Indoor temperature real-time regulation method of heat pipe and control system thereof

Abstract

The application discloses a kind of indoor temperature real-time regulation methods of heat pipe and control system thereof, it is related to heat pipe network dynamic control technical field, solve the problem of hydraulic oscillation and temperature control delay caused by the difficulty of describing multi-order thermal inertia and large thermal capacity in the prior art local valve adjustment ignores network coupling and first-order linear model, the scheme is based on pipe network distributed PDE / lumped parameter hybrid modeling and combined with extended Kalman filter and unscented Kalman filter online identification, using spectral decomposition to generate feedforward decoupling compensation term, construct adaptive multi-model predictive control and iterative learning compensation closed-loop structure, and issue control command according to first pump valve serialization strategy, while dynamically adjusting model weight and prediction time domain;The application significantly improves the global balance ability and temperature tracking precision of heat pipe regulation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dynamic control of heat pipe network, and more particularly to a real-time indoor temperature regulation method for heat pipe and a control system thereof. BACKGROUND

[0002] The real-time indoor temperature regulation of central heating pipe network mainly sends feedback signals into PID, self-correcting fuzzy or dynamic matrix prediction controllers through end temperature sensors, return water temperature and pipe network flow measurement, to drive valve opening and pump speed to realize temperature tracking. In recent years, edge-cloud collaboration and artificial intelligence algorithms have also been introduced to strike a balance between energy saving and comfort, but the long lag, strong coupling and multi-node characteristics of the pipe network pose higher challenges to control accuracy and real-time performance.

[0003] Existing real-time indoor temperature regulation systems mostly rely on the collaboration of modular hardware and online computing platforms. For example, patent CN222783576U installs temperature control valves and drivers on the household pipe, dynamically adjusts the valve core opening through indoor temperature sensor feedback, and realizes household heat monitoring and real-time regulation in combination with heat meters. Its advantages lie in the convenience of deployment and the integration of household metering. Another typical solution, as described in CN119642254A, uses an edge-cloud collaboration architecture, in which the heat exchange station gateway uploads historical weather and indoor temperature data in real time, the cloud platform trains a heating control model based on LSTM neural network algorithms, and then optimizes parameters to be sent to the local heat exchange station for execution. This method can capture load characteristics and perform rolling iterative optimization with the support of big data. In addition, there are dynamic heat dissipation rate adjustment methods based on building thermal parameters (CN119267994A) and indoor temperature feedback linear regression regulation schemes (CN113390126B), which respectively start from the single building model and the relationship between heat supply and indoor temperature mean value of the heat station, and correct the supply water temperature through simple mathematical models. These technologies are effective in small-scale or simulation scenarios, but often focus on the fine design of local valve characteristics or single load models, and lack the ability to model and control global dynamic characteristics (hydraulic strong coupling, thermal large inertia) in complex systems:

[0004] For example, patent CN108916984B proposes to manage branch valves according to heating distance to eliminate horizontal imbalance, but this local flow distribution scheme ignores the mutual restraint between upstream pressure and partition in the pipe network, and distributed regulation often causes pump-valve linkage conflict and hydraulic oscillation, leading to difficulty in maintaining overall balance of the pipe network under load fluctuation, and further causing temperature interference and control quality degradation. Secondly, CN113390126B constructs a linear regression relationship between water supply temperature and average room temperature, but the first-order linear model cannot depict the multi-order lag and large thermal capacity characteristics exhibited by long-distance pipelines and floor heating systems, and the controller output and actual temperature response are severely mismatched, often causing obvious hysteresis, overshoot and oscillation, which not only weakens user comfort, but also causes energy waste and aggravates the wear of the actuator. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application discloses a real-time indoor temperature regulation method for a heat pipe and a control system thereof, aiming to solve the problems in the background art.

[0006] In order to achieve the above technical effects, the present application adopts the following technical solutions:

[0007] A real-time indoor temperature regulation method for a heat pipe, comprising:

[0008] Step 1, based on the node temperature, pressure and flow data collected by the pipe network in real time, an initial full-network model containing second-order thermal inertia and hydraulic coupling terms is constructed by using PDE / lumped parameter hybrid method , and an initial state equation set is output;

[0009] Step 2, using the and sensor data as input, the water impedance and thermal capacity parameters are identified online by alternately using extended Kalman filter and unscented Kalman filter, and an updated model is obtained;

[0010] Step 3, based on the parameter vector and state equation set of the updated model , the network hydraulic coupling coefficient matrix and thermal coupling coefficient matrix are calculated, and the corresponding feedforward decoupling compensation term is generated;

[0011] Step 4, according to the parameter vector of the feedforward decoupling compensation term and , a multi-model prediction framework containing a long-distance pipeline model and an end partition model is constructed, and a weighted prediction model group is output by a model pool weight scheduling method;

[0012] Step 5, by using an online quadratic programming method, the initial control sequence is generated by solving a nonlinear optimization problem containing temperature error, energy consumption and actuator constraint according to , upper and lower limits of the actuator and future load prediction curve;

[0013] Step 6, generating actual pump speed and valve opening control instructions according to the initial control sequence and feedforward decoupling compensation term in the sequence strategy of pump first and valve second combined with dead zone and rate limit;

[0014] Step 7, after control execution, taking temperature residual and flow oscillation data of the previous time as input, dynamically adjusting multi-model weight and prediction time domain length, and outputting updated control configuration parameters.

[0015] As a further technical solution of the present application, a real-time indoor temperature regulation and control system for a heat pipe, comprising:

[0016] A data acquisition module is configured to acquire temperature, flow, pressure and valve position original signals of each heat exchange station and end room, and complete timestamp alignment according to a unified clock to obtain a standardized time series data set ;

[0017] A time series preprocessing module is configured to perform wavelet packet multi-scale filtering and Bayesian interpolation processing on , separate noise and fill in missing values, and obtain a smoothed and denoised time series ;

[0018] A parameter modeling module is configured to map to a pipe network topology and construct a full-network model containing second-order thermal inertia and hydraulic coupling based on a PDE / lumped parameter hybrid strategy ;

[0019] An online identification module is configured to take the interpolated and smoothed and as input, correct water impedance and node thermal capacity parameters through extended Kalman filtering and unscented Kalman filtering alternately, and obtain an updated model ;

[0020] A coupling decoupling module is configured to perform spectral decomposition on the network hydraulic and thermal coupling matrix in , extract subnetwork coupling subspace and generate a feedforward decoupling operator ;

[0021] A predictive control module is configured to construct parallel multi-models in a rolling time domain based on , operator and future load prediction sequence, and solve an initial control vector through linear quadratic programming ;

[0022] An execution monitoring module is configured to issue in the sequence strategy of pump first and valve second, and real-time collect room temperature residual and flow deviation sequence ;

[0023] a compensation iteration module configured to run an iterative learning algorithm with and historical residual library as input to generate bias compensation increment and output a revised control vector ;

[0024] a closed-loop coordination module configured to jointly input , and sub-network interface error into an augmented Lagrangian iteration process to update model parameters and MPC weights and output the next period configuration.

[0025] Based on the above technical solutions, the positive and beneficial effects of the present application are:

[0026] 1. By means of the whole-network PDE / lumped parameter hybrid modeling and online recursive identification, the hydraulic and thermal coupling characteristics of the pipe network are captured in real time and dynamically corrected, so as to maintain the global balance of the upstream pressure and the flow of each sub-zone in the case of severe load fluctuations, and eliminate the pump-valve linkage conflicts and hydraulic oscillations caused by local valve adjustment.

[0027] 2. With the aid of the feedforward decoupling compensation term generated by spectral decomposition and the incremental multi-model predictive control, the multi-order thermal inertia and large thermal capacity effect are simultaneously compensated in the prediction time domain, so that the hysteresis and overshoot of the temperature response are fundamentally inhibited.

[0028] 3. The online quadratic programming, under the premise of considering the upper and lower limits and rate limits of the actuators, unifies the temperature error, energy consumption and actuator constraints into the optimization framework, so that the control sequence is highly matched with the actual dynamics of the pipe network, thereby reducing the mismatch between the controller output and the field response.

[0029] 4. The adaptive weight scheduling and prediction time domain adjustment based on residual monitoring form a closed-loop feedback, and the pipe network can continuously update the model and control parameters under different working conditions, so as to maintain stable temperature control and effectively delay the wear of the actuator in a complex system. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor under the premise of not paying creative labor.

[0031] Figure 1 is the architecture diagram of the indoor temperature real-time regulation method of the heat pipe of the present application;

[0032] Figure 2 is the working method flowchart framework diagram of step 1 of the present application;

[0033] Figure 3 The calculation method step chart for step 3 of the present application;

[0034] Figure 4 The indoor temperature error and control signal change schematic diagram of the present application;

[0035] Figure 5 The indoor temperature real-time regulation control system principle diagram of the heat pipe of the present application. DETAILED DESCRIPTION

[0036] In order to make the purposes, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions of the present application will be described clearly and completely below in combination with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application.

