A feedwater pump closed-loop energy recovery system based on recirculation branch residual pressure driving

Abstract

The application discloses a feed water pump closed-loop energy recovery system based on a recirculation branch residual pressure drive, relates to the technical field of energy recovery of a feed water system of a thermal power generating unit, and discloses a modular energy recovery assembly including a series wide-range hydraulic turbine group which is composed of a high-pressure micro-flow hydraulic turbine and a low-pressure large-flow hydraulic turbine and intelligently switches or cooperatively works according to working condition requirements; a digital twin intelligent control core includes a prediction engine and a decision engine, the prediction engine identifies an operation mode and predicts a future state sequence through a behavior embedding vector, and the decision engine generates an optimal control instruction sequence through a strategy network which is regularly trained through optimal transmission theory. The system realizes cooperative control of the feed water pump rotating speed and the recirculation valve opening degree through execution of the control instruction, forms a triple closed-loop control of a data closed loop, an energy closed loop and a control closed loop. The application improves the energy recovery efficiency, enhances the robustness of the control strategy, and realizes predictive maintenance.

Description

Technical Field

[0001] This invention relates to the field of energy recovery technology for feedwater systems of thermal power units, specifically a closed-loop energy recovery system for feedwater pumps driven by residual pressure in recirculation branches. Background Technology

[0002] The feedwater system of a thermal power unit is an important auxiliary system in a thermal power plant. The feedwater pump, as the core equipment, is responsible for pressurizing the feedwater in the deaerator and sending it to the boiler. During unit operation, when the boiler load decreases, the output of the feedwater pump needs to be reduced accordingly. The traditional approach is to return the excess feedwater to the deaerator through the recirculation pipeline and reduce the feedwater pressure through throttling valves or orifice plates. In this process, a large amount of hydraulic energy is wasted.

[0003] Existing technologies recover some residual pressure energy by installing energy recovery devices such as hydraulic turbines on the recirculation branch, but they still have certain limitations, such as: the high-efficiency operating range of a single hydraulic turbine is narrow, making it difficult to adapt to the operating characteristics of large load fluctuations in thermal power units; the energy recovery system lacks coordinated control with the main feedwater system, making it impossible to achieve global optimization; traditional control methods mainly rely on experience and simple feedback control, lacking the ability to predict future operating conditions; equipment health management remains at the planned maintenance stage, unable to achieve early warning of faults and predictive maintenance; system operation mainly focuses on the efficiency optimization of individual equipment, lacking comprehensive consideration from the perspective of the overall system economy.

[0004] These technological limitations result in limited practical application effectiveness of existing energy recovery systems, low energy recovery efficiency, and room for improvement in system reliability and economy. Especially under the current operating mode where thermal power units frequently participate in grid peak shaving, the operating conditions of the water supply system are more complex and diverse, making traditional energy recovery technologies unable to meet the requirements for efficient, intelligent, and reliable operation.

[0005] Therefore, there is an urgent need for an advanced feedwater pump closed-loop energy recovery system that can adapt to a wide range of operating conditions, possess intelligent predictive decision-making capabilities, achieve system-wide collaborative optimization, and support predictive maintenance. This system must not only achieve efficient energy recovery but also, through digital and intelligent technologies, achieve deep integration and collaborative control with the main feedwater system, providing comprehensive technical support for energy conservation, emission reduction, and intelligent operation of thermal power units. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and propose a closed-loop energy recovery system for a feedwater pump driven by residual pressure in a recirculation branch, so as to solve the above-mentioned problems.

[0007] The objective of this invention is achieved through the following technical solution: a closed-loop energy recovery system for a feedwater pump driven by residual pressure in a recirculation branch, comprising: a modular energy recovery assembly installed on the recirculation pipeline of the feedwater pump; the energy recovery assembly comprising: a series-connected wide-range hydraulic turbine group, consisting of a high-pressure micro-flow hydraulic turbine and a low-pressure large-flow hydraulic turbine connected in series; the high-pressure micro-flow hydraulic turbine is specifically designed for handling small flow and high pressure differential conditions under stable low loads, while the low-pressure large-flow hydraulic turbine is specifically designed for handling large flow and low pressure differential conditions under rapidly decreasing loads; the two turbines can intelligently switch or work collaboratively according to operating conditions.

[0008] A generator and speed-increasing gearbox connected in series with a wide-range hydraulic turbine are used to convert the recovered hydraulic energy into electrical energy;

[0009] The integrated safety bypass module, which includes an isolation valve, a throttling orifice plate, and the original recirculation pipeline, can automatically isolate and recover the assembly and switch back to the original throttling and pressure-reducing circuit within milliseconds in the event of a system failure.

[0010] The digital twin intelligent control core is based on a digital twin model. This model constructs a virtual model that perfectly matches the physical system's piping layout and flowcharts, mapping the physical system's state in real time and predicting operational changes. It includes the following functional engines:

[0011] The prediction engine is configured to receive real-time operating data of the receiving unit, identify and encode behavioral embedding vectors that represent macroscopic operating patterns by learning historical operating trajectories, and adaptively adjust the prediction output of future states based on behavioral embedding vectors as independent inputs. The uncertainty of the prediction is quantified by comparing the consistency of multiple extrapolation results based on historical data of different lengths.

[0012] The decision engine is configured to receive the future state sequence output by the prediction engine. When the prediction uncertainty is lower than a preset threshold, the economic objective is to maximize the difference between the value of the recovered electrical energy and the cost of the water pump system's electrical energy consumption. A smooth and stable optimal control instruction sequence is generated through a policy network trained by the optimal transmission theory regularization. The regularization training ensures the robustness of the policy by penalizing the norm of the control action.

[0013] The system achieves coordinated control of the feedwater pump speed and recirculation valve opening by executing the optimal control command sequence, and feeds back the actual operating status of the system after execution as new real-time data to the prediction engine, forming a triple closed-loop control of data, energy and control.

[0014] The working mode switching mechanism of the series wide-range hydraulic turbine group is as follows: when the system is operating at a stable low load and the recirculation flow rate is less than the first threshold, only the high-pressure micro-flow hydraulic turbine is activated for energy recovery; when the system load drops rapidly and the recirculation flow rate exceeds the second threshold, the low-pressure high-flow hydraulic turbine is automatically integrated or switched to become the main unit for energy recovery; when the recirculation flow rate is between the first and second thresholds, the two turbines work together to achieve the optimal energy recovery efficiency.

[0015] The integrated safety bypass module also includes: an intelligent fault detection unit that monitors the vibration, temperature, speed of the series wide-range hydraulic turbine group and the voltage and current parameters of the generator in real time; an emergency switching execution unit that completes the switching action of closing the isolation valve and opening the bypass valve within 10 milliseconds when abnormal parameters are detected to exceed the safe range or an emergency command is received; and a status indication unit that displays the current operating mode and equipment health status in real time.

[0016] The real-time operational data received by the forecasting engine includes: unit load, boiler feedwater flow rate, feedwater pump operating flow rate, feedwater pump outlet pressure, feedwater pump speed, recirculation pipeline valve opening, deaerator water level, feedwater temperature, power grid dispatch instructions, ambient temperature, and real-time electricity market price information.

[0017] The process of generating behavior embedding vectors includes: an encoder module, which extracts potential behavior pattern features from historical operating trajectories; a prior predictor module, which infers the current behavior pattern based on some historical data; and a decoder module, which verifies the accuracy of the behavior embedding. The behavior embedding vectors can distinguish four different operating modes: stable low load mode, rapid peak shaving mode, unit start-up mode, and unit shutdown mode, and serve as conditional inputs, enabling the dynamic prediction model to adaptively adjust its prediction behavior for different modes.

