance engineering robustness and online availability, step S6 automatically performs a feasibility diagnosis before inversion. When it is detected that the energy of the joint normalized residual vector of the thermal mode or the joint normalized residual vector of the vacuum mode is much higher than the initial statistics (e.g., more than three times the standard deviation of the mean) and does not meet the coding orthogonality self-check, the inversion of this frame is paused and the "only the target partition set and partition control instruction list of the previous frame are used" hold strategy is output. When it is detected that the amplitude mismatch of the column vector of the response matrix exceeds the allowable range, the short-period on / off calibration of the reference capillary is automatically triggered to update the amplitude calibration parameters of the column vector of the vacuum mode response matrix, thereby avoiding long-term deviation accumulation.

[0116] It should also be noted that step S6 includes optional implementation methods:

[0117] The first term is the cross-modal consistency penalty: a column vector registration term is added to the objective function of equation (11). (Equation 21) is used to suppress the inconsistency between the thermal mode and the vacuum mode in determining the same independent control zone. The second term is multi-scale sparsity: based on equation (11), differential operators of different spatial scales are constructed in parallel and added in the form of a weighted sum, thereby taking into account both point leakage and short-segment leakage. Both of these mechanisms have been verified on the prototype to have statistical significance for improvement, and can be directly incorporated into the clauses if necessary.

[0118] Based on the implementation of step S6, the eight independent control zone prototypes obtained the following statistical results during two hours of endurance operation: In the controllable leakage scenario simulated by the reference capillary, the positioning accuracy was 97 percentage points and the positioning resolution was better than that of one independent control zone; in the natural disturbance scenario, the false alarm rate was about 3,000; in the complex scenario where encoding reconstruction and online update of the response matrix occurred simultaneously, the average number of convergence frames was less than four frames, and the vacuum recovery time caused by the execution of the zone control instruction list was shortened by about 30% compared with the baseline strategy.

[0119] In summary, step S6 unifies the joint normalized residuals and the two types of response matrices into an inversion framework with nonnegativity constraints and composite sparse regularization, robustly estimates the sparse leakage vector, and then uses multi-criteria confidence to select the target partition set. Combined with modal indicators, it generates a structured list of partition control instructions and two types of suggestion vectors, providing digital and traceable input for subsequent collaborative setting steps.

[0120] In this embodiment, step S7 will be described in detail. Step S7 is used to coordinately set the partition pumping valve, partition gas supply valve, partition bypass microfluidic valve and heating components of the target partition according to the partition control instruction list, target vector and binary selection vector suggestion generated in step S6, so that the vacuum degree of the partition interlayer cavity and the temperature of the outer wall of the inner tube of the partition meet the preset operating condition window. This step is executed according to the process of "constraint consistency check - two-stage setting solution - quantization and switching scheduling - safety gating and issuance - online evaluation and write-back", and at the implementation level, it combines the optimal solution of constrained actuator allocation, Hamming distance limit of binary switching, rate limit and energy budget control and other mechanisms to ensure that the setting and issuance are completed efficiently and stably within the dynamic range that the vacuum system can withstand.

[0121] Step S7, based on the zonal control instruction list and binary selection vector suggestion from Step S6, executes the "constraint consistency check—two-stage setting solution—quantization and switching scheduling—safety gating and issuance—online evaluation and writeback" process. The first stage applies a Hamming distance constraint to the binary selection vector at the switchable valve network level to control the number of switching operations within a single frame. The second stage integrates the settings of the zonal extraction valve, zonal replenishment valve, zonal bypass microfluidic valve, and heating component power into a constrained optimization solution, ensuring they simultaneously meet valve position boundaries, rate limits, energy budget, and safety gating conditions. The output behavior of this coordinated setting over time can be represented as two time-varying setting curves: valve opening (percentage) and heating component power (relative value). These two curves maintain consistent scheduling at critical moments. This process is as follows: Figure 7 As shown.

[0122] Step S7 calculates the tracking deviation, energy consumption, and switching count after each frame's collaborative setting is completed, and determines whether the current partitioned interlayer cavity vacuum level and the partitioned inner tube outer wall temperature have entered the preset operating condition window. This process of entering the preset operating condition window can be represented by a curve that "gradually reduces the normalized tracking error and keeps it below the upper limit of the preset operating condition window," as shown below. Figure 8 As shown.