[0037] In order to make the purposes, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions of the present application will be described clearly and completely below in combination with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application. Figure 1 For the convenience of understanding the embodiments, first, a heat pipe indoor temperature real-time regulation method disclosed by the embodiments of the present application will be introduced in detail, please refer to the step schematic diagram of the heat pipe indoor temperature real-time regulation method shown in

[0038] Step 1, based on the node temperature, pressure and flow data collected by the pipe network in real time, using PDE / lumped parameter hybrid method to construct the initial whole network model containing second-order thermal inertia and hydraulic coupling term , and output the initial state equation group; specifically, as Figure 2As shown, the system first relies on three types of high-precision sensor networks to continuously monitor the operation state of the pipe network. First, pressure sensors are arranged along the main trunk line and other key nodes. Typical models can be selected from pressure resistance or capacitance sensors with a range of 0~2MPa and an accuracy of 0.1%FS. The sampling frequency is generally between 10Hz and 0.1Hz, which can be flexibly adjusted according to system response requirements. The specific value can be determined by the actual situation, and no limitation is made. The pressure gradient distribution in the region is obtained by the difference between adjacent measuring points, which is used to judge the fluid driving and damping characteristics. Second, flow meters are installed at branch outlets or branch nodes, usually using vortex or electromagnetic schemes. When the flow fluctuation amplitude per unit time is lower than the critical value of turbulent flow (generally taking the lower limit of flow velocity corresponding to Re=2300, about 0.2m / s, the threshold value of flow under the condition of pipe diameter and temperature can be determined by the actual situation, and no limitation is made). The dynamic of the branch can be classified into the lumped parameter domain. Third, distributed temperature sensing uses fiber Bragg grating (FBG) or distributed optical fiber scattering (DAS) technology, and sets measuring points every 0.5m to 5m along the pipeline to capture the axial temperature gradient field; the measuring point resolution and spacing can be selected in combination with the thickness of the insulation layer and the expected temperature gradient sensitivity, and the specific value can be determined by the actual situation, and no limitation is made. Then, the original signal needs to be filtered and denoised, among which the pressure and flow signals are preferably denoised by wavelet threshold to eliminate high-frequency interference; temperature data can be smoothed by using moving average or Kalman filter. All data must be clock-aligned, and PTP (Precision Time Protocol) or IEEE1588 precise synchronization protocol is recommended to achieve microsecond-level timestamp consistency and ensure the correspondence accuracy of space-time information in the subsequent hybrid model.

[0039] After data acquisition is completed, the central processing unit reads the continuous time series of pressure gradient distribution sequence, and introduces the Morris sensitivity analysis algorithm to dynamically partition the pipe network hydraulic domain. The so-called Morris sensitivity analysis is a factor-by-factor screening method based on parameter perturbation. The sensitivity index of pressure gradient change is calculated by orderly perturbing the hydraulic parameters such as pipe diameter, length, and flow resistance coefficient of each pipe section (for example, the relative perturbation amplitude is set to ±10%). If the sensitivity index of a pipe section or node group exceeds the preset threshold value, it is considered that the hydraulic dynamic of the section has a significant coupling effect on the whole network, and should be classified into the PDE domain; otherwise, it is classified into the lumped parameter (Lumped) domain. The threshold value S_{thresh} can be adjusted according to the size of the pipe network, the distribution of pipe diameter, and the fluctuation amplitude of load, and the specific value can be determined by the actual situation, and no limitation is made.

[0040] At the hydraulic solution level, the core of the implementation plan using the PDE / lumped parameter hybrid domain method lies in dynamic domain partitioning. First, Morris global sensitivity analysis is performed on the geometric parameters (inner diameter D, length L, roughness ε) and current flow velocity u of each pipe segment in the pipeline network to assess the contribution of that segment to local pressure gradient changes. When the sensitivity index exceeds a preset threshold (e.g., the normalized sensitivity index exceeds 0.1, which can be flexibly set based on the total number of nodes in the pipeline network and the expected computational load; the specific value can be determined according to the actual situation without limitation), that segment is dynamically assigned to the PDE solution domain; otherwise, it is considered to have limited impact on the overall pressure distribution and is assigned to the Lumped domain. The second-order thermal inertia term, which is reflected in the subsequent coupled state equations as an additional term related to heat capacity and flow velocity acceleration, does not participate in the domain partitioning criterion. The main numerical solution within the PDE domain relies on the method of characteristics to solve the one-dimensional transient flow conservation equation, which is simplified as follows:

[0041]

[0042] in Let represent the pipe cross-sectional area, u be the flow velocity, h be the head, g be the gravitational acceleration, f be the friction factor, and D be the pipe diameter. The method of characteristics (MMR) maintains the physical integrity of information propagation by integrating along the curve in both positive and negative characteristic directions. Furthermore, it only requires storing a few points along the characteristic line, enabling high-precision reconstruction of the main flow within a second-level timescale. To ensure solution stability, the time step Δt and spatial step Δx must satisfy the CFL condition: In addition, when high-frequency oscillations or numerical oscillations occur within the PDE domain, the TVB (Total Variation Bounded) limiter can be enabled to control the fluctuations. However, this strategy, like the CFL condition, can be adjusted automatically according to the system's real-time response requirements and is not limited.

[0043] The lumped domain is modeled using a nodal method to construct an equivalent resistance-capacitance network model, whose basic equations are:

[0044]

[0045] p represents the equivalent node pressure, Q is the flow input, R is the hydraulic impedance, corresponding to the pressure drop ratio caused by coal, water quality, and pipe wall conditions; C is the capacitive coefficient, reflecting the momentum energy storage characteristics of the water body and pipe wall. R and C can be obtained from historical calibration conditions or on-site testing. The model can be obtained through experimental calibration or online identification using on-site sampling data during the initial stage of system operation. This lumped model only needs to handle the flow and pressure relationships between nodes, without requiring spatial discretization, and its computational cost is far lower than that of PDE solutions.

[0046] To achieve tight coupling of the hybrid domains, the system monitors the pressure p_b and flow rate Q_b at the boundary nodes of the PDE domain and the Lumped domain simultaneously and sets the synchronous trigger condition: when , the boundary variables are immediately injected into the opposite model through the data interface to correct it. and The values of p_sync and Q_sync can be determined based on the maximum operating pressure level of the pipe network and the branch flow rate, for example p_sync can be set in the range of 0.005 MPa to 0.02 MPa, and Q_sync can be set in the range of 2 L / min to 10 L / min, which can be determined by the actual situation and is not limited. If the boundary deviation continues to exceed N_sync time steps (usually 3-5 steps), the system will trigger the re-division mechanism to re-evaluate the sensitivity of the pipe segment or region and possibly adjust its domain affiliation to ensure that the model remains consistent with the real state.

[0047] The thermal modeling module uses the axial temperature field T(x, t) obtained by distributed temperature sensing as input and uses Karhunen-Loève decomposition to extract its dominant spatial modes. Specifically, the temperature field is treated as a random process, and by solving the Fredholm integral equation of its spatial covariance function , the eigenfunction and the corresponding eigenvalue are obtained, and the high-dimensional temperature field is reconstructed as:

[0048]

[0049] where = r is the mode truncation number, and the cumulative energy of the current r-th mode reaches 95% or more when the truncation is stopped. It can also be adjusted in combination with the pipe insulation performance and the expected temperature response time, which can be determined by the actual situation and is not limited. After modalization, the thermal dynamics are compressed to an r-dimensional state space, greatly reducing the calculation pressure.

[0050] Online parameter correction relies on the instantaneous heat flux q_w(t) measured by the pipe wall heat flow meter, which is compared with the model-predicted heat flux . When , the parameter update subroutine is triggered, and the pipe wall specific heat C_w and the heat transfer coefficient h are iteratively optimized by extended Kalman filtering or recursive least squares method. q_w can be set between 5 W / m² and 20 W / m², and the specific value varies with the pipe diameter and insulation thickness, which can be determined by the actual situation and is not limited; if the deviation exceeds the limit for M_sync consecutive sampling periods (usually M_sync = 3-5), the filter convergence step can be increased or switched to a suboptimal model update mode to speed up the adjustment.

[0051] In terms of software and hardware architecture, PDE domain dynamic partitioning and solver are deployed on edge computing units, using multi-core CPUs or FPGAs to obtain millisecond-level response; KL decomposition and online parameter identification run on central servers or clouds, combining parallel computing resources for efficient processing. Each module exchanges data through real-time industrial bus protocols such as OPCUA, MQTT or DDS, interfacing with field DCS or SCADA systems. Under real working conditions, whether it is a sudden change in the main pipe transient flow or a sudden cold and hot load disturbance, the system can complete model updating and state estimation within multiple control cycles (usually no more than ten seconds), providing accurate support for subsequent model predictive control (MPC) or optimization scheduling.

[0052] Step 2, using the step 1 in and sensor data as input, alternately using extended Kalman filter and unscented Kalman filter for online identification of water impedance and heat capacity parameters, obtaining updated model ; In the process of online identification of water impedance and node heat capacity parameters, the system first uses the coupled water-thermal state model generated in step one and the real-time collected pressure, flow and temperature data to build a complete state-observation mapping relationship. After low-pass filtering and Kalman smoothing processing, the sensor network is synchronized with the pressure difference, branch flow pulsation and distributed temperature modal coefficient at equal interval time windows (for example, approximately five to ten control periods) to form a smooth observation sequence. The model prediction module uses the last updated parameter value as the initial input to drive the water-thermal numerical simulation, generating a predicted output matching the current observation time step. The difference between the two is the high-dimensional residual vector, and the covariance matrix of the vector within the time window and its Frobenius norm are used as the main basis for judging the degree of nonlinearity of the system.

[0053] When the residual norm is within a pre-set linear approximation interval (typically between 0.1 and 0.2 in normalized dimension), indicating that the deviation of current parameter estimation from the field state is still within the acceptable error band of first-order Taylor linearization, the system automatically selects Extended Kalman Filter (EKF) for rapid iteration. At this time, the nonlinear water-thermal model calculates the local sensitivity in the direction of state variables and parameters respectively by perturbation method or numerical difference method - that is, when each water impedance or heat capacity parameter is increased by 1% at the current estimation point, the rate of change of the predicted output; the finite difference sensitivity matrix formed thereby is equivalent to the numerical approximation of the Jacobian matrix, which not only avoids the high cost of manual derivation of complex models, but also meets the demand for high-precision linearization. Based on the sensitivity matrix, the filter can complete the construction of Kalman gain and parameter correction with extremely low computational cost (usually equivalent to several hundred multiplication and addition operations) after performing a state prediction. After parameter updating, the residual and the deviation of the predicted covariance are used to reconstruct the covariance matching of the noise matrix according to the set forgetting factor (such as 5%), to ensure that EKF can adaptively track the statistical characteristics of process and observation noise.

