A source-load dynamic matching flue gas waste heat pump energy level upgrading system

Abstract

This invention relates to the field of waste heat heat pump system control and optimization technology, specifically disclosing a flue gas waste heat heat pump energy level enhancement system with dynamic source-load matching. This invention achieves latent heat energy recovery by constructing a dynamic thermo-mass interaction model to analyze the phase change characteristics of flue gas. By utilizing a deep time-series prediction network, this invention predicts source-load changes, dynamically adjusts operating strategies, regulates system operation, and improves energy efficiency. Simultaneously, by monitoring the concentration of corrosive components, it adjusts the working fluid ratio in a timely manner, extending equipment life. The system also features dynamic flow compensation and compression stage reconstruction functions, effectively coping with drastic load changes and improving system flexibility and response speed. Furthermore, through a closed-loop correction mechanism, it continuously optimizes the model prediction accuracy, improving overall operating efficiency and heat recovery efficiency, providing an optimized solution for the field of industrial waste heat recovery.

Description

Technical Field

[0001] This invention relates to the field of waste heat heat pump system control and optimization technology, and more specifically, to a flue gas waste heat heat pump energy level enhancement system with dynamic source-load matching. Background Technology

[0002] In complex industrial production environments, waste heat recovery is one of the key technologies for improving energy efficiency. However, the parameters of industrial waste heat sources often exhibit significant dynamism and fluctuations, and the demand on the heat load side is not constant. Current waste heat pump systems face challenges in designing and operating them to match dynamically changing source characteristics in real time, especially the latent heat potential of phase change on the flue gas side with the load demand, making it difficult for the system to achieve the expected energy efficiency level under various operating conditions.

[0003] Patent CN118168194B discloses a heat pump system for deep recovery of waste heat from boiler flue gas, comprising: acquiring a flue gas waste heat recovery rate prediction model; periodically updating the flue gas waste heat recovery rate prediction model based on historical data; optimizing dynamic characteristic parameters based on the flue gas waste heat recovery rate prediction model to obtain an optimal set of characteristic parameters; evaluating the response performance of the dynamic characteristic parameters, analyzing the response speed, response accuracy, and impact on the flue gas waste heat recovery rate of the dynamic characteristic parameters to obtain a dynamic characteristic parameter response performance evaluation index; and taking corresponding measures based on the dynamic characteristic parameter response performance evaluation index. The system utilizes a deep learning model to analyze and obtain the flue gas waste heat recovery rate prediction model, thereby predicting the flue gas waste heat recovery rate; and optimizes the dynamic characteristic parameters based on the flue gas waste heat recovery rate prediction model.

[0004] Patent CN107860153A discloses an energy-saving and water-saving coal-fired boiler wet flue gas deep integrated treatment system and method, including: using the recovered boiler flue gas waste heat to drive an absorption heat pump, realizing the recovery of waste heat from the saturated wet flue gas at the outlet of the desulfurization absorption tower, while condensing and recovering water vapor in the saturated wet flue gas at the outlet of the desulfurization absorption tower, realizing zero water replenishment operation of the desulfurization system, and simultaneously achieving the removal of various pollutants and the treatment of "gypsum rain" and "smoke rain" near the chimney outlet.

[0005] Existing technologies suffer from fundamental deficiencies in the predictive, adaptive, and precise aspects of system control. First, their control logic is either limited to fixed process flows or merely focuses on static parameter optimization based on predicted values, lacking the ability to deeply predict the future dynamic trajectories of both the source and load sides, thus failing to achieve proactive response and opportunistic intervention for critical changing events. Second, their system architecture is fixed, lacking the flexibility to dynamically reconfigure the hardware topology based on real-time energy level enhancement requirements, limiting energy efficiency optimization across a wide operating range. Third, these solutions fail to establish dynamic physical models that deeply couple thermodynamic and mass transfer processes, thus failing to accurately quantify and track the instantaneous potential of the latent heat of phase change in flue gas, and consequently lacking the ability to closely follow fluctuating physical boundaries (such as dew point). Finally, these solutions neglect comprehensive management of the system's internal stability and the effects of multi-physics coupling under drastic changes in operating conditions, lacking dynamic compensation and closed-loop correction mechanisms for maximizing energy efficiency while ensuring operational safety and long-term equipment reliability. Summary of the Invention

[0006] In view of this, in order to solve the problems mentioned in the background technology, a flue gas waste heat heat pump energy level enhancement system with dynamic source-load matching is proposed.

[0007] The objective of this invention can be achieved through the following technical solution: This invention provides a flue gas waste heat heat pump energy level enhancement system with dynamic source-load matching, including: a thermo-mass interaction dynamic model construction module, which acquires real-time physical sensing data and integrates it with the equipment's pre-condition signal to form a source-load environment hybrid dataset, inputs a preset fluid dynamic mechanism architecture to perform parameter optimization matching, and generates a thermo-mass interaction dynamic model.

[0008] The latent heat energy flow instantaneous potential quantification module runs a thermo-mass interaction dynamic model to calculate the instantaneous condensation rate, and combines the flow scalar in the source-load environment mixed dataset to quantify and extract the instantaneous potential of latent heat energy flow.

[0009] The critical event trigger signal generation module inputs the source-load environment hybrid dataset into a deep time series prediction network to fit a multi-dimensional feature trajectory line, and calculates the probability distribution variance by combining the instantaneous potential of latent heat energy flow to generate critical event trigger signals.

[0010] The target loop architecture anchoring module analyzes key event trigger signals, calculates the energy level increase based on multi-dimensional feature trajectory lines, and anchors the target loop architecture.

[0011] The latent heat recovery loop pre-activation module activates the latent heat recovery loop in advance based on the target loop architecture and key event trigger signals, calculates the expected dew point drop slope curve based on multi-dimensional feature trajectory lines, and issues adaptive evaporation pressure parameters.

[0012] The compression stage reconstruction and loop activation module responds to the target loop architecture, reads the boost gap value in the energy level increase, reconstructs the compression stages, and activates the intermediate cooling loop.

[0013] The dynamic flow compensation strategy generation module analyzes the working fluid ratio parameters issued by the sulfur oxide concentration variation rate associated with the dynamic model of thermo-mass interaction, and extracts the slope first derivative of the multi-dimensional feature trajectory line to generate a dynamic flow compensation strategy.

[0014] The thermal-mass interaction dynamic model correction module extracts the actual operating effect features, calculates the temporal residual between the actual operating effect features and the predicted scalar of the multi-dimensional feature trajectory line, and corrects the thermal-mass interaction dynamic model in reverse closed loop.

[0015] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention dynamically analyzes the phase change boundary of flue gas, predicts the condensation range, and drives the latent heat recovery circuit according to the predicted dew point drop slope to accurately track the flue gas cooling trajectory with adaptive evaporation pressure, so that the refrigerant heat absorption cold end temperature continuously follows the flue gas dew point temperature. This mechanism can increase the amount of latent heat captured by flue gas condensation and avoid insufficient or inefficient latent heat recovery caused by inaccurate prediction or static control.

[0016] (2) This invention performs multidimensional prediction of source load dynamics through a deep time series network, and combines the prediction confidence interval with the physical security anti-fluctuation judgment threshold to dynamically decide whether to perform aggressive operation or maintain stable redundancy. This mechanism helps the system maintain operational stability under complex dynamic supply and demand changes, avoids misoperation or system oscillation caused by excessive uncertainty, and ensures that it can switch to the optimal operating mode when conditions permit.

[0017] (3) This invention combines the fingerprint dictionary of the thermo-mass interaction model to analyze the variation rate of corrosive component concentration and uses multi-dimensional feature trajectory lines to capture the sudden jump rate under varying operating conditions, thereby adjusting the working fluid ratio and the flow compensation parameters of the electronic expansion valve in real time. This mechanism can suppress the deterioration process inside the system, reduce the possibility of corrosion, and avoid potential failure risks such as dry burning of the evaporator or liquid slugging of the condenser caused by severe load fluctuations.

[0018] (4) Based on the temporal residual between the actual operating effect characteristics and the predicted scalar, this invention uses the gradient backpropagation algorithm to fine-tune the node logical weights of the thermo-mass interaction dynamic model. This mechanism can continuously optimize the prediction accuracy of the model, improve the ability to represent the real physical process, and thus make the control strategy of the system more accurate under subsequent operating conditions, thereby improving the overall operating efficiency. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the system module structure connection of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 This invention provides a flue gas waste heat heat pump energy level enhancement system with dynamic source-load matching, including: a dynamic model construction module for thermo-mass interaction, a module for quantifying the instantaneous potential of latent heat energy flow, a module for generating key event trigger signals, a module for anchoring the target loop architecture, a module for pre-activating the latent heat recovery loop, a module for reconstructing compression stages and enabling the loop, a module for generating dynamic flow compensation strategies, and a module for correcting the dynamic model for thermo-mass interaction.

