Heat exchanger control method and device based on intelligent feedback mechanism

Abstract

The invention relates to the technical field of heat exchange, in particular to a heat exchanger control method and device based on an intelligent feedback mechanism. According to the method, heat exchange characteristic data of a heat exchanger are collected in real time, and a multi-dimensional state tensor containing space-time correlation characteristics is constructed; performing tensor column decomposition on the multi-dimensional state tensor, and mapping an original high-dimensional space into a low-rank tensor chain structure; the real-time heat exchange working condition of the heat exchanger is evaluated based on the low-rank tensor chain structure, and a working condition evaluation result is obtained; if the working condition evaluation result is a first working condition recognition result, current working parameters of the heat exchanger are regulated and controlled; and if the working condition evaluation result is the second working condition recognition result, the current working parameters of the heat exchanger are kept unchanged. It can be ensured that the heat exchanger always operates in the thermodynamic optimal neighborhood, the energy efficiency ratio is comprehensively increased, and the equipment service life is comprehensively prolonged.

Description

Technical Field

[0001] The present invention relates to the field of heat exchange technology, and in particular to a heat exchanger control method and device based on an intelligent feedback mechanism. Background Art

[0002] As the core heat transfer equipment in the energy system, the control accuracy of the heat exchanger is directly related to the energy efficiency and safety of the industrial process. Traditional control methods mostly use PID regulation or predictive control based on empirical models, but they have significant limitations in complex scenarios such as dynamic changing working conditions and multi-physical field coupling: First, relying on time series analysis of single parameters such as temperature and flow, it is difficult to capture the spatiotemporal correlation characteristics of fluid motion and heat conduction; second, facing the high-dimensional heterogeneous data generated by sensor networks, traditional methods will lose key thermodynamic characteristics through dimensionality reduction, resulting in inaccurate working condition assessment; third, existing control strategies lack the ability to analyze nonlinear coupling mechanisms, which can easily cause control lag when dealing with transient phenomena such as vortex shedding and boundary layer separation. In recent years, although some studies have introduced machine learning algorithms to improve adaptive capabilities, black box models lack physical interpretability and are difficult to meet the time constraints of real-time control. To address the above issues, the industry has attempted to apply tensor decomposition technology to heat transfer system modeling, such as extracting operating condition characteristics through high-order singular value decomposition. However, existing methods still have three defects: first, static tensor modeling cannot adapt to the dynamic characteristics of time-varying operating conditions, resulting in feature drift; second, the law of conservation of thermodynamics is not fully considered during the decomposition process, resulting in physical distortion of the energy transfer coefficient; third, the control strategy is separated from the feature extraction link, failing to form a closed-loop feedback mechanism. Summary of the Invention

[0003] The present invention overcomes the deficiencies of the prior art and provides a heat exchanger control method and device based on an intelligent feedback mechanism.

[0004] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is:

[0005] The first aspect of the present invention discloses a heat exchanger control method based on an intelligent feedback mechanism, comprising the following steps:

[0006] Collect heat transfer characteristic data of the heat exchanger in real time and construct a multi-dimensional state tensor containing spatiotemporal correlation characteristics;

[0007] Performing tensor column decomposition on the multidimensional state tensor to map the original high-dimensional space into a low-rank tensor chain structure;

[0008] Evaluate the real-time heat exchange working condition of the heat exchanger based on the low-rank tensor chain structure to obtain the working condition evaluation result;

[0009] If the operating condition evaluation result is the first operating condition identification result, the current operating parameters of the heat exchanger are regulated;

[0010] If the operating condition evaluation result is the second operating condition identification result, the current operating parameters of the heat exchanger are maintained unchanged.

[0011] Preferably, the heat transfer characteristic data of the heat exchanger is collected in real time to construct a multidimensional state tensor containing spatiotemporal correlation characteristics, specifically:

[0012] The embedded temperature sensor array is used to obtain the three-dimensional temperature field distribution data of the heat exchange surface. The ultrasonic flow meter and the micro-pressure differential sensor are used to synchronously collect the multi-channel fluid velocity parameters and pressure gradient data to form a raw data set with time and space tags.

[0013] Construct a four-dimensional tensor space structure, map the temperature field distribution data into spatial three-dimensional tensor slices, use the fluid velocity parameters as the fourth-dimensional motion mode, and convert the pressure gradient data into the connection weights between tensor slices to form a high-dimensional tensor basis containing spatiotemporal topological relationships;

[0014] Performing a multilinear rank-constrained decomposition on the high-dimensional tensor basis to extract a low-rank characteristic tensor that characterizes the intrinsic thermodynamic characteristics of the heat exchanger;

[0015] A sliding time window is used to intercept the real-time data stream in the original data set, and the real-time data stream is incrementally decomposed online in combination with the modal basis of the low-rank feature tensor to generate a multidimensional state tensor containing spatiotemporal correlation characteristics.

[0016] Preferably, the multidimensional state tensor is subjected to tensor column decomposition to map the original high-dimensional space into a low-rank tensor chain structure, specifically:

[0017] Define the hierarchical connection mode of tensor sequence decomposition according to the coupling relationship between the fluid domain and the solid domain, and initialize the core tensor dimension parameters with physical constraints;

[0018] The multidimensional state tensor is expanded layer by layer using a recursive tensor slicing method, and the rank selection parameters of the sub-tensor chains at each level are generated through a dynamic rank evaluation mechanism to form a low-rank sub-tensor sequence that retains the vortex shedding characteristics;

[0019] The rank selection parameter is input into the nonlinear coupling factor generator, and a modal mapping relationship with a hyperbolic tangent activation function is constructed in combination with the variation law of the thermal boundary layer thickness, and the gradient direction of each order factor matrix is ​​synchronously updated;

[0020] A residual propagation channel is established based on the low-rank sub-tensor sequence, and the decomposed residual is back-injected into the modal space of the core tensor by the alternating direction multiplier method to generate an optimized tensor chain with error correction marks;

[0021] Performing symplectic geometric projection processing on the optimized tensor chain, constructing orthogonalization constraints using the mass conservation law, and dynamically adjusting the energy transfer coefficient between sub-tensors through the error correction flag;

[0022] By integrating the orthogonalization constraints and the modal mapping relationship, a tensor chain structure with thermodynamic interpretability is obtained.

