Control method, system and device for a cogeneration system

Abstract

The application relates to a control method, system and device of a waste heat power generation system. The method comprises the following steps: acquiring running state data of the waste heat power generation system at a current time, inputting the running state data into a pre-constructed time sequence prediction model, and outputting prediction trend information containing a predicted value of power generation at a future time; constructing a state space for an intelligent decision model, wherein the state space comprises current control deviation information, deviation change rate information and a predicted value at the future time in the prediction trend information, and the predicted value at the future time is used as a forward-looking feature; inputting the state space into the intelligent decision model, and outputting a control parameter correction amount for adjusting a system control loop; and issuing the control parameter correction amount to a bottom control system of the waste heat power generation system, so that the control parameter of the system control loop is adaptively adjusted in advance before a power generation deviation actually occurs. The application can obtain a future trend, which is beneficial to real-time adjustment and reduction of control deviation.

Description

Technical Field

[0001] This application relates to the field of waste heat power generation control technology, and in particular to a control method, system and device for a waste heat power generation system. Background Technology

[0002] Waste heat power generation systems typically consist of three core components: a waste heat boiler, a steam turbine, and a generator. The power generation efficiency of the steam turbine is highly dependent on the stability of thermodynamic parameters such as main steam temperature, main steam pressure, and steam flow rate.

[0003] In industrial settings, upstream steelmaking converters, sintering machines, cement kilns, and other production processes operate intermittently, causing the temperature and flow rate of waste heat flue gas to fluctuate frequently over time, which in turn leads to continuous changes in the steam parameters at the outlet of the waste heat boiler.

[0004] To ensure the safe and stable operation of steam turbines, domestic waste heat power generation projects currently widely adopt distributed control systems (DCS) to control steam parameters. The DCS collects process parameters such as temperature, pressure, flow rate, and liquid level, and operators manually set the setpoint (SP value) and output limit of the PID control loop based on their operating experience to keep the steam parameters near the target value.

[0005] In traditional DCS control schemes, the proportional gain, integral time, and derivative time of the PID controller remain unchanged once set. When the steam parameters of the upstream waste heat flue gas deviate from the set value due to process fluctuations, the system relies on the feedback regulation of the PID controller to pull the parameters back to the target range.

[0006] Meanwhile, operators regularly adjust the PID parameters manually based on the trend curves displayed on the DCS operation interface to adapt to the phased changes in operating conditions.

[0007] However, the above control scheme has at least the following problem: the feedback control mechanism has an inherent delay. Summary of the Invention

[0008] Therefore, it is necessary to provide a control method, system, and device for a waste heat power generation system to address the aforementioned technical problems.

[0009] In a first aspect, this application provides a control method for a waste heat power generation system, comprising: Obtain the current operating status data of the waste heat power generation system, including the measured values ​​of steam parameters and the measured value of power generation. The operating status data is input into a pre-built time series prediction model. The time series prediction model is used to predict the changing trend of the operating status data in a predetermined future time domain, and outputs prediction trend information containing the predicted power generation value at future time. A state space for intelligent decision-making models is constructed, which includes current control deviation information, deviation change rate information, and future time prediction values ​​from the prediction trend information. The future time prediction values ​​serve as forward-looking features to characterize the future change trend of power generation. The state space is input into the intelligent decision model, which outputs the control parameter correction amount for adjusting the system control loop based on the joint constraints of the current control deviation information, deviation change rate information and forward-looking features contained in the state space. The control parameter correction is sent to the underlying control system of the waste heat power generation system so as to proactively and adaptively adjust the control parameters of the system control loop before the actual power generation deviation occurs.

[0010] In one embodiment, the time-series prediction model includes: A convolutional-based temporal feature extraction network is used to extract local features of runtime state data in the time dimension. A gated unit based on a recurrent neural network is used to capture long-term dependencies in runtime state data; and A feature weighting network based on an attention mechanism is used to adaptively assign weights to features at different times.

[0011] In one embodiment, the intelligent decision-making model is a parameter optimization model based on reinforcement learning, wherein the policy network of reinforcement learning takes the state space as input and the control parameter correction amount as output.

[0012] In one embodiment, the control parameter correction amount is the adjustment increment of the proportional gain, integral time, and derivative time of the proportional-integral-derivative controller in the system control loop.

[0013] In one embodiment, the control parameter correction amount is sent to the underlying control system, including: The control parameter correction amount is divided into multiple consecutive step increments and written into the underlying control system sequentially at preset time intervals. If the running status data exceeds the safety threshold or communication is interrupted during the writing process, the current writing operation is stopped and the operation is rolled back to the previous set of valid control parameters.

[0014] In one embodiment, the future predetermined time domain is a second-level time window. The time series prediction model receives the operating status data in a sliding window manner and outputs the power generation prediction sequence of multiple consecutive sampling times within the second-level time window.

[0015] In one embodiment, it further includes: The cumulative excess power generation is calculated based on the difference between the actual power generation after optimized control and the preset benchmark power. Based on the preset revenue sharing ratio for energy performance contracting, the revenue data for the energy management party and the revenue data for the owner are generated separately from the cumulative excess power generation.

[0016] In one embodiment, it further includes: When the prediction results of the time series prediction model or the decision results of the intelligent decision model are abnormal, calculate the contribution of each input feature to the prediction results or decision results. Output at least one key feature with the highest contribution to the interactive interface for display.

[0017] Secondly, this application also provides a control system for a waste heat power generation system, comprising: Edge computing gateway, the edge computing gateway is used to perform the methods described above; The underlying control system communicates with the edge computing gateway via an industrial communication protocol. The underlying control system provides operating status data to the edge computing gateway and receives control parameter corrections from the edge computing gateway. The underlying control system includes distributed control systems or programmable logic controllers.