[0023] The thermo-mass interaction dynamic model construction module is connected to the latent heat energy flow instantaneous potential quantification module. The latent heat energy flow instantaneous potential quantification module is connected to the critical event trigger signal generation module. The critical event trigger signal generation module is connected to the target loop architecture anchoring module. Both the critical event trigger signal generation module and the target loop architecture anchoring module are connected to the latent heat recovery loop pre-activation module. The target loop architecture anchoring module is connected to the compression stage reconstruction and loop activation module. Both the thermo-mass interaction dynamic model construction module and the critical event trigger signal generation module are connected to the dynamic flow compensation strategy generation module. Both the critical event trigger signal generation module and the dynamic flow compensation strategy generation module are connected to the thermo-mass interaction dynamic model correction module.

[0024] The thermo-mass interaction dynamic model construction module acquires real-time physical sensing data and integrates it with the equipment's pre-condition signals to form a source-load-environment hybrid dataset. It then inputs a pre-set fluid dynamic mechanism architecture to perform parameter optimization and matching, thereby generating a thermo-mass interaction dynamic model.

[0025] In a specific embodiment of the present invention, the specific process of obtaining real-time physical sensing data and equipment front-end operating condition signals to form a source-load environment hybrid dataset includes: collecting the continuous instantaneous temperature and flow parameters of flue gas at the source end of the waste heat heat pump system, and simultaneously acquiring the target heat supply and target heat supply temperature demand values ​​of the corresponding time slice on the heat load side as real-time physical sensing data.

[0026] Extract the upstream boiler load change signal from the associated production pipeline network as the equipment precondition signal.

[0027] Real-time physical sensor data and equipment front-end operating condition signals are aligned with the system's relative timestamp and fused into a source-load-environment hybrid dataset.

[0028] In a specific embodiment of the present invention, the process of inputting a preset fluid dynamics mechanism architecture for parameter optimization and matching to generate a dynamic model of heat and mass interaction includes: inputting a source-load environment hybrid dataset into a fluid dynamics mechanism architecture that includes multiphase flow theory and heat and mass transfer processes.

[0029] The parameter optimization module based on the Bayesian optimization algorithm is invoked to iteratively match the heat transfer coefficient of the flue gas sidewall and the water vapor diffusion coefficient in the fluid dynamics mechanism framework.

[0030] The iteration stops when the root mean square error between the model prediction and the actual data is lower than the preset convergence threshold, and a dynamic model of thermo-mass interaction is generated to analyze the phase change characteristics of flue gas components.

[0031] Specifically, the real-time acquisition module of the data acquisition and monitoring system first obtains continuous instantaneous flue gas temperature and flow parameters from the source end of the waste heat heat pump system. Specifically, the instantaneous flue gas temperature parameter is sampled at high frequency using a K-type thermocouple array, once per second, and the acquired analog voltage signal is converted into a digital quantity by a 24-bit analog-to-digital converter with an accuracy of 0.1℃. The flue gas flow parameter is acquired using an insertion-type vortex flow meter in conjunction with a differential pressure sensor. The frequency signal output by the vortex flow meter is converted into a digital flow value by a signal conditioning unit with a measurement accuracy of 0.5%, and the acquisition frequency is consistent with the temperature parameter. Simultaneously, the distributed control system on the heat load side synchronously acquires the target heat supply and target heating temperature demand values ​​for the corresponding time slice. The target heat supply is measured by supply and return water temperature sensors (PT100, accuracy 0.1℃) and electromagnetic flow meters (accuracy 0.8%) on the heat load side, and calculated in real time based on the specific heat capacity formula of water. The target heating temperature demand value is directly read from the target value set by the user or the upper-level controller in the DCS system.

[0032] While acquiring the aforementioned real-time sensor data, the system extracts upstream boiler DCS system features, namely boiler load variation signals, from the associated production pipeline network via the OPC unified architecture interface. These boiler load variation signals include, but are not limited to, key operating parameters such as actual boiler steam output, fuel consumption rate, and primary air volume. All collected sensor data and upstream operating signal signals are transmitted to the central data processor. This processor uses a network time protocol server to align all data with system-relative timestamps, ensuring data consistency and controlling time deviation to the millisecond level. Subsequently, the processor-aligned data undergoes data cleaning, formatting, and normalization, and is merged into a source-load-environment hybrid dataset, which is stored in a structured time-series format.

[0033] The source-load environment hybrid dataset is then input into the system's pre-defined fluid dynamics mechanism architecture. This architecture is a numerical simulation framework based on multiphase flow theory and heat and mass transfer processes, including flue gas component transport equations, energy conservation equations, momentum conservation equations, and an integrated condensation phase change kinetic model. The system invokes a parameter optimization module based on a Bayesian optimization algorithm to iteratively match key parameters within the fluid dynamics mechanism architecture. These key parameters include, but are not limited to, the flue gas sidewall heat transfer coefficient, water vapor diffusion coefficient, condensation film thermal resistance, and latent heat of phase change. The goal of the optimization module is to minimize the error between the predicted flue gas internal temperature distribution and component concentration distribution and the actual collected macroscopic data (such as exhaust temperature and dew point temperature) when the flue gas instantaneous temperature and flow parameters from the source-load environment hybrid dataset are input. The optimization process continues until the root mean square error (RMSE) between the model prediction and the actual data is below a preset convergence threshold. After parameter matching is completed, the system generates a dynamic thermo-mass interaction model for analyzing the phase change characteristics of flue gas components. The thermo-mass interaction dynamic model is a high-fidelity numerical simulation model that can accurately predict the diffusion, condensation behavior, and latent heat release characteristics of condensable components such as water vapor in flue gas flow fields.

[0034] Among them, the K-type thermocouple is a temperature sensor based on the Seebeck effect, with a typical temperature range of 0℃ to 1200℃ and an accuracy class of 0.75%. The Pt100 platinum resistance temperature sensor is a temperature sensor that utilizes the resistance change characteristics of platinum resistance with temperature, with a typical temperature range of -200℃ to 850℃ and an accuracy class of 0.1℃. This solution uses a K-type thermocouple array conforming to IEC 60584 standard for stable acquisition of instantaneous flue gas temperature; its long-term stability and vibration resistance meet the requirements of complex industrial environments. The insertion vortex flow meter is an instrument that measures fluid flow based on the Karman vortex street principle. It calculates the flow rate by detecting the frequency of vortex shedding in the fluid, and the measurement accuracy can reach 0.5% to 1.5%FS depending on the model. The electromagnetic flow meter is an instrument that measures the flow rate of conductive fluids based on Faraday's law of electromagnetic induction. Its characteristics include no obstruction, no pressure loss, and high accuracy over a wide range. This solution uses an electromagnetic flow meter with a measurement accuracy of 0.8% for flow monitoring on the heat load side.

[0035] Network Time Protocol (NTP) is a protocol used to synchronize the clocks of various devices in a computer network. By providing a precise time source through a NTP server, it ensures that the system clock synchronization accuracy of all devices is at the millisecond level, effectively avoiding data misalignment caused by inconsistent time sources in distributed systems. OPC Unified Architecture is a cross-platform, service-oriented architecture for data exchange in industrial automation. It provides a secure, reliable, and open data transmission mechanism, enabling seamless information exchange between devices and systems from different manufacturers.

[0036] The fluid dynamics mechanism framework refers to a computational model that is discretized by the finite volume method or the finite element method and solves the mass conservation equation, momentum conservation equation, energy conservation equation and component transport equation.

[0037] Bayesian optimization is a global optimization algorithm, particularly suitable for expensive problems where the objective function is unknown or difficult to differentiate. It finds the global optimum by constructing a surrogate model of the objective function and using a harvesting function to balance exploration and harvesting. In this scheme, it is used to minimize the root mean square error (RMSE) between the predicted output of the thermo-mass interaction dynamic model and the actual measured data. RMSE is a metric that measures the deviation between observed and predicted values; its calculation formula is... ,in These are model predictions. These are actual observed values. This refers to the number of data points. The preset convergence threshold is set to 0.05, meaning that when the RMSE is less than 0.05, the parameter optimization and matching are considered complete. This threshold is set based on the error characteristics of historical operating data of the industrial waste heat recovery system and the system's requirements for prediction accuracy, through multiple offline experiments and expert experience.

[0038] The latent heat energy flow instantaneous potential quantification module runs a thermo-mass interaction dynamic model to calculate the instantaneous condensation rate, and combines the flow scalar in the source-load environment mixed dataset to quantify and extract the instantaneous potential of latent heat energy flow.

[0039] In a specific embodiment of the present invention, the process of calculating the instantaneous condensation rate by running the thermo-mass interaction dynamic model and combining the flow scalar in the source-load environment mixed dataset to quantify and extract the instantaneous potential of latent heat energy flow includes: calculating the concentration of condensable components of water vapor in the flue gas field under different temperature gradient distributions by calculating the state equation inside the thermo-mass interaction dynamic model.

[0040] Extract saturated vapor pressure correlation data of condensable component concentration as a function of pressure, and calculate the initial dew point temperature and instantaneous condensation rate of the corresponding gas-liquid phase change process.

[0041] By integrating the flow scalar and instantaneous condensation rate in the source-load environment hybrid dataset, the instantaneous potential of the latent heat energy flow that can be physically recovered in the current flue gas field is identified through enthalpy difference calculation.

