Method and system for predicting heat dissipation performance of heat exchanger based on multi-parameter fitting

Abstract

The invention relates to the technical field of electric data processing, in particular to a heat exchanger heat dissipation performance prediction method and system based on multi-parameter fitting, and the method comprises the steps that multi-dimensional operation parameters of a heat exchanger are obtained, and the multi-dimensional operation parameters comprise a fluid flow speed sequence and a heat dissipation capacity sequence; according to the method, time dislocation of input and output is adaptively corrected based on cross correlation through the lag offset, and the dynamic evolution index quantifies the dependency of a heat dissipation behavior on a historical path through a phase plane track area. And then, the indexes reflecting the physical characteristics are converted into a weight correction coefficient of a self-attention mechanism and a penalty factor of VMD decomposition, so that the model can dynamically adjust the attention degree of historical data according to real-time working conditions, and the problems that the prediction precision is reduced and the response is lagged when the working conditions violently fluctuate in a traditional black box model are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of electrical data processing. More specifically, this invention relates to a method and system for predicting the heat dissipation performance of heat exchangers based on multi-parameter fitting. Background Technology

[0002] As a core energy conversion component in thermal systems, heat exchangers are commonly used in industries such as energy, chemical, aerospace, and automotive. In actual operation scenarios, in order to achieve precise thermal management monitoring and system optimization, it is usually necessary to quickly and accurately predict the transient heat dissipation performance of the heat exchanger based on real-time operating parameters such as fluid flow rate and inlet / outlet temperature difference collected by sensors.

[0003] However, under complex actual industrial conditions, existing prediction methods face multiple challenges: First, due to the large heat capacity of the heat exchanger's metal frame and heat dissipation fins, its heat dissipation performance exhibits a significant physical delay in response to input conditions (such as flow rate fluctuations), i.e., the thermal inertia effect. During real-time monitoring, this thermal inertia causes an asymmetric misalignment between the flow rate measurement sequence and the heat dissipation response sequence on the time axis. Directly using these misaligned data for prediction will distort the causal chain between the data, leading to significant deviations in the prediction model when operating conditions change drastically.

[0004] Secondly, heat exchangers exhibit significant hysteresis under unsteady flow conditions, such as frequent start-ups and shutdowns or large load fluctuations. The boundary layer development processes differ between the rising and falling velocity phases of the fluid flow, resulting in inconsistent heat dissipation capacity even at the same velocity point. Traditional linear or simple nonlinear fitting methods often overlook this path dependence, failing to capture the impact of historical fluid flow effects on current heat dissipation, thus leading to decreased prediction accuracy under transient conditions. Summary of the Invention

[0005] This invention provides a method and system for predicting the heat dissipation performance of heat exchangers based on multi-parameter fitting. It aims to solve the problem that prediction methods in related technologies are limited by the time sequence misalignment caused by the thermal inertia of the heat exchanger and the hysteresis effect caused by unsteady flow, making it difficult to capture path-dependent characteristics under dynamic operating conditions, thus resulting in a significant decrease in prediction accuracy.

[0006] In a first aspect, the present invention provides a method for predicting the heat dissipation performance of a heat exchanger based on multi-parameter fitting, comprising: acquiring multi-dimensional operating parameters of the heat exchanger, the multi-dimensional operating parameters including a fluid velocity sequence and a heat dissipation sequence; constructing a hysteresis offset characterizing the thermal inertial response delay of the heat exchanger based on the cross-correlation between the fluid velocity sequence and the heat dissipation sequence; performing time alignment of the fluid velocity sequence and the heat dissipation sequence based on the hysteresis offset, and constructing a dynamic evolution index characterizing the unsteady flow hysteresis effect based on the area enclosed by the trajectories formed by the time-aligned fluid velocity sequence and the heat dissipation sequence in the phase plane; constructing a weight correction coefficient for guiding the model attention allocation by combining the dynamic evolution index, the hysteresis offset, the rate of change of the fluid velocity sequence, and the average temperature difference between the inlet and outlet fluids of the heat exchanger; and inputting the multi-dimensional operating parameters into a preset prediction model for processing to obtain a predicted value of the heat dissipation performance of the heat exchanger, wherein the processing of the prediction model includes: embedding the weight correction coefficient into the weight scoring function of a self-attention mechanism to adjust the model's attention to historical data. By constructing hysteresis offset and dynamic evolution index and converting them into weight correction coefficients for attention mechanism, the prediction model can accurately identify and compensate for the time misalignment and path dependence between fluid flow rate and heat dissipation response under complex working conditions such as chemical and energy industries (e.g., drastic flow fluctuations or frequent equipment start-ups and shutdowns). This effectively solves the problem of causal chain distortion caused by physical delay and significantly improves the accuracy and real-time performance of transient heat dissipation performance prediction.