[0123] To formally describe the core solution step S7, different symbols are introduced and defined for the relevant technical names within the formula range: the control vector containing the zone vacuum pump speed setting, zone extraction valve opening setting, zone replenishment valve opening setting, zone bypass microfluidic valve opening setting, and heating component power setting is denoted as... The target vector given in step S6 is denoted as... The actuator-to-state linear mapping matrix that maps the control vector to the "vacuum degree of the partitioned sandwich cavity and the temperature deviation space of the inner tube of the partition" is denoted as: The lower and upper bounds of the control vector are denoted as follows: and The binary selection vector of the switchable valve network is denoted as... The switchable valve network binary selection vector of the previous frame is denoted as... The control vector of the previous frame is denoted as... The rate limiting vector is denoted as The energy budget cap is denoted as The unit energy coefficient of the power step of the heating component is denoted as... ; Record the frame duration as .

[0124] Regarding constraint consistency checks, step S7 first confirms the feasibility of the target vector based on the measured values ​​of the vacuum degree of the partitioned interlayer cavity and the outer wall temperature of the inner tube in the current frame, according to the preset operating condition window. Based on this, soft constraint relaxation variables are constructed so that a solution can still be obtained even when the target vector is temporarily unreachable or experiences strong disturbances. Therefore, a relaxation variable vector is introduced into the formula. and its weighting coefficients The penalty term is reflected in the objective function. To avoid handover jitter, step S7 also checks whether the difference between the proposed binary selection vector from step S6 and the switchable valve network binary selection vector of the previous frame satisfies the handover constraint that "the Hamming distance in a single cycle is not greater than 1". If it does not satisfy the constraint, the proposed vector is projected to the nearest node to satisfy the upper limit of the number of handovers.

[0125] Regarding the two-stage solution setup, step S7 preferably employs a two-stage strategy of "binary selection first, continuous quantity allocation later." In the first stage, at the switchable valve network level, the switchable valve network binary selection vector for the current frame is determined based on the target partition set and the proposed binary selection vector, and a Hamming distance constraint is applied. To formalize this constraint, the Hamming distance is defined in the formula using a standard mathematical expression. ,in This represents the zero-norm count. The second stage solves the constrained actuator allocation problem at a continuously set level. The goal is to make the mapped state deviation as close as possible to the target vector while satisfying constraints such as boundary, rate, and energy, and to limit the magnitude and rate of change of the control variable. The quadratic form of this problem in the formula is: , , Consistent with s in terms of connectivity and connectivity isolation constraints. ,in To control the energy weighting coefficient, The variable rate weighting coefficient is represented by 1, and 1 represents a column vector with all elements equal to 1. This form ensures that, given a binary selection of the switchable valve network, continuously setting the problem as a convex quadratic programming problem allows for rapid attainment of the global optimum. Prototype experiments show that, under conditions of eight independent control zones, a frame length of 0.4 s, and boundary range and rate constraints, the average solution time for quadratic programming is less than 15 ms, meeting the requirements for online applications.

[0126] In terms of quantization and switching scheduling, step S7 quantizes the continuous solution of the partitioned bypass microfluidic valve and the heating component according to a preset discrete set, and applies a dual-edge compensation pulse at the instant of switching the switchable valve network to ensure that the valve core is in place. At the same time, a phase-aligned synchronization mark is used to ensure that the binary switching and the power step are strictly in phase. To formalize the quantization operation, a discrete set is defined in the formula. and (Units are relative nominal values), and the corresponding components of the continuous solutions are mapped to set elements using the nearest neighbor rule. To limit the switching to be too frequent, step S7 checks the difference with the previous frame again after quantization. If multiple variables change across levels simultaneously in a single frame, one item is selected for execution according to priority order, and the rest are delayed to the next frame. The priority order can be configured as follows: "Settings related to the vacuum degree of the partitioned interlayer cavity exceed the limit take precedence over settings related to the temperature of the outer wall of the inner tube of the partition, gas extraction takes precedence over gas replenishment, and valve position takes precedence over power."

[0127] Regarding safety gating and distribution, step S7 introduces the gating matrix and health status indicator into the execution path. When the vacuum level of the partitioned interlayer cavity is detected to be close to the lower or upper limit, or when the temperature of the outer wall of the inner tube of the partition is detected to be close to the upper limit, the gating matrix forces the switchable valve network to remain in a safe state of isolation or connection with the main exhaust branch in this frame, and limits the power setting value of the heating component to below the threshold. When the health status indicator indicates that an actuator is abnormal (e.g., valve position sticking or insufficient stroke), step S7 masks the corresponding channel of the actuator and uses the "equivalent substitution" method to complete the setting of the approach target vector through the combination of other actuators. During the two-hour disturbance operation of the prototype, after adopting the gating and masking strategy, the impact of abnormal frames on the target time was reduced from an average of +2.1 frames to +0.6 frames, avoiding setting instability.