[0042] Specifically, the processor first invokes the state equation solver within the thermo-mass interaction dynamic model. The state equations include equations for the conservation of mass, energy, and momentum, as well as component transport equations describing the diffusion and phase change of water vapor in the flue gas. Based on the parameters optimized and matched in the thermo-mass interaction dynamic model building module, the thermo-mass interaction dynamic model calculates the local condensable component concentration of water vapor under different temperature gradient distributions within the flue gas flow field for the instantaneous temperature and flow parameters of the flue gas in the current time slice of the source-load environment mixing dataset through iterative calculations and numerical simulations (e.g., using the finite volume method or finite element method). The condensable component concentration is output as a mass fraction; for example, the mass fraction distribution of water vapor at different cross-sections and axial positions of the flue gas duct is precisely quantified and stored as the calculation results.

[0043] Next, the processor extracts saturated vapor pressure correlation data showing the change in condensable component concentration with pressure from a pre-stored physicochemical property database. The processor inputs the local pressure and condensable component concentration (converted to water vapor partial pressure) in the flue gas flow field, and uses the saturated vapor pressure correlation data to accurately calculate the initial dew point temperature at which flue gas condensation begins. Simultaneously, the processor runs the phase transition kinetics sub-model within the thermo-mass interaction dynamic model, combining parameters such as the supersaturation, temperature gradient, and condensation nucleus density of the current flue gas region to obtain the instantaneous condensation rate of water vapor in the flue gas flow field. The initial dew point temperature is a single scalar value, while the instantaneous condensation rate represents the total mass of water vapor condensed in the flue gas flow field per unit time, in kg / s.

[0044] Finally, the processor integrates the flue gas flow rate scalar (in m³ / s) and instantaneous condensation rate (in kg / s) from the source-load environment hybrid dataset. By directly substituting the instantaneous condensation rate and applying the enthalpy difference calculation principle, the processor identifies and quantifies the instantaneous potential of the latent heat energy flow that can be physically recovered in the current flue gas flow field. Specifically, the enthalpy difference calculation multiplies the instantaneous condensation rate by the latent heat of vaporization of water vapor at the initial dew point temperature, thus obtaining the latent heat energy flow in kW.

[0045] (1);

[0046] in: This indicates that water vapor at temperature The saturated vapor pressure at that point is expressed in kPa. This indicates the local temperature at which water vapor begins to condense, and is expressed in °C. These are constants, and they are based on... The properties and selected temperature range can be obtained from standard chemistry handbooks or physicochemical databases.

[0047] (2);

[0048] in: It represents the instantaneous potential of latent heat energy flow, and the unit is kW (kilowatt) or kJ / s (kilojoules per second). This represents the instantaneous condensation rate, measured in kg / s (kilograms per second). This value is calculated using the phase transition kinetics model within the thermo-mass interaction dynamic model. This represents the latent heat of vaporization of water vapor at the initial dew point temperature, expressed in kJ / kg (kilojoules per kilogram). This value is obtained by consulting standard thermodynamic property tables (e.g., steam tables based on the IAPWS-IF97 formula).

[0049] The condensable component concentration refers to the mass fraction of water vapor in the flue gas flow field. This concentration is calculated by the thermo-mass interaction dynamic model based on the flue gas inlet conditions and its heat and mass transfer processes within the flow field, and is expressed as a dimensionless ratio between 0 and 1. Saturated vapor pressure correlation data are sourced from internationally recognized thermodynamic property databases, such as standard data tables and fitting formulas provided by the National Institute of Standards and Technology (NIST) or the International Association for the Properties of Water and Steam, used to accurately describe the equilibrium vapor pressure of water vapor at different temperatures. The initial dew point temperature refers to the temperature at which water vapor in the flue gas begins to condense into liquid water droplets under the current total flue gas pressure. This temperature is calculated based on the correspondence between water vapor partial pressure and saturated vapor pressure and is a key physical quantity for determining the feasibility of latent heat recovery. The instantaneous condensation rate refers to the mass of water vapor that transforms from a gaseous state to a liquid state per unit time, with units of kg / s. This rate is calculated by the condensation kinetics sub-model within the thermo-mass interaction dynamic model based on parameters such as local supersaturation of the flue gas, mass transfer coefficient, and effective condensation surface area. The flue gas flow rate scalar is the flue gas volumetric flow rate defined in the thermo-mass interaction dynamic model construction module, with units of m³ / s, used to characterize the macroscopic scale of flue gas flow. The latent heat of vaporization of water refers to the energy required for a unit mass of water to change from a liquid to a gaseous state or vice versa at constant temperature and pressure. Its value varies with temperature and pressure; in this scheme, it is dynamically obtained from industrial standard tables of thermodynamic properties of water and steam, such as IAPWS-IF97, based on the initial dew point temperature.

[0050] The critical event trigger signal generation module inputs the source-load environment hybrid dataset into a deep time series prediction network to fit a multi-dimensional feature trajectory line, and calculates the probability distribution variance by combining the instantaneous potential of latent heat energy flow to generate critical event trigger signals.

[0051] In a specific embodiment of the present invention, the process of generating a key event trigger signal by combining the variance of the probability distribution of the instantaneous potential of latent heat energy flow includes: inputting the source-load environment hybrid dataset into a deep time series prediction network to perform multi-step forward propagation calculations, and fitting a multi-dimensional feature trajectory line of continuous change of source-end energy supply and load-end energy consumption within a future preset time window.

[0052] Extract time node markers from multi-dimensional feature trajectory lines to characterize the bottoming out of flue gas and its descent into the condensation temperature zone or the step jump in heat load demand.

[0053] By combining the probability distribution variance of the segment where the instantaneous potential of latent heat energy flow is calculated, a key event trigger signal carrying a prediction confidence interval is generated.

[0054] Specifically, the central data processor first inputs a hybrid source-load environment dataset, which includes instantaneous flue gas temperature, flue gas flow rate, target heating supply, target heating temperature demand, and boiler load variation signals, into a pre-trained deep time-series prediction network. The deep time-series prediction network employs a Long Short-Term Memory (LSTM) architecture, and its model weights have been optimized and trained using historical operating data. The processor performs multi-step forward propagation calculations, that is, within a preset future time window (e.g., 30 minutes in the future, with a time step of 1 minute), it performs sequential predictions of various parameters in the hybrid source-load environment dataset. This calculation process recursively generates the predicted value for each time step based on the internal state of the LSTM and the current input, fitting a multi-dimensional feature trajectory line showing the continuous changes in source-side energy supply and load-side energy consumption within the preset future time window. The multidimensional feature trajectory lines are stored in the form of a two-dimensional array, where each row represents a future time step and each column represents a predicted feature variable, such as the predicted flue gas instantaneous temperature sequence, the predicted flue gas flow rate sequence, the predicted target heating heat sequence, the predicted target heating temperature sequence, and the predicted comprehensive energy supply sequence combining the instantaneous potential of latent heat energy flow.

[0055] Next, the processor performs real-time analysis on the generated multi-dimensional feature trajectory lines to extract time node identifiers representing changes in key operational logic. Specifically, for trajectory lines representing instantaneous temperature changes in flue gas, the system identifies whether they have bottomed out and fallen into a preset condensation temperature range. This identification is achieved through a sliding window averaging method and stationarity detection; if continuous... The predicted instantaneous flue gas temperature at each time step is below the condensation threshold. If this occurs, it is determined to be a critical event of "bottoming out and falling into the condensation temperature range". Simultaneously, for the trajectory representing changes in heat load demand, the processor calculates the difference between adjacent time steps. If the rate of change of the target heat supply or target heat supply temperature demand exceeds a preset step response threshold... (For example, in) Its change within each time step exceeds If any of these events occur, they are considered critical events indicating a "step increase in heat load demand." Once these events are detected, the corresponding future time points are marked as time node identifiers.

[0056] Finally, the processor combines the instantaneous potential of latent heat energy flow to calculate the probability distribution variance of the segment where the time node identifier is located, in order to quantify the uncertainty of the prediction and generate a key event trigger signal carrying the prediction confidence interval. Specifically, for each identified time node identifier, the processor examines a fixed-size time window around it (e.g., before and after). Within a given time step, the probability distribution of the predicted instantaneous potential value sequence of latent heat energy flow is calculated. This is achieved by calculating the variance of the deviation between the predicted values ​​and actual historical statistical data within this time window. This measures the uncertainty of the forecast. Based on this, the forecast value is used as the mean, and... Using the square root of the variance as the standard deviation, construct a Gaussian distribution model and calculate a... Predicted confidence interval Finally, the system combines and encapsulates the time node identifier, its corresponding main predicted event type, such as "entry into condensation temperature zone" and "load step," and the predicted confidence interval to generate a key event trigger signal, which is then published to the downstream control module via the internal communication bus.

[0057] ;

[0058] in: This represents the variance of the probability distribution of the predicted instantaneous potential value sequence of latent heat energy flow within the segment marked by the time node. This represents the number of predicted data points within the time window used to calculate the variance. This number is equal to... . Indicates the first time within the time window Predicted instantaneous potential value of latent heat energy flow at each time step. This represents the average value of the instantaneous potential prediction of latent heat energy flow within a time window.