[0007] Furthermore, the method for constructing the hysteresis offset at each time step includes: calculating the sum of the products of the flow velocity at each time step within the sliding window and the heat dissipation compensated based on the baseline delay step size. The hysteresis offset is proportional to the sum of the products and also positively correlated with the difference between the maximum and minimum values ​​of the flow velocity sequence within the sliding window. By dynamically constructing the hysteresis offset index using the cross-correlation within the sliding window, the time alignment deviation can be adaptively adjusted according to the fluctuations in industrial operating conditions (such as irregular flow oscillations), thereby providing a more interpretable time axis correction basis for the neural network at the physical level and ensuring a rigorous correspondence between the input and output sequences in energy logic.

[0008] Furthermore, the method for calculating the baseline delay step size at each time point includes: within the sliding window of the current time point, determining the time delay that maximizes the cross-correlation function value between the fluid velocity sequence and the heat dissipation sequence as the baseline delay step size. Determining the baseline delay step size by maximizing the cross-correlation function value achieves automatic quantification of the thermal hysteresis time under specific physical structures of the heat exchanger, ensuring that the subsequent feature extraction process can accurately pinpoint the key time points of the thermal inertial response and reducing prediction noise caused by insufficient data alignment accuracy.

[0009] Furthermore, the method for constructing the dynamic evolution index at each time step includes: within the sliding window of the current time step, calculating the logarithmic sum of the absolute values ​​of the vector cross products between adjacent discrete points formed by the time-compensated fluid velocity sequence and the heat dissipation sequence; the dynamic evolution index is proportional to the product of the logarithmic sum and the hysteresis offset. By calculating the dynamic evolution index through the vector cross product of the phase plane trajectory, the influence of the fluid boundary layer development history on the current heat dissipation is accurately quantified, i.e., the degree of hysteresis loop opening. This allows the system to identify the difference in heat dissipation capacity even at the same flow rate during the acceleration and deceleration phases, greatly enhancing its ability to capture unsteady and complex flow characteristics.

[0010] Furthermore, the method for constructing the weight correction coefficient includes: dividing the product of the dynamic evolution index and the tangent of the hysteresis offset by the hyperbolic cosine function value of the product of the rate of change of the fluid velocity sequence and the average temperature difference between the inlet and outlet fluids of the heat exchanger, to obtain the basic weight; adding the basic weight to a preset parameter of instantaneous response sensitivity to obtain the weight correction coefficient. By combining physical indicators with parameters such as temperature difference and rate of change to construct a nonlinear weight correction coefficient, and introducing a hyperbolic cosine function into the denominator to simulate the heat conduction saturation effect, the prediction model can automatically increase its focus on historical data during high-load energy accumulation phases, effectively mitigating the problem of model prediction lag or misjudgment during sudden thermal shocks.

[0011] Furthermore, the processing of the prediction model also includes: performing variational mode decomposition on the multidimensional operating parameters to obtain multiple intrinsic mode components containing heat transfer characteristics at different scales; wherein, the variational mode decomposition process utilizes the hysteresis offset as an adaptive penalty factor to enhance the extraction capability of low-frequency mode components dominated by thermal inertia. By using the hysteresis offset as an adaptive penalty factor in variational mode decomposition (VMD), the model is forced to enhance the extraction of low-frequency heat dissipation characteristics dominated by thermal inertia, thereby clearly separating the long-period components reflecting the core heat dissipation law of the heat exchanger even in industrial environments containing high-frequency sensor noise, thus improving the robustness of feature engineering.

[0012] Furthermore, the hysteresis offset is used as an adaptive penalty factor, including: when the value of the hysteresis offset is greater than a preset threshold, increasing the retention weight of low-frequency intrinsic mode components during variational mode decomposition. By actively increasing the retention weight of low-frequency intrinsic mode components when the hysteresis offset is large, the model can more robustly lock in physically meaningful trend features when operating conditions fluctuate drastically and thermal hysteresis is prominent, avoiding the loss of key thermal response signals due to being mistakenly identified as noise during decomposition, and ensuring the feature integrity of the model under harsh operating conditions.