[0128] Regarding online evaluation and write-back, step S7 performs a closed-loop evaluation of the implemented settings after each frame and writes back key quantities to steps S6 and S4 as priors and weights for the next frame. To quantify this evaluation, the tracking deviation of the current frame is defined in the formula as... The energy consumption of this frame is defined as And define the switching count as .

[0129] The above three quantities, along with the Boolean flag indicating whether the preset operating condition window has been entered, are written into the buffer as the evaluation result. This buffer is used to adaptively adjust the weighting coefficients and rate limits in the next frame. Prototype data shows that, on eight independent control zone devices, after adopting quadratic programming with rate limits and energy budgets, the median time to enter the preset operating condition window was reduced from the baseline of 3.2 frames to 2.1 frames, the median switching count decreased from 1.8 to 1.0, while energy consumption remained at [value missing]. Below 70%; on long-segment devices with twenty independent control zones, the hierarchical two-stage strategy can still control the median time to reach the target within 3 frames.

[0130] It should be noted that step S7 also includes the following optional implementation methods:

[0131] First, prioritization of zones under resource constraints: When the pumping capacity or energy budget is insufficient to simultaneously satisfy multiple target zones, the target vector or weight coefficients in the quadratic programming are dynamically adjusted based on a weighted score composed of the sparse leakage vector amplitude, the extent of vacuum exceedance in the zone's interlayer cavity, and the temperature deviation of the inner tube's outer wall within the zone, prioritizing the satisfaction of high-risk zones. Second, robust tube constraint: When there is identification uncertainty in the actuator-to-state linear mapping matrix, a state deviation uncertainty set is introduced, and its external estimated radius is used as an additional constraint or cost term to ensure that the boundary is not exceeded even under identification deviation.

[0132] In summary, step S7, through constraint consistency checks, binary selection under Hamming distance constraints, continuous setting of quadratic planning constrained by boundaries, rate and energy budgets, quantization and switching scheduling, security gating and execution masking, and frame-level online evaluation and write-back, forms an online, traceable and engineering-implementable collaborative setting and distribution mechanism, which ensures that the vacuum degree of the partition interlayer cavity and the temperature of the outer wall of the inner tube of the partition are stably entered and maintained within the preset operating condition window.

[0133] In this embodiment, step S8 will be described in detail. Step S8 is used to update the parameters of the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model online, and then apply stability projection constraints to the updated parameters before returning to step S2 for repeated execution. Its core objective is to continuously correct the parameter vectors and covariance matrices of the two types of models under the condition of uninterrupted joint excitation and synchronous demodulation, so that the deviation between prediction and measurement remains controllable, and the numerical stability and engineering usability of the algorithm link are maintained even in the presence of disturbances, noise and local degradation. Step S8 executes a small loop of "constructing regression vector - calculating robust residual - recursive least squares update - stability projection - forgetting factor adaptation - health weight write-back" for each independent control partition, and executes a slow loop of "coupling term sparsification - column vector registration - model order fine-tuning" at the cluster level (hierarchical structure of long-distance devices). In order to clarify each quantity within the formula range, the parameters and regression quantities of the two types of models are first defined. The partitioned heat flow prediction model is defined in the first step. The parameter vector of each independent control zone is denoted in the formula as follows: The parameter vector of the partitioned vacuum conduction prediction model in the k-th independent control partition is denoted as... Let the corresponding regression vector be denoted as The covariance matrix of the partitioned vacuum conduction prediction model is denoted as... The frame-level prediction values ​​output in step S4 are denoted as follows: and The frame-level measurement values ​​output in step S3 are recorded as follows: and .

[0134] Step S8 performs online parameter updates for the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model: First, robust residuals are constructed based on frame-level residuals and health weights. Then, recursive least squares with a forgetting factor are used to recursively update the parameter vector and covariance matrix of each independent control partition. Stability projection is performed on the parameters in each frame to ensure that the parameters satisfy causal mask constraints, sign / amplitude constraints, and model convergence constraints. Step S8 also adaptively adjusts the forgetting factor based on the residual energy, enabling the system to converge quickly during the disturbance phase and remain stable during the steady-state phase. This online update process can be represented by two curves: "the forgetting factor gradually converges over time, and the estimated model parameters tend to stabilize over time," as shown in the figure. Figure 9 As shown.