[0059] Deep time series prediction networks are a special type of neural network capable of capturing long-term dependencies and nonlinear patterns in time series data. Long Short-Term Memory (LSTM) networks, a variant of Recurrent Neural Networks (RNNs), effectively solve the gradient vanishing / exploding problem of traditional RNNs by introducing gating mechanisms (input gate, forget gate, output gate), making them superior in processing and predicting long-sequence data. This scheme uses a three-layer LSTM structure, with each layer containing 256 hidden units, and uses activation functions, trained using the Adam optimizer. Multi-step forward propagation computation refers to the neural network model generating predictions for multiple future time steps given a series of historical inputs, rather than simply predicting the next time step. The preset time window is set to 30 minutes in this scheme, with a time step size of 1 minute. This setting was determined after comprehensively considering the response time of industrial production processes, the computational load of the prediction model, and actual operational requirements, ensuring the real-time performance and accuracy of the prediction.

[0060] The condensation temperature range refers to the temperature range within which water vapor in flue gas begins to condense, typically defined as the area below the flue gas dew point temperature. Condensation threshold. The thermo-mass interaction dynamic model is set up to calculate the average dew point temperature based on historical data, minus a safety margin. ,For example This is to ensure that condensation risks can be identified in advance during forecasting. Step response threshold. Used to identify significant changes in heat load demand. In this scheme, if... The range of change in the target heat supply or target heating temperature demand value within a continuous time step. (For example, the absolute change) exceeds its current value If the probability distribution variance is not specified, it is determined to be a step jump. The variance of the probability distribution is a statistical measure of the dispersion of a set of data. In this scheme, the deviation of the instantaneous potential prediction series of latent heat energy flow from the mean value within the time window is calculated to assess the accuracy and stability of the prediction. The prediction confidence interval refers to the interval within which the true value of the estimated parameter may fall at a given confidence level. In this scheme, the confidence level... Set as , indicating that there is The predicted true value will fall within this interval. Time window Set as Each time step examines the events before and after the event node. Minute-by-minute forecast data.

[0061] The target loop architecture anchoring module analyzes key event trigger signals, calculates the energy level increase based on multi-dimensional feature trajectory lines, and anchors the target loop architecture.

[0062] In a specific embodiment of the present invention, the specific process of parsing the key event trigger signal, calculating the energy level improvement based on the multi-dimensional feature trajectory line, and anchoring the target loop architecture includes: comparing the predicted confidence interval carried in the key event trigger signal with the preset physical security anti-fluctuation judgment threshold.

[0063] If the predicted confidence interval is lower than the physical safety anti-fluctuation judgment threshold, a command to maintain the current baseline operating condition is sent to the underlying driver to act as a system stability redundancy dead zone.

[0064] When the predicted confidence interval is greater than or equal to the physical security anti-fluctuation judgment threshold, the energy level improvement magnitude corresponding to the supply and demand sides is calculated according to the multi-dimensional feature trajectory line, and the target loop architecture containing the hardware-level topology switching path is anchored in the feedforward expert library according to the energy level improvement magnitude.

[0065] Specifically, the processor first receives a critical event trigger signal. This signal encapsulates the event type, the predicted instantaneous potential value of latent heat energy flow, and its prediction confidence interval, indicating a specific future time point. The processor parses this signal and extracts the lower limit of the prediction confidence interval. Subsequently, the processor retrieves the physical security anti-vibration judgment threshold built into the controller from the memory pre-stored in the system configuration module. This threshold represents the minimum credible latent heat recovery potential required for the system to perform aggressive control operations involving hardware-level topology switching. The processor will predict the lower bound of the confidence interval. Physical security anti-vibration judgment threshold Compare them.

[0066] If the lower limit of the prediction confidence interval is... Below the physical safety anti-vibration judgment threshold This indicates that the predicted future events and their potential have a high degree of uncertainty, insufficient to support aggressive control measures that might introduce system fluctuations or risks. In this case, the processor sends a command to the underlying driver via the Industrial Ethernet protocol to maintain the current baseline operating conditions. This command explicitly prohibits the underlying actuator from actively changing the current compressor speed, electronic expansion valve opening, or circulation loop valve state, locking the system in the current known and stable operating state. This serves as a system stability redundancy dead zone, effectively avoiding misoperation caused by uncertain predictions.

[0067] When the lower limit of the prediction confidence interval Greater than or equal to the physical security anti-vibration judgment threshold If the prediction is accurate, it indicates that the prediction has sufficient reliability and the system can be safely adjusted. At this point, the processor, based on the multi-dimensional feature trajectory, calculates the corresponding energy level increase on both the supply and demand sides at the time nodes marked by the key event trigger signal. The energy level improvement comprehensively considers the predicted total recoverable heat on the flue gas side (including sensible and latent heat) and the predicted changes in demand for target heating temperature and heat supply on the heat load side, quantifying it as the expected increase in the target output water temperature of the heat pump system. Finally, the processor calculates the energy level improvement. The event types contained in the key event trigger signals, such as "flue gas entering the condensation temperature zone" or "a step increase in heat load demand," are used as query keys for matching in the feedforward expert database. The feedforward expert database is a knowledge base composed of predefined rules, storing the mapping relationship between different energy level enhancement demands and specific hardware-level topology switching paths. Upon successful matching, the processor anchors the target loop architecture identifier containing the corresponding hardware-level topology switching path for subsequent execution.

[0068] The physical safety anti-fluctuation threshold is a key threshold determined during the design phase of a heat pump system based on a series of risk assessments and performance tests. This threshold represents the lowest recoverable latent heat energy flow the system can tolerate without introducing additional operational risks. Its setting is based on ensuring that the system maintains stable operation and complies with safety specifications even under hardware-level topology switching conditions caused by prediction errors or external disturbances. For example, through system fatigue testing and failure mode analysis on an experimental platform, a threshold below [a certain value] is determined. The latent heat recovery potential is not worth the additional hardware switchover costs and potential risks, therefore... Set as The instruction to maintain the current baseline operating condition refers to a structured set of control commands used to instruct actuators in the heat pump system to maintain their current operating state. For example, for a variable frequency compressor, the instruction is to maintain the current frequency; for an electronic expansion valve, the instruction is to maintain the current opening degree; and for various electric valves, the instruction is to maintain the current open / closed state. This instruction ensures that the system avoids any proactive adjustments that might disrupt the current equilibrium state when prediction uncertainty is high.

[0069] Energy level increase ( (This is a comprehensive quantitative indicator based on multi-dimensional characteristic trajectory lines, predicting the expected increase in hot water temperature on the heat load side and the potential increase in output power provided by the recoverable latent heat on the flue gas side over a future period. Specifically defined as...) .in, Represents the expected increment of the predicted target output water temperature, in °C; Represents the instantaneous potential of the predicted latent heat energy flow in the key event trigger signal, in kW; Temperature rise conversion factor (unit: Its value is determined by the system water flow rate and specific heat capacity, and is used to convert latent heat power into an equivalent temperature rise contribution. In this scheme, it is set to 0.05-0.15 based on rated operating conditions. This coefficient ensures the consistency of dimensions at both ends of the formula.

[0070] The feedforward expert library is a rule mapping table loaded into the non-volatile memory of the central data processor during system initialization. This library associates input conditions (such as "energy level improvement" falling within a specific range and "event type") with specific "target loop architecture" identifiers using predefined IF-THEN rules. The establishment of this expert library relies on detailed thermodynamic simulation optimizations performed during the heat pump system design, multi-condition experimental data analysis, and the experience summarized by senior industry experts.

[0071] Hardware-level topology switching path: This refers to the sequence of underlying physical device actions that the system needs to perform to implement the target refrigerant cycle architecture after it has been anchored. This includes switching specific electric valves, adjusting pump operating parameters, switching compressor stage connection methods, or activating / deactivating certain auxiliary heat exchangers and bypass pipes to physically change the refrigerant cycle flow path and thermodynamic operating mode.

[0072] Target cycle architecture: This refers to the optimal thermodynamic cycle configuration mode determined by the heat pump system within a specific future time window, based on the energy level improvement, the event type indicated by key event trigger signals, and the matching results from the feedforward expert library. This architecture defines key operating topologies such as the refrigerant flow direction, the number of compressor stages, the connection method of the heat exchangers, and whether to enable latent heat recovery loops or intermediate cooling loops.

[0073] The latent heat recovery loop pre-activation module activates the latent heat recovery loop in advance based on the target loop architecture and key event trigger signals, calculates the expected dew point drop slope curve based on multi-dimensional feature trajectory lines, and issues adaptive evaporation pressure parameters.

[0074] In a specific embodiment of the present invention, the process of calculating the expected dew point drop slope curve based on the multi-dimensional feature trajectory line and issuing adaptive evaporation pressure parameters includes: before the phase change precursor time window indicated by the key event trigger signal officially arrives, the pipeline valve cutting action is performed to connect the latent heat recovery loop containing the direct contact two-phase flow heat exchanger to the main flow path of the system in parallel.

[0075] The expected dew point decrease slope curve shown in the multi-dimensional feature trajectory line is calculated, and adaptive evaporation pressure parameters that match and fit it are deployed in the latent heat recovery loop.

[0076] By using adaptive evaporation pressure parameters to guide the bottom-side throttling action of the latent heat recovery loop, the actual heat absorption cold end temperature of the system follows the flue gas cooling trajectory to continuously strip away and capture condensation latent heat.