[0013] Furthermore, the preset prediction model is a long short-term memory neural network model that incorporates a self-attention mechanism. By employing an LSTM model with an integrated self-attention mechanism, the modeling ability of long short-term memory networks for temporal dependencies is combined with the focusing ability of attention mechanisms for key physical states, establishing a deep nonlinear mapping between multidimensional physical features and heat dissipation performance, and realizing closed-loop fitting from multidimensional raw data to high-dimensional physical perception.

[0014] Furthermore, after obtaining the multi-dimensional operating parameters of the heat exchanger, the process also includes preprocessing these parameters. This preprocessing includes: using a moving average filtering algorithm to remove noise, using the Laida criterion to remove outliers, and using linear interpolation to fill in missing values. Through these preprocessing methods, such as moving average filtering, the Laida criterion, and linear interpolation, impulse noise and isolated points generated by industrial field sensors due to electromagnetic interference and signal drops are effectively eliminated. This provides a high-quality, continuous data foundation for subsequent physical index construction and deep learning model training, reducing prediction bias caused by data quality issues.

[0015] In a second aspect, a heat exchanger heat dissipation performance prediction system based on multi-parameter fitting is also provided, including a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the heat exchanger heat dissipation performance prediction method based on multi-parameter fitting described in any of the above embodiments.

[0016] Beneficial Effects: To address the issues of thermal inertia and hysteresis effects caused by metal heat capacity in industrial heat exchangers under unsteady flow conditions, two physical descriptive indices were constructed: hysteresis offset and dynamic evolution index. The hysteresis offset adaptively corrects the time misalignment of input and output based on cross-correlation, while the dynamic evolution index quantifies the dependence of heat dissipation behavior on historical paths through the phase plane trajectory area. Most importantly, these physical characteristic-reflecting indices are transformed into weight correction coefficients for a self-attention mechanism and penalty factors for VMD decomposition. This allows the model to dynamically adjust its focus on historical data based on real-time operating conditions (such as energy accumulation state and thermal shock intensity), effectively solving the problems of decreased prediction accuracy and delayed response in traditional black-box models under drastic fluctuations in operating conditions. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating a performance prediction flowchart according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the predicted and compared heat dissipation performance under dynamic operating conditions according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the loss function curve during the training process of the improved algorithm according to an embodiment of the present invention. Detailed Implementation

[0018] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] like Figure 1 As shown, S101: Data acquisition and preprocessing.

[0020] In this embodiment, multi-dimensional operational data of the heat exchanger is first collected synchronously. Specifically, PT100 temperature sensors are installed on the inlet and outlet pipes of the hot and cold sides of the heat exchanger to acquire the inlet and outlet temperature data of the fluid in real time; simultaneously, an electromagnetic flowmeter is used to monitor the fluid mass flow rate in real time. All the above sensor data are transmitted through a data acquisition (DAQ) card. The sampling frequency is synchronously transmitted to the host computer. Combined with the inherent structural parameters of the heat exchanger, such as fin spacing and tube bundle arrangement, they together form the original time-series dataset.

[0021] To ensure data quality and eliminate the influence of different units of measurement on subsequent model training, preprocessing of the original dataset is necessary. Specifically, a moving average filtering algorithm is first used to process the original sequences acquired by the sensors to eliminate high-frequency noise disturbances. Subsequently, the Laida criterion ( The filtered data is examined according to the criteria to identify and remove outliers. Missing values ​​due to removal or signal loss are filled using linear interpolation to ensure the integrity of the data sequence. Finally, all time-series data are normalized to scale the parameter values ​​of different physical quantities to a uniform scale. Within the specified interval, it facilitates subsequent processing by the neural network model.

[0022] S102: Thermal response hysteresis characteristic analysis and construction of hysteresis offset.

[0023] In this embodiment, considering the heat capacity characteristics of the heat exchanger's metal frame, its heat dissipation performance exhibits a significant physical delay in response to changes in input conditions, i.e., thermal inertia. This characteristic leads to an asymmetric misalignment between the acquired input parameter sequence and the output response sequence on the time axis. To accurately quantify this physical time difference, this step constructs a hysteresis offset that reflects the duration of thermal hysteresis.

[0024] First, select the current time. A sliding window containing a segment of historical data. In this embodiment, the number of samples in the sliding window... Preferred Then extract the flow rate sequence within that window. With heat dissipation sequence .