[0018] The specific method for uncertainty quantification is as follows: use historical data of three different lengths (5 minutes, 10 minutes, and 15 minutes) to predict the state at the same point in time; calculate the variance between the three prediction results as the uncertainty index; when the uncertainty index exceeds the preset threshold, the system automatically reduces the prediction time window length or switches to a conservative control strategy.

[0019] The economic objective is calculated as follows: the power generation revenue is obtained by multiplying the recovered power generation capacity by the power generation duration and the real-time electricity price, the power consumption cost is obtained by subtracting the additional power consumption of the water pump multiplied by the running time and the electricity cost, and the equipment maintenance cost allocation is subtracted to obtain the system net revenue. The decision engine generates control instructions with the optimization objective of maximizing the net revenue per unit time.

[0020] The optimal control command sequence specifically includes: a target speed setpoint sequence for one or more feedwater pumps with an accuracy of ±1 rpm; a target opening setpoint sequence for one or more regulating valves on the recirculation pipeline with an accuracy of ±0.1%; and a start-stop and switching command sequence for the series-connected wide-range hydraulic turbine group. The time step of the optimal control command sequence is 1 minute, and the prediction time window is 30 minutes.

[0021] The digital twin model specifically includes: a thermodynamic calculation module, which calculates the pressure, temperature, and enthalpy changes of the water supply system in real time; a fluid dynamics simulation module, which simulates the flow distribution and pressure loss in the pipeline; an equipment performance simulation module, which simulates the working characteristics of the series wide-range hydraulic turbine, generator, and water pump; and an economic benefit assessment module, which calculates the economic benefits and costs of the system in real time. The synchronization error between the digital twin model and the physical system is less than 5%.

[0022] It also includes an active health management module, which specifically includes: an equipment condition monitoring submodule, which assesses the health status of the equipment by analyzing the vibration spectrum of the series wide-range hydraulic turbine group, the electrical parameters of the generator, and the bearing temperature data; a fault early warning submodule, which predicts the potential failure time based on the equipment degradation trend and issues maintenance warnings 7-30 days in advance; a maintenance plan generation submodule, which automatically generates equipment maintenance work orders and spare parts demand lists; and a performance optimization submodule, which dynamically adjusts control parameters according to the aging of the equipment to maintain optimal performance.

[0023] The beneficial effects of this invention are:

[0024] This invention achieves efficient processing of recirculation flow under different operating conditions through a series-connected wide-range hydraulic turbine assembly. The high-pressure, low-flow hydraulic turbine is specifically designed for stable, low-load conditions with low flow and high pressure differential, while the low-pressure, high-flow hydraulic turbine is specifically designed for rapid load reduction conditions with high flow and low pressure differential. The two turbines can intelligently switch or work collaboratively according to operating requirements, greatly expanding the system's efficient operating range and effectively solving the technical problem of narrow efficiency range inherent in traditional single turbines.

[0025] The digital twin intelligent control core receives real-time operating data of the unit through a prediction engine, learns historical operating trajectories, identifies and encodes behavioral embedding vectors representing macroscopic operating modes, and can distinguish four different operating modes: stable low load mode, rapid peak shaving mode, unit start-up mode, and unit shutdown mode. This enables the dynamic prediction model to adaptively adjust its prediction behavior for different modes, significantly improving prediction accuracy and adaptability.

[0026] The decision engine generates a smooth and stable sequence of optimal control commands through a policy network trained by the optimal transport theory regularization. The regularization training ensures the robustness of the policy by penalizing the norm of the control actions, effectively avoiding the control command oscillation problem that may occur in traditional control methods, and ensuring the stability and reliability of the coordinated control of the feedwater pump speed and the recirculation valve opening.

[0027] The system uses consistency comparison of multiple extrapolation results based on historical data of different lengths to quantify the uncertainty of prediction. It uses historical data of different time lengths to make state predictions at the same time point and calculates the variance between multiple prediction results as an uncertainty index. When the uncertainty index exceeds a preset threshold, the system automatically reduces the prediction time window length or switches to a conservative control strategy, which effectively improves the system's adaptability to complex working conditions.

[0028] The integrated safety bypass module includes an intelligent fault detection unit, an emergency switching execution unit, and a status indication unit. It can monitor the vibration, temperature, speed, and voltage and current parameters of the series wide-range hydraulic turbine unit in real time. When abnormal parameters are detected to exceed the safe range or an emergency command is received, it completes the switching action of closing the isolation valve and opening the bypass valve within milliseconds. This ensures that the original throttling and voltage reduction circuit can be seamlessly switched back in the event of a system failure, without affecting the safe operation of the unit.

[0029] The digital twin model includes a thermodynamic calculation module, a fluid dynamics simulation module, an equipment performance simulation module, and an economic benefit assessment module. It can calculate the pressure, temperature, and enthalpy changes of the water supply system in real time, simulate the flow distribution and pressure loss in the pipeline, simulate the working characteristics of the series wide-range hydraulic turbine, generator, and water pump, and calculate the economic benefits and costs of the system in real time. The synchronization error between the digital twin model and the physical system is controlled within a very small range, providing a reliable virtual mapping basis for the precise control of the system.

[0030] The proactive health management module analyzes the vibration spectrum of the series wide-range hydraulic turbine unit, the electrical parameters of the generator, and the bearing temperature data through the equipment status monitoring submodule to assess the health status of the equipment. The fault early warning submodule predicts the potential failure time based on the equipment degradation trend and issues maintenance warnings in advance. The maintenance plan generation submodule automatically generates equipment maintenance work orders and spare parts demand lists. The performance optimization submodule dynamically adjusts control parameters according to the aging of the equipment to maintain optimal performance, realizing a fundamental shift from planned maintenance to predictive maintenance.

[0031] The system achieves coordinated control of the feedwater pump speed and recirculation valve opening by executing the optimal control command sequence, and feeds back the actual operating status of the system after execution as new real-time data to the prediction engine, forming a triple closed-loop control of data, energy and control. This realizes a technological leap from traditional open-loop energy recovery to intelligent closed-loop coordinated control, and significantly improves the overall performance and operating efficiency of the system.

[0032] The decision engine takes maximizing the difference between the value of recovered electrical energy and the cost of electrical energy consumption of the water pump system as its economic objective. It calculates the power generation revenue by multiplying the recovered power generation capacity by the power generation duration and the real-time electricity price, subtracts the electricity consumption cost by multiplying the additional power consumption of the water pump by the running time and the electricity cost, and then subtracts the equipment maintenance cost allocation to obtain the system net revenue. The control command is generated with the optimization objective of maximizing the net revenue per unit time, thus achieving the economic optimal balance between energy recovery and system operation. Attached Figure Description

[0033] Figure 1 The system architecture of this invention Figure 1 ;

[0034] Figure 2 The system architecture of this invention Figure 2 ;

[0035] Figure 3 The system architecture of this invention Figure 3 . Detailed Implementation

[0036] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0037] It should be noted that the directional concepts of "left", "right", "up", "down", "front", "back", "inner", and "outer" in the following scheme are all relative directions, and will not be listed one by one here.

[0038] Example 1:

[0039] like Figure 1 As shown, this embodiment provides a closed-loop energy recovery system for feedwater pumps based on residual pressure drive in the recirculation branch. Through hardware architecture design, this system achieves efficient recovery and utilization of residual pressure energy in the recirculation branch of the feedwater pump of a thermal power unit.