[0023] Preferably, the real-time heat exchange working condition of the heat exchanger is evaluated based on the low-rank tensor chain structure to obtain the working condition evaluation result, specifically:

[0024] Extracting modal parameters and energy transfer coefficients of the core tensor from the low-rank tensor chain structure, and combining them with error correction marks fed back by the residual propagation channel to generate dynamic eigenvectors reflecting the strength of thermodynamic coupling;

[0025] Based on the dynamic feature vector, a spatiotemporal convolution kernel is constructed, and feature convolution is performed on the sub-tensor sequence in the tensor chain through a sliding time window to extract three types of time-varying evaluation indicators: temperature gradient change rate, flow velocity fluctuation spectrum, and pressure pulsation correlation;

[0026] Projecting the time-varying evaluation index into a preset symplectic geometric space, and calculating the dynamic deviation between the current operating condition and the optimal operating mode by geodesic distance;

[0027] If the dynamic deviation between the current operating condition and the optimal operating mode is greater than a preset deviation threshold, a first operating condition identification result is generated;

[0028] If the dynamic deviation between the current operating condition and the optimal operating mode is not greater than the preset deviation threshold, a second operating condition identification result is generated.

[0029] Preferably, the time-varying evaluation index is projected into a preset symplectic geometric space, and the dynamic deviation between the current operating condition and the optimal operating mode is calculated by geodesic distance, specifically:

[0030] Acquire in advance a historical optimal operating condition dataset of the heat exchanger when performing the current heat exchange task; generate a Lie group symmetric structure of the optimal operating mode based on the historical optimal operating condition dataset, and construct a symplectic geometric manifold basis that includes temperature gradient, flow velocity stability, and pressure balance constraints;

[0031] Decomposing the time-varying evaluation index into a conserved component and an unbalanced component, projecting the unbalanced component onto the symplectic geometric manifold basis through tangent space mapping, and generating a tangent vector with physical conservation properties;

[0032] Defining an affine connection coefficient on the symplectic geometric manifold basis, and calculating the shortest geodesic path connecting the current operating point and the optimal operating mode point in combination with the motion trajectory of the tangent vector;

[0033] The shortest geodesic path is integrated to obtain a dynamic deviation parameter; the dynamic deviation parameter is input into a Sigmoid activation function for nonlinear calibration to generate an interpretable quantized value of the working condition deviation.

[0034] Preferably, if the operating condition evaluation result is the first operating condition identification result, the current operating parameters of the heat exchanger are regulated, specifically:

[0035] Decomposing the quantized value of the operating condition deviation into a temperature field distortion component, a flow velocity instability component, and a pressure imbalance component, and generating a weight correction coefficient for each component through an error correction marker of the residual propagation channel;

[0036] According to the product relationship between the weight correction coefficient and the energy transfer coefficient, a three-dimensional optimization space basis including a heat transfer efficiency improvement rate, a pressure drop constraint boundary, and an energy consumption gradient limit is constructed;

[0037] Based on the thermodynamic conservation properties of the symplectic geometric manifold basis, implicit constraints are generated to prevent parameter out-of-bounds, and the three-dimensional optimization space basis is mapped into a feasible domain tensor satisfying a Lie group symmetric structure;

[0038] Constructing a rolling optimization objective with time lag compensation on the feasible domain tensor, updating the control parameter search direction based on the error gradient fed back through the residual propagation channel using a particle swarm optimization algorithm, and solving the optimal control quantity increment sequence;

[0039] Inputting the optimal control amount increment sequence into a preset fluid inertia delay model for simulation analysis to obtain the valve opening correction amount, pump speed adjustment amount and flow distribution coefficient;

[0040] The heat exchanger is regulated and processed according to the valve opening correction amount, pump speed adjustment amount and flow distribution coefficient.

[0041] The second aspect of the present invention discloses a heat exchanger control device based on an intelligent feedback mechanism, the control device includes a memory and a processor, the memory stores a heat exchanger control method program, when the heat exchanger control method program is executed by the processor, any step of the heat exchanger control method based on the intelligent feedback mechanism is implemented.

[0042] The present invention solves the technical defects existing in the background technology, and has the following beneficial effects: the present invention uses the low-rank characteristics of tensor chains to compress redundant information, thereby improving the analytical efficiency of the spatiotemporal correlation of heat exchange characteristics; through the intelligent feedback mechanism, the abstract working condition evaluation results are converted into control decisions in real time, breaking through the traditional threshold method's perception blind spot for gradual performance degradation, and realizing early identification of hidden anomalies such as the initial stage of scaling and microscale flow instability; at the same time, combining the dual-mode response of dynamic parameter control and steady-state maintenance, while reducing the frequency of invalid control, it ensures that the heat exchanger always operates in the thermodynamically optimal neighborhood, thereby comprehensively improving the energy efficiency ratio and equipment life. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying any creative work, they can also obtain drawings of other embodiments based on these drawings.