[0135] Regarding robust residuals and health weights: Step S8 first calculates the frame-level residuals for both models and then uses a robust weighting function to suppress the influence of abnormal frames. Within the formula range, the thermal mode residual is defined as... (Equation 22), the vacuum modal residual is defined as: (Equation 23). Based on the health status identifier formed in step S3, the frame-level health status weight is defined in the formula as follows: (Equation 24), where and It is obtained from the median absolute deviation of the reference frame. The effective residual used for updating is denoted in the formula as follows. , (Equation 25).

[0136] Regarding recursive least squares update with a forgetting factor: Step S8 applies recursive least squares with a forgetting factor to both types of models to deduce parameters and covariance, and adaptively adjusts the forgetting factor in each frame. To avoid sign confusion, different forgetting factors are used for the thermal mode and the vacuum mode. The forgetting factor for the thermal mode is denoted in the formula as follows: The vacuum mode forgetting factor is denoted as Taking thermal modes as an example, the standard update-style writing is as follows:

[0137] (Equation 26)

[0138] (Equation 27)

[0139] (Equation 28);

[0140] in This is the thermal mode gain vector; This represents the projection operator, which projects parameters onto the thermal modal feasible set. The update of the vacuum modes is given in isomorphic form:

[0141] (Equation 29)

[0142] (Formula 30)

[0143] (Equation 31).

[0144] Regarding stability projection and physical priors: Step S8 projects the parameters onto a feasible set to prevent offline physical priors from being corrupted by online data "drift". The feasible set contains three types of constraints: first, causal masking constraints, allowing only terms determined as "physically reachable" in steps S2 and S4 to be non-zero; second, sign and magnitude constraints, corresponding to the sign and upper and lower bounds of parameters such as heat conduction and conductivity terms; and third, stability constraints, applying absolute sum constraints to the autoregressive part of the partitioned heat flow prediction model to ensure convergence. Taking thermal modes as an example, if the autoregressive coefficient subvector is denoted in the formula as... Its projection can be calculated according to (Equation 32), where This is a constant not exceeding 1, used to ensure bounded input and bounded output. The conduction-dependent parameter subvector of the vacuum mode is denoted in the formula as... Non-negative and upper bound projections can be applied: (Equation 33). The above projection operations are all included in the projection operators of equations (27) and (30).

[0145] Regarding the adaptive adjustment of the forgetting factor: To adaptively compromise between perturbation and steady state, step S8 adjusts the forgetting factor according to the residual energy ratio of the nearest window and projects it onto the closed interval to ensure numerical stability. This is given in the formula:

[0146] , (Equation 34);

[0147] (Equation 35);

[0148] (Equation 36);

[0149] The overline indicates a moving average; and It is the energy normalization constant; and For smoothing coefficients; It is an interval projection function; and The preferred values ​​are 0.97 and 0.995, respectively.

[0150] Regarding coupling term sparsification and column vector registration: For applications involving long-distance devices or hierarchical coding, step S8 performs coefficient sparsification and column vector registration at the cluster level to reduce overfitting and maintain cross-modal consistency. The coupling term coefficient sub-vectors for the thermal mode and the vacuum mode are denoted in the formula as follows: and Perform group sparsification:

[0151] (Equation 37);

[0152] in Simultaneously, utilizing the response matrix column vector registration concept from step S4, the formula is as follows:

[0153] Using Equation 38 as the target, the small-step correction criterion periodically fine-tunes the scaling factor vector to ensure that the thermal mode and the vacuum mode maintain consistent sensitivity to the same "leakage location". This slow loop is executed every few tens of frames without affecting intra-frame real-time performance.

[0154] Regarding model order fine-tuning and soft reset, step S8 provides online information criteria to fine-tune the order, avoiding the accumulation of systematic errors caused by a fixed order when operating conditions change. The information criteria are defined as follows: (Equation 39), where For candidate order, For sample size, This represents the complexity penalty coefficient. Every few tens of frames, an attempt is made to increase or decrease the order by one within a narrow neighborhood. If the information criterion improves, it is retained; otherwise, it reverts to the original order. To handle extreme perturbations, step S8 sets a soft reset. When the condition number of the covariance matrix exceeds a threshold or the residual energy continuously exceeds the limit, a soft reset is performed. (Equation 40) performs a gentle return to normal, where and For initialization parameters, It uses a small weight. This mechanism can quickly restore stability without losing all learning outcomes.

[0155] Regarding write-back and closed-loop coupling: At the end of each frame, step S8 writes the latest parameter vectors, covariance matrix diagonals, forgetting factor, health weights, sparsification results of coupling terms, and registration ratio coefficients of both models into the buffer. It also feeds back weight suggestions and credibility scores to step S6, and small-step correction suggestions for the "actuator-to-state" linear mapping matrix to step S7. All write-back records include timestamps, encoded frame numbers, and synchronization markers, supporting subsequent playback and comparison.