[0040] In this embodiment, the modular energy recovery assembly is installed on the recirculation pipeline of the feedwater pump, utilizing the hydraulic energy that would otherwise be wasted through the throttling valve. The entire system adopts a modular integrated design concept, which not only facilitates installation and maintenance, but more importantly, ensures stable and reliable operation of the system under various complex operating conditions through its hardware architecture.

[0041] The core component of the modular energy recovery assembly is a series-connected wide-range hydraulic turbine unit. This turbine unit overcomes the technical bottleneck of the narrow efficiency range of traditional single hydraulic turbines, employing an architecture consisting of a high-pressure micro-flow hydraulic turbine and a low-pressure high-flow hydraulic turbine connected in series. The high-pressure micro-flow hydraulic turbine features a blade design with an optimized blade count and precisely controlled blade angle, specifically designed for stable low-load, low-flow, high-pressure differential operating conditions. The turbine's internal flow channel adopts a tapered design with a relatively small inlet cross-section, effectively handling minute flow rates under high-pressure differential conditions, ensuring high energy conversion efficiency even at low flow rates. The turbine rotor is made of high-strength alloy materials, and the bearing system is designed for high pressure resistance, ensuring long-term stable operation under high-pressure differential conditions.

[0042] High-pressure micro-flow hydraulic turbine: suitable for working conditions with pressure difference ≥8MPa and flow rate ≤50t / h.

[0043] Low-pressure, high-flow hydraulic turbine: suitable for operating conditions with pressure differentials of 2-8 MPa and flow rates of 50-500 t / h.

[0044] The energy conversion efficiency calculation of high-pressure micro-flow hydraulic turbines is based on a multi-dimensional efficiency evaluation algorithm:

[0045]

[0046] in, For high-pressure turbine real-time efficiency The rated efficiency is given by the given parameters, and Q is the actual flow rate (m³ / h). H represents the rated flow rate (m³ / h) and H represents the actual head (m). Where n is the rated head (m) and n is the actual rotational speed (rpm). Rated speed (rpm) These are correction factors for flow rate, head, and rotational speed, respectively. Wear correction factor based on runtime:

[0047]

[0048] in, The actual running time (h) The wear strength coefficient is... The wear nonlinear index, Design life (h).

[0049] Low-pressure, high-flow hydraulic turbines employ a completely different design philosophy. Their blade design prioritizes high-flow-rate handling capacity, featuring fewer blades but larger individual blade areas. Optimized blade angles are used to handle high-flow-rate, low-pressure-difference conditions during rapid load reduction. The turbine's flow channel design is a gradually expanding structure with a large inlet cross-section, enabling rapid acceptance of large influxes of recirculated water and preventing back pressure caused by flow surges. The turbine casing is made of cast iron, providing excellent shock resistance, and the rotor dynamic balancing accuracy is strictly controlled to ensure stable operation under high-flow-rate conditions.

[0050] The impact flow handling capacity of a low-pressure, high-flow hydraulic turbine is calculated using a dynamic response algorithm:

[0051] ;

[0052] in, This represents the maximum processing flow rate (m³ / h) for a low-pressure turbine. The rated flow rate (m³ / h) for a low-pressure turbine. Let t be the inlet flow rate (m³ / h) and t be the time (s). This is the flow surge gain coefficient. To switch the trigger flow threshold (m³ / h), The standard deviation of flow rate fluctuation (m³ / h) is given.

[0053] The two hydraulic turbines are interconnected via a series connection mechanism. This mechanism not only achieves a mechanical connection but, more importantly, enables the two turbines to intelligently switch or work collaboratively according to operating conditions through a built-in intelligent switching system. The intelligent switching system includes a flow detection sensor, a differential pressure detection sensor, and an electronic control unit. The flow detection sensor monitors the water flow rate in the recirculation pipeline in real time, the differential pressure detection sensor monitors the pressure difference before and after the turbine, and the electronic control unit determines the current operating conditions based on the detection data and automatically selects the most suitable turbine operating mode.

[0054] The operating mode switching mechanism of the tandem wide-range hydraulic turbine unit is highly intelligent. The intelligent switching algorithm is based on multi-parameter fuzzy decision theory:

[0055] ;

[0056] The pattern membership function is:

[0057] ;

[0058] ;

[0059]

[0060] in, For the selected operating mode, Let i be the membership function of pattern i. Let Q represent the efficiency of mode i, and Q represent the flow rate (m³ / h). The pressure difference is (MPa). The optimal flow rate (m³ / h) for a high-pressure turbine. The optimal flow rate (m³ / h) for a low-pressure turbine. Let $\Delta P_{HP,min}$ be the corresponding standard deviation, and $\Delta P_{HP,min}$ be the minimum differential pressure (MPa) of the high-pressure turbine. This represents the maximum pressure differential (MPa) for a low-pressure turbine. The cumulative function of the standard normal distribution. This is the equipment reliability weighting factor.

[0061] Closely connected to the series-connected wide-range hydraulic turbine unit is the generator and speed-increasing gearbox system, which is responsible for converting the recovered hydraulic energy into electrical energy. The speed-increasing gearbox employs a planetary gear transmission structure, with its input shaft directly connected to the output shaft of the hydraulic turbine. Through gear ratio design, it increases the relatively low speed of the hydraulic turbine to the rated speed of the generator. The gears inside the gearbox undergo carburizing and quenching to ensure transmission accuracy and service life. The bearings are configured with deep groove ball bearings, capable of withstanding certain radial and axial loads. The gearbox housing is equipped with a lubricating oil circulation system, ensuring smooth and reliable gear transmission through forced circulation lubrication.

[0062] The generator adopts a synchronous generator design, with copper wire windings for the stator and a permanent magnet structure for the rotor, avoiding the complexity of the excitation system. The generator casing has a good protection rating, and an internal temperature monitoring device ensures stable operation under various environmental conditions. The generator output is equipped with a voltage regulator and power converter to convert the AC power generated by the generator into standard electrical energy that meets grid requirements, achieving seamless integration with the plant's power system.

[0063] The electromagnetic torque of the generator is calculated using an improved magnetic field coupling algorithm:

[0064] ;

[0065] in, The electromagnetic torque is (N·m). For extreme logarithms, For the dq axis flux linkage (Wb), Let dq be the current (A). For reluctance torque correction factor:

[0066] in, For the dq axis inductance (H), For permanent magnet flux linkage (Wb), A correction function to account for magnetic circuit saturation.

[0067] The integrated safety bypass module cleverly integrates the isolation valve, orifice plate, and existing recirculation piping into a unified safety protection system. The isolation valve features a fast-response design, a stainless steel body, and a spherical valve core. Driven by a pneumatic or electric actuator, it quickly closes upon receiving a switching command, completely isolating the energy recovery assembly from the main pipeline. The orifice plate, serving as a backup pressure-reducing element, has its orifice size precisely calculated to ensure a pressure reduction effect comparable to the original orifice valve during emergency switching, maintaining pressure balance in the water supply system. The existing recirculation piping is retained and optimized in the integrated design; its inner diameter and routing have been redesigned to ensure rapid restoration to its original operating state in emergencies.