[0054] If the residual norm exceeds the linear interval, it means that the model nonlinear effect is significant, and simple first-order linearization will bring too large error, at which time the system seamlessly switches to Unscented Kalman Filter (UKF). UKF does not rely on explicit derivation, but generates a set of weighted Sigma points by a sampling point generation strategy based on the current state estimation covariance, with a state dimension n multiplied by 2 plus one. The generation process of the sampling points is based on Cholesky decomposition as the numerical basis, and the concentration degree of Sigma points in the state space is controlled by adjusting parameters a and K (such as a=0.1, K=3−n). Then each Sigma point is input into the complete nonlinear simulation model to obtain the corresponding predicted Sigma point and observation Sigma point, and then the pre-set weight is combined to synthesize the prior mean and prior covariance, and finally the UKF gain is calculated and the state parameters are corrected in combination with the real-time residual. UKF has natural robustness to strong nonlinear scenarios and can maintain stable estimation accuracy in the presence of large disturbances or model structure mutations.

[0055] After each EKF or UKF completes parameter updating, the system immediately adopts the Rauch-Tung-Striebel-based posterior smoothing strategy to reverse fuse the latest estimation with the previous result to eliminate the estimation jump that may be caused by filter switching or nonlinear mutation. Posterior smoothing calculates the smoothing gain based on the sensitivity matrix obtained by the previous EKF or the Sigma propagation information of UKF, and makes a weighted correction to the forward estimation, and then outputs the smoothed and continuous water impedance and heat capacity parameters in time series. This process can be completed in a few milliseconds on the edge computing node.

[0056] Meanwhile, online noise reconstruction runs in parallel after each filtering iteration. By mixing the outer product of the prediction residuals of the current cycle with the current noise matrix at a set ratio (e.g., 5% and 95%), the system continuously updates the covariance estimates of process noise and observed noise, enabling the filter to adaptively respond to uncertainties such as heat source load fluctuations, pipeline wear, and changes in the external environment. If the statistical characteristics of the on-site noise change abruptly, such as a sudden increase in flow meter signal noise, the reconstruction ratio can be dynamically increased to accelerate estimation convergence. The specific reconstruction rate can be flexibly set within the range of 1% to 10%, without any hard limit.

[0057] After EKF and UKF run alternately and complete smoothing, the latest water impedance and heat capacity parameters are brought back to the water-heat coupling simulation model from step one to predict key indicators such as end-point return water temperature and maximum mains pressure differential. If the maximum relative error between these predictions and real-time observations exceeds the pre-set tolerance (e.g., pressure tolerance of 1% of full scale, temperature tolerance of 0.5K), the system determines that the current online parameter identification has lost convergence or the model has drifted, and then triggers the model reconstruction process. Model reconstruction first reinitializes the noise covariance to the design calibration value and reverts the parameters to be estimated to the recent smoothing average level. If convergence is still not possible, global optimization methods such as particle swarm optimization or genetic algorithms are run in parallel on the central server to find the optimal parameter space through multiple iterations. Finally, the new parameters are distributed to the edge nodes to complete the model reconstruction action described in the manual.

[0058] In the above process, it should be noted that the residual norm criterion is used for filter switching; the perturbation ratio provides a balance between accuracy and numerical stability for the sensitivity numerical approximation; the forgetting factor determines the adaptive rate of noise reconstruction; the smoothing gain ensures that posterior smoothing achieves a trade-off between debouncing and preserving the dynamic response; and the tolerance of consistency detection corresponds to the actual control requirements. These values ​​can all be adjusted according to the pipeline network scale, flow rate and temperature difference fluctuations, and the performance of the edge computing hardware. Specific adjustments can be made based on the site environment and project requirements, without rigid limitations. Through the alternating execution of EKF and UKF, the organic combination of noise adaptive reconstruction, posterior smoothing, and model consistency detection, the system achieves real-time, accurate, and robust online identification of water impedance and node thermal capacity parameters, providing reliable and continuously self-correcting model support for subsequent model predictive control and optimized scheduling.

[0059] Step 3: Based on the updated model Given the parameter vectors and state equations, calculate the network hydraulic coupling coefficient matrix and thermal coupling coefficient matrix, and generate the corresponding feedforward decoupling compensation terms; such as Figure 3 As shown, specifically, the calculation steps in step 3 include:

[0060] s100, based on the water resistance parameter and the node heat capacity parameter in the updated model output in step 2, a hydraulic coupling coefficient matrix is constructed by a node admittance matrix analysis method wherein is the pressure difference of the node , is the flow of the branch ; the node admittance matrix analysis method solves the sensitivity of the pipe network node pressure gradient to the flow change by the Jacobi iteration method, and outputs a diagonal dominant sparse matrix ;

[0061] s200, singular value decomposition is performed on the state matrix according to the state equation set of the heat capacity chain state space model, and a reduced order matrix is reconstructed by retaining the first r significant singular values , a heat flux partial derivative relationship of layered temperature is extracted, and a thermal coupling coefficient matrix is generated wherein is the temperature state of the kth layer, is the heat flux;

[0062] s300, the non-diagonal norm of the hydraulic coupling matrix is calculated , if it is greater than a preset threshold , a pressure disturbance suppression term is calculated, and then a valve opening correction amount is generated by a feedforward compensator, and the formula is:

[0063] (1)

[0064] In formula (1), is a gain matrix, which is used to map the pressure fluctuation to the valve opening correction amount; is a real-time pressure gradient residual error, with a unit of Pa / m, which is used to reflect the unoffset oscillation energy;

[0065] s400, the condition number of the thermal coupling matrix is calculated synchronously , when is greater than a preset threshold , a temperature lag compensation term is generated by a thermal inertia decoupling algorithm based on a heat flux-temperature transfer function pseudo-inverse matrix, and the formula expression is:

[0066] (2)

[0067] In formula (2), is a heat flux-temperature transfer function pseudo-inverse matrix, which is used to offset the multi-order lag effect of building heat capacity; is a target heat flux deviation;

[0068] S500, based on the real-time pipeline pressure oscillation spectrum energy , and thermal inertia residual ratio The compensation terms are weighted and fused to generate the final feedforward decoupling instruction.

[0069] In the scenario of real-time temperature control of indoor heating pipe networks, step 3 requires constructing the hydraulic coupling coefficient matrix of the network based on the model M1 output in step 2 and relying on the updated water impedance parameters and nodal heat capacity parameters. With thermodynamic coupling coefficient matrix Based on these two factors, a feedforward decoupling compensation term is generated to address the uncertainties and cross-interferences caused by the water-heat synergistic coupling. The entire process must be completed within each control cycle (typically 1s~5s) to meet the real-time requirements of indoor temperature control. The numerical thresholds and matrix truncation orders must take into account both computational resources and overall control accuracy. The specific values ​​can be determined based on the scale of the on-site pipeline network, the magnitude of load fluctuations, and edge computing capabilities, without any limitations.

[0070] Hydraulic coupling coefficient matrix The calculation first relies on the inverse operation of the nodal admittance matrix. This matrix Essentially, it reflects the hydraulic admittance between any two nodes in the pipeline network under frequency domain or static conditions, constructed through online updates of water impedance parameters; its inverse matrix corresponds to the sensitivity mapping of pipeline network node pressure difference to flow rate. System call After obtaining the initial coupling matrix, to more realistically characterize the sensitivity under transient conditions, the Jacobi iteration method is used to fine-tune the derivative of the pressure gradient with respect to flow rate change. Specifically, for each branch... At the current traffic benchmark Add a small disturbance (For example, take 5 L / h or 1% FS), calculate the pressure difference increment at each node using a rapid hydraulic simulation module or a lookup table method. During the iteration, the perturbation-response pair is fed back to the initial point until the change converges to a preset precision (e.g., the error is less than 0.01 Pa under a change of 0.1 Pa), forming a numerically approximate first-order partial derivative.

[0071]

[0072] Thus fill the matrix It is to be noted that the "iterative convergence accuracy" in the present application is distinguished from the 0.1% error standard commonly used in offline simulation, which can be adjusted in the range of 0.01 Pa to 1 Pa according to the pressure grade of the pipe network and the control accuracy requirement, and can be determined by the field measurement accuracy, without limitation. After the above steps are completed, the obtained AHA_HAH is usually a row-diagonal dominant sparse matrix, the diagonal elements of which dominate the "self-flow - self-pressure difference" coupling of each node, and the remaining non-diagonal elements reflect the auxiliary coupling between nodes.

[0073] The heat coupling coefficient matrix is extracted based on the heat capacity chain state space model. This model describes the layered or zoned temperature state (e.g. multi-point temperature measurement data from the outlet of the heat exchange station, the terminal heat dissipation surface and the return water inlet) and the heat flux input (flow and temperature difference product of hot water or steam) through a system of linear ordinary differential equations:

[0074]

[0075] wherein the matrix contains the thermal inertia and thermal diffusion characteristics of each pipe and indoor heat exchange equipment, and the matrix describes the direct driving effect of the heat flux input on the state change. In order to reduce the dimension and extract the main coupling mode, the system performs singular value decomposition on :

[0076]

[0077] The first r singular values (corresponding to an energy contribution rate of about 95%) are retained, and thus the reduced matrix is reconstructed: The corresponding columns are taken.