[0156] In one specific embodiment, eight independently controlled partition prototypes were used with a frame length of 32, a sampling frequency of 200 Hz, a forgetting factor range of [0.97, 0.995], and soft reset parameters. After running for two hours with a setting of 0.05, the median goodness-of-fit of the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model reached over 0.93 and 0.90, respectively. In the sudden disturbance (five-fold amplitude step) scenario, after adopting adaptive forgetting and stability projection, the median number of frames required to recover to 90% of the goodness-of-fit level before the disturbance was reduced from eight frames at the baseline to three frames. In the mild actuator degradation (valve position viscosity simulation) scenario, the health weight reduced the median parameter drift by about 40%, and no covariance divergence occurred. Hierarchical operation verification was performed using long-segment prototypes with twenty independent control partitions. Coupling term sparsification and column vector registration reduced the condition number of the inverted column space by about 30%, maintaining the convergence of subsequent sparse solutions.

[0157] It should be noted that step S8 also includes:

[0158] Partitioned cluster-level shared prior: Within the same cluster, a set of prior means and upper bounds of covariance of the coupling kernel are shared. By small-step correction and projection binding, the number of parameters is reduced and statistical significance is improved.

[0159] Uncertain domain robust projection: Introducing a “mapping uncertainty radius” into the feasible set restricts the allowed parameter perturbations to a bounded set centered on offline identification, which is equivalent to superimposing a “radius upper limit” on the projection steps after equations (27) and (30).

[0160] In summary, step S8, through robust residual construction, recursive least-squares update with a forgetting factor, stability and physical prior projection, adaptive forgetting adjustment, sparsification of coupling terms and column vector registration, model order fine-tuning and soft reset, and frame-level write-back for upstream and downstream steps, constitutes the adaptive learning closed loop of the "vacuum insulation pipe control method based on adaptive heat flow prediction algorithm". This ensures that the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model maintain prediction accuracy, numerical stability and physical consistency during long-term online operation.

[0161] Step S0 will be described in detail in this implementation. It is used to construct a stable, traceable, and identifiable operating baseline and to provide a quantified, consistent starting point for steps S1 to S8. Step S0 covers vacuum pumping and degassing pretreatment, parallel system and on / off calibration of reference capillaries, full-link sensor calibration, linearization acquisition of actuator-state mapping, establishment of encoding and timing synchronization baseline, joint noise statistics and gating safety domain setting, initial column vector construction of response and control matrices, online update and initial domain setting of projection constraints, etc. In the encoding and timing synchronization baseline establishment stage, step S0 loads a set of binary encoded reference frames to synchronously trigger the switching state of the switchable valve network and the power step of the heating component, and introduces synchronization markers to align frame boundaries at the acquisition end, thereby ensuring that subsequent identification and control are executed under the same time reference; the relationship between this binary encoded channel and the synchronization marker on the time axis is illustrated by a set of stepped waveform sequences appearing in parallel and maintaining a fixed relative phase, such as... Figure 2 As shown.

[0162] Firstly, regarding the vacuum evacuation and degassing pretreatment, step S0 uses a combination of staged evacuation and temperature-rise degassing to establish a low outgassing baseline. In this embodiment, a rough vacuum pump is first used to evacuate the partitioned interlayer cavity of the vacuum insulation pipe to... The pressure rise was then measured in Pa increments. A high-vacuum pump was then switched on, and the outer jacket wall temperature was gradually increased to the 50–70 °C range over four hours to accelerate desorption, until the pressure rise rate of the partitioned interlayer cavity stabilized below 0.5 Pa / min. For subsequent quantification, the total degassing time is defined below within the formula range as... The pressure rise rate is defined as... Define the allowed upper limit as The degassing criterion is expressed by the standard mathematical expression as follows: , ,in and , The minimum degassing time is (preferably not less than 2 hours).

[0163] Regarding the parallel system and on / off calibration of the reference capillary, step S0 involves setting a detachable reference capillary interface in each independent control zone. Preferably, a metal-sealed VCR connection is used, and a torque-limiting nut is configured to ensure consistency during repeated assembly. In this embodiment, two combinations of equivalent conductance of the reference capillary are selected to enhance the calibration dynamic range: the lower equivalent conductance is denoted in the formula as... (typical value) The high-grade equivalent flux is denoted in the formula as... (typical value) Both are connected to the same reference capillary interface via a micro-switching valve and are switched on and off by the coded switching of the switchable valve network. To obtain the column vector magnitude and time constant of the vacuum modal response matrix, step S0 sequentially executes a four-step on / off sequence of "off – low-level on – off – high-level on" in each independent control zone, and records the frame-level response of the vacuum degree of the corresponding zone's interlayer cavity and the vacuum degree of the adjacent zone's interlayer cavity. To formalize the construction of this column vector, within the formula range, the first... The steady-state residual signature generated by the on / off state of the reference capillary in each independent control zone is denoted as follows: The vacuum modal response matrix is ​​denoted as Then the column vector construction is written as ,in This indicates the number of independently controlled partitions (used only in formulas).