[0044] Figure 1 The figure is a flow chart of an overall method of heat exchanger control method based on intelligent feedback mechanism;

[0045] Figure 2 A partial flow chart of a heat exchanger control method based on an intelligent feedback mechanism;

[0046] Figure 3 This is a schematic diagram of a heat exchanger control device based on an intelligent feedback mechanism. DETAILED DESCRIPTION

[0047] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0048] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0049] like Figure 1 As shown, the first aspect of the present invention discloses a heat exchanger control method based on an intelligent feedback mechanism, comprising the following steps:

[0050] S102, collecting heat transfer characteristic data of the heat exchanger in real time, and constructing a multidimensional state tensor containing spatiotemporal correlation characteristics;

[0051] S104, performing tensor column decomposition on the multidimensional state tensor, mapping the original high-dimensional space into a low-rank tensor chain structure;

[0052] S106. Evaluate the real-time heat exchange working condition of the heat exchanger based on the low-rank tensor chain structure to obtain a working condition evaluation result;

[0053] S108: If the operating condition evaluation result is the first operating condition identification result, the current operating parameters of the heat exchanger are regulated;

[0054] S110: If the operating condition evaluation result is the second operating condition identification result, the current operating parameters of the heat exchanger are maintained unchanged.

[0055] It should be noted that the present invention uses the low-rank characteristics of tensor chains to compress redundant information and improve the efficiency of analyzing the spatiotemporal correlation of heat exchange characteristics; through the intelligent feedback mechanism, the abstract working condition evaluation results are converted into control decisions in real time, breaking through the traditional threshold method's perception blind spot for gradual performance degradation, and realizing early identification of hidden anomalies such as the initial stage of scaling and microscale flow instability; at the same time, combined with the dual-mode response of dynamic parameter control and steady-state maintenance, while reducing the frequency of invalid control, it ensures that the heat exchanger always operates in the thermodynamically optimal neighborhood, thereby comprehensively improving the energy efficiency ratio and equipment life.

[0056] Preferably, the heat transfer characteristic data of the heat exchanger is collected in real time to construct a multi-dimensional state tensor containing spatiotemporal correlation characteristics, such as Figure 2 As shown, specifically:

[0057] S202, acquiring three-dimensional temperature field distribution data of the heat exchange surface through an embedded temperature sensor array, and synchronously collecting multi-channel fluid velocity parameters and pressure gradient data in combination with an ultrasonic flow meter and a micro-pressure differential sensor to form a raw data set with time and space tags;

[0058] S204. Construct a four-dimensional tensor space structure, map the temperature field distribution data into three-dimensional spatial tensor slices, use the fluid velocity parameters as the fourth-dimensional motion mode, and convert the pressure gradient data into connection weights between tensor slices to form a high-dimensional tensor basis containing spatiotemporal topological relationships;

[0059] S206, performing multilinear rank-constrained decomposition on the high-dimensional tensor basis to extract a low-rank characteristic tensor that characterizes the intrinsic thermodynamic characteristics of the heat exchanger;

[0060] It should be noted that the rank constraints of each dimension are initialized based on the physical structure of the heat exchanger. A hierarchical decomposition is then used to decompose the high-dimensional tensor into multiple low-rank sub-tensors while preserving the heat conduction path data. A residual feedback mechanism is introduced to propagate the decomposition error back to the input layer, dynamically adjusting the rank parameters of the sub-tensors to optimize feature preservation accuracy. The sub-tensors are then orthogonalized to eliminate redundant information between dimensions and ensure independent representation of thermodynamic properties. Finally, the core features of all sub-tensors are fused, combined with an incremental update mechanism based on real-time data streams, to generate a dynamically evolving low-rank feature tensor.

[0061] S208. Use a sliding time window to intercept the real-time data stream in the original data set, and perform online incremental decomposition on the real-time data stream in combination with the modal basis of the low-rank feature tensor to generate a multidimensional state tensor containing spatiotemporal correlation characteristics.

[0062] It should be noted that by setting the length of the sliding time window and intercepting the latest fragments of the real-time data stream at set time intervals, the pre-trained low-rank feature basis is then called to quickly decompose the data within the window, separating the feature components related to the essence of thermodynamics. The decomposed feature fragments are then superimposed in chronological order to form a continuously evolving dynamic feature sequence. Finally, the feature sequence is weightedly fused with the historical modal basis to generate a multidimensional state tensor that retains both real-time dynamics and long-term laws. The historical modal basis refers to the core feature combination extracted from the long-term operating data of the heat exchanger that reflects its typical thermodynamic behavior laws and serves as a benchmark for real-time data analysis.

[0063] In summary, to address the technical issues of the lack of spatiotemporal correlation between multi-source heterogeneous data and the difficulty in extracting real-time dynamic features in traditional heat exchanger control, this method constructs a four-dimensional tensor space to fuse the spatiotemporal topological relationships of temperature fields, flow rates, and pressure gradients. This allows for precise characterization of the multi-physics coupling characteristics of the heat transfer process, thereby improving the accuracy of thermodynamic state assessment under complex operating conditions.

[0064] Preferably, the multidimensional state tensor is subjected to tensor column decomposition to map the original high-dimensional space into a low-rank tensor chain structure, specifically:

[0065] Define the hierarchical connection mode of tensor sequence decomposition according to the coupling relationship between the fluid domain and the solid domain, and initialize the core tensor dimension parameters with physical constraints;

[0066] The fluid domain refers to the area of ​​heat exchanger fluid flow and heat transfer, while the solid domain refers to the solid material region that makes up the heat exchanger structure. The two exchange energy through a coupling of thermal conduction and convection. The core tensor is a low-dimensional tensor unit generated during the tensor sequence decomposition process that carries the key thermodynamic parameters of the heat exchanger (such as temperature gradient and energy transfer coefficient). It serves as the core hub connecting the sub-tensor chains at each level, preserving the essential correlation characteristics of multidimensional data through dimensional compression.