[0078] Subsequently, the heat coupling coefficient matrix is defined as the instantaneous partial derivative of each input flux to each temperature state: It is to be noted that the "singular value energy contribution rate" in the present application is distinguished from the variance explanation rate of traditional principal component analysis, which takes into account the influence of thermal lag dynamics on spatial coupling; the singular value decomposition can be selected based on the direct QR algorithm or the randomization SVD acceleration scheme, and can be determined by the number of nodes m and the computing capacity, without limitation.

[0079] After the construction of the coupling matrix is completed, the system enters the generation stage of the feedforward decoupling compensation term. First, for hydraulic disturbance suppression, the Frobenius norm of the off-diagonal energy of the coupling matrix, i.e. the remaining matrix after excluding the diagonal elements, is measured: When this norm exceeds a predetermined threshold (For example, take the dimensionless quantity of 0.1~0.5, which is used to evaluate the degree of cross-node coupling, which can be set according to the size of the pipe network and the coupling strength, not limited), indicating that the flow-pressure difference coupling between nodes is significant, and needs to be suppressed through feedforward control. The control quantity is generated according to the formula , where the vector represents the instantaneous pressure gradient residual of each node (unit: Pa / m, which can be obtained by dividing the difference between the readings of the adjacent two pressure sensors by the installation spacing), and the matrix is the feedforward gain matrix, which is usually determined by engineers based on the inverse or pseudo-inverse of combined with a proportional adjustment coefficient (value 0.5~2.0, which can be flexibly set according to the response speed of the regulator and the stability requirements of the system, not limited) :

[0080]

[0081] The calculated represents the correction amount that each valve opening or pump frequency needs to make. In actual operation, the controller will write these instructions to the target register of each valve position controller or frequency converter through MODBUS or OPC UA protocol, and the actuator will complete the opening or pump speed adjustment at a fixed ramp rate (such as 0.1% / s) to avoid mechanical chattering.

[0082] The trigger condition for thermal decoupling compensation is based on the condition number of the matrix

[0083]

[0084] When >ϵ2 (usually take 10³~10 4 , to measure the numerical ill-conditioned degree of thermal coupling matrix, which can be set according to the simulation test results, not limited), indicating that there is a high-order lag or amplification effect in the coupling of heat flux to temperature, and the thermal inertia decoupling algorithm needs to be started. This algorithm relies on the pseudo-inverse matrix of the temperature transfer function to offset the multi-order lag, and the formula is ,(2)

[0085] where the vector represents the target deviation of the heat exchange station or the terminal heat load (unit can be kW or m³ / h), and the matrix ∈ is usually obtained by system identification or empirical model, reflecting the multi-order delay relationship from heat flux input to temperature response. In actual operation, the controller will convert each into the water temperature increment of the corresponding heat exchange station or the heat pump load adjustment instruction, and issue it through the building automation BACnet or Modbus interface.

[0086] For the balance between hydraulic and thermal compensation, the system adopts dynamic weighted fusion: . Wherein is the integral of the power spectral density of the trunk pressure signal in the oscillation frequency band (such as 0.1Hz2Hz) of the pipe network, representing the hydraulic suppression requirement; is defined as the norm of the difference between the latest temperature prediction and the measured value (unit K); when ≫thermal inertia residual, close to 1, the system focuses more on hydraulic suppression; otherwise, it focuses more on thermal compensation. It should be noted that the "spectrum energy integration" in this application is different from the instantaneous energy measurement, which must be realized through real-time FFT or band-pass filtering and energy detection module, and the calculation window length can be selected between 5s30s, which can be determined according to the dynamic characteristics of the pipe network and the control cycle synchronization strategy, and is not limited.

[0087] The final fused feedforward decoupling instruction contains valve opening or pump speed offset and heat exchange water temperature difference or power offset , which is sent to each actuator in the form of a continuous signal through the industrial bus to realize water-heat collaborative decoupling control of the indoor heating pipe network. The implementation of the whole step 3 not only covers the complete link from model data acquisition, sensitivity matrix calculation to feedforward compensation generation and dynamic weight fusion, but also specifically describes the triggering conditions, numerical thresholds, matrix operation methods and execution interfaces, and provides flexible selection space where necessary with "specifically can be determined according to the actual situation, not limited" to ensure that the scheme has good applicability and engineering feasibility in indoor heating pipe systems of different sizes and topologies

[0088] It should be noted that the "node admittance matrix analysis method" in this application is different from the existing static hydraulic admittance method, which can obtain the pressure-flow coupling sensitivity between pipe network nodes through iterative calculation under real-time working conditions, and then output the mapping coefficient required for hydraulic decoupling; the "thermal capacity chain state space model" is different from the traditional single pipe heat balance model, which can uniformly describe the thermal inertia of multiple pipe sections and heat exchangers in the form of a state matrix, making it easy to extract the main coupling mode through singular value decomposition; the "thermal inertia decoupling algorithm" is different from the conventional PID or first-order lag compensation mechanism, which can automatically generate multiple-order thermal lag offset based on the pseudo-inverse matrix of the inverted temperature transfer function; "pressure oscillation spectrum energy" is different from instantaneous pressure measurement value, which is the integral of the power spectral density in a specific oscillation frequency band after frequency domain analysis of the trunk or branch pressure signal of the pipe network, which can quantify the degree of system oscillation; "dynamic weighted relaxation" is different from the static weighting coefficient in control theory, which adjusts online according to the real-time measured ratio of hydraulic oscillation energy to thermal inertia residual, ensuring that the feedforward compensation can suppress oscillation without losing temperature response.

[0089] Step 4, constructing a multi-model prediction framework containing long pipeline model and terminal partition model according to the parameter vector of the feedforward decoupling compensation term and , outputting a weighted prediction model group through model pool weight scheduling method ; in the indoor heat pipe network temperature real-time regulation scene, the implementation of step 4 first relies on the key working condition indicators continuously collected by the system online: the flow mutation rate of any main or branch of the pipe network (the unit can be m³ / s², and when the value exceeds 2 m³ / s², it is determined to be a severe flow disturbance) and the spatial variance of the terminal temperature gradient (the dimension is °C² / m², and when the value exceeds 0.5 °C² / m², it indicates significant temperature unevenness). The working condition characteristic classifier monitors the above two indicators, and when the flow mutation rate , the system automatically determines it as "hydraulic dominant working condition", at which time the long pipeline model based on transmission line theory is called. The model takes the one-dimensional wave equation as the core: , where P(x, t) and Q(x, t) are the pressure and flow along the pipe coordinate , respectively, is the fluid density, is the wave speed (which can be online calibrated according to the pipe material, filling fluid and pressure rating). In actual operation, the system feeds back the feedforward decoupling compensation generated in step 3 to the boundary condition module of the model to adjust the initial pressure and flow boundary values, so that the model prediction can immediately reflect the compensation action. At the working condition switching moment, the model adds the corresponding compensation to the original boundary pressure or flow value, and through parallel operation by finite difference or characteristic line method in the next period, the predicted pressure waveform is output.

[0090] If the terminal temperature gradient variance exceeds the threshold value of 0.5 °C² / m², the system determines it as "thermal dominant working condition" and switches to the terminal partition model. The model is based on the distributed heat capacity chain state equation , where the indices and traverse the same region or adjacent partitions, is the equivalent heat capacity of the node, is the equivalent thermal resistance between nodes, which can be obtained from the online identification results or initial installation calibration data in step 3. The system also superimposes the feedforward compensation on the heat exchanger output temperature or heat flow boundary to drive the model state to converge to the new working condition value. The model uses the implicit Euler or Gear method to ensure numerical stability.

[0091] Regardless of which dominant model is used, the system will keep the prediction results of the other model in parallel for residual comparison and dynamic weighting. When the dominant model outputs a residual When the residual is generated, the normalized residual amplitude is calculated. To achieve online updating of model weights, the system adjusts the weights using a recursive least squares method with a forgetting factor , where the forgetting factor λ ∈ (0, 1) can be set typically to 0.01-0.1 to balance the influence of new and old residuals, which can be set according to the dynamic frequency of the pipe network and the measurement noise level, without limitation. In this way, the model weights will automatically tend to the model that best adapts to the current working condition, ensuring that the prediction can smoothly transition and the error is controlled in the working condition switching or coexistence scenario.

[0092] Finally, the multi-model prediction output is fused in the form of a weighted sum, and the fusion result is further used as the input of model predictive control (MPC) to optimize the scheduling of heat source outlet water temperature, pump speed and valve opening. When the action is executed, the system will simultaneously issue the control increment after weighted fusion to the corresponding execution unit, and confirm the instruction arrival and execution feedback through SCADA or BACnet network. If it is found that there is a deviation between the issued and executed feedback, the compensation fault-tolerant mechanism is automatically triggered to re-adjust the weights or amplify the compensation amount.

[0093] In the present application, the “working condition adaptive model pool” is different from the traditional single model or static switching mechanism, which can trigger the most suitable model online according to the real-time flow mutation and temperature gradient characteristics, and maintain the continuity of prediction and control through weighted fusion; the “covariance matching method” is different from the standard least squares fitting, which introduces a forgetting factor in the recursive process to balance the influence of historical and current residuals on model weights; the “long pipeline model” and the “end partition model” are different from the uncoupled model, which respectively based on the transmission line theory and the distributed thermal capacity chain theory, accurately depict the dominant dynamics of water and heat in different spatial scales and working conditions; the “normalized residual” is different from the direct residual value, which divides the residual by the preset accuracy or range to realize comparable error measurement between different models.