[0068] The integrated safety bypass module also includes comprehensive intelligent fault detection capabilities. The intelligent fault detection unit uses multiple sensors to monitor key operating parameters of the series-connected wide-range hydraulic turbine assembly in real time. Vibration sensors, mounted on the turbine bearing housing, monitor the vibration frequency and amplitude during turbine operation, identifying fault signs such as imbalance, bearing wear, or blade damage through spectral analysis. Temperature sensors are distributed across the turbine bearings, windings, and key components to monitor temperature changes and promptly detect overheating. The speed sensor employs a non-contact design to monitor the actual turbine speed, ensuring it remains within a safe range. For the generator system, voltage and current sensors monitor the generator's output voltage amplitude, frequency, and current magnitude in real time, using parameter analysis to determine if the generator is operating normally.

[0069] The fault detection algorithm employs a comprehensive diagnostic method that fuses information from multiple sensors.

[0070] ;

[0071] in Let N be the probability of failure, and N be the total number of sensors. For the first The weight of each sensor, For sensor signals, This represents the normal value for sensor k. Let k be the standard deviation of the sensor signal. The activation function is sigmoid. The dynamic weight adjustment algorithm is as follows:

[0072]

[0073] in, for The weight of time sensor k, Let K be the weight of sensor k at time t. To diagnose the loss function for weights gradient, For learning rate, For diagnostic loss function, This is the regularization parameter.

[0074] The emergency switching execution unit is the core of the safety bypass module. When the intelligent fault detection unit detects any abnormal parameters exceeding the preset safety range, or when the system receives an emergency shutdown command from the main control system, the emergency switching execution unit immediately initiates the preset safety program. The entire switching process is controlled within ten milliseconds. First, the isolation valve entering the energy recovery assembly is closed, cutting off the recirculated water flow into the turbine assembly. Simultaneously, the bypass valve is opened, directing the recirculated water flow into the backup throttling and pressure-reducing circuit.

[0075] The action sequence optimization algorithm for emergency switching is as follows:

[0076] ;

[0077] in, The optimal switching time (s) is... To switch the start time (s), Let be the integral variable (s). For pressure deviation, For the switch completion time (s), The time penalty factor is used. The valve opening coordination control algorithm is as follows:

[0078] ;

[0079] in, The isolation valve opening degree (%) Bypass valve opening degree (%) Maximum opening (%) The maximum opening degree (%) of the bypass valve. Let t be the time when the fault occurred (s), and t be the current time (s). and These are the time constants (s) for turning off and turning on, respectively.

[0080] The status indicator unit provides operators with intuitive system status information, including a multi-color LED indicator array and an LCD screen. The indicator array displays the system's basic operating status through different colors and flashing patterns: solid green indicates normal system operation, flashing yellow indicates the system is in a switching state, and solid red indicates a system fault requiring manual intervention. The LCD screen displays detailed operating parameters in real time, including turbine speed, power generation, system efficiency, and sensor values, while also showing the current operating mode and equipment health status assessment results.

[0081] The entire modular energy recovery assembly connects to the existing feedwater pump recirculation pipeline via standardized flange connections, requiring no major modifications to the existing system during installation. The assembly housing features a modular design, allowing each major component to be independently disassembled and maintained, significantly reducing maintenance difficulty and costs. The housing is made of corrosion-resistant stainless steel or corrosion-coated carbon steel, ensuring long-term stable operation in the high-temperature and high-humidity environment of the power plant.

[0082] When the thermal power unit is operating normally, the feedwater pump pressurizes the feedwater in the deaerator and sends it to the boiler. Simultaneously, to prevent cavitation in the feedwater pump under low load, a portion of the feedwater returns to the deaerator via a recirculation pipeline. In conventional solutions, the pressure energy of this recirculated water is directly converted into heat energy and lost through a throttling valve. In this embodiment, the recirculated water first enters the modular energy recovery assembly. Based on the current flow rate, the intelligent switching system automatically selects a suitable hydraulic turbine for energy recovery.

[0083] When the unit is in a stable operating state with minimal load changes, the recirculation flow rate is relatively small. In this condition, the high-pressure micro-flow hydraulic turbine undertakes the primary energy recovery task. Utilizing its optimized blade design and flow channel structure, the high-pressure micro-flow turbine efficiently converts the pressure energy of the water flow into mechanical rotational energy, driving a connected generator to produce electricity through a speed-increasing gearbox. Because the turbine is specifically designed for low-flow, high-pressure differential operating conditions, it can maintain a high energy conversion efficiency under these conditions.

[0084] When the power grid dispatch requires the generating units to rapidly reduce load, or when the generating units need to significantly reduce output for other reasons, the demand for feedwater decreases sharply, leading to a substantial increase in recirculation flow. The intelligent switching system detects the surge in flow through flow sensors and immediately activates the low-pressure, high-flow hydraulic turbine or switches it to the main turbine. The low-pressure, high-flow turbine, with its high flow handling capacity and optimized blade design, can effectively handle the sudden increase in recirculation water flow, ensuring that system efficiency does not decrease or equipment damage due to excessive flow. During this process, two turbines may operate simultaneously; through intelligent flow allocation, the entire turbine group always operates within its optimal efficiency range.

[0085] Throughout the entire operation, the integrated safety bypass module continuously monitors the system's safety status. Various sensors collect key data in real time, such as turbine vibration, temperature, speed, and generator electrical parameters, and the intelligent fault detection unit analyzes this data in real time. Once any abnormality is detected, such as excessive turbine vibration, overheated bearing temperature, abnormal speed, or abnormal generator voltage or current, the emergency switching execution unit responds immediately, completing a safe switch within ten milliseconds to ensure the safe and reliable operation of the feedwater pump system.

[0086] Through the design of a series-connected wide-range hydraulic turbine unit, the system can maintain high-efficiency energy recovery performance over an extremely wide operating range. Compared with a single turbine design, the efficiency improvement is significant, and the energy recovery rate is greatly enhanced. Secondly, the modular integrated design makes the system easy to install and maintain, requires minimal modification to existing equipment, and has relatively low implementation costs. Thirdly, the integrated safety bypass module design ensures high system safety and reliability. Even in the event of a failure in the energy recovery system, it can quickly switch to standby mode without affecting the normal operation of the main feedwater system, eliminating the concerns of power plant operators regarding the reliability of new technologies.

[0087] Furthermore, the application of the intelligent switching mechanism enables the system to automatically adjust its operating mode according to actual working conditions without manual intervention, thus improving the system's automation level and operational stability. The intuitive status display provided by the status indicator unit facilitates monitoring and maintenance by operators, reducing the complexity of operation management. The highly integrated design of the entire hardware system also brings practical advantages such as small footprint, short installation cycle, and minimal commissioning workload, making it particularly suitable for the actual application environment of power plants.

[0088] This embodiment converts previously wasted surplus pressure energy into usable electrical energy, directly bringing economic benefits to the power plant while reducing plant power consumption and improving the overall economic efficiency of the unit. From an environmental perspective, the system improves energy utilization efficiency, reduces energy waste, and meets the development requirements of energy conservation and emission reduction.

[0089] Example 2:

[0090] like Figure 1 and 2 As shown, this embodiment, based on the modular energy recovery assembly hardware of Embodiment 1, further provides a specific implementation scheme for the digital twin intelligent control core, focusing on the predictive control technology based on artificial intelligence algorithms. This embodiment upgrades traditional passive energy recovery to active intelligent optimization control by constructing a predictive decision dual-engine architecture, maximizing system economic benefits and significantly improving operational reliability.

[0091] The core technology of this embodiment lies in the design and implementation of a digital twin intelligent control core. This control core adopts a multi-layered modeling architecture, with the bottom layer being a device-level precise simulation model, the middle layer being a system-level integrated simulation model, and the top layer being a process-level global optimization model. Millisecond-level synchronization with the physical system is achieved through a high-speed data interface, ensuring consistency between the virtual and real worlds.