[0067] The multidimensional state tensor is expanded layer by layer using a recursive tensor slicing method, and the rank selection parameters of the sub-tensor chains at each level are generated through a dynamic rank evaluation mechanism to form a low-rank sub-tensor sequence that retains the vortex shedding characteristics;

[0068] It should be noted that, taking the plate heat exchanger as an example, the collected 100×100×20×300 four-dimensional state tensor (spatial coordinates X / Y / Z×time series) is recursively sliced:

[0069] (1) Cut the first layer sub-tensor (100 × 100 × 20 × 10) along the time axis, calculate the first five singular values ​​after HOSVD decomposition, and the proportion is 92%. Set the rank parameter R1 = 5 of this layer to generate the sub-tensor chain T1 (100 × R1 × 10) containing the main fluid temperature zone migration pattern;

[0070] (2) T1 is sliced ​​twice along the Z axis (100×5×2), and the second layer singular value distribution is detected to have a bimodal characteristic (the first three items account for 88%). R2=3 is dynamically set to capture the vortex shedding characteristics caused by boundary layer separation, and the sub-tensor chain T2 (100×R2×2) is output;

[0071] (3) When decomposing T2 to the third level, residual energy monitoring revealed that R3 = 2 retained 95% of the vortex frequency components (0.5-2 Hz), ultimately forming a T3 (100 × 2) low-order sequence. The second channel tensor slice successfully retained the Karman vortex street spectrum characteristics corresponding to a Reynolds number of Re = 2500, as verified by FFT. The rank parameters at each level were dynamically adjusted by real-time monitoring of the Frobenius norm change rate of the velocity gradient matrix within the sub-tensor slice, with an error tolerance of ±3%.

[0072] The rank selection parameter is input into the nonlinear coupling factor generator, and a modal mapping relationship with a hyperbolic tangent activation function is constructed in combination with the variation law of the thermal boundary layer thickness, and the gradient direction of each order factor matrix is ​​synchronously updated;

[0073] The rank parameters of each level obtained by decomposition (such as R1=5, R2=3, etc.) and the real-time monitored thermal boundary layer thickness data (such as a dynamic range of 0.2mm-1.5mm) are input into the generator together; the input parameters are nonlinearly transformed by the hyperbolic tangent function to generate a modal weight matrix reflecting the flow velocity-temperature coupling strength (for example, the weight value is constrained to be in the interval [-1, 1]); the update direction of the weight matrix is ​​calculated according to the current thermal boundary layer thickening rate (such as 0.05mm per second). If a sudden increase in the boundary layer is detected (such as exceeding 0.1mm / s), the gradient update step size is increased to accelerate the response; the updated factor matrix is ​​substituted into the tensor slicing processing of the next time window (such as t+1 second data), and the gradient direction is reversely fine-tuned by comparing the deviation between the predicted vortex frequency (such as 1.8Hz) and the actual sensor data (such as 2.0Hz); when the modal weight fluctuation is less than ±2% in three consecutive time windows, the current mapping relationship is locked.

[0074] A residual propagation channel is established based on the low-rank sub-tensor sequence, and the decomposed residual is back-injected into the modal space of the core tensor by the alternating direction multiplier method to generate an optimized tensor chain with error correction marks;

[0075] The reconstruction error of each decomposed sub-tensor (e.g., the 100×2 tensor at the third level) is calculated, and the residual energy is quantified using the Frobenius norm (for example, correction is triggered when the residual ratio exceeds 5%). A bidirectional data channel is established between adjacent sub-tensors (e.g., a 3×2 connection matrix is ​​constructed between the second layer R2=3 and the third layer R3=2), and the residuals are distributed to each channel according to their weights. The residual data (e.g., the missing part of the vortex shedding feature in the 0.5-2 Hz frequency band) is decomposed into gradient components using the alternating direction multiplication method and superimposed into the corresponding modal space of the core tensor (e.g., T3(100×2)). Binary flags are embedded in the optimized tensor chain according to the residual correction amplitude (e.g., velocity field correction of ±0.3 m / s) (e.g., 0010 indicates that local flow overload requires priority correction). The marked tensor chain is input into the downstream symplectic geometry projection module. If the energy conservation deviation is still detected to exceed 2%, the residual weights are returned and redistributed until the thermodynamic constraints are met.

[0076] Performing symplectic geometric projection processing on the optimized tensor chain, constructing orthogonalization constraints using the mass conservation law, and dynamically adjusting the energy transfer coefficient between sub-tensors through the error correction flag;

[0077] By integrating the orthogonalization constraints and the modal mapping relationship, a tensor chain structure with thermodynamic interpretability is obtained.

[0078] According to the law of conservation of mass, the difference in the inflow / outflow energy of each sub-tensor in the optimized tensor chain is calculated (for example, the energy corresponding to a flow rate of 1.2 m / s in a certain area must match the deviation of ±0.05 m / s in the adjacent area), and an orthogonalized energy balance matrix is ​​constructed; when the error mark detects that the local energy loss exceeds 5% (for example, the energy gap mark 0011 in the vortex area), the transfer coefficient between the sub-tensors is adjusted according to the constraint direction of the conservation matrix (for example, the heat conduction coefficient is corrected from 0.8 to the range of 0.85±0.02); the independence constraint rules of the orthogonalized matrix (for example, the energy exchange amount of adjacent areas must be ≤ 20% of the total energy) are combined with the modal The weight relationship of the mapping (such as the 0.6 coupling strength generated by the hyperbolic tangent function) is matrix-multiplied to generate the fused thermodynamic association rules; the fused tensor chain is jointly checked for vortex-temperature gradient (such as the temperature gradient at the center of the vortex must be ≥15°C / m). If a rule conflict is detected (such as the gradient in a certain area is only 12°C / m), the coefficient weight is returned for redistribution; when the energy conservation deviation is ≤1.5% within 10 consecutive time windows and the temperature gradient meets the standard, the current tensor chain structure is locked, and the final output can be clearly interpreted as a characteristic tensor chain of physical phenomena such as "high-speed vortex enhanced heat transfer zone" and "boundary layer stagnation zone".