[0094] Step 5, solving a nonlinear optimization problem containing temperature error, energy consumption and actuator constraints by an online quadratic programming method according to , actuator upper and lower limits and future load prediction curve to generate an initial control sequence; it should be noted that the “online quadratic programming method” in the present application is different from the traditional real-time optimization or rolling optimization, which not only aims at the multivariable control problem of the water-heat coupled system of the pipe network, but also unifies the temperature tracking error, energy consumption index and actuator physical constraints into the same quadratic optimization framework, which can generate a smooth and implementable control sequence in a short time; for details, please refer to Figure 4 ​The comparison shows the indoor temperature error versus the control signal variation within the same time window.

[0095] In implementation, firstly, the weighted prediction model group generated in the previous step and the heat load and cold load prediction curves in a certain time domain in the future (for example, the next ten-minute control window, divided into several steps) are taken as inputs, and the latest actuator states, including the opening of each valve, the rotating speed of the pump frequency, and the water supply temperature of the heat source, are called. After parallel simulation based on the prediction model group, the system obtains the multi-model joint prediction of device behavior and indoor terminal temperature at each prediction step. In order to meet the multi-objective control requirements, the system forms a temperature tracking error sequence with the difference between the predicted temperature and the set temperature, and the sum of the control quantity fluctuation and the cumulative energy consumption as the energy consumption penalty term, and weaves them into a quadratic objective function according to the pre-set comfort weight and energy efficiency weight. In this process, users can set the relative weight coefficients of comfort and energy efficiency according to the building type and energy consumption cost, and the values of which are not limited by this application.

[0096] When the objective function and the constraint condition are clear, the system needs to integrate the upper and lower limits of the actuators: each valve has physical limits between fully closed and fully open, each pump group has safety limits between the minimum and maximum frequency, and the heat source also has device and comfort boundaries between the minimum and maximum outlet water temperature. In order to prevent the control command from exceeding the actuator capacity, the algorithm adds these constraints to the optimization problem in the form of inequalities, and also considers the maximum change amplitude of the control quantity at adjacent time, in order to avoid water hammer shock or thermal shock caused by sudden changes. Specific configuration can be made according to the technical indicators of response time and maximum slope rate in the actuator manual, and the specific values can be determined according to the performance and service life requirements of the on-site device, which are not limited.

[0097] The optimization problem is essentially a quadratic programming (QP) problem with multi-section dynamic prediction coupling. In order to realize real-time solution under limited computing resources, this application proposes a solution strategy combining Krylov subspace projection and preconditioned conjugate gradient. First, the system constructs a set of linear equations according to the KKT condition, which covers the Hessian matrix, the gradient vector and the constraint Jacobian matrix. Since the number of pipe network nodes and the number of future prediction steps are usually large, the direct solution of this linear system is extremely costly. Therefore, the system generates a Krylov subspace only for the current gradient and Hessian in each optimization period: an equal-dimension basis set much smaller than the original variable number is generated by the initial residual vector and the Hessian-residual product in a loop. In this process, no full matrix storage is required, only matrix-vector multiplication is required multiple times, which fully utilizes the sparse matrix characteristics produced by the physical coupling structure of the pipe network. The subspace dimension can be flexibly selected between tens and fifty according to the performance of the edge computing node and the control period requirement, which can be determined according to the on-site hardware and system size, and is not limited.

[0098] After obtaining the Krylov subspace basis, the system projects the KKT system into this subspace, forming a small-scale equivalent linear system, and applies the preconditioned conjugate gradient method for iterative solution. The preconditioner is usually selected as the approximate inverse matrix based on diagonal or incomplete LU decomposition to further reduce the condition number and accelerate convergence. In each iteration, the step size is estimated by the Barzilai-Borwein formula, which does not require multiple objective function evaluations like traditional line search, thus greatly improving the iteration speed while ensuring the effectiveness of the search direction.

[0099] During the iteration process, the system continuously monitors the consistency of the constraints. When it detects that part of the constraints conflict—such as the temperature tracking target and the minimum energy consumption target cannot be satisfied simultaneously within the given actuator limits— the system automatically enables the relaxation factor mechanism to dynamically relax the inequality boundary. This relaxation is achieved by adding or subtracting a certain percentage of the relaxation amount to the boundary value, and the buffer factor size can be adjusted online according to the severity of the conflict and the tolerance of comfort. The specific value can be determined by the safety specifications and comfort standards, and is not limited. The relaxation range minimization principle is to restore the feasible solution in real time, and also avoids excessive relaxation leading to a significant decrease in comfort.

[0100] In the conjugate gradient iteration, if the residual reduction rate is always below a preset threshold (e.g. the residual decreases by only 10% or less per step, which can be set according to the pre-set iteration efficiency target, and is not limited) after a certain number of iterations, the algorithm will trigger the heuristic restart mechanism: the current residual is taken as the new search direction, the conjugate direction is reset, and the Krylov subspace basis is updated to escape the possible long slot dilemma and ensure that the optimization can continue to effectively advance. The restart criterion and restart frequency can also be flexibly adjusted according to the dynamic characteristics of the pipe network and the control frequency, and are not limited.

[0101] When the preconditioned conjugate gradient iteration reaches the termination condition, including the residual norm being below an absolute threshold or the iteration step number reaching the maximum limit, the algorithm maps the optimal solution in the reduced subspace back to the original decision variable space to form a complete control sequence To prevent mechanical or thermal shock caused by sudden execution, the system further applies smoothing filtering to each control quantity before issuing the control sequence: after subtracting the execution value of the previous cycle and multiplying by a smoothing coefficient (usually 0.2-0.5, which can be determined according to the comfort jitter tolerance, and is not limited), the final execution instruction is obtained. These instructions are issued to specific actuators through industrial protocols, such as the supply water temperature set value of the boiler or heat pump, the target frequency of the pump frequency converter, and the opening percentage of the electronic proportional valve, and continuous execution feedback is obtained to ensure the reliability of the closed loop.

[0102] The sensitivity analysis table of parameters of the online quadratic programming method, as shown in Table 1, is used to evaluate the influence of each parameter on the convergence performance and comfort:

[0103] Table 1 Sensitivity analysis table

[0104] Parameter Name Optional Range Impact on Convergence Performance Impact on Comfort Flow Threshold 0.1-1.0 m³ / h Smaller threshold may result in slower convergence Too small may lead to excessive control, affecting comfort Forgetting Factor λ 0.9-0.999 Larger λ means more influence from historical data, potentially more stable convergence Too large λ may result in sluggish system response, affecting temperature tracking BB Step Size α 0.5-2.0 α affects search step size, appropriate value accelerates convergence Inappropriate α may lead to increased control signal fluctuations Krylov Subspace Dimension 10-100 Higher dimension means worse dimension reduction, increased computation Too low dimension may affect convergence precision, leading to increased temperature error

[0105] In addition, it should be noted that the "weighted prediction model group" in the present application is distinguished from single model prediction, and is composed of the weighted outputs of multiple sub-models, and can provide continuous and robust state prediction in the working condition switching or coexistence scene; the "Krylov subspace projection" is distinguished from simple dimension reduction or eigenvalue decomposition, and uses the quadratic Hessian information of the objective function to construct a low-dimensional search subspace, thereby greatly reducing the calculation amount of solving the KKT system; the "preconditioned conjugate gradient method" is distinguished from the conjugate gradient without preconditioning, and accelerates the convergence of the Krylov method in the pipe network coupling optimization problem by selecting a preconditioning matrix; the "BB step length" is distinguished from the fixed step length or line search step length, and estimates the approximate optimal step length in each iteration by the Barzilai-Borwein formula, thereby significantly improving the iteration efficiency; the "heuristic restart" is distinguished from the fixed period restart, and dynamically determines when to reset the conjugate direction based on the residual error reduction rate, so as to avoid search stagnation or walking into a pathological direction.

[0106] Step 6, according to the serialization strategy of "pump first and valve second", and combined with the dead zone and the rate limit, the actual pump speed and the valve opening control instruction are generated according to the initial control sequence and the feedforward decoupling compensation term;

[0107] The core of step 6 is to convert these intentions into specific instruction sets that can be directly issued to the pump frequency converter and the electronic proportional valve, which not only meets the physical limits and response characteristics of the actuators, but also ensures that the water-heat coupling compensation logic is landed in the control action. This process mainly includes instruction extraction and classification, timestamp sorting and trigger criterion, dead zone judgment and action delay, rate planning and feedforward correction, and labeled instruction output.

[0108] First, the system parses two types of instructions from the initial control sequence: one type of pump speed instruction represents the target speed or power of each pump group frequency converter; the other type of valve opening instruction , representing the target opening percentage of each control valve. During the analysis process, the controller will specifically superimpose a feed-forward decoupling compensation term on the valve instruction set to ensure that the valves can compensate for the pressure or flow coupling errors introduced after the pump speed changes without adjusting the pump frequency again. It should be noted that the "feed-forward decoupling compensation term" in this application is different from the traditional feedback correction. It pre-compensates for coupling interference at the instruction level, reducing the burden on the closed-loop controller. The specific value of the compensation can be dynamically determined in combination with the matrix product result generated in step 3, and is not limited.

[0109] After the analysis is complete, all instructions will be marked with a timestamp, i.e. each pump speed or valve opening instruction carries a time label , indicating that the instruction should be issued for execution after the current time (typically an integer multiple of the control period). Through the timestamp marker, the controller can strictly order the pump speed instructions and valve instructions: first, arrange the pump speed instructions in ascending order and send them to the pump frequency inverter interface in turn; after sending a pump speed instruction, the system will call the pressure gradient detector module in real time to monitor the pressure gradient and its rate of change of adjacent measuring points in the main pipe. When the pressure change rate enters the steady-state dead zone range (e.g. =0.001 MPa / s), the system triggers the valve opening instruction with the corresponding timestamp; otherwise, the valve instruction will be delayed until the steady-state condition is met or the maximum delay time (typically 5s, which can be determined according to the length of the pipe network and the response requirement, and is not limited). This strategy of first pump and then valve combined with the dead zone criterion can prevent the pressure wave caused by pump speed fluctuations from reflecting back to the main pipe when the valve is not closed or opened in time, causing water hammer.