[0077] Specifically, the processor first receives the target loop architecture identifier and a critical event trigger signal. From the critical event trigger signal, the processor extracts the time node identifier (e.g., 20 minutes in the future) for the event type "flue gas entering the condensation temperature zone". To ensure that the switch is completed before the condensation physical time window officially arrives, the processor calculates a pre-activation time margin. (For example: The processor (in minutes) subtracts this margin from the time node identifier to obtain the actual pipeline valve-cutting action execution time. At the action execution time, the processor sends a series of pipeline valve-cutting commands to the underlying actuators via the industrial Ethernet network. These commands activate the latent heat recovery loop of the direct contact two-phase flow heat exchanger in the corresponding target circulation architecture. Specifically, the command controller (e.g., a programmable logic controller, PLC) drives the electric ball valve assembly to perform opening and closing actions in a preset sequence, connecting the latent heat recovery loop, which was originally in a non-operational state, to the main flow path of the system in parallel. The parallel connection method ensures that before the flue gas enters the main heat exchanger, a portion of the flue gas flow first passes through the latent heat recovery loop for pre-cooling and latent heat capture.

[0078] Next, the processor extracts the predicted flue gas dew point temperature sequence within a specific time window before and after the time node marker from the multi-dimensional feature trajectory line. For example, it extracts the predicted flue gas dew point temperature sequence before the time node marker. Minutes later The processor calculates the predicted dew point temperature data points within a given time window using first-order difference and moving average processing. This calculates the expected dew point drop slope curve for that time window. The expected dew point drop slope curve represents the future trend of flue gas condensation temperature over time, expressed in °C / min. Subsequently, based on the expected dew point drop slope curve, the processor generates adaptive evaporation pressure parameters that match the curve, either by querying an internally stored optimized lookup table or through an online optimization algorithm. These adaptive evaporation pressure parameters are a series of dynamically changing evaporation pressure setpoints designed to ensure that the refrigerant evaporation temperature in the latent heat recovery loop accurately tracks the predicted flue gas dew point temperature drop trajectory.

[0079] Finally, the processor sends adaptive evaporation pressure parameters in real time to the electronic expansion valve controller inside the latent heat recovery loop. The electronic expansion valve controller receives the parameters and, based on its built-in proportional-integral-derivative (PID) control algorithm, dynamically adjusts the opening of the electronic expansion valve, thereby guiding the bottom-side throttling action of the latent heat recovery loop. This throttling action controls the refrigerant flow rate and evaporation pressure entering the evaporator in real time, ensuring that the actual heat absorption cold-end temperature of the latent heat recovery loop (i.e., the refrigerant evaporation temperature) follows the predicted flue gas cooling trajectory. Through this dynamic tracking, the system can strip away and capture the latent heat of condensation in the flue gas throughout the entire flue gas condensation process, avoiding a decrease in latent heat recovery efficiency due to temperature mismatch.

[0080] ;

[0081] in, This indicates the expected slope of the dew point decrease, expressed in °C / min. Indicates after the time node marker The predicted flue gas dew point temperature at the minute mark, as shown in °C, within the multi-dimensional feature trajectory line. Indicates before the time node marker The predicted flue gas dew point temperature at the minute mark, as shown in °C, within the multi-dimensional feature trajectory line. This indicates the moment of the time node. This represents the length of time, in minutes, that extends forward from the time node marker within the time window used for slope calculation. This indicates the length of time, in minutes, that extends backward from the time node marker within the time window used for slope calculation.

[0082] Among them, the pre-activation time margin ( ): This is a time setting that takes into account factors such as valve action response time, system pipeline pressurization time, and data processing delay; for example, a value of The aim is to ensure that the latent heat recovery loop is in operation before the flue gas actually enters the condensation temperature zone. Hardware-level switching and pre-conditioning are completed within minutes. Electric ball valve assembly: This is an array of ball valves driven by electric actuators, used to achieve high-speed, high-precision switching of the flow path. This solution selects those with… Electric ball valve with rapid opening and closing function. Direct contact two-phase flow heat exchanger: a heat exchange device where the refrigerant directly contacts the flue gas, and while water vapor condenses in the flue gas, the refrigerant evaporates and absorbs heat. Its characteristics include high heat exchange efficiency, but it has high requirements for the compatibility and corrosion resistance of the refrigerant and flue gas components. System main flow path: refers to the main heat exchange path through which the flue gas flows during the sensible heat recovery stage of the flue gas waste heat heat pump system. First-order difference calculation: a method of numerically approximating the derivative of a discrete function, used to calculate the rate of change between adjacent data points. Moving average processing: smoothing the data by calculating the average value of data within a time window to reduce the impact of noise on the slope calculation. Expected dew point drop slope curve: derived through numerical differentiation calculation based on the predicted flue gas dew point temperature change trend in the multi-dimensional characteristic trajectory line. For example, in this scheme... Set as minute, Set as Minutes. Adaptive evaporation pressure parameters: These are a series of real-time dynamically updated evaporation pressure reference values, obtained through table lookup or iterative calculation, designed to ensure that the saturated evaporation temperature of the refrigerant in the evaporator closely tracks changes in the flue gas dew point temperature. Electronic expansion valve controller: This is a dedicated controller in the heat pump system responsible for receiving control signals and adjusting the opening of the electronic expansion valve. Its built-in PID control algorithm enables precise regulation of the refrigerant flow. Actual system heat absorption cold end temperature: This refers to the evaporation temperature of the refrigerant in the latent heat recovery loop, which is typically obtained by measuring the refrigerant temperature at the evaporator outlet using a high-precision temperature sensor.

[0083] The compression stage reconstruction and loop activation module responds to the target loop architecture, reads the boost gap value in the energy level increase, reconstructs the compression stages, and activates the intermediate cooling loop.

[0084] In a specific embodiment of the present invention, the specific process of responding to the target loop architecture, reading the boost gap value in the energy level increase range, reconstructing the compression stage and enabling the intermediate cooling circuit includes: reading the boost gap value in the energy level increase range that exceeds the preset single-stage conventional refrigeration cycle temperature zone boundary, and triggering a multi-stage connection command to pump the heat captured by the latent heat recovery circuit to a higher grade level.

[0085] By performing a hard-wired contact switching action, the conventional single-stage series work path of the actuator assembly is reconstructed into a multi-stage compression stage with a nonlinear enthalpy-increasing effect.

[0086] Along with the topology change of the multi-stage compression, the intermediate cooling circuit with built-in flash thermostatic expansion valve is activated to inject liquid refrigerant in the multi-stage compression chamber in the reverse direction to suppress excessive exhaust temperature and improve the near isothermal properties of the compression process.

[0087] Specifically, the processor first reads the energy level increase. This energy level increase quantifies the target temperature rise that the heat pump system needs to achieve. The processor compares this target temperature rise with the preset boundary of a single-stage conventional refrigeration cycle temperature range. Comparison is needed. The temperature zone boundary of a single-stage conventional refrigeration cycle represents the maximum temperature rise that a heat pump system can achieve in single-stage compression mode. If the energy level rise... Breakthrough in temperature range boundaries of single-stage conventional refrigeration cycles Then the boost gap value is calculated. This pressure gap indicates that the current heat load demand exceeds the capacity of a single-stage compression cycle. Subsequently, the processor triggers a multi-stage connection instruction based on the pressure gap value, pumping the heat captured in the latent heat recovery loop to a higher-grade level. The multi-stage connection instruction is a logical signal indicating that the system needs to activate multi-stage compression mode.

[0088] Based on a multi-level connected instruction and anchored target loop architecture, the processor drives a series of hardwired contact switching actions through a digital output interface. Specifically, the processor sends instructions to a power control unit integrated with a programmable logic controller (PLC), which controls industrial relays and contactors to change the power supply circuit of the compressor motor windings and the energized state of the electric valves in the refrigerant lines. These actions reconstruct the conventional single-stage series work path of the execution assembly into a multi-stage compression process with nonlinear enthalpy enhancement. For example, if the original system is a single-stage compression, it is converted into a two-stage series compression by switching valves and electrical connections, allowing the exhaust gas from the first-stage compressor to directly enter the intake side of the second-stage compressor, and introducing an intermediate cooling mechanism through an intermediate pressure section.

[0089] As the topology changes with each compression stage, the processor synchronously sends a command to the flash thermostatic expansion valve controller to activate the built-in flash thermostatic expansion valve, thereby activating the intercooling circuit. The built-in flash thermostatic expansion valve throttles a portion of the high-pressure liquid refrigerant from the condenser to the intermediate pressure, causing it to flash and produce a mixture of gaseous and liquid refrigerant. Subsequently, the processor precisely controls the opening of the flash thermostatic expansion valve to inject the flashed liquid refrigerant in reverse into the intercooler between the multiple compression chambers or directly into the suction line of the second-stage compressor. The liquid refrigerant evaporates and absorbs heat in the intercooler or mixes directly with the high-temperature first-stage exhaust gas in the suction line, thus suppressing excessively high exhaust temperatures and improving the near-isothermal properties of the compression process. This ensures appropriate suction temperatures in subsequent compression stages, thereby improving compression efficiency and extending compressor life.