[0079] In summary, this method generates a tensor chain structure that combines high-precision dimensionality reduction and thermodynamic mechanism expression capabilities through fluid-solid coupling constraints and symplectic geometry projection, providing a physically meaningful and dynamically adaptive feature expression basis for real-time working condition evaluation.

[0080] Preferably, the real-time heat exchange working condition of the heat exchanger is evaluated based on the low-rank tensor chain structure to obtain the working condition evaluation result, specifically:

[0081] Extracting modal parameters and energy transfer coefficients of the core tensor from the low-rank tensor chain structure, and combining them with error correction marks fed back by the residual propagation channel to generate dynamic eigenvectors reflecting the strength of thermodynamic coupling;

[0082] Taking the plate heat exchanger as an example, its low-rank tensor chain structure contains three core tensor levels after decomposition: extract the modal parameters from the first layer of core tensors (dimension 100×5×10), including the main mode amplitude of the temperature gradient (such as 0.35-1.2W / (m·K)) and the corresponding flow velocity coupling coefficient (such as 0.8-1.5m / s / ℃), and read the energy transfer coefficient matrix (such as the heat conduction weight 0.65±0.03); obtain the error correction mark of the current time window through the residual channel (such as binary code 0011 indicates that the local temperature gradient residual exceeds 5% and the energy coefficient deviation is ±0.02), and convert it into a normalized correction factor (such as [0.95, 1.05, 0.98, 1.03]); the modal parameters (main modal amplitude 1.02, coupling coefficient 1.2), energy coefficient 0.68 and correction factor are combined in a weighted ratio (e.g., 0.4:0.3:0.3) to generate a four-dimensional dynamic eigenvector (e.g., [1.02, 1.2, 0.68×1.05, 0.68×0.98]), which reflects the thermal-fluid coupling intensity of the current vortex zone (coordinates X=30-50, Y=20-40) in real time; the eigenvector is updated every 5 seconds. When it is detected that the error mark triggers the flow velocity correction three times in a row (e.g., mark position 11XX), the weight of the flow velocity coupling coefficient is automatically increased to 0.4 to ensure the sensitivity of the eigenvector to turbulent mutations.

[0083] Based on the dynamic feature vector, a spatiotemporal convolution kernel is constructed, and feature convolution is performed on the sub-tensor sequence in the tensor chain through a sliding time window to extract three types of time-varying evaluation indicators: temperature gradient change rate, flow velocity fluctuation spectrum, and pressure pulsation correlation;

[0084] Taking a plate heat exchanger as an example, the dynamic feature vectors generated in real time (dimension 1×4, such as [1.02, 1.2, 0.71, 0.67]) are used to construct a spatiotemporal convolution kernel: a three-dimensional kernel structure (time×space×feature channel) is designed, with the time dimension covering a 5-second window (step length 2 seconds), the spatial dimension bound to the coordinates of the vortex active area (X=30-50, Y=20-40), and the feature channels corresponding to three types of parameters: temperature, flow rate, and pressure; a convolution operation is performed on the sub-tensor sequence (such as a 100×2 tensor of 5 consecutive time windows) every 2 seconds. When calculating the temperature gradient change rate, the maximum gradient difference within 3 consecutive seconds (such as the increase from 0.8W / (m·K) to the maximum gradient difference) is extracted. to 1.1 W / (m·K), with a change rate of 0.1 W / (m·K) / s); perform time-frequency conversion on velocity sub-tensor slices (such as a 100×2 matrix) to capture the 0.8-1.5 Hz main frequency fluctuation (corresponding to the Karman vortex street characteristics), and simultaneously calculate the covariance matrix of the pressure tensor (inlet and outlet areas) (such as a 0.35-0.62 correlation coefficient) to reflect the pulsation propagation characteristics; finally, the three types of indicators (temperature change rate 0.1, velocity main frequency 1.2 Hz, pressure covariance 0.5) are normalized to the [0-1] interval, combined into a time-varying evaluation vector [0.67, 0.8, 0.5], and synchronously stored in the symplectic geometric space mapping queue.

[0085] Projecting the time-varying evaluation index into a preset symplectic geometric space, and calculating the dynamic deviation between the current operating condition and the optimal operating mode by geodesic distance;

[0086] If the dynamic deviation between the current operating condition and the optimal operating mode is greater than a preset deviation threshold, a first operating condition identification result is generated;

[0087] If the dynamic deviation between the current operating condition and the optimal operating mode is not greater than the preset deviation threshold, a second operating condition identification result is generated.

[0088] In summary, this method obtains the dynamic deviation index of the operating condition and identifies abnormal operating conditions such as heat transfer efficiency degradation and vortex instability of the heat exchanger in real time, providing a decision-making basis with both timeliness and reliability for the intelligent control system, effectively preventing the aggravation of scaling or structural fatigue damage.