[0018] Thirdly, this application also provides a control device for a waste heat power generation system, comprising: The data acquisition module is used to acquire the current operating status data of the waste heat power generation system. The operating status data includes the measured values ​​of steam parameters and the measured values ​​of power generation. The prediction module has a built-in time-series prediction model, which is used to predict the trend of changes in operating status data in a predetermined future time domain and output prediction trend information containing the predicted power generation value at future time. The decision optimization module has a built-in intelligent decision model. The decision optimization module is used to construct the state space of the intelligent decision model. The state space includes current control deviation information, deviation change rate information, and future time prediction values ​​in the prediction trend information. The future time prediction values ​​are used as forward-looking features to characterize the future change trend of power generation. Based on the joint constraints of the current control deviation information, deviation change rate information, and forward-looking features contained in the state space, the intelligent decision model outputs the control parameter correction amount for adjusting the control loop of the system. The execution module is used to send control parameter corrections to the underlying control system of the waste heat power generation system, so as to proactively and adaptively adjust the control parameters of the system control loop before the actual power generation deviation occurs.

[0019] The control method, system, and apparatus for the waste heat power generation system provided in this application enable the system to perceive the current operating condition of the waste heat power generation system in real time by acquiring operating status data including measured values ​​of steam parameters and measured values ​​of power generation. Based on this, the operating status data is input into a pre-constructed time series prediction model. This time series prediction model uses the operating status data at the current and historical times to predict the changing trend of the operating status data in a predetermined future time domain and outputs prediction trend information including the predicted power generation value at future times, so that the system can know the changing trend of power generation at the current time in the future.

[0020] Based on this, a state space for the intelligent decision-making model is constructed. This state space simultaneously contains current control deviation information, deviation change rate information, and the predicted value at a future time from the aforementioned predicted trend information. This future time predicted value is used as a forward-looking feature characterizing the future trend of power generation. Thus, the state space includes both real-time feedback information characterizing the current control deviation and its changing trend, and forward-looking prediction information characterizing the future trend of power generation, enabling the intelligent decision-making model to simultaneously process current real-time deviation information and future trend information within the same decision-making framework. Subsequently, the aforementioned state space is input into the intelligent decision-making model, which is based on the current control deviation information, deviation change rate information, and forward-looking feature contained in the state space. The intelligent decision-making model makes decisions based on joint constraints. Since the state space simultaneously contains current control deviation information, deviation change rate information, and forward-looking features, the output control parameter correction is constrained not only by the current real-time feedback state but also by the future power generation change trend. Therefore, the output control parameter correction is itself a decision result that incorporates forward-looking prediction information. Finally, the control parameter correction is sent to the underlying control system of the waste heat power generation system. Since the control parameter correction is obtained based on state space decision containing forward-looking features, the future power generation change trend has been considered in advance when it is sent to the underlying control system, so that the adjustment of the control parameters of the system control loop is completed before the actual power generation deviation occurs.

[0021] Therefore, this application enables the system to predict future operating conditions by using a time-series prediction model. By incorporating the predicted values ​​of future moments as forward-looking features into the state space of the intelligent decision-making model, the decision-making process integrates forward-looking information. By sending the control parameter correction amount obtained from the decision based on forward-looking information to the underlying control system before the deviation occurs, the adjustment of control parameters is made in advance of the actual occurrence of the deviation. Thus, future trends can be known, which is conducive to real-time adjustment of actions and reduction of control deviation. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A schematic flowchart illustrating a control method for a waste heat power generation system provided in an embodiment of this application; Figure 2 A flowchart illustrating a control parameter correction issuance step provided in an embodiment of this application; Figure 3 A schematic diagram of the control system of a waste heat power generation system provided in this application embodiment; Figure 4 The control effect comparison curves provided in the embodiments of this application; Figure 5 A schematic diagram of the structure of a control device for a waste heat power generation system provided in an embodiment of this application; Figure 6 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0025] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0026] In one exemplary embodiment, Figure 1 A flowchart illustrating a control method for a waste heat power generation system provided in this application embodiment is shown below. Figure 1 As shown, a control method for a waste heat power generation system is provided. Taking the application of this method to a terminal as an example, it can be understood that this method can also be applied to a server, and can also be applied to a system including a terminal and a server, and can be realized through the interaction between the terminal and the server.

[0027] The terminal can be, but is not limited to, various personal computers, laptops, smartphones, and tablets.

[0028] A server can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.

[0029] In this embodiment, the method includes the following steps S101 to S105.

[0030] in: S101. Obtain the operating status data of the waste heat power generation system at the current moment. The operating status data includes the measured values ​​of steam parameters and the measured values ​​of power generation.

[0031] Among them, the operating status data refers to the collection of various process parameters that can reflect the operating conditions of the waste heat power generation system at the current moment. The measured values ​​of steam parameters include one or more of the measured values ​​of main steam temperature, main steam pressure and steam flow. The measured value of power generation refers to the actual output power of the steam turbine generator at the current moment.

[0032] For example, the terminal establishes a communication connection with the underlying control system of the waste heat power generation system and reads the current operating status data from the real-time database of the underlying control system according to a preset sampling period.

[0033] The operating status data includes, but is not limited to, measured values ​​of steam parameters and measured values ​​of power generation. The terminal will parse and store the operating status data it reads according to a preset data format, so as to serve as input data for subsequent processing steps.

[0034] As an example, the terminal establishes a communication connection with the distributed control system through the OPC unified architecture protocol, and reads the current time values ​​of measuring points such as main steam temperature, main steam pressure and power generation from the real-time database of the distributed control system at a sampling period of 1 second.

[0035] The terminal associates and stores the read data according to the timestamp and measurement point name to form a record of the current operating status data.

[0036] S102. Input the operating status data into the pre-built time series prediction model, use the time series prediction model to predict the change trend of the operating status data in the future predetermined time domain, and output the prediction trend information containing the predicted power generation value at future time.

[0037] Among them, the time series prediction model refers to a machine learning model that has been pre-trained and can output the predicted value of future time based on the input historical time series data. The time series prediction model receives the running status data including the current time and the historical time in a sliding window manner, and outputs the predicted value of each time in the future within a predetermined time domain.

[0038] The future time domain refers to the time interval spanning a certain period of time in the future, starting from the current moment.

[0039] Forecast trend information refers to a set of time-series data output by a time-series forecast model, which contains the predicted power generation values ​​for future times, and is used to characterize the trend of power generation changes over a period of time.

[0040] For example, the terminal combines the current running status data obtained in S101 and the running status data stored in historical time into a time series data sequence in chronological order, and inputs the time series data sequence into a pre-built time series prediction model.

[0041] The time-series forecasting model extracts features and models time-series dependencies from the input time-series data sequence, and outputs a sequence of predicted power generation values ​​within a predetermined time domain.