[0090] Among them, the temperature zone boundary of a single-stage conventional refrigeration cycle ( This refers to the maximum difference between the condensing and evaporating temperatures that a heat pump system can achieve while meeting the design COP (coefficient of performance) requirements, using only a single-stage compressor (and only conventional suction superheat control and condensing subcooling control). This boundary value is typically determined by thermodynamic calculations and experimental testing. For example, under a specific refrigerant, the maximum temperature rise difference between the outlet water temperature and the source water temperature in single-stage compression mode can be set as follows: Boost gap value ( ): This is the difference between the energy level increase and the temperature range boundary of a single-stage conventional refrigeration cycle. It quantifies the portion of the single-stage compression capacity that is insufficient to meet the target temperature rise, and its unit is °C. This value is a key criterion for switching the drive system from single-stage compression to multi-stage compression. Multi-stage connection instruction: This is a binary control signal (e.g., 00b represents single-stage, 01b represents two-stage, and 10b represents three-stage), which is issued by the processor based on... Whether it is greater than zero is used to generate and send it to the underlying relays and valve actuators.

[0091] Hard-wired contact switching action: This refers to physically switching the operating mode of the compressor motor and the physical position of the flow control valve in the refrigerant pipeline by changing the electrical wiring and the opening and closing state of the relay, thereby changing the topology of the refrigerant circulation loop. Multi-stage compression: This refers to decomposing the compression process into two or more stages, each using a single compressor or a combination of different cylinders from a multi-cylinder compressor. For example, two-stage compression can achieve a higher compression ratio, resulting in a higher condensing temperature and a greater temperature rise. Nonlinear enthalpy increase effect: In multi-stage compression cycles with intercooling, the supercooling of the coolant and the intercooling effect reduce the compressor's suction superheat, thus increasing the compressor's volumetric cooling capacity and coefficient of performance (COP) at the same compression ratio. The thermodynamic path is represented by an increase in the effective area of ​​the compression line region on the pressure-enthalpy diagram.

[0092] Built-in flash thermostatic expansion valve: This refers to a dedicated thermostatic expansion valve integrated within the heat pump system, equipped with flash function. It rapidly throttles the incoming liquid refrigerant, causing part of it to flash into vapor and cooling the remaining liquid refrigerant. These two components are then directed to different paths: saturated vapor is typically sent directly to the suction port of the second-stage compressor or the intermediate pressure vessel, while the subcooled liquid refrigerant is used for intercooling or further throttled for the evaporator. Intercooling circuit: In a multi-stage compression system, this circuit cools the refrigerant gas discharged from the previous stage compressor, lowering its temperature before it enters the suction port of the next stage compressor. Common intercooling methods include flash cooling, liquid injection cooling, or cooling using an intermediate heat exchanger. Reverse injection of liquid refrigerant: This refers to directly injecting a portion of the liquid refrigerant obtained from the flash process or throttled by an auxiliary expansion valve into the multi-stage compression chamber (usually the interstage of a two-stage compressor or the suction port of a two-stage compressor), lowering the gas temperature through heat absorption during refrigerant evaporation. This helps to suppress overheating caused by non-isentropic processes.

[0093] The dynamic flow compensation strategy generation module analyzes the working fluid ratio parameters issued by the sulfur oxide concentration variation rate associated with the dynamic model of thermo-mass interaction, and extracts the slope first derivative of the multi-dimensional feature trajectory line to generate a dynamic flow compensation strategy.

[0094] In a specific embodiment of the present invention, the specific process of extracting the first derivative of the slope of the multi-dimensional feature trajectory line to generate a dynamic flow compensation strategy includes: parsing the sulfur oxide concentration variation rate associated in the preset thermal-mass interaction dynamic model fingerprint dictionary, and issuing supplementary working fluid ratio parameters for perfluorocarbon anti-corrosion fluid to the variable capacity mixing tank.

[0095] The first derivative of the slope of the multi-dimensional feature trajectory line in the transition state period is extracted as a correction factor, and a dynamic flow compensation strategy for limiting the mechanical damping and opening / closing dead zone of the electronic expansion valve is iteratively generated.

[0096] Apply a dynamic flow compensation strategy to regulate the dynamic flow response characteristics of the bottom heat exchange components, thereby reducing the risk of evaporator dry burning and refrigerant depletion caused by severe load fluctuations, as well as the risk of liquid slugging and backflow in the condenser.

[0097] Specifically, the processor first parses the associated sulfur oxide concentration variation rate from the constructed thermal-mass interaction dynamic model fingerprint dictionary. The fingerprint dictionary is a pre-built database that records the empirical correlation between sulfur oxide concentration and the internal corrosion rate of the system under different flue gas components. The processor uses data mining algorithms to extract historical sulfur oxide concentration data from the fingerprint dictionary and calculate its statistical variation rate (e.g., standard deviation or coefficient of variation) within a specific time window. If the sulfur oxide concentration variation rate exceeds a preset threshold... This indicates an increased risk of corrosion. The processor, based on the variability rate, consults a corrosion prevention strategy lookup table and sends supplementary working fluid mixing parameters for a perfluorocarbon corrosion inhibitor to the controller integrated in the variable-capacity mixing tank. These parameters, expressed as a volume percentage, indicate the mixing ratio of refrigerant and corrosion inhibitor, used to establish a protective layer or neutralize corrosive substances in the refrigerant circuit.

[0098] Next, the processor extracts the generated multi-dimensional feature trajectory lines. It focuses on analyzing the changes in the trajectory lines within the transition state period, which is defined as the period before and after the identified key event trigger signal. A time window of minutes is defined. Within the transition state period, the processor performs numerical differentiation calculations on the sequence of key parameters (e.g., flue gas flow rate, target heat supply) of the multi-dimensional characteristic trajectory to obtain the first derivative of its slope. The first derivative characterizes the dynamic rate of the system response, i.e., the variable operating condition jump rate. The processor uses the variable operating condition jump rate as a correction factor. The processor utilizes a correction factor. By combining internal characteristic curve data provided by the electronic expansion valve manufacturer with online system operating parameters, a dynamic flow compensation strategy is iteratively generated to limit the mechanical damping and opening / closing dead zone of the electronic expansion valve. Specifically, the strategy involves adjusting the PID controller parameters of the electronic expansion valve (e.g., proportional gain). Integral Time Differential time and dead zone width The dynamic setting value.

[0099] Finally, the processor sends the dynamic flow compensation strategy to the electronic expansion valve controller via the industrial fieldbus and applies the strategy to regulate the dynamic flow response characteristics of the underlying heat exchange components. The strategy ensures that the refrigerant flow accurately and smoothly responds to instantaneous load fluctuations by adjusting the real-time opening of the electronic expansion valve. Specifically, when a rapid load decrease is predicted, the strategy preemptively reduces the opening of the electronic expansion valve to prevent excessive refrigerant from entering the evaporator, thereby reducing the risk of evaporator dry burning and refrigerant depletion caused by drastic load fluctuations. When a rapid load increase is predicted, the strategy preemptively increases the opening of the electronic expansion valve to ensure sufficient refrigerant and finely controls the valve closing speed to prevent refrigerant droplets from flowing back to the compressor too quickly, thereby reducing the risk of condenser liquid slugging and backflow.

[0100] ;

[0101] in: This represents the rate of variation in sulfur oxide concentration and is dimensionless. This indicates the number of data points used for calculation, representing the number of samples of historical SOx concentration data in the fingerprint dictionary. Indicates the first fingerprint in the dictionary Historical SOx concentration values, in ppm. This represents the average historical SOx concentration within the time window, expressed in ppm.

[0102] ;

[0103] in: This represents the sudden jump rate under varying operating conditions, i.e., the key parameters in the multi-dimensional characteristic trajectory line during time. The first derivative of the slope at that point. This represents the predicted value of a key parameter in a multi-dimensional feature trajectory line at consecutive time points. and The sampled value at that location. The time step is in minutes.

[0104] Among them, the thermal-mass interaction dynamic model fingerprint dictionary is a structured knowledge base containing historical analysis data of flue gas chemical composition, corrosion research results, and the correlation between sulfur oxide content and equipment material degradation, stored in non-volatile memory. Sulfur oxide concentration variation rate ( ): refers to a preset sliding time window (e.g., The statistical dispersion of sulfur oxide concentration in flue gas relative to its average value over a period of time (hours), for example, the coefficient of variation calculated using the formula above. Its threshold. Set as This indicates that when the coefficient of variation exceeds At this point, the risk of corrosion increases significantly. This threshold is set based on corrosion engineering experience and accelerated aging test data for materials. Variable capacity mixing tank: This is a tank with precise metering and mixing functions, capable of storing and transporting two or more fluids (such as refrigerant and corrosion inhibitors) proportionally according to instructions. Perfluorocarbon corrosion inhibitors: These are chemically stable, high-temperature resistant, and corrosion-resistant synthetic fluids. They form a protective layer in the refrigerant circuit or react with corrosive substances to reduce their activity, but do not significantly adversely affect the thermodynamic properties of the refrigerant. The working fluid ratio parameter is usually set at... The volume concentration range, the specific value of which is determined by the corrosion variability rate and system design.