[0089] Preferably, the time-varying evaluation index is projected into a preset symplectic geometric space, and the dynamic deviation between the current operating condition and the optimal operating mode is calculated by geodesic distance, specifically:

[0090] Acquire in advance a historical optimal operating condition dataset of the heat exchanger when performing the current heat exchange task; generate a Lie group symmetric structure of the optimal operating mode based on the historical optimal operating condition dataset, and construct a symplectic geometric manifold basis that includes temperature gradient, flow velocity stability, and pressure balance constraints;

[0091] Among them, the historical optimal operating condition data set refers to the set of optimal values ​​of key parameters such as temperature gradient, flow rate stability and pressure balance recorded in the historical operation of the heat exchanger under the same heat exchange task, including the operating parameter benchmarks under thermodynamically verified high-efficiency, stable and low-energy consumption conditions.

[0092] Taking the plate heat exchanger as an example, its historical optimal data set includes 100 sets of operating parameters (temperature gradient 0.8-1.2 W / (m·K), flow rate standard deviation ≤ 0.15 m / s, inlet and outlet pressure difference ≤ 5 kPa):

[0093] Select the main mode of temperature gradient (amplitude 1.05W / (m·K)), velocity covariance matrix (diagonal term 0.12m 2 / s 2 ) and the pressure balance tensor (inlet / outlet covariance 0.6) as Lie group generators to construct a 4×4 parameter space transformation matrix, where the temperature gradient corresponds to a rotation transformation (θ = 0.35 rad), the velocity stability corresponds to a translation operation (Δv = 0.1 m / s), and the pressure balance constraint shear transformation (γ = 0.05);

[0094] The above generator matrix is ​​tensor-producted with the symplectic structure matrix (which satisfies the energy conservation condition) to form an 8-dimensional symplectic geometric basis, in which the temperature gradient constraint is embedded in the 1-2 dimensional orthogonal subspace (allowing ±0.2 W / (m·K) fluctuation), the flow rate stability is assigned to the 3-5 dimensional symplectic dual space (flow rate fluctuation spectrum bandwidth ≤ 0.3 Hz), and the pressure balance is mapped to the 6-7 dimensional covariance matrix (pressure pulsation correlation ≥ 0.55).

[0095] By projecting actual operating data (such as a temperature gradient of 0.95 W / (m·K) and a flow velocity standard deviation of 0.18 m / s) onto the substrate, it is verified that when the pressure difference exceeds the limit (e.g., 6.2 kPa), the corresponding coordinate point exceeds the permitted boundary of the symplectic flow (the 7th-dimensional coordinate value is greater than 1.5), triggering a constraint alarm, thus proving the substrate's automatic rejection of non-equilibrium states.

[0096] Decomposing the time-varying evaluation index into a conserved component and an unbalanced component, projecting the unbalanced component onto the symplectic geometric manifold basis through tangent space mapping, and generating a tangent vector with physical conservation properties;

[0097] Defining an affine connection coefficient on the symplectic geometric manifold basis, and calculating the shortest geodesic path connecting the current operating point and the optimal operating mode point in combination with the motion trajectory of the tangent vector;

[0098] Taking a plate heat exchanger as an example, its symplectic geometry manifold basis has been constructed into an 8-dimensional structure (2-dimensional temperature gradient constraint, 3-dimensional flow velocity stability, and 3-dimensional pressure balance):

[0099] Based on the main direction of the temperature gradient (0.8-1.2W / (m·K)) and the velocity covariance matrix (diagonal item 0.12m 2 / s 2 ), construct the affine connection coefficient tensor, in which the temperature gradient direction component allows orthogonal fluctuations of ±0.02, and the velocity stability component is used to constrain the propagation direction of velocity fluctuations;

[0100] The coordinates of the current operating point are (0.95, 1.1, 0.15, 0.6, 0.55, 0.58, 1.2, 0.9), and the coordinates of the optimal mode point are (1.05, 1.0, 0.12, 0.65, 0.6, 0.6, 1.0, 1.0). The direction of the tangent vector movement is adjusted iteratively through the connection coefficient, and the coordinate offset is corrected (for example, the step size of the third dimension is adjusted from 0.03 to 0.025 to match the flow velocity constraint);

[0101] When the pressure balance component (6th-7th dimension) exceeds the limit (e.g., current point 1.2 > threshold 1.0), the corresponding connection coefficient is automatically increased so that the geodesic bypasses the non-physical area and ultimately obtains the shortest geodesic path;

[0102] The shortest geodesic path is integrated to obtain a dynamic deviation parameter; the dynamic deviation parameter is input into a Sigmoid activation function for nonlinear calibration to generate an interpretable quantized value of the working condition deviation.

[0103] In summary, this method converts the deviation of the working conditions of complex multi-physical field coupling into quantitative indicators with clear physical meanings (such as "vortex energy loss exceeds the standard" and "pressure energy transmission instability"), thereby achieving accurate graded early warning of abnormal conditions such as heat transfer efficiency attenuation and flow instability, and providing a judgment basis for the control strategy that is both mathematically rigorous and engineering operability.

[0104] Preferably, if the operating condition evaluation result is the first operating condition identification result, the current operating parameters of the heat exchanger are regulated, specifically:

[0105] Decomposing the quantized value of the operating condition deviation into a temperature field distortion component, a flow velocity instability component, and a pressure imbalance component, and generating a weight correction coefficient for each component through an error correction marker of the residual propagation channel;

[0106] According to the product relationship between the weight correction coefficient and the energy transfer coefficient, a three-dimensional optimization space basis including a heat transfer efficiency improvement rate, a pressure drop constraint boundary, and an energy consumption gradient limit is constructed;

[0107] Based on the thermodynamic conservation properties of the symplectic geometric manifold basis, implicit constraints are generated to prevent parameter out-of-bounds, and the three-dimensional optimization space basis is mapped into a feasible domain tensor satisfying a Lie group symmetric structure;

[0108] Constructing a rolling optimization objective with time lag compensation on the feasible domain tensor, updating the control parameter search direction based on the error gradient fed back through the residual propagation channel using a particle swarm optimization algorithm, and solving the optimal control quantity increment sequence;

[0109] Inputting the optimal control amount increment sequence into a preset fluid inertia delay model for simulation analysis to obtain the valve opening correction amount, pump speed adjustment amount and flow distribution coefficient;

[0110] The heat exchanger is regulated and processed according to the valve opening correction amount, pump speed adjustment amount and flow distribution coefficient.