[0042] The terminal uses the predicted value sequence as predictive trend information for decision processing in subsequent steps.

[0043] As an example, the terminal uses a 60-second sliding window to combine the running status data of the current moment and the previous 59 seconds, a total of 60 sampling moments, into an input sequence, which is then input into the time series prediction model.

[0044] The time-series forecast model outputs a sequence of predicted power generation values ​​for each second within the next 60 seconds.

[0045] The terminal will use a sequence of predicted power generation values ​​for 60 prediction times as prediction trend information.

[0046] S103. Construct a state space for the intelligent decision-making model. The state space includes current control deviation information, deviation change rate information, and future time prediction values ​​in the prediction trend information. The future time prediction values ​​serve as forward-looking features to characterize the future change trend of power generation.

[0047] Among them, intelligent decision-making models refer to pre-built machine learning models that can output decision-making actions based on input state information.

[0048] The state space refers to the set of input features that an intelligent decision-making model relies on when making decisions.

[0049] Current control deviation information refers to the difference between the current control target setpoint and the current measured value of the actual operating parameters.

[0050] The rate of change of deviation information refers to the rate at which the control deviation information at the current moment changes relative to the control deviation information at the previous moment.

[0051] Forward-looking features refer to feature values ​​extracted from predicted trend information to characterize the future trend of power generation changes. They enable intelligent decision-making models to consider both the current real-time state and future trends when making decisions.

[0052] For example, the terminal calculates the current control deviation information based on the control target setpoint and the actual measured values ​​of the operating parameters in the current operating status data obtained in S101.

[0053] The terminal calculates the rate of change of deviation based on the control deviation information at the current moment and the control deviation information stored at the previous moment.

[0054] The terminal extracts the predicted power generation value for future time moments from the predicted trend information output by S102.

[0055] The terminal combines the current control deviation information, the deviation change rate information, and the predicted value at future time moments into a state space.

[0056] As an example, the terminal calculates the difference between the current main steam pressure setpoint and the measured main steam pressure to obtain the current pressure deviation value, and calculates the difference between the pressure deviation value at the current time and the previous time to obtain the deviation change rate.

[0057] The terminal extracts the predicted power generation value for the first second in the future from the predicted trend information.

[0058] The terminal combines the current pressure deviation value, the deviation change rate value, and the predicted power generation value in the first second of the future into the state space of the intelligent decision-making model.

[0059] S104. Input the state space into the intelligent decision model. Based on the current control deviation information, deviation change rate information and forward-looking characteristics contained in the state space, the intelligent decision model outputs the control parameter correction amount for adjusting the system control loop.

[0060] Among them, the control parameter correction amount refers to the amount of change in the control parameter values ​​in the system control loop output by the intelligent decision model.

[0061] Joint constraints refer to the fact that when an intelligent decision-making model outputs a control parameter correction, it is simultaneously constrained by three factors contained in the state space: the current control deviation information, the deviation change rate information, and the forward-looking features. In other words, the control parameter correction is obtained by jointly mapping the current control deviation information, the deviation change rate information, and the forward-looking features through the intelligent decision-making model.

[0062] For example, the terminal inputs the state space constructed in S103 into the intelligent decision model in vector form. The intelligent decision model performs joint processing on the current control deviation information, deviation change rate information and forward-looking features in the state space according to the pre-trained mapping relationship, and outputs the control parameter correction amount.

[0063] The control parameter correction output by the intelligent decision-making model simultaneously reflects the correction needs based on the current control deviation, the correction needs based on the rate of change of the deviation, and the forward-looking correction needs based on the prediction of future trends.

[0064] As an example, the intelligent decision-making model is a deep reinforcement learning model based on the Actor-Critic architecture. The terminal inputs the state vector containing the current control deviation value, the deviation change rate value, and the predicted power generation value at future time moments into the Actor network of the deep reinforcement learning model. The Actor network outputs the proportional gain correction, integral time correction, and derivative time correction.

[0065] S105. The control parameter correction amount is sent to the bottom control system of the waste heat power generation system so as to make forward adaptive adjustments to the control parameters of the system control loop before the actual power generation deviation occurs.

[0066] The underlying control system refers to the industrial control system that directly executes the process control of the waste heat power generation system, including distributed control systems or programmable logic controllers.

[0067] The system control loop refers to the control loop in the underlying control system used to control parameters related to power generation.

[0068] Proactive adaptive adjustment refers to adjusting control parameters in advance based on predictions of future trends before actual deviations in power generation occur, so that the system is already in a state of adaptive control parameters when the deviations actually occur.

[0069] For example, the terminal sends the control parameter correction amount output by S104 to the underlying control system through an industrial communication protocol.

[0070] After receiving the control parameter correction, the underlying control system applies it to the corresponding control parameters in the system control loop.

[0071] Since the control parameter correction is based on state-space decisions that include forward-looking features, the adjustment is completed before the actual power generation deviation occurs.

[0072] As an example, the terminal writes the proportional gain correction, integral time correction, and derivative time correction to the corresponding parameter registers of the PID controller in the distributed control system via the OPC unified architecture protocol. The distributed control system then adjusts the control behavior of its PID controller based on the new parameter values ​​after they are written.

[0073] In this embodiment, by acquiring the current operating status data of the waste heat power generation system, the system gains real-time operating condition awareness. By inputting the operating status data into a time-series prediction model, the system predicts the changing trend of the operating status data within a predetermined future time domain and outputs prediction trend information containing the predicted power generation value at future times, enabling the system to know the future power generation trend at the current moment. By constructing a state space containing current control deviation information, deviation change rate information, and the predicted value at future times as a forward-looking feature, the intelligent decision-making model simultaneously obtains real-time feedback information and future trend prediction information. By inputting the state space into the intelligent decision-making model, the intelligent decision-making model is subject to the joint constraints of current control deviation information, deviation change rate information, and forward-looking features when making decisions, and outputs a control parameter correction amount that integrates forward-looking information. By sending the control parameter correction amount down to the underlying control system, the adjustment of control parameters is made in advance of the actual occurrence of power generation deviation.