[0105] Transitional state period ( This refers to the time period during which the system's operating state transitions from one steady state to another, during which system parameters (such as flue gas flow rate and target heat supply) change at a relatively high rate. In this scheme, it is set as the period before and after the identified key event trigger signal. A minute-by-minute time window is used to capture sufficient information about market jumps. First derivative of the slope: represents the instantaneous rate of change of the trajectory line at each time point within the transition period. Correction factor ( ): This is a dimensionless multiplier whose value is positively correlated with the magnitude of the sudden jump rate under varying operating conditions. The correction factor is used to adjust the gain parameters and dead zone width of the electronic expansion valve PID controller. For example, can be Decision, among which These are empirical constants. Mechanical damping and dead zone of an electronic expansion valve (EEV): These are inherent mechanical characteristics of the EEV. Damping causes a lag in valve response, while the dead zone refers to the range within which the valve motor does not operate under small signal changes. These characteristics can affect flow control accuracy under highly dynamic operating conditions. Dynamic flow compensation strategies aim to counteract these adverse effects by adjusting controller parameters. Dynamic flow compensation strategy: Specifically, this refers to the PID controller parameters of the electronic expansion valve (e.g., proportional gain). Integral Time Differential time And the real-time adjustment scheme for dead zone width values, these parameters are based on correction factors. Dynamically updated.

[0106] Evaporator dry burning due to refrigerant deficiency: This refers to insufficient refrigerant flow into the evaporator caused by a sudden drop in load or a delayed response of the electronic expansion valve. The evaporator's heat transfer surface temperature becomes too high without refrigerant evaporation to absorb heat, potentially damaging the evaporator or causing the compressor to overheat. Condenser liquid slugging backflow risk: This refers to a sudden increase in load or an excessively rapid instantaneous closure of the electronic expansion valve, causing a large amount of incompletely evaporated liquid refrigerant to flow back to the compressor's suction side, potentially damaging the compressor valve plates or piston.

[0107] The thermal-mass interaction dynamic model correction module extracts the actual operating effect features, calculates the temporal residual between the actual operating effect features and the predicted scalar of the multi-dimensional feature trajectory line, and corrects the thermal-mass interaction dynamic model in reverse closed loop.

[0108] In a specific embodiment of the present invention, the process of calculating the temporal residual between the actual operating effect characteristics and the predicted scalar of the multi-dimensional feature trajectory line, and then correcting the dynamic model of heat-mass interaction in reverse closed loop includes: after passing the action time axis assigned by the key event trigger signal, reading the actual outlet heat release temperature and effective heat supply settlement amount at the system load end through the bus.

[0109] The effective heating settlement amount, the measured shaft power input of the system as power consumption loss, and the heat release from condensation liquefaction that is physically intercepted are calculated and combined according to preset fixed weights to form the actual operating effect characteristics.

[0110] The temporal residual between the actual operating effect characteristics and the predicted scalar of the multi-dimensional feature trajectory is calculated. The temporal residual is used to fine-tune the node logical weights of the thermo-mass interaction dynamic model through gradient backpropagation, thereby converging and narrowing the prediction confidence interval generated by the secondary prediction.

[0111] Specifically, the processor first continuously monitors the system timeline. After the timeline of the action assigned by the critical event trigger signal has passed, for example, after a critical event occurs and the system has stabilized... Minutes later, the processor initiates the data acquisition and evaluation program. The processor reads the actual outlet water heat release temperature at the system load end via the CANopen bus at a high frequency (e.g., 10 times per second), measured by a PT1000 high-precision platinum resistance temperature sensor installed at the outlet of the heating pipeline. Simultaneously, the processor combines this with real-time flow data from the electromagnetic flowmeter, and calculates the flow rate over an evaluation period (e.g., within a given evaluation cycle) through numerical integration. The cumulative heat output by the system to the load side (minutes) is the effective heating settlement amount.

[0112] Simultaneously, the processor calculates complete system energy efficiency indicators. Using a three-phase power quality analyzer connected to the compressor motor power supply circuit, the processor collects and calculates the measured shaft power input as power loss with a millisecond-level response speed. Furthermore, using an ultrasonic flow meter and temperature sensor installed on the condensate drainage pipe of the latent heat recovery circuit, the processor measures the physically intercepted condensate liquefaction water volume and its temperature in real time, and calculates the actual heat released during the condensation liquefaction process per unit time based on the enthalpy of water. The processor weights and sums the effective heat supply (positive contribution), the measured shaft power power loss (negative contribution), and the physically intercepted condensate liquefaction heat release (positive contribution) according to preset fixed weights, assembling a characteristic of the system's overall energy efficiency based on actual operating performance. .

[0113] Finally, the processor will display the actual running performance characteristics. Corresponding prediction scalar generated by multi-dimensional feature trajectory prediction Compare the two and calculate the time residuals. Predicting scalars It is used in the prediction phase and The calculations were performed using the exact same weighting formula and prediction parameters. The processor then calculates the timing residuals. As input to the loss function, the nodal logical weights of the dynamic model of heat-mass interaction are fine-tuned using a gradient backpropagation algorithm based on the Adam optimizer. Nodal logical weights are adjustable parameters in the model used to correct physical formulas or empirical correlations. For example, correction factors related to the convective heat transfer coefficient or condensation mass transfer coefficient on the flue gas side are updated via gradient descent. This closed-loop correction process continues, aiming to converge and narrow the prediction confidence interval generated by the secondary prediction (i.e., the prediction of the next similar operating condition), thereby improving the prediction accuracy and control performance of the entire system.

[0114] ;

[0115] in: This represents the characteristics of the operational effect, which can be the actual operational effect characteristics. Or multi-dimensional feature trajectory line prediction scalar The unit is kW. This indicates the effective heating settlement amount (or its predicted value), in kW. This represents the measured shaft power as power loss (or its predicted value), in kW. This represents the heat released by the physical interception of condensation liquefaction (or its predicted value), expressed in kW. These are the corresponding fixed weights, dimensionless, and satisfying... The heat released during condensation liquefaction is listed separately here. This is to specifically increase the weight of latent heat recovery in model optimization, encouraging the system to make more use of latent heat, rather than simply calculating physical energy balance.

[0116] ;

[0117] in: This represents the timing residual, in kW. The actual operating effect characteristics are represented by formula (7) using measured values. The multidimensional feature trajectory line prediction scalar is calculated using the predicted value from formula (7).

[0118] ;

[0119] in: and These represent the values ​​of the logical weight of a node in the dynamic model of thermal-mass interaction before and after the update. The learning rate is a hyperparameter that controls the update step size. This represents the gradient of the loss function (mean squared error) with respect to the node logical weights, calculated using the gradient backpropagation algorithm.

[0120] The action time axis refers to the assessment point in time after the predicted key event occurs and corresponding controls are implemented, from which the system enters a new steady state. Stable operating time ( ) set as Minutes, a value greater than the system's thermodynamic response time after a typical operating condition switch, ensures that the collected data represents the performance under the new steady state. PT1000 high-precision platinum resistance temperature sensor: its... The resistance value at that time is Compared to the PT100, it offers higher resolution and accuracy, making it suitable for applications with stringent temperature measurement requirements. Effective heating billing volume ( ): Within an evaluation cycle, by The calculation shows that, among which The specific heat capacity of water, The density of water, For traffic, The supply and return water temperature difference. Actual shaft power is used to calculate power consumption loss. ): This refers to the actual electrical power consumed by the compressor, calculated in real time by a three-phase power quality analyzer by measuring voltage, current, and power factor. The heat released during the physical interception of condensation liquefaction ( ):Depend on The calculation shows that, among which The mass flow rate of the condensate is measured by an ultrasonic flow meter. This represents the latent heat of vaporization at this condensation temperature. (Fixed weights:) These are dimensionless constants pre-set based on the economic benefits and energy efficiency targets of the system design. For example, if the system's COP is the primary optimization objective, then the weights can be set as follows: This setting emphasizes the contribution of effective heat supply and latent heat recovery, while imposing a penalty on power losses. Multidimensional feature trajectory line prediction scalar ( ): This refers to the prediction phase, used in calculations. The exact same formula (7), but the input value is the predicted future. The calculated overall performance prediction value. Time series residuals ( ): This is the quantified deviation between actual and predicted performance, and is the core error signal driving model self-optimization. Gradient backpropagation: This is a standard algorithm for calculating gradients in neural networks and differentiable models. It propagates the error of the output layer backward layer by layer, thereby calculating the contribution of each adjustable parameter to the total error. Node logical weights ( (): This refers to adjustable correction coefficients in dynamic models of heat-mass interaction, based on empirical or semi-empirical formulas. For example, in calculating the convective heat transfer coefficient of flue gas to a wall. At that time, the model may adopt In the form of, This refers to the node logical weight, whose initial value is... This closed-loop correction process is used to fine-tune the model so that it better reflects the actual physical process.