[0110] For the specific execution of the valve opening instruction, the system uses a segmented S-type rate planner to generate a smooth opening curve. The planner divides the change from the current opening to the target opening into three stages: acceleration stage (opening increment uniformly accelerated from 0 to maximum rate ), constant speed stage (maintaining ), and deceleration stage (uniformly decelerated from to 0). The maximum allowed change rate is determined by the pipe hydraulic diameter and the current flow Q, which is usually based on the empirical formula

[0111]

[0112] where is an empirical coefficient (e.g. =0.1~0.5 (can be adjusted according to valve actuator characteristics and system water hammer tolerance, without limitation), ensuring faster opening changes are allowed for large-diameter, high-flow-rate pipes, while reducing the rate of change to avoid impact for small-diameter, low-flow-rate pipes. The sliding window constraint calculates the actual rate within each small time step (e.g., 100ms) and ensures...

[0113]

[0114] When the system detects a sudden change in traffic ( ,For example =2 This indicates that the coupling compensation may be insufficient, and the controller will enable the feedforward decoupling compensation term to correct the rate threshold: Multiply by a magnification factor >1 (values ​​range from 1.1 to 1.5) to respond more quickly to sudden disturbances. The specific factor can be determined by the system based on safety constraints and comfort requirements, and is not limited.

[0115] After the rate curve is generated, the valve opening commands are also queued according to timestamps. Each opening curve is discretized into several steps at the underlying actuator interface and sent in batches to balance the controller's communication bandwidth and the actuator's minimum action resolution. For actuators that support batch writing to registers, the system can send the node values ​​of the entire S-curve at once and start hardware self-execution; for devices that do not support batch writing, the next opening value is sent through a timer interrupt at each small time step to ensure complete tracking of the opening curve.

[0116] After all pump speed and valve commands are sent and executed, the controller will re-acquire the actual feedback values ​​of the actuator in the next control cycle, including the pump speed achievement rate and the actual valve opening, and compare them with the command values ​​to calculate the execution error. If the execution error exceeds the set threshold (pump speed error > 2%FS or opening error > 1%FS, which can be set according to the actuator accuracy and is not limited), the system will automatically amplify the corresponding compensation amount in step 3 or step 5 to ensure the robustness of the closed loop.

[0117] From the perspective of system behavior, step 6 ensures that pump speed changes are stabilized first through a sequential strategy of pumping before valves, and then valve actions compensate for coupling pressure errors; steady-state dead zone judgment avoids valve malfunctions before pressure waves subside; S-shaped rate planning and sliding window constraints achieve smooth and controllable valve dynamics; a feedforward rate correction mechanism addresses sudden flow disturbances; and timestamp marking and delay mechanisms ensure sequential execution and real-time synchronization. The final output control command set not only meets the physical characteristics and safety constraints of the equipment, but also accurately implements the decoupling compensation and optimization strategies generated in steps 3 and 5, providing a solid execution guarantee for temperature control in the next cycle.

[0118] It should be noted that the "pump-first, valve-later sequential strategy" in this application differs from the traditional simultaneous drive mode. It can eliminate the transient coupling of mutual interference between pump speed and valve through the time sequence, thereby ensuring the predictability of pressure wave propagation and flow distribution. The "steady-state dead zone" is different from the dead zone in general control. It is defined as the system considering it to have entered a steady state when the pressure change rate of the main pipeline is within a certain small range (such as ±0.001MPa / s), and can safely execute the downstream valve action without causing secondary oscillations. The "S-shaped rate planner" is different from linear ramp rate control. It generates smooth opening changes through a speed curve of three stages: acceleration, constant speed, and deceleration, in order to reduce water hammer and mechanical shock. The "pressure gradient detector" is different from single-point pressure monitoring. It compares the pressure difference between adjacent measuring points in real time and calculates the gradient change rate to determine whether the system is in an allowable state.

[0119] Step 7: After control execution, using the temperature residual and flow oscillation data from the previous moment as input, dynamically adjust the multi-model weights and prediction time domain length, and output the updated control configuration parameters. Specifically, the difference between the predicted and measured indoor temperature from the previous moment constitutes the temperature residual sequence. The flow sensor synchronously records the instantaneous flow rate of each branch or main trunk. The system first takes the residual sequence as input and calls the Hilbert-Huang transform module to process it. Empirical Mode Decomposition (EMD): The algorithm decomposes the original residual signal into several IMF components through the construction and iterative selection of extreme point envelopes. (k),…, (k) and residual trend terms. Subsequently, the time-domain energy is calculated for each IMF.

[0120]

[0121] The energy spectral weights are then obtained by normalization.

[0122]

[0123] The system allocates the energy percentage of all low-frequency IMFs (typically those with larger numbers and time-domain periods exceeding twice the control period). (For example, 0.6~0.8, which can be set according to the building's thermal insulation characteristics and thermal inertia time constant, without limitation) Compare. If the proportion of this low-frequency energy exceeds the threshold, it is determined that the temperature residual is mainly dominated by thermal inertia hysteresis. In this case, the time domain length of the multi-model prediction needs to be extended to capture the slowly changing trend, thereby avoiding the omission of significant thermal hysteresis components in short-window predictions. The prediction time domain length extension mechanism is to extend the current prediction window length... According to heat capacity time constant Incremental correction of 1.2 times:

[0124]

[0125] wherein The ratio of the node heat capacity parameter and the heat exchange coefficient obtained from the online identification in step two represents the time for the system to reach the 1 / e level of steady state. It should be noted that the "thermal capacity time constant" in the present application is different from the general RC circuit constant, and it specifically refers to the overall thermal inertia of the heat pipe network system. The expansion range (1.2 times) and the final upper limit prediction window can be flexibly adjusted according to the system calculation capacity and the control accuracy requirement (such as a maximum of 20 minutes), and are not limited.

[0126] At the same time, the system takes the flow oscillation data as input, and for a certain typical measurement point or multi-point combined signal calculates the autocorrelation function

[0127]

[0128] and normalizes it to . The main lobe width of the autocorrelation function is defined as the delay time required for the signal to decay from to the critical level (such as 1 / e or 0.5). If (typically between 1s and 5s, which can be set according to the response characteristics of the single pump or valve and the oscillation period, and is not limited), it indicates that the flow oscillation has high frequency components and short duration, and it is determined that the hydraulic disturbance is dominant in this area. At this time, the multi-model weight needs to be redistributed to reduce the weight of the hydraulic model in this area and improve the confidence of the thermal model. The system extracts the main frequency component of the flow signal in the Hilbert instantaneous phase and divides the phase-locked subnetwork, regarding the branches or partitions with phase synchronization and amplitude mutation as the same associated subnetwork. The weight of the corresponding model (hydraulic dominant model) in the associated subnetwork is linearly down-regulated:

[0129]

[0130] and the down-regulation amplitude ∈(0.1,0.3) is evenly distributed to all involved areas in the subnetwork; at the same time, the weight of the corresponding thermal model = +β in the area is adjusted. It should be noted that the "oscillation phase recognition" in the present application is different from the simple time domain oscillation detection, which combines Hilbert transform and phase continuity criterion, and can distinguish between synchronous and asynchronous behaviors of multi-area flow oscillation; the weight adjustment ratio β can be adjusted in the range of 0.05-0.5 according to the oscillation intensity and control target, and is not limited.

[0131] For the regions where the oscillation criterion is not triggered, the system increases the confidence of the thermal model in the overall weight distribution: i.e. all the remaining model weights are proportionally renormalized and the thermal model weight is multiplied by an amplification factor (1+γ), γ∈(0.05,0.2) representing the increase in trust in the thermal dynamics model. After the weight adjustment of all regions, a renormalization is needed in the same control period to guarantee . This dynamic updating mechanism enables the model pool to maintain an immediate preference for different physical processes in scenarios where hydraulics and thermodynamics dominate or coexist, improving the robustness of prediction and control.

[0132] After the update, the system injects the new model pool weight vector and the prediction time domain parameter set into the multi-model prediction and online quadratic programming module, providing the latest structured configuration for the next step of prediction modeling and control optimization. This process is automatically completed in each control period (which can be periodically executed between 30s~120s) to ensure that the system can quickly adjust the prediction window and model trust under various disturbance conditions such as indoor thermal load mutation, main valve regulation, or heating source switching, achieving high-precision and dynamic adaptive regulation of the heating pipe network.

[0133] It should be noted that the "Hilbert-Huang Transform" in this application is different from the traditional Fourier or wavelet time-frequency analysis. It can adaptively decompose several intrinsic mode functions (IMF) and extract instantaneous frequency and amplitude information under the background of nonlinear and non-stationary signals, reflecting the multi-scale oscillation characteristics implied in the indoor temperature residual signal. The "intrinsic mode function energy spectrum" is different from the frequency spectrum density. It refers to the proportion of the time energy (i.e. IMF square integral) of each IMF to the total energy, which can quantitatively reflect the low-frequency lag and high-frequency fluctuation components. The "prediction time domain length" is different from the control period length. It refers to the future time window size used by the model in online quadratic programming or multi-model prediction, which can affect the prediction accuracy and computational burden. The "autocorrelation function decay rate" is different from the correlation coefficient. It reflects the memory length and coupling duration of the signal by the delay time required for the autocorrelation sequence peak to decay to 1 / e or 1 / 2. The "main lobe width" is different from the concept of frequency domain main lobe. It corresponds to the half width of the main peak in the autocorrelation function time domain, which can indicate the duration of flow oscillation. The "oscillation phase recognition" is different from ordinary phase recognition. It groups the flow oscillations of each subregion in the pipe network through Hilbert instantaneous phase analysis or phase envelope method, so as to identify the coupled regions in the multi-model weight redistribution.