[0074] The above steps together constitute a complete control link from trend prediction and forward-looking feature injection to advance adjustment, enabling the system to perceive future operating condition changes in advance and make pre-adjustments, thereby reducing the lag effect of feedback control, reducing the oscillation amplitude of steam parameters, and making the steam turbine operate more stably near the target operating point.

[0075] In one exemplary embodiment, the time-series prediction model includes: A convolutional-based temporal feature extraction network is used to extract local features of runtime state data in the time dimension. A gated unit based on a recurrent neural network is used to capture long-term dependencies in runtime state data; and A feature weighting network based on an attention mechanism is used to adaptively assign weights to features at different times.

[0076] Among them, the temporal feature extraction network based on convolution operation refers to the neural network structure that uses convolution operation to process temporal data. The convolution operation extracts local temporal patterns between adjacent time points in the input sequence by sliding the convolution kernel in the time dimension.

[0077] A gated unit based on a recurrent neural network refers to a recurrent neural network structure that uses a gating mechanism to selectively memorize and forget temporal information. The gated unit controls the transmission of information between time steps by updating and resetting the gate.

[0078] Feature weighted networks based on attention mechanisms refer to neural network structures that use attention calculation to adaptively assign weights to features at each time step in the input sequence. The attention mechanism determines the importance weight of each feature to the output result by calculating the correlation score between the query vector and the key vector.

[0079] For example, after the terminal inputs the running status data into the time series prediction model, the time series feature extraction network based on convolution operation first performs a time dimension convolution operation on the input sequence to extract the local change patterns between adjacent time moments and output local time series features.

[0080] The gated unit based on the recurrent neural network receives local temporal features, performs recursive processing along the time steps, captures long-term dependencies in the running state data, and outputs temporal features containing long-range memory information.

[0081] The attention-based feature weighting network calculates attention weights for the features output by the gating unit at each time step, and uses the weighted aggregated features as the final feature representation of the time-series prediction model to output the prediction results.

[0082] As an example, the temporal feature extraction network is a temporal convolutional network containing three dilated causal convolutional layers, each containing 128 channels, with dilation coefficients set to 1, 2, and 4 respectively; the gating unit is a bidirectional gated recurrent unit, containing two layers, each containing 256 units; and the attention mechanism is an eight-head self-attention mechanism.

[0083] The operational status data is processed sequentially by a temporal feature extraction network, a gating unit, and an attention mechanism, and then mapped to predicted values ​​through a fully connected layer.

[0084] In this embodiment, a temporal convolutional network is used to extract local features of the running status data in the time dimension, a gated recurrent unit is used to capture the long-term dependencies of the running status data, and an attention mechanism is used to adaptively assign weights to features at different times. This enables the temporal prediction model to more accurately capture the temporal change patterns in the running status data, which is beneficial to improving the accuracy of prediction and thus providing a more reliable information foundation for subsequent forward-looking decision-making.

[0085] In an exemplary embodiment, the intelligent decision-making model is a parameter optimization model based on reinforcement learning, wherein the policy network of reinforcement learning takes the state space as input and the control parameter adjustment amount as output.

[0086] Reinforcement learning refers to a machine learning paradigm in which an agent selects actions based on state information and receives reward signals through interaction with the environment, with the goal of maximizing cumulative rewards.

[0087] A policy network is a neural network used in reinforcement learning to map states to actions. The policy network outputs deterministic actions or a probability distribution of actions.

[0088] For example, the terminal uses the state space as input to the policy network of reinforcement learning. The policy network performs a nonlinear mapping on the input state space and outputs an action value corresponding to the control parameter correction amount.

[0089] The policy network optimizes its network parameters in advance through a training process that interacts with the environment, so that the output control parameter corrections enable the system to obtain greater cumulative rewards.

[0090] As an example, the intelligent decision-making model is a deep deterministic policy gradient model based on the Actor-Critic architecture. The Actor network serves as the policy network, taking the state space as input and outputting continuous PID parameter corrections.

[0091] In this embodiment, a parameter optimization model based on reinforcement learning is used as the intelligent decision-making model. This enables the intelligent decision-making model to continuously optimize its decision-making strategy through continuous interaction with the environment, which is conducive to outputting control parameter corrections that are more suitable for the current working conditions, thereby improving the control effect.

[0092] In an exemplary embodiment, the control parameter correction amount is the adjustment increment of the proportional gain, integral time, and derivative time of the proportional-integral-derivative controller in the system control loop.

[0093] In industrial process control, a proportional-integral-derivative (PID) controller uses a combination of proportional, integral, and derivative control actions to adjust deviations. The proportional gain determines the controller's response magnitude to the current deviation, the integral time determines the speed at which the controller eliminates accumulated deviations, and the derivative time determines the controller's responsiveness to the rate of change of the deviation. The adjustment increment refers to the change in the control parameter value relative to the current operating value.

[0094] For example, the control parameter corrections output by the intelligent decision-making model include the proportional gain adjustment increment, the integral time adjustment increment, and the derivative time adjustment increment.

[0095] The terminal adds the proportional gain adjustment increment, integral time adjustment increment, and derivative time adjustment increment to the current proportional gain value, integral time value, and derivative time value, respectively, to obtain the updated control parameter values.

[0096] As an example, the action space output by the intelligent decision model is a three-dimensional continuous action, corresponding to the relative rates of change of the proportional gain Kp, integral time Ki, and derivative time Kd, respectively, with the relative rates of change ranging from -20% to +20%.

[0097] In this embodiment, by limiting the adjustment amount of the control parameters to the adjustment increments of the proportional gain, integral time, and derivative time, the adjustment of the control parameters is made incrementally based on the current control parameter values. This helps to maintain the stability of the control system and avoid system oscillations caused by sudden parameter changes.

[0098] In one exemplary embodiment, Figure 2A flowchart illustrating a control parameter correction step provided in this application embodiment is shown in the figure. The step of sending the control parameter correction to the underlying control system includes: S201. Divide the control parameter correction amount into multiple continuous step increments and write them into the underlying control system sequentially at preset time intervals. S202. If the running status data exceeds the safety threshold or communication is interrupted during the writing process, the current writing operation is stopped and the operation is rolled back to the previous set of valid control parameters.

[0099] Among them, the step increment refers to the correction amount written each time after the total control parameter correction amount is divided equally or proportionally according to the preset number of steps.