[0111] Taking the plate heat exchanger as an example, when the quantified value of the operating condition deviation is 0.65 (threshold 0.6), the first operating condition identification is triggered:

[0112] The deviation is decomposed into a temperature field distortion component of 0.3 (the temperature difference in the main channel X = 20-40 area exceeds 3°C), a flow velocity instability component of 0.25 (the fluctuation frequency of 1.8Hz exceeds the limit of 0.3Hz), and a pressure imbalance component of 0.1 (the inlet and outlet pressure difference of 6kPa exceeds the threshold of 5kPa). The weight coefficients (temperature 0.4, flow velocity 0.35, pressure 0.25) are generated according to the error mark 0110.

[0113] The heat transfer efficiency improvement rate is set to 5%-8%, the pressure drop constraint boundary is ≤15% of the initial value, and the energy consumption gradient limit is ≤2% per 10 minutes. Multiplying with the energy coefficient (0.65 / 0.72 / 0.58) generates the base boundary (for example, the efficiency improvement upper limit is 0.72×8%=5.76%).

[0114] Using the symplectic manifold basis constraints (temperature gradient fluctuation ±0.2 W / (m·K) and pressure pulsation correlation ≥0.6), the basis is mapped into a three-dimensional feasible domain tensor (40×30×25 grid points), with each grid point associated with a Lie group symmetry parameter (e.g., temperature rotation transformation angle θ≤0.4 rad).

[0115] Set a 10-second rolling time window (including a 3-second valve response delay and a 5-second pump speed inertia delay) and search within the feasible region for a control variable combination that increases heat transfer efficiency by 5.2%, increases pressure drop by ≤12%, and maintains an energy consumption gradient of 1.8% / 10 minutes.

[0116] After simulation, the fluid inertia model parameters (inertia time constant 8 seconds, pipeline damping coefficient 0.05) output valve opening correction +8% (step size 2% / minute), pump speed reduced by 50rpm (adjusted in 5 times), flow distribution coefficient main channel 0.7 / bypass 0.3, and the deviation after regulation was reduced to 0.55.

[0117] In summary, in order to solve the technical problems of the lack of physical constraints in the coordinated optimization of multiple parameters in traditional heat exchanger control, and the easy occurrence of secondary instability and dynamic response hysteresis caused by parameter adjustment, this method decomposes the deviation into temperature, flow rate, and pressure components and constructs a three-dimensional optimization space basis. Under the mathematical framework that strictly follows the law of conservation of thermodynamics, it realizes the solution of the coordinated control quantity of multiple actuators such as valves and pump speeds, ensuring that parameter adjustment is always within the physical feasible domain of heat transfer efficiency, pressure drop, and energy consumption. In combination with time lag compensation and inertia delay simulation, it breaks through the limitation of traditional PID control's poor adaptability to sudden working conditions, effectively avoids problems such as aggravated scaling and vortex instability caused by over-adjustment oscillation or response lag, and improves the safety of the control process and system stability.

[0118] like Figure 3 As shown, the second aspect of the present invention discloses a heat exchanger control device 8 based on an intelligent feedback mechanism, wherein the control device includes a memory 60 and a processor 80, wherein a heat exchanger control method program is stored in the memory 60. When the heat exchanger control method program is executed by the processor 80, any step of the heat exchanger control method based on the intelligent feedback mechanism is implemented.

[0119] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0120] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0121] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0122] Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0123] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0124] The above are only specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A heat exchanger control method based on an intelligent feedback mechanism, characterized in that: The following steps are involved: Collect heat transfer characteristic data of the heat exchanger in real time and construct a multi-dimensional state tensor containing spatiotemporal correlation characteristics; Performing tensor column decomposition on the multidimensional state tensor to map the original high-dimensional space into a low-rank tensor chain structure; Evaluate the real-time heat exchange working condition of the heat exchanger based on the low-rank tensor chain structure to obtain the working condition evaluation result; If the operating condition evaluation result is the first operating condition identification result, the current operating parameters of the heat exchanger are regulated; If the operating condition evaluation result is the second operating condition identification result, the current operating parameters of the heat exchanger are maintained unchanged.

2. The heat exchanger control method based on intelligent feedback mechanism according to claim 1, characterized in that: Collect the heat transfer characteristic data of the heat exchanger in real time and construct a multi-dimensional state tensor containing spatiotemporal correlation characteristics, specifically: The embedded temperature sensor array is used to obtain the three-dimensional temperature field distribution data of the heat exchange surface. The ultrasonic flow meter and the micro-pressure differential sensor are used to synchronously collect the multi-channel fluid velocity parameters and pressure gradient data to form a raw data set with time and space tags. Construct a four-dimensional tensor space structure, map the temperature field distribution data into spatial three-dimensional tensor slices, use the fluid velocity parameters as the fourth-dimensional motion mode, and convert the pressure gradient data into the connection weights between tensor slices to form a high-dimensional tensor basis containing spatiotemporal topological relationships; Performing a multilinear rank-constrained decomposition on the high-dimensional tensor basis to extract a low-rank characteristic tensor that characterizes the intrinsic thermodynamic characteristics of the heat exchanger; A sliding time window is used to intercept the real-time data stream in the original data set, and the real-time data stream is incrementally decomposed online in combination with the modal basis of the low-rank feature tensor to generate a multidimensional state tensor containing spatiotemporal correlation characteristics.