[0025] Secondly, to determine the optimal time alignment between the two, the cross-correlation function is used to calculate the current time. Reference delay step Specifically, Integers within the specified range are used sequentially as delay step sizes to calculate the cross-relationship between the flow velocity sequence and the heat dissipation sequence under that delay. The delay step size that maximizes the cross-relationship value is determined as the current time step. Reference delay step .

[0026] Finally, based on the above analysis, the time interval is constructed. hysteresis offset The rationale behind this indicator is that when the heat exchanger's operating conditions fluctuate drastically and its heat dissipation response is irregular, the thermal inertia effect is more pronounced, and the time alignment deviation is also greater. Its calculation formula is as follows: In the formula, For a moment The hysteresis offset, For a moment The flow rate; For a moment After delay step size Heat dissipation after alignment; The number of samples in the sliding window; For a moment The reference delay step size; and They are time points The maximum and minimum values ​​of the velocity sequence within the window; This is a preset minimum constant, and its preferred value is... This is used to prevent the denominator from being zero; When the thermal inertia of the heat exchanger due to the wall heat capacity is more pronounced, changes in the input flow rate are smoothed and delayed by the wall heat storage behavior, resulting in a shift in the root mean square of the aligned sequence product. If the operating conditions fluctuate drastically at this point, i.e. As the value of increases, the hysteresis offset index will increase significantly. This index accurately characterizes the time alignment deviation of the heat exchanger under high thermal inertia through nonlinear mapping, reflecting the degree of causal chain distortion caused by heat conduction hysteresis, and is more physically sensitive than traditional methods.

[0027] S103: Analysis of hysteresis intensity in unsteady flow and construction of dynamic evolution index.

[0028] In this embodiment, in addition to thermal inertia, the heat exchanger also exhibits a hysteresis effect under unsteady flow conditions. According to the theory of unsteady convective heat transfer, when the fluid velocity undergoes a cycle of rising and falling, even at the same velocity point, the heat dissipation performance trajectory of the heat exchanger will not coincide due to different historical effects of boundary layer development, thus forming a closed hysteresis loop. To quantify this path-dependent characteristic, this step constructs a dynamic evolution index reflecting the degree of opening of the hysteresis loop.

[0029] First, the hysteresis offset calculated in step S102 is used to perform time compensation on the original flow velocity sequence and heat dissipation sequence to correct the time axis misalignment caused by thermal inertia, thereby reconstructing a working condition evolution trajectory that better reflects the real physical process in the phase plane.

[0030] Subsequently, the construction time The dynamic evolution index. This index is constructed based on the fact that the area enclosed by the trajectory lines formed by a series of discrete points in the phase plane can be approximated by the sum of vector cross products. The size of this area intuitively reflects the degree of opening of the hysteresis loop. Its calculation formula is as follows: In the formula, For a moment The dynamic evolution index, For a moment The hysteresis offset; and After time compensation, at time 10:00 Flow rate and heat dissipation; and After time compensation, at time 10:00 Flow rate and heat dissipation; The number of samples in the sliding window; For the value to be The constant.

[0031] As can be seen from the formula, when the fluid inside the heat exchanger is in an unsteady flow state, and the difference in heat transfer efficiency between the acceleration and deceleration sections causes the phase plane trajectories to enclose a large area, the absolute value of the vector cross product in the molecule... The sum will increase accordingly. This causes the value of the dynamic evolution exponent to increase as well, indicating that the current heat dissipation behavior of the system is more deeply influenced by historical paths, and the hysteresis loop is more pronounced. By introducing a logarithmic accumulation method, the influence of instantaneous sampling noise is smoothed out, while amplifying the asymmetric characteristics in the evolution process.

[0032] S104: Spatiotemporal energy accumulation analysis and construction of weight correction coefficients.

[0033] In this embodiment, the heat dissipation process of the heat exchanger is essentially the continuous accumulation and release of energy in the spatiotemporal domain. The heat dissipation capacity at the current moment is affected not only by the current temperature difference but also by the heat accumulation on the wall surface in previous time periods. To enable the model to perceive the decay of historical contributions, this step constructs a weight correction coefficient to guide the self-attention mechanism.

[0034] First, at the current moment Within a sliding window, the rate of change of the flow velocity is obtained by taking the first derivative of the velocity sequence. Simultaneously, the average temperature difference between the inlet and outlet fluids of the heat exchanger is extracted.