[0164] Regarding the end-to-end calibration of the sensors, step S0 performs joint calibration of the heat flow sensor, temperature sensor, and vacuum sensor using zero-point, sensitivity, and cross-sensitivity methods. For the heat flow sensor, step S0 applies a known step power to the outer wall of the inner tube within the partition, records the heat flow through the partition tube wall and the temperature of the outer wall of the inner tube within the partition, and obtains the gain and bias of the heat flow sensor using a least-squares method; within the formula range, the original output sequence of the heat flow sensor is denoted as... The step power of the heating component is denoted as The calibrated heat flow is recorded as Its linear relationship is written as ,in For the heat flow sensor gain, For bias. For the vacuum sensor, step S0 uses a combination of a thermal conductivity vacuum gauge and a cold cathode vacuum gauge to construct a segmented, smooth fused reading; within the formula range, the thermal conductivity vacuum gauge reading is recorded as... Record the cold cathode vacuum gauge reading as The fusion threshold is denoted as , Integration of readings in writing:

[0165] ;

[0166] in The weights are generated by the interpolation function. This is the lower limit switching point. Prototype data shows that, within the 0.5–10 Pa range, the relative error between the vacuum level of the fused partitioned sandwich cavity and the reference gauge is better than ±5%. For the temperature sensor, step S0 performs linear calibration and records the calibration coefficient using two-point temperatures (ambient temperature and low-power steady-state temperature of the heating element) to eliminate sensor channel differences.

[0167] In obtaining the linearization of the actuator-state mapping, step S0 injects small-amplitude step-slope excitations into the partitioned extraction valve, partitioned replenishment valve, partitioned bypass microfluidic valve, and heating component, respectively. The state deviation, composed of "vacuum degree of the partitioned sandwich cavity – temperature of the outer wall of the partitioned inner tube," is measured. Under linear approximation, the initial value of the linear mapping matrix from actuator to state is obtained through least squares solution. To represent this in the formula, the state deviation vector is denoted as... The control increment vector is denoted as Let the linear mapping matrix from the actuator to the state be denoted as Then there is .

[0168] Regarding the establishment of the coding and timing synchronization baseline, step S0, without triggering overheating or over-sampling, first loads a set of binary coded reference frames to check orthogonality, Hamming distance constraints, and phase consistency. For quantization, the reference frame length is denoted as [length] within the formula range below. Let the set of binary encoded reference sequences be denoted as The normalized correlation between sequences is denoted as Its acceptance criteria are for any have Simultaneously, the switching count of the switchable valve network of adjacent symbols is checked to ensure that the Hamming distance does not exceed 1. On the prototype, a reference frame length of 32 is used, and the following statistics are collected: The maximum value is 0 or 1 / 32, and the phase drift does not exceed 5% of the symbol duration within one hour, satisfying the consistency of identification and control timing in subsequent steps.

[0169] Regarding the joint noise statistics and gated safety domain setting, step S0, under the self-test state of "all switchable valve networks in isolation and heating component power switches at zero," collects multiple frames of zoned pipe wall heat flux, zoned inner pipe outer wall temperature, zoned outer sleeve inner wall temperature, and zoned interlayer cavity vacuum degree. It estimates the frame-level noise standard deviation and sets the scale factor and covariance shaping window for step S5 accordingly, while simultaneously calculating the upper and lower thresholds of the gated safety domain. For clarity, the thermal modal noise standard deviation will be denoted as [insert standard deviation here] within the formula range. The standard deviation of vacuum modal noise is denoted as The upper and lower limits of the safe vacuum level of the partitioned interlayer cavity are respectively denoted as: and The upper limit of the safe temperature of the outer wall of the pipe in the zone is denoted as: Then the gating criterion writing If the condition is not met, the gating mask is triggered and a pause instruction is sent back to steps S2 and S7.