[0100] The preset time interval refers to the time interval between two consecutive write operations.

[0101] The safety threshold refers to the preset upper and lower limit values ​​used to determine whether the operating status data is within a safe range.

[0102] For example, when the terminal sends the control parameter correction amount to the underlying control system, it first decomposes the total correction amount into multiple step increments, and then writes the step increments into the underlying control system step by step according to a preset time interval.

[0103] After each write operation is completed, the terminal reads the current operating status data of the underlying control system and determines whether it exceeds the preset safety threshold. If it does, the subsequent write operation is stopped and the control parameters are rolled back to the valid parameter values ​​before the write started.

[0104] If an interruption in communication with the underlying control system is detected, writing will also stop and parameters will be rolled back.

[0105] As an example, the terminal divides the PID control parameter correction into 3 to 5 steps and writes them sequentially to the underlying control system at a speed of 5 to 10 seconds between each step. If the steam pressure exceeds the safety limit or the communication times out after any step is written, the writing is stopped immediately and the control parameters are restored to the values ​​before this batch of writing.

[0106] In this embodiment, by dividing the control parameter correction amount into multiple continuous step increments and writing them sequentially to the underlying control system at preset time intervals, the change in control parameters is gradual rather than abrupt, which helps to avoid system oscillations caused by parameter abrupt changes. By detecting whether the running status data exceeds the safety threshold or whether communication is interrupted during the writing process, and stopping the writing and reverting the parameters when an abnormality occurs, it helps to ensure the operational safety of the system during the control parameter adjustment process.

[0107] In an exemplary embodiment, the future predetermined time domain is a second-level time window. The time series prediction model receives the operating status data in a sliding window manner and outputs the power generation prediction sequence of multiple consecutive sampling times within the second-level time window.

[0108] The second-level time window refers to the prediction time domain length measured in seconds.

[0109] The sliding window approach refers to a time series prediction model receiving a fixed-length segment of historical time series data as input each time, and adding the latest data to the window and removing the oldest data at each new sampling time, while keeping the window length constant.

[0110] The power generation prediction sequence refers to the ordered set of power generation prediction values ​​output by the time series prediction model, corresponding to multiple consecutive sampling times within a time window.

[0111] For example, the terminal continuously acquires operating status data according to a preset sampling period and maintains historical data for a recent period using a sliding window of fixed length.

[0112] In each control cycle, the terminal inputs the historical operating status data within the current sliding window into the time series prediction model. The time series prediction model outputs the predicted power generation value corresponding to each sampling moment within the next second-level time window, forming a prediction sequence.

[0113] In the next control cycle, the terminal adds the latest acquired operating status data to the sliding window and removes the oldest data before inputting it into the time series prediction model for prediction.

[0114] As an example, the terminal maintains the operating status data sequence of the most recent 60 seconds with a sliding window length of 60 seconds and a sampling period of 1 second. The time series prediction model takes the 60-second data sequence as input and outputs the predicted power generation value for each second in the next 60 seconds.

[0115] In this embodiment, the time-series prediction model receives operating status data in a sliding window manner and outputs a power generation prediction sequence of multiple consecutive sampling moments within a second-level time window. This enables the system to continuously obtain updated prediction information, which is beneficial for achieving a refined perception of future trends and thus providing a more accurate reference for forward-looking decision-making.

[0116] In one exemplary embodiment, it further includes: The cumulative excess power generation is calculated based on the difference between the actual power generation after optimized control and the preset benchmark power. Based on the preset revenue sharing ratio for energy performance contracting, the revenue data for the energy management party and the revenue data for the owner are generated separately from the cumulative excess power generation.

[0117] The preset benchmark power refers to the reference power value determined based on historical operating data or contractual agreements, used to measure whether the power generation exceeds the limit.

[0118] Cumulative excess power generation refers to the integral of the portion of actual power generation exceeding the preset benchmark power over time within a preset time interval.

[0119] The revenue sharing ratio in energy performance contracting refers to the pre-agreed ratio of excess profits between the energy management party and the owner in the energy performance contracting model.

[0120] For example, during the process of optimizing the control of the waste heat power generation system using the control method provided in this application, the terminal continuously obtains the actual power generation at the current moment, reads the preset reference power value, and calculates the difference between the actual power generation and the reference power.

[0121] The terminal accumulates and integrates the difference over time to obtain the cumulative excess power generation.

[0122] The terminal allocates the excess revenue corresponding to the cumulative excess power generation into revenue data for the energy management party and revenue data for the owner, based on the preset revenue sharing ratio for energy performance contracting.

[0123] As an example, the terminal calculates the cumulative excess power generation according to the following formula: ,in, Let τ be the actual power generation at time τ. The preset reference power is the power generated under the same external operating conditions in the historical operating data of the previous year.

[0124] The terminal calculates the EMC revenue based on the EMC revenue sharing ratio R: EMC revenue = excess power generation × electricity price × R, and the owner revenue = excess power generation × electricity price × (1-R).

[0125] In this embodiment, the cumulative excess power generation is calculated based on the difference between the actual power generation after optimized control and the preset benchmark power. The revenue data of the energy management party and the owner party are generated according to the preset contract energy management sharing ratio. This enables the EMC benefits to be calculated in real time and automatically, which is conducive to providing transparent and timely benefit data support for contract energy management projects.

[0126] In one exemplary embodiment, it further includes: When the prediction results of the time series prediction model or the decision results of the intelligent decision model are abnormal, calculate the contribution of each input feature to the prediction results or decision results. Output at least one key feature with the highest contribution to the interactive interface for display.

[0127] Anomalies refer to the state in which the prediction or decision results deviate from the expected range, including predicted values ​​exceeding the reasonable range or the correction amount of the control parameters of the decision output exceeding the safety boundary.

[0128] Contribution refers to the quantitative indicator of the degree of influence of each input feature on the output result, and is used to characterize the importance of the feature in the model output.

[0129] Key features refer to one or more input features that rank highly in terms of contribution.

[0130] For example, after the time-series prediction model outputs a prediction result or the intelligent decision-making model outputs a decision result, the terminal performs an anomaly determination on the prediction result or the decision result.