[0121] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A flue gas waste heat heat pump energy level enhancement system with dynamic source-load matching, characterized in that, include: The thermo-mass interaction dynamic model construction module acquires real-time physical sensing data and integrates it with the equipment's pre-condition signals to form a source-load-environment hybrid dataset. It then inputs a pre-set fluid dynamic mechanism architecture to perform parameter optimization and matching, thereby generating a thermo-mass interaction dynamic model. The latent heat energy flow instantaneous potential quantification module runs a thermo-mass interaction dynamic model to calculate the instantaneous condensation rate, and combines the flow scalar in the source-load environment mixed dataset to quantify and extract the instantaneous potential of latent heat energy flow. The critical event trigger signal generation module inputs the source-load environment hybrid dataset into a deep time series prediction network to fit a multi-dimensional feature trajectory line, and calculates the probability distribution variance by combining the instantaneous potential of latent heat energy flow to generate a critical event trigger signal. The target loop architecture anchoring module analyzes key event trigger signals, calculates the energy level improvement based on multi-dimensional feature trajectory lines, and anchors the target loop architecture. The latent heat recovery loop pre-activation module activates the latent heat recovery loop in advance based on the target loop architecture and key event trigger signals, calculates the expected dew point drop slope curve based on multi-dimensional feature trajectory lines, and issues adaptive evaporation pressure parameters. The compression stage reconstruction and loop activation module responds to the target loop architecture, reads the boost gap value in the energy level increase, reconstructs the compression stage and activates the intermediate cooling loop; The dynamic flow compensation strategy generation module analyzes the sulfur oxide concentration variation rate associated with the dynamic model of thermo-mass interaction and sends working fluid ratio parameters, and extracts the slope first derivative of the multi-dimensional feature trajectory line to generate a dynamic flow compensation strategy. The thermal-mass interaction dynamic model correction module extracts the actual operating effect features, calculates the temporal residual between the actual operating effect features and the predicted scalar of the multi-dimensional feature trajectory line, and corrects the thermal-mass interaction dynamic model in reverse closed loop.

2. The flue gas waste heat heat pump energy level enhancement system with dynamic source-load matching according to claim 1, characterized in that, The specific process of fusing real-time physical sensor data with equipment pre-condition signals to form a source-load-environment hybrid dataset includes: The system collects the instantaneous temperature and flow parameters of the continuous flue gas at the source end of the waste heat heat pump system, and simultaneously obtains the target heat supply and target heat supply temperature demand values ​​of the corresponding time slice on the heat load side as real-time physical sensing data. Extract the upstream boiler load change signal from the associated production pipeline network as the equipment precondition signal; Real-time physical sensor data and equipment front-end operating condition signals are aligned with the system's relative timestamp and fused into a source-load-environment hybrid dataset.

3. The source-load dynamic matching flue gas waste heat heat pump energy level enhancement system according to claim 1, characterized in that, The specific process of performing parameter optimization and matching on the input pre-set fluid dynamic mechanism framework to generate a dynamic model of thermo-mass interaction includes: Input the source-load environment hybrid dataset into a fluid dynamics mechanism framework that includes multiphase flow theory and heat and mass transfer processes; The parameter optimization module based on Bayesian optimization algorithm is invoked to iteratively match the heat transfer coefficient of flue gas sidewall and the water vapor diffusion coefficient in the fluid dynamics mechanism framework; The iteration stops when the root mean square error between the model prediction and the actual data is lower than the preset convergence threshold, and a dynamic model of thermo-mass interaction is generated to analyze the phase change characteristics of flue gas components.

4. The source-load dynamic matching flue gas waste heat heat pump energy level enhancement system according to claim 1, characterized in that, The specific process of calculating the instantaneous condensation rate using the dynamic model of thermo-mass interaction, and combining it with the flow scalar in the source-load environment hybrid dataset to quantify and extract the instantaneous potential of latent heat energy flow includes: By calculating the state equations within the dynamic model of thermo-mass interaction, the concentration of condensable components of water vapor in the flue gas field under different temperature gradient distributions is calculated. Extract saturated vapor pressure correlation data of condensable component concentration as a function of pressure, and calculate the initial dew point temperature and instantaneous condensation rate of the corresponding gas-liquid phase change process. By integrating the flow scalar and instantaneous condensation rate in the source-load environment hybrid dataset, the instantaneous potential of the latent heat energy flow that can be physically recovered in the current flue gas field is identified through enthalpy difference calculation.

5. The flue gas waste heat heat pump energy level enhancement system with dynamic source-load matching according to claim 1, characterized in that, The specific process of calculating the probability distribution variance by combining the instantaneous potential of latent heat energy flow to generate key event trigger signals includes: The source-load environment hybrid dataset is input into a deep time series prediction network to perform multi-step forward propagation calculations and fit a multi-dimensional feature trajectory line of continuous change of source-end energy supply and load-end energy consumption within a future preset time window. Extract time node markers that represent the bottoming out of flue gas and falling into the condensation temperature zone or the step jump in heat load demand from multi-dimensional feature trajectory lines. By combining the probability distribution variance of the segment where the instantaneous potential of latent heat energy flow is calculated, a key event trigger signal carrying a prediction confidence interval is generated.

6. The source-load dynamic matching flue gas waste heat heat pump energy level enhancement system according to claim 5, characterized in that, The specific process of analyzing key event trigger signals, calculating the energy level increase based on multi-dimensional feature trajectory lines, and anchoring the target cyclic architecture includes: Compare the predicted confidence interval carried in the key event trigger signal with the preset physical security anti-fluctuation judgment threshold; If the predicted confidence interval is lower than the physical safety anti-fluctuation judgment threshold, then send an instruction to the underlying driver to maintain the current baseline operating condition as a system stability redundancy dead zone. When the predicted confidence interval is greater than or equal to the physical security anti-fluctuation judgment threshold, the energy level improvement magnitude corresponding to the supply and demand sides is calculated according to the multi-dimensional feature trajectory line, and the target loop architecture containing the hardware-level topology switching path is anchored in the feedforward expert library according to the energy level improvement magnitude.

7. The flue gas waste heat heat pump energy level enhancement system with dynamic source-load matching according to claim 6, characterized in that, The specific process of calculating the expected dew point decrease slope curve based on the multi-dimensional feature trajectory line and issuing adaptive evaporation pressure parameters includes: Before the phase change precursor time window indicated by the critical event trigger signal officially arrives, the pipeline valve cutting action will connect the latent heat recovery loop, which includes the direct contact two-phase flow heat exchanger, to the main flow path of the system in parallel. Calculate the expected dew point drop slope curve shown in the multi-dimensional feature trajectory line, and deploy adaptive evaporation pressure parameters that match and fit it into the latent heat recovery loop; By using adaptive evaporation pressure parameters to guide the bottom-side throttling action of the latent heat recovery loop, the actual heat absorption cold end temperature of the system follows the flue gas cooling trajectory to continuously strip away and capture condensation latent heat.

8. The flue gas waste heat heat pump energy level enhancement system with dynamic source-load matching according to claim 1, characterized in that, The specific process of the target loop architecture in response, reading the boost gap value in the energy level increase, reconstructing the number of compression stages, and enabling the intermediate cooling loop includes: Read the pressure gap value that exceeds the preset temperature zone boundary of the single-stage conventional refrigeration cycle in the energy level increase range, and trigger a multi-stage connection command to pump the heat captured by the latent heat recovery circuit to a higher grade level. By performing a hard-wired contact switching action, the conventional single-stage series work path of the actuator assembly is reconstructed into a multi-stage compression stage with a nonlinear enthalpy-increasing effect. Along with the topology change of the multi-stage compression, the intermediate cooling circuit with built-in flash thermostatic expansion valve is activated to inject liquid refrigerant in the multi-stage compression chamber in the reverse direction to suppress excessive exhaust temperature and improve the near isothermal properties of the compression process.

9. The source-load dynamic matching flue gas waste heat heat pump energy level enhancement system according to claim 5, characterized in that, The specific process of extracting the first derivative of the slope of the multi-dimensional feature trajectory line to generate the dynamic flow compensation strategy includes: The sulfur oxide concentration variation rate associated with the preset thermal-mass interaction dynamic model fingerprint dictionary is analyzed, and supplementary working fluid ratio parameters for perfluorocarbon anti-corrosion fluid are issued to the variable capacity mixing tank. The first derivative of the slope of the multi-dimensional feature trajectory line in the transition state period is extracted as a correction factor, and a dynamic flow compensation strategy for limiting the mechanical damping and opening / closing dead zone of the electronic expansion valve is iteratively generated. Apply a dynamic flow compensation strategy to regulate the dynamic flow response characteristics of the bottom heat exchange components, thereby reducing the risk of evaporator dry burning and refrigerant depletion caused by severe load fluctuations, as well as the risk of liquid slugging and backflow in the condenser.

10. The source-load dynamic matching flue gas waste heat heat pump energy level enhancement system according to claim 5, characterized in that, The specific process of calculating the temporal residual between the actual operational performance characteristics and the predicted scalar of the multi-dimensional feature trajectory, and then using the reverse closed-loop correction of the dynamic model of thermo-mass interaction includes: After the critical event trigger signal has passed its designated timeline, the actual outlet water heat release temperature and effective heating settlement amount at the system load end are read via the bus. The effective heating settlement amount, the measured shaft power input of the system as power consumption loss, and the heat release from condensation liquefaction that is physically intercepted are calculated and combined according to preset fixed weights to form the actual operating effect characteristics. The temporal residual between the actual operating effect characteristics and the predicted scalar of the multi-dimensional feature trajectory is calculated. The temporal residual is used to fine-tune the node logical weights of the thermo-mass interaction dynamic model through gradient backpropagation, thereby converging and narrowing the prediction confidence interval generated by the secondary prediction.

Images