[0035] Secondly, the construction time Weighting adjustment coefficient This index aims to abstract the complex spatiotemporal energy field into a weighted guiding signal that can be used for neural network regulation. Its construction is based on the fact that when the system is in a high-load accumulation phase (large temperature difference and strong hysteresis), the heat dissipation output is more dependent on the past energy accumulation state. Its calculation formula is as follows: In the formula, For a moment Weighting correction coefficient, For a moment The dynamic evolution index, For a moment The hysteresis offset; For a moment The rate of change of flow velocity; For a moment The average temperature difference between the inlet and outlet fluids of the heat exchanger; As an energy saturation adjustment coefficient, and as a preferred option, its reference value is [value missing]. ; To determine the instantaneous sensitivity of heat dissipation to flow velocity, calculations are performed at specific times. The ratio of heat dissipation to flow rate change is obtained; It is a hyperbolic cosine function.

[0036] As the formula shows, when the system is in a state of high lag (large lag offset) and strong hysteresis (large dynamic evolution exponent), the numerator term will collectively increase the value of the weight correction coefficient. Meanwhile, the denominator... The term simulates the saturation effect of heat conduction. When the product of flow rate change and temperature difference reaches a certain level, its growth tends to level off, limiting the unlimited growth of the weight correction coefficient. Therefore, if the weight correction coefficient is large, it means that the heat dissipation output at the current moment is highly dependent on the past energy accumulation state, and the model should pay more attention to historical information.

[0037] S105: Algorithm Improvement and Heat Dissipation Performance Prediction.

[0038] In one embodiment, the physical indices constructed above are embedded into the VMD-LSTM prediction model to enhance the model's physical perception capabilities.

[0039] Specifically, in the VMD decomposition phase of the algorithm, the hysteresis offset is used as an adaptive penalty factor. For example, when the hysteresis offset exceeds a preset threshold (e.g., ...), ... When the system determines that it is currently in a state dominated by strong thermal inertia, the retention weight of the low-frequency modal component (IMF) should be increased, with the threshold value set to 0.6. This allows the VMD decomposition process to adaptively lock in and enhance the signal characteristics dominated by thermal inertia.

[0040] Subsequently, in the self-attention layer of the LSTM network, weight correction coefficients are embedded into the calculation formula for the attention score. Specifically, the input to this neural network is the multi-scale modality matrix obtained after VMD decomposition. When calculating the weight scoring function in the self-attention layer, the weight correction coefficients are treated as a gain term or modulation factor in the scoring result. In this way, when the model detects high-intensity thermal hysteresis and energy accumulation characteristics (i.e., ... When the value is large, it can automatically increase the attention ratio to the relevant historical memory units, thereby achieving accurate and dynamic focusing and extraction of lagging physical features.

[0041] During the prediction process, the multidimensional time series (including inlet and outlet temperature sequences, fluid mass flow rate sequences, etc.) preprocessed in step S101 is fed into the prediction system as a multidimensional input signal. The system first uses an adaptive VMD algorithm to decompose the input signal in each dimension, obtaining intrinsic mode components (IMFs) with different center frequencies and containing heat transfer characteristics at different scales. This transforms the original fluctuating, non-stationary multidimensional signal into a multi-scale mode matrix that reflects thermal inertia. Subsequently, this mode matrix is ​​embedded at the embedding time. Weighting adjustment coefficient In the self-attention neural network, the self-attention layer serves as the core of feature fitting. Its input is the aforementioned multi-scale modality matrix. After iterative convergence, the weight correction coefficients are applied to the weight scoring function of the self-attention mechanism to correct the attention allocation, thereby achieving targeted enhancement of physically lagging features in the hidden state space (such as...). Figure 3 The diagram shows the loss function curve during the iterative convergence process. The output of the neural network is the predicted heat dissipation performance value after feature fusion, such as at time step [time value missing]. The system calculates the heat dissipation power and total thermal resistance. In this way, the prediction system achieves a closed-loop process from multidimensional raw data to physical feature quantification and then to dynamic weighted fitting, enabling high-precision online prediction of the transient heat dissipation performance of the heat exchanger while ensuring the model possesses physical awareness (e.g., heat dissipation power and total thermal resistance). Figure 2 (A comparison chart showing predicted heat dissipation performance under dynamic operating conditions).

[0042] The present invention also provides a heat exchanger heat dissipation performance prediction system based on multi-parameter fitting. The system includes a processor and a memory, the memory storing computer program instructions. When the processor executes the computer program instructions, it implements the heat exchanger heat dissipation performance prediction method based on multi-parameter fitting according to the first aspect of the present invention.