3. The heat exchanger control method based on intelligent feedback mechanism according to claim 1, characterized in that: The multidimensional state tensor is decomposed into a tensor column, and the original high-dimensional space is mapped into a low-rank tensor chain structure, specifically: Define the hierarchical connection mode of tensor sequence decomposition according to the coupling relationship between the fluid domain and the solid domain, and initialize the core tensor dimension parameters with physical constraints; The multidimensional state tensor is expanded layer by layer using a recursive tensor slicing method, and the rank selection parameters of the sub-tensor chains at each level are generated through a dynamic rank evaluation mechanism to form a low-rank sub-tensor sequence that retains the vortex shedding characteristics; The rank selection parameter is input into the nonlinear coupling factor generator, and a modal mapping relationship with a hyperbolic tangent activation function is constructed in combination with the variation law of the thermal boundary layer thickness, and the gradient direction of each order factor matrix is ​​synchronously updated; A residual propagation channel is established based on the low-rank sub-tensor sequence, and the decomposed residual is back-injected into the modal space of the core tensor by the alternating direction multiplier method to generate an optimized tensor chain with error correction marks; Performing symplectic geometric projection processing on the optimized tensor chain, constructing orthogonalization constraints using the mass conservation law, and dynamically adjusting the energy transfer coefficient between sub-tensors through the error correction flag; By integrating the orthogonalization constraints and the modal mapping relationship, a tensor chain structure with thermodynamic interpretability is obtained.

4. The heat exchanger control method based on intelligent feedback mechanism according to claim 1, characterized in that: The real-time heat exchange working condition of the heat exchanger is evaluated based on the low-rank tensor chain structure to obtain the working condition evaluation results, specifically: Extracting modal parameters and energy transfer coefficients of the core tensor from the low-rank tensor chain structure, and combining them with error correction marks fed back by the residual propagation channel to generate dynamic eigenvectors reflecting the strength of thermodynamic coupling; Based on the dynamic feature vector, a spatiotemporal convolution kernel is constructed, and feature convolution is performed on the sub-tensor sequence in the tensor chain through a sliding time window to extract three types of time-varying evaluation indicators: temperature gradient change rate, flow velocity fluctuation spectrum, and pressure pulsation correlation; Projecting the time-varying evaluation index into a preset symplectic geometric space, and calculating the dynamic deviation between the current operating condition and the optimal operating mode by geodesic distance; If the dynamic deviation between the current operating condition and the optimal operating mode is greater than a preset deviation threshold, a first operating condition identification result is generated; If the dynamic deviation between the current operating condition and the optimal operating mode is not greater than the preset deviation threshold, a second operating condition identification result is generated.

5. The heat exchanger control method based on intelligent feedback mechanism according to claim 4, characterized in that: The time-varying evaluation index is projected into a preset symplectic geometric space, and the dynamic deviation between the current operating condition and the optimal operating mode is calculated by geodesic distance, specifically: Acquire in advance a historical optimal operating condition dataset of the heat exchanger when performing the current heat exchange task; generate a Lie group symmetric structure of the optimal operating mode based on the historical optimal operating condition dataset, and construct a symplectic geometric manifold basis that includes temperature gradient, flow velocity stability, and pressure balance constraints; Decomposing the time-varying evaluation index into a conserved component and an unbalanced component, projecting the unbalanced component onto the symplectic geometric manifold basis through tangent space mapping, and generating a tangent vector with physical conservation properties; Defining an affine connection coefficient on the symplectic geometric manifold basis, and calculating the shortest geodesic path connecting the current operating point and the optimal operating mode point in combination with the motion trajectory of the tangent vector; Performing integration processing on the shortest geodesic path to obtain a dynamic deviation parameter; The dynamic deviation parameter is input into the Sigmoid activation function for nonlinear calibration to generate an interpretable quantified value of the working condition deviation.

6. The heat exchanger control method based on intelligent feedback mechanism according to claim 5, characterized in that: If the operating condition evaluation result is the first operating condition identification result, the current operating parameters of the heat exchanger are regulated, specifically: Decomposing the quantized value of the operating condition deviation into a temperature field distortion component, a flow velocity instability component, and a pressure imbalance component, and generating a weight correction coefficient for each component through an error correction marker of the residual propagation channel; According to the product relationship between the weight correction coefficient and the energy transfer coefficient, a three-dimensional optimization space basis including a heat transfer efficiency improvement rate, a pressure drop constraint boundary, and an energy consumption gradient limit is constructed; Based on the thermodynamic conservation properties of the symplectic geometric manifold basis, implicit constraints are generated to prevent parameter out-of-bounds, and the three-dimensional optimization space basis is mapped into a feasible domain tensor satisfying a Lie group symmetric structure; Constructing a rolling optimization objective with time lag compensation on the feasible domain tensor, updating the control parameter search direction based on the error gradient fed back through the residual propagation channel using a particle swarm optimization algorithm, and solving the optimal control quantity increment sequence; Inputting the optimal control amount increment sequence into a preset fluid inertia delay model for simulation analysis to obtain the valve opening correction amount, pump speed adjustment amount and flow distribution coefficient; The heat exchanger is regulated and processed according to the valve opening correction amount, pump speed adjustment amount and flow distribution coefficient.

7. A heat exchanger control device based on an intelligent feedback mechanism, characterized in that: The control device includes a memory and a processor, wherein a heat exchanger control method program is stored in the memory. When the heat exchanger control method program is executed by the processor, the steps of the heat exchanger control method based on the intelligent feedback mechanism as described in any one of claims 1 to 6 are implemented.

Images