[0092] The core of digital twin intelligent control includes two collaborative core functional engines: a prediction engine and a decision engine. These two engines use advanced artificial intelligence algorithms to achieve intelligent system prediction and optimized control.

[0093] The prediction engine is configured to receive real-time operating data from the generating units, and a multi-source heterogeneous data fusion processing platform has been built. This real-time operating data includes multi-dimensional information such as unit load, boiler feedwater flow rate, feedwater pump operating flow rate, feedwater pump outlet pressure, feedwater pump speed, recirculation pipeline valve opening, deaerator water level, feedwater temperature, grid dispatch instructions, ambient temperature, and real-time electricity market price information. The data acquisition system adopts a layered distributed architecture, with sensor nodes connected to the upper-level data processing center via various communication methods such as industrial Ethernet and wireless networks, ensuring the real-time performance and reliability of data transmission.

[0094] The core algorithm of the prediction engine is based on deep learning and time series analysis techniques. It learns from historical operating trajectories to identify and encode behavioral embedding vectors that represent macroscopic operating patterns. The generation process of behavioral embedding vectors is a key technology of the prediction engine, which includes three core components: an encoder module, a prior predictor module, and a decoder module.

[0095] The encoder module employs a variational autoencoder architecture to extract latent behavioral pattern features from historical running trajectories. It maps high-dimensional running data sequences to a low-dimensional latent space using a deep neural network. The encoder utilizes a multi-scale feature extraction strategy, adaptively focusing on important information at different time scales through an attention mechanism.

[0096] The generation of behavior embedding vectors employs a three-module collaborative coding algorithm:

[0097] The variational coding loss function of the encoder module is:

[0098]

[0099] in Let be the encoder loss function. For encoder parameters, For input data, For data distribution, For the encoder posterior distribution, As latent variables, As a prior distribution, For balance coefficient, Let KL divergence be a metric.

[0100] The prior predictor module is based on a hybrid architecture of recurrent neural networks and Transformers, possessing powerful sequence modeling capabilities. This module employs Bayesian inference methods, not only outputting predictions of behavioral patterns but also quantifying the uncertainty of the predictions.

[0101] The multimodal behavior recognition algorithm of the prior predictor is as follows:

[0102] ;

[0103] in For the output distribution of the prior predictor, These are the parameters of the prior predictor. Let H be the historical trajectory of length h, and K be the number of mixed components. The weight of the k-th component. It follows a normal distribution. Let be the mean of the k-th component. The covariance matrix of the k-th component

[0104] The decoder module is used to verify the accuracy of the behavior embedding. It employs a generative adversarial network training strategy to ensure that the generated trajectories are highly realistic and diverse.

[0105] The behavioral embedding vectors can distinguish four different operating modes: stable low-load mode, rapid peak-shaving mode, unit start-up mode, and unit shutdown mode. These embedding vectors serve as conditional inputs, enabling the dynamic prediction model to adaptively adjust its predictive behavior for different modes. The prediction model employs a conditional Transformer architecture, inputting the behavioral embedding vectors as conditional information into the model and capturing long-term dependencies from historical data through a self-attention mechanism.

[0106] The prediction engine also quantifies the uncertainty of predictions by comparing the consistency of multiple extrapolation results based on historical data of different lengths. The specific method for uncertainty quantification is as follows: use historical data of three different lengths (5 minutes, 10 minutes, and 15 minutes) to make state predictions at the same time point; calculate the variance between the three prediction results as an uncertainty index; when the uncertainty index exceeds a preset threshold, the system automatically reduces the prediction time window length or switches to a conservative control strategy.

[0107] The uncertainty quantization algorithm employs a multi-time window ensemble method:

[0108] ;

[0109] in for The uncertainty in predicting a given moment, where w is the length of the historical data window (in minutes). These are the state prediction values ​​based on the window length w. The variance function

[0110] The decision engine is configured to receive the future state sequence output by the prediction engine. When the prediction uncertainty is below a preset threshold, the decision engine begins to perform intelligent optimization calculations. The decision engine aims to maximize the difference between the value of recovered electrical energy and the cost of electrical energy consumption of the water pump system. It generates a smooth and stable optimal control command sequence through a policy network trained with optimal transmission theory regularization.

[0111] The economic objective is calculated as follows: the power generation revenue is obtained by multiplying the recovered power generation capacity by the power generation duration and the real-time electricity price, the power consumption cost is obtained by subtracting the additional power consumption of the water pump multiplied by the running time and the electricity cost, and the equipment maintenance cost allocation is subtracted to obtain the system net revenue. The decision engine generates control instructions with the optimization objective of maximizing the net revenue per unit time.

[0112] The formula for calculating the economic objective function is:

[0113] ;

[0114] in Let the economic objective function be... Here, H represents the control input at time t, and H represents the prediction time domain length. As a discount factor, for Electricity revenue at any given time (in yuan), for The operating cost of the water pump at any given time (in yuan). for Maintenance cost per moment (RMB);

[0115] The decision engine employs a deep reinforcement learning algorithm based on optimal transfer theory. This algorithm introduces an optimal transfer regularization term on the basis of the traditional deep deterministic policy gradient algorithm, which ensures the robustness of the policy by penalizing the norm of the control action.

[0116] The optimization objective of the optimal transmission regularization strategy is:

[0117]

[0118] in Let be the policy loss function. For policy network parameters, For a policy network, 'a' represents the action and 's' represents the state. For the dominant function, The optimal transmission regularization coefficient is... For optimal transmission regularization term

[0119] The optimal control command sequence specifically includes: a target speed setpoint sequence for one or more feedwater pumps with an accuracy of ±1 rpm; a target opening setpoint sequence for one or more regulating valves on the recirculation pipeline with an accuracy of ±0.1%; and a start-up, shutdown, and switching command sequence for a series-connected wide-range hydraulic turbine unit. The time step of the optimal control command sequence is 1 minute, and the prediction time window is 30 minutes.

[0120] The algorithm for generating control instruction sequences employs a hierarchical optimization strategy:

[0121]

[0122] in Let be the optimal pump speed (rpm) at time t, and n be the pump speed (rpm). Reference speed (rpm) The change in rotational speed (rpm) These are the weighting coefficients;

[0123] The feedwater pump speed control command is determined through comprehensive analysis of the feedwater system's flow demand, pressure requirements, and the optimal operating point of the energy recovery system. Speed ​​control employs a composite control strategy of feedforward and feedback. Feedforward control adjusts the speed in advance based on predicted load changes, while feedback control performs fine-tuning based on actual operating deviations.

[0124] Valve opening control commands optimize the recirculation flow distribution to achieve the optimal operating state of the hydraulic turbine unit. The control algorithm, based on the turbine characteristic curve and flow-pressure relationship, calculates the optimal flow distribution scheme that maximizes turbine efficiency and implements this scheme by adjusting the corresponding valve openings.

[0125] Turbine start-up, shutdown, and switching commands pre-plan turbine operating modes based on predicted load change trends and recirculation flow variations. The control algorithm analyzes future operating conditions to predict the timing and method of turbine switching, enabling seamless switching and coordinated operation between different turbines.

[0126] The system achieves coordinated control of the feedwater pump speed and recirculation valve opening by executing an optimal control command sequence. The executed system operating status is then fed back to the predictive engine as new real-time data, forming a triple closed-loop control system consisting of a data loop, an energy loop, and a control loop. The data loop ensures the control system can perceive changes in system state in real time; the energy loop enables efficient energy recovery and recycling; and the control loop continuously optimizes system performance through a continuous prediction-decision-execution-feedback loop.