[0043] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0044] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0045] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for predicting the heat dissipation performance of a heat exchanger based on multi-parameter fitting, characterized in that, include: The multidimensional operating parameters of the heat exchanger are obtained, including the fluid velocity sequence and the heat dissipation sequence. Based on the cross-correlation between the fluid velocity sequence and the heat dissipation sequence, a hysteresis offset characterizing the thermal inertial response delay of the heat exchanger is constructed. The fluid velocity sequence and heat dissipation sequence are time-aligned based on the hysteresis offset, and a dynamic evolution index characterizing the unsteady flow hysteresis effect is constructed based on the area enclosed by the trajectories formed by the time-aligned fluid velocity sequence and heat dissipation sequence in the phase plane. By combining the dynamic evolution index, the hysteresis offset, the rate of change of the fluid velocity sequence, and the average temperature difference between the inlet and outlet fluids of the heat exchanger, a weight correction coefficient is constructed to guide the model attention allocation. The multidimensional operating parameters are input into a preset prediction model for processing to obtain the predicted value of the heat exchanger's heat dissipation performance. The processing of the prediction model includes: embedding the weight correction coefficient into the weight scoring function of the self-attention mechanism to adjust the model's attention to historical data.

2. The method for predicting heat exchanger heat dissipation performance based on multi-parameter fitting according to claim 1, characterized in that, The methods for constructing the hysteresis offset at each time step include: The sum of the products of the flow velocity at each moment within the sliding window and the heat dissipation after compensation based on the baseline delay step size is calculated. The hysteresis offset is proportional to the sum of the products and is also positively correlated with the difference between the maximum and minimum values ​​of the flow velocity sequence within the sliding window.

3. The method for predicting heat exchanger heat dissipation performance based on multi-parameter fitting according to claim 2, characterized in that, The method for calculating the reference delay step size at each time point includes: Within the sliding window at the current moment, the time delay that maximizes the cross-correlation function value between the fluid velocity sequence and the heat dissipation sequence is determined as the baseline delay step.

4. The method for predicting heat exchanger heat dissipation performance based on multi-parameter fitting according to claim 1, characterized in that, Methods for constructing dynamic evolution indices include: Within the sliding window at the current moment, the logarithmic sum of the absolute values ​​of the vector cross products between adjacent discrete points formed by the time-compensated fluid velocity sequence and the heat dissipation sequence is calculated; the dynamic evolution index is proportional to the product of the logarithmic sum and the hysteresis offset.

5. The method for predicting heat exchanger heat dissipation performance based on multi-parameter fitting according to claim 1, characterized in that, The method for constructing the weight correction coefficients includes: The basic weight is obtained by dividing the product of the dynamic evolution index and the tangent of the hysteresis offset by the hyperbolic cosine function value of the product of the rate of change of the fluid velocity sequence and the average temperature difference between the inlet and outlet fluids of the heat exchanger; the basic weight is then added to the preset instantaneous response sensitivity parameter to obtain the weight correction coefficient.

6. The method for predicting heat exchanger heat dissipation performance based on multi-parameter fitting according to claim 1, characterized in that, The processing steps of the prediction model also include: Variational mode decomposition is performed on the multidimensional operating parameters to obtain multiple intrinsic mode components containing heat transfer characteristics at different scales; wherein, the variational mode decomposition process uses hysteresis offset as an adaptive penalty factor to enhance the extraction capability of low-frequency mode components dominated by thermal inertia.

7. The method for predicting heat exchanger heat dissipation performance based on multi-parameter fitting according to claim 6, characterized in that, Using the hysteresis offset as an adaptive penalty factor includes: When the value of the hysteresis offset is greater than a preset threshold, the retention weight of low-frequency intrinsic mode components is increased during variational mode decomposition.

8. The method for predicting heat exchanger heat dissipation performance based on multi-parameter fitting according to claim 1, characterized in that, The preset prediction model is a long short-term memory neural network model that includes a self-attention mechanism.

9. The method for predicting heat exchanger heat dissipation performance based on multi-parameter fitting according to claim 1, characterized in that, After obtaining the multidimensional operating parameters of the heat exchanger, the following is also included: The multidimensional operating parameters are preprocessed, including: removing noise using a moving average filtering algorithm, removing outliers using the Laida criterion, and filling in missing values ​​using linear interpolation.

10. A heat exchanger heat dissipation performance prediction system based on multi-parameter fitting, comprising a processor and a memory, characterized in that, The memory stores a computer program, and the processor executes the computer program to implement the heat exchanger heat dissipation performance prediction method based on multi-parameter fitting as described in any one of claims 1-9.

Images