[0134] Please refer to Figure 5 ​, as a possible implementation of a multi-parameter coupled intelligent control system for vanadium ore smelting process, the specific working principle is: in actual operation, each module is closely coordinated through the defined data interface and message bus between each module to ensure that the output of each link can be directly used as the input of the downstream link. Specifically, the data acquisition module first completes the original measurement of temperature, flow, pressure and valve position through the edge node, and after local filtering and clock synchronization, the standardized time series data set D1 is pushed to the central message bus; the time series preprocessing module takes the message bus as the subscription port, automatically pulls D1, performs wavelet packet multi-scale denoising and Bayesian interpolation filling on it, and the smoothed and denoised sequence S1 is forwarded to the parameter modeling module through the same bus.

[0135] The parameter modeling module immediately calls the pipe network topology database service after receiving S1, loads the static topological relationship and physical property parameters of the nodes and pipe segments, and uses the PDE / lumped parameter hybrid strategy to build the whole network model M0. When M0 is built, its state equation set and parameter vector θ0 will be packaged as a "model initialization" message and sent to the input queue of the online identification module.

[0136] The online identification module listens to the "model initialization" and "smooth sequence" two inputs, first based on the interpolated S1 and M0, performs state prediction and residual calculation, and automatically switches the EKF or UKF sub-process according to the amplitude threshold of the residual covariance matrix, outputs the latest water impedance and heat capacity parameters θ1, and the updated model M1. This set of updated model and parameters are then published to the coupled and decoupled module through the "model update" message.

[0137] The coupled and decoupled module extracts the network hydraulic and thermal coupling matrices from M1 after receiving M1, and performs spectral decomposition on them respectively, selects the dominant coupling subspace, and generates the feedforward decoupling operator C. This operator, as a decoupling compensation item before quadratic programming, is pushed to the predictive control module together with M1 through the "decoupling operator" message.

[0138] The predictive control module monitors the "decoupling operator" message, and also pulls the future load curve L(t+i) from the external load prediction service. It takes M1, C, L as input, and outputs the initial control vector U0 containing decoupling compensation through parallel multi-model adaptive integration and linear quadratic programming solver. U0 is immediately sent to the execution monitoring module as a "initial control sequence" message.

[0139] After receiving U0, the execution monitoring module first splits it into pump speed instruction and valve opening instruction, and sends them to the field pump control unit and valve drive unit according to the timestamp. The pump control unit only responds to the opening instruction of the valve drive unit after executing the pump speed instruction and verifying that the main pipe pressure has entered the dead zone through the pressure gradient detector. The execution monitoring module continuously collects the feedback of the equipment, including real-time pump speed and actual valve travel, and sends these feedbacks to the compensation iteration module in the form of residual sequence R.

[0140] The compensation iteration module subscribes to the "execution feedback" message, runs the iterative learning control algorithm using the residual sequence R and the historical residual library, generates the compensation increment ΔU_ILC, and superimposes ΔU_ILC on U0 of the next cycle to form the modified control vector U', which is then input to the closed-loop coordination module again through the "modified control" message to achieve pre-compensation for future control sequences.

[0141] The closed-loop coordination module listens to both "modified control" and "subnet interface error" messages, the latter coming from the interface flow and pressure monitors of each local subnet. The closed-loop coordination module takes ΔU_ILC, R, and interface error E as inputs, coordinates the model parameters and MPC weights of multiple subnets through the augmented Lagrangian iteration algorithm, and outputs the new model parameter set θ2 and MPC weight vector ω after convergence, together with the control configuration of the next cycle, through the "new cycle configuration" message back to the prediction control module, forming an end-to-end closed-loop adaptive regulation process.

[0142] The push and subscription of each message described above run on a reliable industrial message bus (such as OPCUA or Kafka industrial version) to ensure the real-time and reliability of data transmission. All modules can be horizontally expanded to adapt to different scales of smelting systems, from single furnace island to complex conditions of multiple furnace parallel connection, to realize end-to-end intelligent regulation and control of multi-parameter coupling in vanadium ore smelting process.

[0143] In implementation, temperature, flow, pressure, and valve position sensors distributed in each heat exchange station and terminal room continuously upload raw signals with unified clock timestamps to the data collection module. After aligning these signals in memory according to the timestamp, the collection module pushes them to the time series preprocessing module in the form of standardized time series data sets. Upon receiving the aligned raw sequence, the time series preprocessing module starts the wavelet packet multi-scale filter to separate high-frequency mechanical vibration noise and electrical interference, and uses Bayesian interpolation method to fill in missing values caused by maintenance or communication delay, outputting a set of smooth, noise-reduced, and complete time series signals for downstream use.

[0144] The parameter modeling module continuously receives the set of trusted time series from the preprocessing module, and maps each measurement point to the nodes or segments of the model according to the physical topology of the pipe network, uses a PDE / lumped parameter hybrid strategy to construct an initial full-network state equation set containing second-order thermal inertia and hydraulic coupling terms, and then submits the model together with the latest flow and temperature data to the online identification module. The online identification module takes the model as the basis for prediction, uses smoothed time series data as observation input, and continuously corrects water impedance and node thermal capacity parameters through the alternating operation of extended Kalman filtering and unscented Kalman filtering, and synchronously returns the updated coupled model to the parameter modeling module and the subsequent coupling and decoupling module, so that the models used by each module can reflect the latest physical characteristics on site.

[0145] After obtaining the latest model, the coupling and decoupling module first extracts the network hydraulic coupling coefficient matrix and the thermal coupling coefficient matrix from the model, and performs spectral decomposition respectively - the node admittance matrix inverse operation combined with the Jacobi iteration generates the hydraulic subspace, and the singular value decomposition reconstructs the thermal subspace; then the two subspaces are jointly operated to form a feedforward decoupling operator, and the operator and the updated model are pushed to the predictive control module together, so as to compensate for the coupling disturbance in the rolling time domain optimization.

[0146] The predictive control module takes the decoupling operator, the updated model and the future load prediction sequence as input, constructs a multi-model parallel prediction framework, generates a weighted prediction result, and solves the initial control vector in a rolling time domain quadratic programming manner, and then delivers the initial sequence to the execution monitoring module. After receiving the batch of control instructions, the execution monitoring module serializes the pump speed instruction and the valve opening degree instruction according to the sequence strategy of "pump first and valve second", timestamps, sorts and issues them, and at the same time monitors the response state with the help of the pressure gradient detector and the flow meter, and triggers the valve action after the steady state criterion is met, and then feeds back the executed room temperature residual error and flow deviation signal to the compensation iteration module in real time.

[0147] The compensation iteration module takes the latest temperature residual error, flow deviation and historical residual error library as the training set, runs the iterative learning algorithm (such as recursive least squares or online reinforcement learning) to generate a deviation compensation increment, and pushes the corrected control vector back to the execution monitoring module to complete the closed-loop compensation; at the same time, it reports the compensation increment and the executed residual error to the closed-loop coordination module.

[0148] The closed-loop coordination module inputs the compensation increment, the interface error of each subnetwork and the current model weight into the augmented Lagrangian iteration process, adjusts the Lagrange multiplier and the relaxation factor dynamically, updates the model parameters and the MPC weight, and issues the control configuration parameters of the next round back to the predictive control module and the execution monitoring module, and starts the closed-loop optimization cycle of the next period.

[0149] While specific embodiments of the application have been described above, it will be appreciated that those skilled in the art within the scope of the application can make modifications, substitutions and changes in the form and details of the method and system described above without departing from the spirit and scope of the application. For example, it is well within the scope and spirit of the application to combine various of the illustrative steps of the above methods to carry out essentially the same function to achieve essentially the same result in substantially the same way. Accordingly, the scope of the application should be determined sole by the appended claims.