[0131] When the judgment result is abnormal, the terminal calls the contribution analysis algorithm to calculate the contribution value of each input feature to the current output result, and outputs at least one key feature with the highest contribution and its corresponding numerical information to the interactive interface for display.

[0132] As an example, the terminal uses the SHAP algorithm to calculate the SHAP contribution value of each input feature for each output of the time series prediction model or intelligent decision model. When the predicted value deviates from the preset range or the parameter correction of the decision output exceeds the safety boundary, the terminal sends the three features with the highest SHAP contribution values ​​and their contribution values ​​to the human-computer interaction interface for visualization.

[0133] In this embodiment, when the prediction or decision results are abnormal, the contribution of each input feature is calculated and the key feature with the highest contribution is output to the interactive interface for display. This helps operators understand the reasons for the abnormal model output, thereby assisting in judgment and intervention decisions.

[0134] In some exemplary embodiments, the method provided in this application is deployed on an edge computing gateway, which communicates with the underlying control system via an industrial communication protocol. The underlying control system is a distributed control system or a programmable logic controller.

[0135] The edge computing gateway is equipped with an AI-accelerated inference module, in which time-series prediction models and intelligent decision-making models are deployed.

[0136] Edge computing gateways interact with the underlying control system through any one of the following protocols: OPC unified architecture protocol, Modbus protocol, or Profinet protocol.

[0137] In a specific example, the edge computing gateway uses NVIDIA Jetson Orin NX as the AI ​​inference module, and connects to the ZKTeco ECS-700 distributed control system through the OPC unified architecture protocol. The time-series prediction model and intelligent decision model are deployed and run in Jetson Orin NX.

[0138] In another specific example, for a waste heat generator set with a small installed capacity, the edge computing gateway uses a Raspberry Pi as the main control unit and an NVIDIA Jetson Nano as the AI ​​inference coprocessor. It is connected to the Hollysys MACS-S distributed control system via the Modbus protocol. The time series prediction model is compressed to less than 100KB by knowledge distillation and INT8 quantization and deployed in the Jetson Nano.

[0139] In some exemplary embodiments, the method provided in this application is implemented using an edge-cloud collaborative architecture.

[0140] A model training platform is set up in the cloud. The model training platform is used to pre-train the time series prediction model and the intelligent decision-making model using historical operating status data, and then sends the trained model parameters to the edge computing gateway.

[0141] Edge computing gateways utilize model parameters to run sequence prediction models and intelligent decision-making models, and use real-time collected operational status data to perform online incremental fine-tuning of model parameters during operation.

[0142] The cloud also has a time-series database to store historical running status data and model running log data uploaded from the edge computing gateway, in order to support regular iterative training and performance analysis of the model.

[0143] In a specific example, the cloud training platform is equipped with an NVIDIA A100 graphics processor. It uses 90 days of historical running status data to train the time series prediction model offline, with a training time of about 6 hours. After training is completed, the model parameters are sent to the edge computing gateway. In actual operation, the edge computing gateway uses the latest accumulated running data to perform online incremental fine-tuning of the model parameters once a week.

[0144] In some exemplary embodiments, after acquiring the operating status data, a data preprocessing step is also included: performing data filtering and timestamp alignment processing on the measured values ​​of steam parameters and measured values ​​of power generation in the acquired operating status data.

[0145] Data filtering is used to eliminate sensor noise and sampling errors, while timestamp alignment is used to unify data from different data sources or different acquisition channels to the same time base.

[0146] In a specific example, the Kalman filter algorithm is used to filter the operating status data. The state transition matrix and observation matrix of the Kalman filter are pre-calibrated according to the thermodynamic process characteristics of the waste heat power generation system.

[0147] In another specific example, the timestamp alignment process uses a linear interpolation method to uniformly interpolate measurement data from different sampling frequencies onto the same time grid.

[0148] Data preprocessing helps improve the quality of input data, thereby enhancing the accuracy and stability of time series forecasting model prediction results.

[0149] In some exemplary embodiments, the state space also includes exhaust steam temperature parameters of the turbine in the waste heat power generation system, metal wall temperature parameters, and current instantaneous power generation efficiency estimates.

[0150] The current instantaneous power generation efficiency estimate is approximately calculated by the ratio of the generator's active power to the equivalent heat source mass flow rate and the specific enthalpy drop of the waste heat flue gas.

[0151] Specifically, the terminal uses existing measurement data from the distributed control system to obtain the inlet flue gas temperature and exhaust gas temperature of the waste heat boiler, calculates the input heat power in combination with the flue gas mass flow rate, and then obtains the instantaneous power generation efficiency estimate by dividing the generator active power by the input heat power.

[0152] Incorporating the current instantaneous power generation efficiency estimate into the state space helps the intelligent decision-making model to more comprehensively perceive the current thermodynamic performance of the unit, thereby outputting control parameter corrections that are more suitable for the current unit operating conditions.

[0153] In some exemplary implementations, the intelligent decision-making model is pre-trained in a digital twin simulation environment and fine-tuned online using real-time operational data during actual operation.

[0154] The digital twin simulation environment is built based on the thermodynamic process mechanism model of the waste heat power generation system. It is used to simulate the changes in operating status data under different operating conditions and the impact of control parameter adjustments on power generation.

[0155] During the pre-training process, the intelligent decision-making model interacts with the simulation environment in the digital twin simulation environment, outputs control parameter corrections based on the state space, and optimizes the parameters of the policy network based on the reward signals fed back from the simulation environment.

[0156] After pre-training, the parameters of the policy network are deployed to the terminal, and the policy network is incrementally fine-tuned by periodically collecting running data during actual operation.

[0157] In a specific example, the reward function includes a power tracking error term, a steam parameter stability term, and an action smoothness term. The power tracking error term is calculated based on the square of the difference between the actual power generation and the target power generation. The steam parameter stability term is calculated based on the square of the deviations of the main steam temperature and main steam pressure from their respective set values. The action smoothness term is calculated based on the square of the difference between the control parameter corrections of two consecutive outputs. All terms are weighted and summed to form the total reward function.

[0158] By combining pre-training in a digital twin simulation environment with online fine-tuning, it is beneficial to accelerate the training convergence of intelligent decision-making models and continuously adapt to changes in actual working conditions.

[0159] In some exemplary embodiments, the preset reference power is a fixed reference power or a dynamic reference power.