[0170] Regarding the initial column vector construction of the response matrix and control matrix, in step S0, in addition to constructing the vacuum mode response matrix by referencing the capillary on / off state, the column vector of the thermal mode response matrix is ​​also constructed by a small-power heating step. For clarity, the following will refer to the first... The steady-state thermal mode residual signature caused by a small-power step in each independent control zone is denoted as follows: The thermal modal response matrix is ​​denoted as Its column vector construction is written as .

[0171] Regarding the initial domain setting for online updates and projection constraints, step S0 sets the initial parameter vector, initial covariance matrix, initial forgetting factor, and upper bound of the projection feasible region for the recursive least squares algorithm for the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model, respectively. For consistency, the initial parameter vector of the thermal mode will be denoted as follows within the formula range: Let the initial parameter vector of the vacuum mode be denoted as The initial covariance matrix of the thermal mode and the vacuum mode is denoted as... and Let the initial forgetting factor be denoted as (Preferred value 0.985), the upper bound vectors of the feasible parameter sets for the thermal mode and the vacuum mode are respectively denoted as... and Then the initialization relation is written as , , , , And establish the projection operator ,in and This is the lower bound vector. Two hours of prototype operation data show that, after adopting this initial domain setting, the online update in step S8 converges to a stable interval within twenty frames, and no covariance divergence phenomenon is observed.

[0172] It should be noted that step S0 also includes: firstly, an adaptive selection mechanism for dual reference capillaries, that is, in the online stage, a more suitable reference capillary level is automatically selected based on the energy of the joint normalized residual of the current frame vacuum mode to perform short-cycle recalibration, thereby maintaining the amplitude consistency of the vacuum mode response matrix; secondly, a polarity-phase joint search mechanism for encoded reference frames, that is, in each reinitialization, the polarity and phase of the binary encoded reference sequence are discretely searched, and the combination that minimizes the cross-correlation sidelobes is selected and saved as a new version of the reference frame.

[0173] In summary, step S0 establishes a unified starting point for the "vacuum insulation tube control method based on adaptive heat flow prediction algorithm" through vacuum pumping and degassing pretreatment, reference capillary parallel system and on / off calibration, sensor full-link calibration, actuator-state mapping linearization, encoding and timing synchronization baseline, joint noise statistics and gating domain setting, response matrix and control matrix column vector construction, and online update and projection constraint initial domain setting. This ensures that subsequent steps S1 to S8 can operate stably on a consistent numerical scale, an identifiable excitation-measurement link, and a traceable data baseline.

[0174] It should be noted that the embodiments of the present invention have better implementability and are not intended to limit the present invention in any way. Any person skilled in the art may use the above-disclosed technical content to change or modify it into equivalent effective embodiments. However, any modifications or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A method for controlling vacuum insulation pipes based on an adaptive heat flow prediction algorithm, characterized in that, Includes the following steps: Step S1: Divide the vacuum heat insulation pipe into N independent control zones along the axial direction. A zoned sandwich cavity is formed between the inner tube and the outer tube of each independent control zone. A monitoring unit and an execution unit are set in each independent control zone. The monitoring unit includes a heat flow sensor, a temperature sensor and a vacuum sensor. The execution unit includes a heating component, a zoned air extraction valve, a zoned air replenishment valve and a zoned bypass microfluidic valve. Step S2: Construct a switchable valve network, which is formed by valve groups located at both ends of each independent control zone. During the sampling period, the connection relationship of the valve network and the power switch of the heating component are switched according to a predetermined binary encoding sequence to form a zone-identifiable joint excitation. Step S3: Synchronously collect zoned data, which includes zoned pipe wall heat flow, zoned inner pipe outer wall temperature, zoned outer pipe inner wall temperature, and zoned interlayer cavity vacuum degree, and generate thermal mode measurement vectors and vacuum mode measurement vectors respectively. Step S4: Establish a zoned heat flow prediction model and a zoned vacuum conduction prediction model. The zoned heat flow prediction model uses the historical data of the joint excitation and the thermal mode measurement vector as input and output mapping, and the zoned vacuum conduction prediction model uses the historical data of the joint excitation and the vacuum mode measurement vector as input and output mapping. Step S5: Generate thermal residual vector and vacuum residual vector based on the predicted output and measured output of the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model, and perform joint normalization processing on the thermal residual vector and the vacuum residual vector to obtain joint normalized residual. Step S6: Using the joint normalized residual as the observation, solve the sparse leakage vector to obtain the target partition set, and generate the partition control instruction corresponding to the target partition set; Step S7: Send the partition control command to the execution unit of the target partition to coordinate the partition air extraction valve, the partition air replenishment valve, the partition bypass microfluidic valve and the heating component to make the vacuum degree of the partition interlayer cavity and the temperature of the outer wall of the partition inner tube meet the preset working condition window. Step S8: Update the parameters of the partitioned heat flow prediction model and the partitioned vacuum conduction prediction model online, apply stability projection constraints to the updated parameters, and then return to step S2 for repeated execution.