[0127] During system startup, the prediction engine learns from historical data to build a behavioral pattern library and identifies the characteristic patterns of different operating modes. Once the system is running normally, the prediction engine continuously receives real-time operating data, uses behavior recognition algorithms to determine the current operating mode, and generates corresponding behavior embedding vectors.

[0128] When prediction uncertainty is low, the decision engine receives the prediction results and generates the optimal control command sequence based on the economic objective function and various constraints using a deep reinforcement learning algorithm. When a change in operating mode is detected, the prediction engine updates the behavior embedding vector in a timely manner and adjusts the prediction strategy. When prediction uncertainty is high, the system automatically switches to a conservative control mode, prioritizing system stability and security.

[0129] By applying behavioral embedding technology, the system gains the ability to deeply understand operating modes, achieving a leap from traditional mathematical model-based control to data-driven intelligent control. Secondly, multi-time-window uncertainty quantification technology provides the system with powerful risk assessment capabilities, significantly improving its robustness.

[0130] Furthermore, the control optimization algorithm based on optimal transmission theory ensures the smoothness and continuity of control commands, avoiding control oscillations and equipment shocks that may occur with traditional optimization methods. In addition, the economic goal-oriented control strategy enables the system to dynamically adjust the control strategy according to economic factors such as real-time electricity prices and operating costs, maximizing economic benefits while ensuring technical performance.

[0131] This embodiment integrates cutting-edge artificial intelligence technologies such as deep learning, reinforcement learning, and optimal transport theory into the field of industrial control. The algorithm adopts a standardized software architecture, possessing strong engineering applicability and scalability. The intelligent control system significantly reduces reliance on the professional skills of operators and improves the standardization of operation management.

[0132] Example 3:

[0133] like Figures 1 to 3 As shown, this embodiment, based on the modular energy recovery assembly of Embodiment 1 and the intelligent control algorithm of Embodiment 2, further provides a deep implementation scheme for digital twin modeling technology and a complete functional architecture for an active health management system. It focuses on the deep integration of virtual mapping technology with the physical system and the engineering application of predictive maintenance technology. By constructing a high-fidelity digital twin model and realizing a full lifecycle health management system, this embodiment upgrades the traditional periodic maintenance mode to a state-based predictive maintenance mode, achieving a significant improvement in system reliability and a substantial reduction in operation and maintenance costs.

[0134] The core technology of this embodiment lies in the deep modeling technology of the digital twin model. Building upon the prediction and decision-making functions described in Embodiment 2, this digital twin model achieves significant technological breakthroughs in areas such as accurate mapping of physical systems, multi-physics coupled simulation, and real-time economic benefit assessment. The digital twin model adopts a hierarchical and progressive modeling architecture, abstracting and integrating from basic physical phenomena to complex system behaviors layer by layer, ensuring that the model possesses both microscopic accuracy and macroscopic consistency.

[0135] The digital twin model comprises four interconnected core modules, each undertaking a specific simulation calculation task. Through data exchange and coupled computation between the modules, a comprehensive and accurate model of the physical system is achieved. These four modules are the thermodynamics calculation module, the fluid dynamics simulation module, the equipment performance simulation module, and the economic benefit assessment module. These modules form an organic, integrated simulation environment through standardized data interfaces and a unified time synchronization mechanism.

[0136] The thermodynamics calculation module is responsible for the precise calculation of the thermodynamic processes of the water supply system, calculating the changes in pressure, temperature, and enthalpy in the system in real time, providing fundamental thermodynamic state information for the entire digital twin model. Based on classical thermodynamics theory and modern computational fluid dynamics methods, this module constructs a thermodynamic calculation system covering all thermal equipment, including water pumps, recirculation pipelines, hydraulic turbines, and heat exchangers.

[0137] The multiphysics coupling algorithm in the thermodynamic calculation module employs a spatiotemporal decomposition method:

[0138] The transient thermodynamic calculation based on the modified Reynolds transport equation is as follows:

[0139] ;

[0140] in Let be the density (kg / m³), h be the specific enthalpy (J / kg), and t be the time (s). Let k be the velocity vector (m / s), k be the thermal conductivity (W / (m·K)), and T be the temperature (K). This is the viscous dissipation term (W / m³). For enthalpy source term (W / m³);

[0141] The overall heat transfer coefficient calculation employs a multi-mode coupled algorithm:

[0142] ;

[0143] in The overall heat transfer coefficient (W / (m²·K)) The convective heat transfer coefficient (W / (m²·K)) The radiative heat transfer coefficient (W / (m²·K)) The thermal conductivity of the wall (W / (m·K)) The wall thickness is in meters (m). Contact correction factor

[0144] The fluid dynamics simulation module is specifically designed for accurate modeling of fluid flow behavior in pipeline systems, simulating flow distribution and pressure loss within the pipeline, and providing detailed flow field information for the flow analysis of the entire system. Based on computational fluid dynamics theory and numerical methods, this module constructs a three-dimensional flow field simulation model covering all flow regions, including the main feedwater pipeline, recirculation branches, turbine internal channels, and valve channels.

[0145] The fluid dynamics simulation module employs a high-precision numerical solution method with adaptive mesh refinement.

[0146] The turbulence correction to the Reynolds-averaged Navier-Stokes equations is as follows:

[0147] in Let be the fluid density (kg / m³), and t be the time (s). Spatial coordinates (m), For velocity components (m / s);

[0148] The multi-condition adaptive calculation method for pipeline resistance is as follows:

[0149] in Total pressure loss (Pa), Let be the adaptive resistance coefficient of the i-th pipeline segment. Where is the fluid density (kg / m³), and v is the flow velocity (m / s);

[0150] The equipment performance simulation module is specifically responsible for the accurate modeling of the operating characteristics of each key piece of equipment, simulating the working characteristics of series wide-range hydraulic turbine units, generators, and water pumps, providing accurate equipment performance predictions for system optimization control.

[0151] The multidimensional similarity criterion prediction algorithm for turbine performance is as follows: ;

[0152] in For turbine efficiency, For flow coefficient, For speed coefficient, This is the head coefficient. It is the Reynolds number;

[0153] The dimensionless parameters include key parameters such as flow coefficient, rotational speed coefficient, head coefficient, and Reynolds number. The equipment performance simulation also considers the long-term effects of factors such as wear, deposition, and cavitation on performance.

[0154] The multi-mechanism coupled degradation model of equipment aging is as follows:

[0155] ;

[0156] in Let be the degradation efficiency at time t. Let t be the initial efficiency, t be the running time (h), and m be the degradation mechanism number. The characteristic time (h) for the m-th mechanism is... For the load of the m-th mechanism, Let be the critical load for the m-th mechanism. is the index of the m-th mechanism;

[0157] The economic benefit assessment module is specifically responsible for the real-time calculation and dynamic evaluation of the system's economic performance, calculating the system's economic benefits and costs in real time, and providing a basis for decision-making for economic optimization control.

[0158] The dynamic return-risk assessment algorithm for the economic benefit assessment module is as follows: ;

[0159] in Where is the net present value (in yuan), and T is the project period (in years). Let r be the cash flow (in yuan) in year t, and r be the discount rate. Initial investment (RMB);

[0160] The digital twin model achieves comprehensive and accurate modeling of the physical system through the collaborative computation of four core modules, with a synchronization error of less than 5% between the digital twin model and the physical system. To achieve real-time synchronization with the physical system, the digital twin model establishes a high-speed data interface and a real-time communication mechanism, and adopts multi-threaded parallel computing technology to ensure the real-time requirements of complex simulation calculations.