[0160] When a fixed reference power is used, the fixed reference power is determined based on the design parameters of the waste heat power generation system or the value agreed upon in the contract, and remains unchanged throughout the year's operating cycle.

[0161] When a dynamic reference power is used, the dynamic reference power is updated according to a preset time period based on historical operating data. Specifically, the predetermined percentile of the power generation under the same external operating conditions in the historical operating data of the previous year is taken as the reference power for the current time period, where the predetermined percentile is the 50th percentile.

[0162] In a specific example, the baseline power is jointly agreed upon by both parties to the energy performance contract based on the guaranteed value of the turbine's thermal performance and the design flue gas parameters, and is fixed in the system configuration file.

[0163] In another specific example, the baseline power is updated monthly, taking the median power generation from the same historical operating data of the same month in the same year as the baseline power for the current month, in order to adapt to the impact of seasonal ambient temperature changes on waste heat power generation efficiency.

[0164] In some exemplary embodiments, the inputs to the time-series prediction model may also include one or more of the following operating status data: waste heat flue gas temperature parameters, waste heat flue gas flow parameters, waste heat boiler drum liquid level parameters, turbine exhaust temperature parameters, and turbine metal wall temperature parameters.

[0165] Among them, the waste heat flue gas temperature parameters include the inlet flue gas temperature and the exhaust gas temperature of the waste heat boiler, and the waste heat flue gas flow parameters are the mass flow rate or volume flow rate of the flue gas entering the waste heat boiler.

[0166] Using the above parameters as input to the time series prediction model helps the prediction model to more comprehensively perceive the thermal state of the waste heat power generation system, thereby improving the accuracy of power generation prediction.

[0167] In some exemplary embodiments, the system control loop is a proportional-integral-derivative controller loop.

[0168] The control parameters are the control parameters in the proportional-integral-derivative controller loop.

[0169] The proportional-integral-derivative controller loop is used to regulate one or more of the following control objectives: main steam temperature, main steam pressure, main steam flow rate, and steam drum liquid level.

[0170] The control parameter correction is applied to the corresponding actuator through the proportional-integral-derivative controller loop. The actuator includes one or more of the following: desuperheating water regulating valve, flue gas damper, feedwater pump frequency converter, and main steam regulating valve.

[0171] In one exemplary embodiment, Figure 3 A schematic diagram of the control system of a waste heat power generation system provided in this application embodiment is shown below. Figure 3 As shown, this application also provides a control system for a waste heat power generation system, including:

[0172] Edge computing gateway, the edge computing gateway is used to perform the methods described above; The underlying control system communicates with the edge computing gateway via an industrial communication protocol. The underlying control system provides operating status data to the edge computing gateway and receives control parameter corrections from the edge computing gateway. The underlying control system includes distributed control systems or programmable logic controllers.

[0173] In this context, an edge computing gateway refers to an embedded computing device deployed on the field side of a waste heat power generation system, operating close to the underlying control system. The edge computing gateway has a built-in AI-accelerated inference module used to run sequence prediction models and intelligent decision-making models, and executes the data acquisition, prediction, decision-making, and data distribution steps described above. The edge computing gateway establishes a communication connection with the underlying control system through industrial communication protocols, enabling real-time data interaction.

[0174] The underlying control system refers to the industrial control system that directly performs process control in the waste heat power generation system. It is used to collect real-time data from field sensors and actuators, and to receive and execute control commands from the edge computing gateway.

[0175] The underlying control system provides operational status data to the edge computing gateway as input for the gateway's prediction and decision-making. At the same time, the underlying control system receives control parameter corrections from the edge computing gateway and applies them to the corresponding system control loops.

[0176] The underlying control system includes distributed control systems or programmable logic controllers.

[0177] Among them, distributed control systems are suitable for complex process control scenarios with large scale and multiple loops, while programmable logic controllers are suitable for small-scale scenarios with logic control as the main function.

[0178] By separating the edge computing gateway from the underlying control system, AI computing capabilities can be integrated into the existing industrial control system in a non-intrusive manner without requiring hardware modifications to the underlying control system.

[0179] Optionally, in some implementations, the system also includes a cloud training platform that is communicatively connected to an edge computing gateway, used to pre-train the time series prediction model and the intelligent decision-making model using historical operating status data, and to send the trained model parameters to the edge computing gateway.

[0180] Optionally, in some implementations, the edge computing gateway is further provided with a security execution module, which is used to perform security verification and gradual writing processing on the control parameter correction amount before sending the control parameter correction amount to the underlying control system.

[0181] In an exemplary embodiment, a 50MW waste heat power generation EMC project of a certain project is used as the application scenario to verify the control method of the waste heat power generation system provided in this application.

[0182] The waste heat power generation system is equipped with an N50-3.82 condensing steam turbine generator set. The waste heat comes from the flue gas from the steelmaking converter and the sintering machine. The underlying control system is the central control ECS-700 distributed control system.

[0183] Edge computing gateways are deployed in the distributed control system's computer room and communicate with the distributed control system via the OPC unified architecture protocol.

[0184] For example, after deploying a time-series prediction model and an intelligent decision-making model in an edge computing gateway, the output of the edge computing gateway is connected to the PID control loop of the distributed control system.

[0185] During the 14-day trial run in automatic mode, the edge computing gateway continuously executes the control method according to the preset control cycle, and the distributed control system adjusts the proportional gain, integral time, and derivative time of the PID controller in real time according to the control parameter correction amount issued by the edge computing gateway.

[0186] As an example, the time series prediction model adopts a hybrid architecture of three dilated causal convolutional layers (with dilation coefficients of 1, 2, and 4, and 128 channels per layer), two bidirectional gated recurrent units (256 units per layer), and an eight-head self-attention mechanism. It receives operating status data with a 60-second sliding window and outputs a power generation prediction sequence for the next 60 seconds, with a model inference latency of less than 5 milliseconds.

[0187] The intelligent decision-making model adopts a deep deterministic policy gradient model based on the Actor-Critic architecture. The state space includes current control deviation information, deviation change rate information, and future power generation prediction. The action space outputs the proportional gain, integral time, and derivative time adjustment increments, with the adjustment increments ranging from -20% to +20%.

[0188] Figure 4 A comparison curve of control effects provided in the embodiments of this application.