2. The method for controlling vacuum insulation pipes based on adaptive heat flow prediction algorithm according to claim 1, characterized in that, The predetermined binary encoding sequence in step S2 is a Walsh encoding sequence or an equivalent orthogonal binary sequence with a length of 16 to 64. Each bit of the encoding sequence drives the connection state of the switchable valve network and the power switch of the heating component, respectively.

3. The method for controlling vacuum insulation pipes based on adaptive heat flow prediction algorithm according to claim 1, characterized in that, The joint excitation in step S2 also includes applying a linear sweep frequency microfluidic signal to the partitioned bypass microfluidic valve, with a sweep frequency range of 0.05 Hz to 2 Hz, and synchronizing with the binary power sequence of the heating component, so as to obtain mutually independent thermal modal response and vacuum modal response in step S3.

4. The method for controlling vacuum insulation pipes based on adaptive heat flow prediction algorithm according to claim 1, characterized in that, The solution for the sparse leakage vector in step S6 satisfies the following equation: in, A sparse leakage vector of dimension N; The response matrix is ​​obtained by the partitioned heat flow prediction model; The response matrix is ​​obtained by the partitioned vacuum conduction prediction model. Let be the thermal residual vector; Let be the vacuum residual vector; and Positive scalar weights; and Let them represent the 2-norm and the 1-norm, respectively.

5. The method for controlling vacuum insulation pipes based on adaptive heat flow prediction algorithm according to claim 4, characterized in that, The response matrix and The online parameter update employs a recursive least squares method with an adaptive forgetting factor. The forgetting factor is updated in each sampling period according to the following formula and projected onto the closed interval 0.97, 0.995: in, The current forgetting factor; For a fixed weight constant, satisfying ; It is a scalar function obtained based on the joint normalized residual; A projection operator that limits the input value to the interval 0.97, 0.

995.

6. The method for controlling a vacuum insulation pipe based on an adaptive heat flow prediction algorithm according to claim 1, characterized in that, The cooperative setting in step S7 is obtained by solving the constrained actuator allocation problem, which satisfies the following equation: in, It is a control vector that includes the zone vacuum pump speed setting value, the zone extraction valve opening degree, the zone replenishment valve opening degree, the zone bypass microfluidic valve opening degree, and the heating component power setting value; This is the linear mapping matrix from the actuator to the state; The target vector is determined by the sparse leakage vector and the preset operating condition window; It is a positive scalar; and Define the upper and lower bounds of the vector; Let m be the binary selection vector of the switchable valve network.

7. The method for controlling a vacuum insulation pipe based on an adaptive heat flow prediction algorithm according to claim 4, characterized in that, Before performing step S1, an initialization step S0 is included, in which a reference capillary is connected in parallel to each of the independent control zones, and the equivalent conductivity of the reference capillary is... And under the encoding switch in step S2, the on / off calibration of the reference capillary is performed. Used for the response matrix The column vectors are used for magnitude scaling.

8. The method for controlling a vacuum thermal insulation pipe based on an adaptive heat flow prediction algorithm according to claim 1, characterized in that, Prior to step S3, time alignment is performed on the data collected from the partition, and the partition delay parameter for time alignment is... Determine using the following formula: in, In order to delay Next Multi-channel predicted sequences for each partition The corresponding multi-channel measured sequence; Represents the set of real numbers; This indicates the vector in the channel dimension. and The sum of the products of each term.

9. The method for controlling a vacuum insulation pipe based on an adaptive heat flow prediction algorithm according to claim 1, characterized in that, The joint normalization process in step S5 includes applying two-dimensional adaptive filtering along the time index and partition index to the thermal residual vector and the vacuum residual vector respectively, and normalizing the amplitude using their respective calibration standard deviations. The time window length of the two-dimensional adaptive filtering is 32 to 128, and the partition window length is 3 to 7.

10. The method for controlling a vacuum insulation pipe based on an adaptive heat flow prediction algorithm according to claim 6, characterized in that, The constrained actuator allocation problem allows the binary selection vector s of the switchable valve network to switch at most once with a Hamming distance of 1 in each sampling period, and discretizes the opening degree of the partitioned bypass microfluidic valve into a four-level set {0, 0.25, 0.5, 0.75}, and the power setting value of the heating component into a four-level set {0, 0.2, 0.5, 0.8}, with the unit being the relative rated value.

Images