[0161] Another key aspect of this embodiment lies in the design and implementation of the proactive health management module. This module achieves comprehensive perception and proactive management of equipment health status through advanced condition monitoring technology, intelligent diagnostic algorithms, and predictive maintenance strategies. The proactive health management module comprises four functional sub-modules: equipment condition monitoring, fault early warning, maintenance plan generation, and performance optimization. These sub-modules form a complete equipment health management ecosystem through standardized data interfaces and a unified health management platform.

[0162] The equipment condition monitoring submodule is responsible for comprehensively sensing the operating status of the equipment. It assesses the health status of the equipment by analyzing the vibration spectrum of the series wide-range hydraulic turbine unit, the electrical parameters of the generator, and bearing temperature data. This submodule establishes a multi-parameter, multi-level condition monitoring system, which collects key status information of the equipment in real time through a sensor network and extracts characteristic indicators of the equipment's health status through advanced signal processing and data analysis techniques.

[0163] For vibration monitoring of tandem wide-range hydraulic turbine units, the equipment condition monitoring submodule deploys a multi-channel vibration sensor network, arranging high-precision accelerometers and displacement sensors in multiple directions, including radial, axial, and tangential directions, of the turbine. The submodule uses signal processing techniques such as Fast Fourier Transform, wavelet analysis, and Empirical Mode Decomposition to extract time-domain, frequency-domain, and time-frequency features from the vibration signals.

[0164] The multi-scale wavelet packet decomposition algorithm for vibration signals is as follows: ;

[0165] in For the k-th wavelet coefficient of the n-th node in the j-th layer, These are the wavelet filter coefficients. These are the wavelet coefficients of the previous layer;

[0166] For monitoring the electrical parameters of the generator, the equipment condition monitoring submodule has established a comprehensive monitoring system for multiple electrical quantities, which monitors in real time the generator's basic electrical parameters such as voltage, current, power, power factor, and frequency, as well as insulation condition parameters such as insulation resistance, partial discharge, and dielectric loss.

[0167] For bearing temperature monitoring, the equipment condition monitoring submodule installs high-precision temperature sensors at various key bearing locations to monitor the bearing's operating temperature in real time. The submodule establishes a baseline model for bearing temperature, considering the influence of factors such as ambient temperature, load level, and lubricating oil temperature on the bearing temperature.

[0168] The multi-sensor information fusion algorithm of the equipment condition monitoring submodule is as follows:

[0169] ;

[0170] in Let A be the evidence function after fusion, and let A be a proposition. For the i-th sensor, Let n be the number of sensors, and assign a basic probability to the i-th sensor. It is an empty set;

[0171] Based on the results of equipment status monitoring, the fault early warning submodule uses advanced fault prediction algorithms and degradation modeling technology to predict the potential failure time based on the equipment degradation trend and issue maintenance early warnings seven to thirty days in advance.

[0172] The degradation trend prediction algorithm of the fault early warning submodule is as follows:

[0173] Nonlinear state-space model based on particle filtering:

[0174]

[0175]

[0176] in Let k be the state vector at time k. This is the state transition function. for Time-based control input, For process noise, Let k be the observation value at time k. For the observation function, To observe noise

[0177] The probability density function for remaining life prediction:

[0178] ;

[0179] in Based on observation sequences The remaining lifetime probability density, The remaining lifetime is given in hours (h), and N is the number of particles. Let be the weight of the i-th particle. This represents the state of the i-th particle;

[0180] The fault early warning submodule establishes a multi-level early warning mechanism, setting different warning levels based on the urgency and impact of the fault. Level 1 warning corresponds to an impending serious fault, requiring immediate shutdown for maintenance; Level 2 warning corresponds to a moderate fault that may affect equipment performance; and Level 3 warning corresponds to early signs of fault, requiring enhanced monitoring.

[0181] The maintenance plan generation submodule automatically generates equipment maintenance work orders and spare parts requirement lists based on the prediction results of the fault early warning submodule and the equipment operation plan.

[0182] The multi-objective decision algorithm for maintenance plan optimization is as follows:

[0183] ;

[0184] Where x is the maintenance decision variable. There are three objective functions. Maintenance costs (in yuan), The downtime is in hours. As a risk indicator;

[0185] Maintenance work orders are generated based on standardized maintenance operation procedures and process specifications. The submodule establishes a standardized template library for maintenance operations, containing maintenance operation instructions for different equipment and different fault types. Spare parts requirement lists are generated based on equipment failure mode analysis and historical spare parts consumption data; the submodule establishes a spare parts demand prediction model.

[0186] The performance optimization submodule is specifically responsible for the continuous improvement and optimization control of equipment performance, and dynamically adjusts control parameters according to the aging of the equipment to maintain optimal performance.

[0187] The adaptive parameter tuning algorithm of the performance optimization submodule is as follows:

[0188] ;

[0189] in For the control sequence, N is the prediction time domain length. For time k, pair The prediction output at any given time. for Reference value at any time, for Time-based control input, For terminal status prediction, Weight matrix

[0190] Feedforward control for equipment performance degradation compensation:

[0191] ;

[0192] in Let be the compensation control quantity at time t. For proportional gain, For integral gain, For differential gain, Let be the efficiency deviation at time t. Let be the integral variable (s);

[0193] The performance optimization submodule also has an operation strategy optimization function, which dynamically adjusts the system's operation strategy based on the current health status and performance level of the devices. When the performance of a device degrades, the submodule will correspondingly increase the load sharing of other devices or adjust the overall operation mode.

[0194] The proactive health management module achieves closed-loop control of equipment health management through the collaborative work of four functional sub-modules, forming a complete equipment health management chain from status monitoring to fault early warning, and from maintenance planning to performance optimization.

[0195] During the system initialization phase, the digital twin model establishes an initial virtual model using CAD drawings, design parameters, and experimental data. Then, through data docking with the physical system, the model is checked and verified.

[0196] During normal operation, the digital twin model continuously receives real-time data from the physical system and updates the virtual system's status in real time through the collaborative computation of the four core modules. The proactive health management module works continuously during system operation, and the equipment status monitoring submodule collects and analyzes equipment status data in real time. Once abnormal signs are detected, a fault early warning process is immediately triggered.

[0197] When the system faces changes in operating conditions or abnormal events, the digital twin model can respond quickly and provide decision support. The proactive health management module enhances the frequency and accuracy of condition monitoring when abnormal events occur, promptly detecting and addressing potential equipment failures.

[0198] High-fidelity digital twin models provide unprecedented transparency and predictability for system operation. Operators can gain a deeper understanding of the system's internal working status through virtual models, greatly improving the scientific nature and precision of operation control.

[0199] Secondly, the proactive health management system realizes a fundamental shift from traditional planned maintenance to predictive maintenance. By identifying early signs of failure and predicting remaining lifespan, the system can formulate targeted maintenance plans before equipment failure occurs, avoiding production losses and safety risks caused by sudden failures.

[0200] Furthermore, the collaborative computing of the four digital twin modules provides powerful simulation analysis capabilities for system optimization. Operators can test various operating strategies and control schemes in a virtual environment and evaluate the effectiveness and risks of different schemes.