Claims

1. A method for real-time indoor temperature regulation of a heat pipe; characterized by: Comprise: Step 1, based on the real-time acquisition of node temperature, pressure and flow data of the pipe network, an initial full-network model containing second-order thermal inertia and hydraulic coupling terms is constructed by using PDE / lumped parameter hybrid method and output the initial state equation; the working principle of the PDE / lumped parameter hybrid method is that the pressure gradient distribution of the pipe network nodes is collected in real time through the pressure sensor network, and based on the dynamic partition mechanism of the hydraulic domain of Morris sensitivity analysis, the main pipe meeting the pressure gradient threshold is classified into the PDE domain, and the characteristic line method is used to discretely solve the transient flow conservation equation; at the same time, the branch pulsation amplitude is identified through the flow meter data, if it is lower than the critical value of turbulent flow, the branch subnetwork is classified into the Lumped domain, and the node method is used to construct the lumped parameter model; the PDE domain and the Lumped domain are based on the real-time exchange of pressure and flow interface conditions at the coupling boundary to trigger two-way data synchronization; for thermal modeling, the axial temperature distribution is obtained through the distributed temperature sensor, the KL decomposition method is used to extract the dominant spatial mode, and it is reconstructed into a low-dimensional state space equation; at the same time, the pipe wall heat capacity parameter is corrected online through the measured value of the heat flow meter, and finally the coupled state equation is output by fusing the hydraulic PDE solution and the thermal dimension reduction model. Step 2, in the step 1 and temperature, pressure, flow data as input, alternately use extended Kalman filter and unscented Kalman filter to identify water impedance and heat capacity parameters on line, obtain updated model ; Step 3, updating the model based on the parameter vector and the state equation set The network hydraulic coupling coefficient matrix and the thermal coupling coefficient matrix are calculated, and the corresponding feedforward decoupling compensation term is generated. Step 4, constructing a multi-model prediction framework containing a long pipeline model and an end section model according to the parameter vector of the feedforward decoupling compensation term and outputting a weighted prediction model group through a model pool weight scheduling method ; Step 5. Solving a nonlinear optimization problem including temperature error, energy consumption and actuator constraints by an online second-order programming method according to , actuator upper and lower bounds and future load prediction curve to generate an initial control sequence; The working steps of the online second-order programming method for solving the nonlinear optimization problem containing temperature error, energy consumption and actuator constraints include: By weighting the prediction model group A state prediction equation is generated, and a quadratic objective function is constructed to minimize the control amount fluctuation and minimize the temperature tracking error. Inequality constraint matrix is constructed based on actuator upper and lower bounds, and future load prediction curve is introduced as the right side item of equality constraint; Krylov subspace projection method is used to compress the dimension of decision variables, and preconditioned conjugate gradient method is used to solve the KKT system after dimension reduction; If constraint conflict is detected, inequality boundary tolerance is dynamically adjusted by relaxation factor, and BB step is used to accelerate iteration convergence; Residual norm is calculated after each iteration, and heuristic restart mechanism is triggered if residual drop rate is lower than preset threshold, and conjugate direction vector is reinitialized; Step 6, according to the serialization strategy of pump first and valve second, and combined with dead zone and rate limit, the actual pump speed and valve opening control command are generated according to the initial control sequence and feedforward decoupling compensation item; Step 7, after control execution, the temperature residual and flow oscillation data at the previous time are input, the multi-model weight and prediction time domain length are dynamically adjusted, and the updated control configuration parameters are output.

2. The method according to claim 1, wherein the method is characterized by: The steps of the alternating identification method of the extended Kalman filter and the unscented Kalman filter include: By initial full network model Generate state prediction values, compare with interpolated and smoothed sensor data, and output residual sequence; Triggering the extended Kalman filter when the Frobenius norm of the residual covariance matrix is lower than a preset threshold Performing first-order Taylor linearization processing, constructing the Kalman gain with the state Jacobian matrix, and recursively calculating the water impedance and node thermal capacity parameters in the prediction-update cycle to output the extended Kalman filter correction value; If the Frobenius norm of the residual covariance matrix exceeds the preset threshold, the weighted Sigma point is generated by the unscented Kalman filter, and the unscented transformation is used to calculate the prior covariance on the state prediction and observation mapping respectively, and then the observation residual is used to update the parameter vector; After each filtering is completed, the noise matrix of the extended Kalman filter and the unscented Kalman filter is dynamically adjusted based on the observation noise and process noise covariance reconstruction algorithm; The outputs of the extended Kalman filter and the unscented Kalman filter of adjacent periods are fused by the posterior smoothing algorithm to filter out sudden errors and maintain parameter continuity, and the updated model is output after model consistency detection.

3. The method according to claim 1, wherein the method comprises the steps of: determining the temperature of the heat pipe; and adjusting the temperature of the heat pipe to a desired temperature. The calculation step of step 3 includes: s100. Based on the water impedance parameters and nodal heat capacity parameters in the updated model output from step 2, construct the hydraulic coupling coefficient matrix using the nodal admittance matrix analysis method. ,in For nodes pressure difference, branch road The flow rate; the node admittance matrix analysis method solves for the sensitivity of the pressure gradient at the pipeline nodes to flow rate changes using the Jacobi iteration method, and outputs a diagonally dominated sparse matrix. ; s200, performing singular value decomposition on the state matrix to retain the first r significant singular values to reconstruct a reduced order matrix is the temperature state of the kth layer, and​​​​ s300, calculating a hydraulic coupling matrix non-diagonal norm of , if greater than a preset threshold , calculating a pressure disturbance suppression term, then generating a valve opening correction amount through a feedforward compensator, the formula is: （1） In equation (1), G is a gain matrix for mapping pressure fluctuations to valve opening corrections; Real-time pressure gradient residual in Pa / m, reflecting uncancelled oscillation energy. S400, Simultaneous calculation of thermo-coupling matrix condition number ,exist Greater than the preset threshold At that time, the temperature hysteresis compensation term is generated based on the pseudo-inverse matrix of the heat flux temperature transfer function using the thermal inertia decoupling algorithm. The formula expression is as follows: （2） In equation (2), is the heat flux-temperature transfer function pseudo-inverse matrix, which is used to counteract the multi-order lag effect of building thermal capacity; is the target heat flux deviation; s500, according to the real-time pipe network pressure oscillation frequency spectrum energy , the ratio of the thermal inertia residual , the final feedforward decoupling instruction is generated by weighting and fusing the compensation term, and the fusion formula is: .​ 4. The method of claim 1, wherein the method comprises: The multi-model prediction framework determines that the hydraulic dominant working condition is when the pipe network flow mutation rate is greater than 0.1 , and determines that the thermal dominant working condition is when the end temperature gradient variance is greater than 0.1 ; the long pipeline model is based on the fluid mechanics wave equation and simulates the pressure wave propagation process by solving the one-dimensional unsteady Navier-Stokes equation; the end partition model adopts the variable order thermal capacity chain state space architecture and integrates the thermal coupling coefficient matrix and the compensation term in step 3 to analyze the temperature response lag effect caused by the thermal inertia of the building envelope.

5. The method of claim 1, wherein the method comprises: The multi-model prediction framework modifies the initial parameters of each model based on the feedforward decoupling compensation item output by step 3, and calculates the model residual sensitivity by covariance matching method, wherein: Long pipeline model residual ; End partition model residual ; wherein and respectively represent the measured and predicted pressure deviation norm; and respectively represent the measured and predicted temperature deviation norm; then a recursive least square algorithm with a forgetting factor is used to update the model weight, and a weighted prediction model group is output; the update formula is: （3） In equation (3), is the weighted weight for the model pool, and are the normalized residuals for the first and the second scenarios, respectively.

6. The method of claim 1, wherein the method comprises: The working principle of step 6 is: The pump speed command set and the valve opening command set are extracted from the initial control sequence, so that the feedforward decoupling compensation item acts on the valve opening set to offset the coupling interference; The time stamp marking mechanism is used to sort the commands, and the pump speed command is executed preferentially, and the pressure gradient detector is used to monitor the pressure propagation state of the main pipe; If the pressure change rate enters the steady-state dead zone, the valve command enable signal is triggered, otherwise the valve action is delayed; For the valve opening command, a segmented S-shaped rate planner is used, the maximum allowed change slope is calculated based on the pipe segment hydraulic diameter, and the instantaneous adjustment rate is constrained by a sliding window; When a flow mutation is detected, the feedforward decoupling compensation item is used to modify the rate limit threshold; The actual pump speed and valve opening control command set with time sequence label are output.

7. The method of claim 1, wherein the method comprises: The dynamic updating method of step 7 is: the intrinsic mode function energy spectrum of the temperature residual sequence at the last time is extracted by Hilbert-Huang transform; if the low-frequency IMF energy proportion exceeds the preset threshold, it is determined that the error is dominated by thermal inertia, and the prediction time domain length expansion mechanism is used to correct by 1.2 times increment of the thermal capacity time constant; at the same time, the autocorrelation function decay rate is calculated based on the flow oscillation data, if the main lobe width is lower than the preset critical value, the multi-model weight redistribution mechanism is triggered, the correlation sub-network is identified through the oscillation phase, and the hydraulic model weight of the corresponding area is reduced; The confidence of the thermal model in the non-oscillation area is improved; and finally the updated model pool weight vector and the prediction time domain parameter set are output.

8. A real-time indoor temperature regulation control system for a heat pipe, characterized by: The indoor temperature real-time regulation method of the heat pipe in any one of claims 1-7, comprising: A data acquisition module is configured to acquire temperature, flow, pressure and valve position original signals of each heat exchange station and end room, and to complete time stamp alignment according to a unified clock to obtain a standardized time series data set ; The timing preprocessing module is used for... Wavelet packet multi-scale filtering and Bayesian interpolation are performed to separate noise and fill in missing values, resulting in a smoothed and denoised time series. ; a parameter modeling module for mapping to the pipe network topology and building a full network model with second order thermal inertia and hydraulic coupling based on a PDE / lumped parameter hybrid strategy a parameter modeling module for mapping to the pipe network topology and building a full network model with second order thermal inertia and hydraulic coupling based on a PDE / lumped parameter hybrid strategy ; An online identification module is configured to identify the water impedance and the node thermal capacity parameters by using the interpolated and smoothed and water impedance and node thermal capacity parameters as inputs, and alternately correcting the water impedance and the node thermal capacity parameters by using the extended Kalman filter and the unscented Kalman filter to obtain an updated model . A coupling and decoupling module is used for Spectral decomposition is performed on the network hydraulic and thermal coupling matrix, a sub-network coupling subspace is extracted, and a feedforward decoupling operator is generated ; a prediction control module configured to construct parallel multi-models in a rolling horizon based on , operator and future load forecast sequence, solve initial control vector by linear quadratic programming ; The execution monitoring module is configured to issue a sequence strategy of pump first and valve second and collect the room temperature residual error and the flow deviation sequence in real time ; a compensation iteration module, configured to run an iterative learning algorithm with the historical residual library as input to generate a deviation compensation increment and output a corrected control vector ; a closed-loop coordination module for updating the model parameters and the MPC weights by jointly inputting the 、 and the subnet interface error into an augmented Lagrangian iteration procedure, and outputting the configuration for the next period.

Images