[0201] This embodiment successfully advances digital twin technology from proof of concept to engineering application, achieving real-time synchronization and deep integration between the virtual and physical worlds. The application of multiphysics coupling simulation technology enables the digital twin model to possess simulation accuracy and applicability.

[0202] Digital twin technology enables remote monitoring and unattended operation; proactive health management technology transforms maintenance work from passive response to proactive planning; and intelligent optimization technology enables the system to autonomously adapt to changes in operating conditions.

[0203] The reduction in equipment failure rate decreases unplanned downtime losses; the implementation of predictive maintenance reduces maintenance and spare parts costs; and the application of performance optimization techniques improves system efficiency and energy utilization.

[0204] The application of predictive maintenance technology extends equipment lifespan and reduces the environmental impact of abandoned equipment; the application of performance optimization technology improves energy efficiency and reduces carbon emissions; the application of digital operation and maintenance technology reduces the frequency of on-site operations, thereby reducing personnel safety risks and environmental impact.

[0205] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A closed-loop energy recovery system for a feedwater pump driven by residual pressure in a recirculation branch, characterized in that, include: A modular energy recovery assembly is installed on the recirculation pipeline of the feedwater pump. The energy recovery assembly includes a series wide-range hydraulic turbine group, which is composed of a high-pressure micro-flow hydraulic turbine and a low-pressure large-flow hydraulic turbine connected in series. The high-pressure micro-flow hydraulic turbine is used to handle the small flow and high pressure difference conditions under stable low load, and the low-pressure large-flow hydraulic turbine is used to handle the large flow and low pressure difference conditions when the load drops rapidly. The two turbines can intelligently switch or work together according to the operating conditions. The generator and speed-increasing gearbox connected to the series-connected wide-range hydraulic turbine are used to convert the recovered hydraulic energy into electrical energy; The integrated safety bypass module, which includes an isolation valve, a throttling orifice plate, and the original recirculation pipeline, can automatically isolate and recover the assembly and switch back to the original throttling and pressure-reducing circuit within milliseconds in the event of a system failure. The digital twin intelligent control core is based on a digital twin model. This model constructs a virtual model that is completely consistent with reality based on the piping layout and flowchart of the physical system. It maps the physical system state in real time and predicts changes in operating conditions, and includes the following functional engines: The prediction engine is configured to receive real-time operating data of the receiving unit, identify and encode behavioral embedding vectors that represent macroscopic operating patterns by learning historical operating trajectories, adaptively adjust the prediction of future states based on the behavioral embedding vectors as independent inputs, and output a sequence of future states. The uncertainty of the prediction is quantified by comparing the consistency of multiple extrapolation results based on historical data of different lengths. The decision engine is configured to receive the future state sequence output by the prediction engine. When the prediction uncertainty is lower than a preset threshold, it takes maximizing the difference between the value of recovered electrical energy and the cost of electrical energy consumption of the water pump system as the economic objective. It generates a smooth and stable optimal control instruction sequence through a policy network trained by the optimal transmission theory regularization. The regularization training ensures the robustness of the policy by penalizing the norm of the control action. The system achieves coordinated control of the feedwater pump speed and recirculation valve opening by executing the optimal control command sequence, and feeds back the actual operating status of the system after execution as new real-time data to the prediction engine, forming a triple closed-loop control of data, energy and control.

2. The system according to claim 1, characterized in that, The working mode switching mechanism of the series wide-range hydraulic turbine group is as follows: when the system is operating at a stable low load and the recirculation flow rate is less than the first threshold, only the high-pressure micro-flow hydraulic turbine is activated for energy recovery; when the system load drops rapidly and the recirculation flow rate exceeds the second threshold, the low-pressure high-flow hydraulic turbine is automatically integrated or switched to become the main force for energy recovery. When the recirculation flow rate is between the first and second thresholds, the two turbines work together to achieve optimal energy recovery efficiency.

3. The system according to claim 1, characterized in that, The integrated safety bypass module also includes: an intelligent fault detection unit that monitors the vibration, temperature, speed of the series wide-range hydraulic turbine group and the voltage and current parameters of the generator in real time; an emergency switching execution unit that, when an abnormal parameter is detected to exceed the safe range or an emergency command is received, completes the switching action of closing the isolation valve and opening the bypass valve within 10 milliseconds; and a status indication unit that displays the current operating mode and equipment health status in real time.

4. The system according to claim 1, characterized in that, The real-time operating data received by the prediction engine includes: unit load, boiler feedwater flow rate, feedwater pump operating flow rate, feedwater pump outlet pressure, feedwater pump speed, recirculation pipeline valve opening, deaerator water level, feedwater temperature, power grid dispatch instructions, ambient temperature, and real-time electricity market price information.

5. The system according to claim 1, characterized in that, The generation process of the behavior embedding vector includes: an encoder module for extracting potential behavior pattern features from historical operating trajectories; a prior predictor module for inferring the current behavior pattern based on some historical data; and a decoder module for verifying the accuracy of the behavior embedding. The behavior embedding vector can distinguish four different operating modes: stable low load mode, rapid peak shaving mode, unit start-up mode, and unit shutdown mode, and serves as a conditional input, enabling the dynamic prediction model to adaptively adjust its prediction behavior for different modes.

6. The system according to claim 1, characterized in that, The specific method for quantifying uncertainty is as follows: using historical data of three different lengths (5 minutes, 10 minutes, and 15 minutes) to predict the state at the same point in time; calculating the variance between the three prediction results as an uncertainty index; when the uncertainty index exceeds a preset threshold, the system automatically reduces the prediction time window length or switches to a conservative control strategy.

7. The system according to claim 1, characterized in that, The economic target is calculated as follows: the power generation revenue is obtained by multiplying the recovered power generation capacity by the power generation duration and the real-time electricity price, the power consumption cost is obtained by subtracting the additional power consumption of the water pump multiplied by the running time and the electricity cost, and the equipment maintenance cost is subtracted to obtain the system net revenue. The decision engine generates control commands with the optimization objective of maximizing the net revenue per unit time.

8. The system according to claim 1, characterized in that, The optimal control command sequence specifically includes: a target speed setpoint sequence for one or more feedwater pumps with an accuracy of ±1 rpm; a target opening setpoint sequence for one or more regulating valves on the recirculation pipeline with an accuracy of ±0.1%; and a start-stop and switching command sequence for the series-connected wide-range hydraulic turbine group. The time step of the optimal control command sequence is 1 minute, and the prediction time window is 30 minutes.

9. The system according to claim 1, characterized in that, The digital twin model specifically includes: a thermodynamic calculation module for real-time calculation of pressure, temperature, and enthalpy changes in the water supply system; a fluid dynamics simulation module for simulating flow distribution and pressure loss in the pipeline; an equipment performance simulation module for simulating the working characteristics of a series wide-range hydraulic turbine unit, generator, and water pump; and an economic benefit evaluation module for real-time calculation of the system's economic benefits and costs. The synchronization error between the digital twin model and the physical system is less than 5%.

10. The system according to claim 1, characterized in that, It also includes an active health management module, which specifically includes: an equipment condition monitoring submodule, which assesses the health status of the equipment by analyzing the vibration spectrum of the series wide-range hydraulic turbine group, the electrical parameters of the generator, and the bearing temperature data; a fault early warning submodule, which predicts the potential failure time based on the equipment degradation trend and issues maintenance warnings 7-30 days in advance; a maintenance plan generation submodule, which automatically generates equipment maintenance work orders and spare parts demand lists; and a performance optimization submodule, which dynamically adjusts control parameters according to the aging of the equipment to maintain optimal performance.

Images