[0189] like Figure 4 As shown, the graphs are arranged in a left-right column layout. The horizontal axis represents time in seconds, and the vertical axis represents the main steam pressure, main steam temperature, and power generation. The three graphs on the left are parameter change curves using a traditional distributed control system with manual adjustment, while the three graphs on the right are parameter change curves using the control method of this application. Figure 4 In the diagram, the dashed line represents the parameter change curve using a traditional distributed control system with manual adjustment, the solid line represents the parameter change curve using the control method of this application, and the dotted line represents the set value or average value of each parameter.

[0190] The comparison clearly shows that after adopting the control method of this application, the fluctuation range of main steam pressure and main steam temperature is significantly reduced, and the stability of power generation is significantly improved.

[0191] In this embodiment, through the above verification, after adopting the control method provided in this application, the main steam pressure fluctuation decreased from ±0.3 MPa to ±0.08 MPa, a reduction of 73.3%; the main steam temperature fluctuation decreased from ±15 degrees Celsius to ±4 degrees Celsius, a reduction of 73.3%; the turbine heat rate decreased from 11,850 kJ / kWh to 11,520 kJ / kWh, a reduction of approximately 2.8%; and the average power generation increased from 43.2 MW to 44.4 MW, an increase of approximately 2.8%.

[0192] Daily power generation increased from 1.037 million kWh to 1.066 million kWh, resulting in an annual increase of approximately 9.6 million kWh and an annual increase in revenue of approximately 4.8 million yuan.

[0193] The above data fully demonstrates that the control method provided in this application can effectively reduce the hysteresis effect of feedback control, reduce the oscillation amplitude of steam parameters, and enable the steam turbine to operate more stably near the target operating point.

[0194] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows.

[0195] Unless otherwise specified in this document, there is no strict order in which these steps are performed, and they may be performed in any other order.

[0196] Moreover, at least some steps in the flowcharts involved in the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0197] The control device for the waste heat power generation system provided in the embodiments of this application is described below. The control device for the waste heat power generation system has the same inventive concept as the control method for the waste heat power generation system described above. The solution to the problem provided by the device is similar to the solution described in the above method. Therefore, the specific limitations of one or more control device embodiments for the waste heat power generation system provided below can be referred to the limitations of the control method for the waste heat power generation system above. The control device for the waste heat power generation system described below and the control method for the waste heat power generation system described above can be referred to each other, and will not be repeated here.

[0198] In one exemplary embodiment, Figure 5 This is a schematic diagram of the structure of a control device for a waste heat power generation system provided in an embodiment of this application, as shown below. Figure 5 As shown, the control device 50 of the waste heat power generation system includes:

[0199] The data acquisition module 510 is used to acquire the operating status data of the waste heat power generation system at the current moment. The operating status data includes the measured values ​​of steam parameters and the measured values ​​of power generation.

[0200] The prediction module 520 has a built-in time-series prediction model. The time-series prediction model is used to predict the trend of changes in operating status data in a predetermined future time domain and output prediction trend information containing the predicted power generation value at future time.

[0201] The decision optimization module 530 has a built-in intelligent decision model. The decision optimization module is used to construct the state space of the intelligent decision model. The state space includes current control deviation information, deviation change rate information, and future time prediction values ​​in the prediction trend information. The future time prediction values ​​are used as forward-looking features to characterize the future change trend of power generation. Based on the joint constraints of the current control deviation information, deviation change rate information, and forward-looking features contained in the state space, the intelligent decision model outputs the control parameter correction amount for adjusting the control loop of the system.

[0202] The execution module 540 is used to send the control parameter correction amount to the underlying control system of the waste heat power generation system so as to make forward adaptive adjustments to the control parameters of the system control loop before the actual power generation deviation occurs.

[0203] In an exemplary embodiment, the temporal prediction model includes: a convolution-based temporal feature extraction network for extracting local features of runtime state data in the time dimension; a recurrent neural network-based gating unit for capturing long-term dependencies in runtime state data; and an attention-based feature weighting network for adaptively assigning weights to features at different times.

[0204] In an exemplary embodiment, the intelligent decision-making model is a parameter optimization model based on reinforcement learning, wherein the policy network of reinforcement learning takes the state space as input and the control parameter adjustment amount as output.

[0205] In an exemplary embodiment, the control parameter correction amount is the adjustment increment of the proportional gain, integral time, and derivative time of the proportional-integral-derivative controller in the system control loop.

[0206] In an exemplary embodiment, the execution module 540 is used to divide the control parameter correction amount into multiple consecutive step increments and write them sequentially into the underlying control system at preset time intervals. If the running status data exceeds the safety threshold or communication is interrupted during the writing process, the current writing operation is stopped and the operation is rolled back to the previous set of valid control parameters.

[0207] In an exemplary embodiment, the future predetermined time domain is a second-level time window. The time series prediction model receives the operating status data in a sliding window manner and outputs the power generation prediction sequence of multiple consecutive sampling times within the second-level time window.

[0208] In an exemplary embodiment, the system further includes a benefit calculation module, which calculates the cumulative excess power generation based on the difference between the actual power generation after optimized control and the preset benchmark power; and generates revenue data for the energy management party and revenue data for the owner party based on the cumulative excess power generation according to the preset contract energy management revenue sharing ratio.

[0209] In an exemplary embodiment, the system further includes an interpretability analysis module, which calculates the contribution of each input feature to the prediction result or decision result when the prediction result of the time series prediction model or the decision result of the intelligent decision model is abnormal; and outputs at least one key feature with the highest contribution to the interactive interface for display.

[0210] Each module in the control device of the aforementioned waste heat power generation system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0211] In one exemplary embodiment, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the control method for any of the waste heat power generation systems described above.

[0212] In one exemplary embodiment, this application also provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the control method for any of the waste heat power generation systems described above.

[0213] In one exemplary embodiment, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the control method for any of the waste heat power generation systems described in the above embodiments.

[0214] Indicatively, such as Figure 6 As shown, Figure 6 This is a schematic diagram of the internal structure of a computer device 600 provided in an embodiment of this application. The computer device 600 can be provided as a server.

[0215] Reference Figure 6 The computer device 600 includes a processor 602, which further includes one or more processors, and a memory resource represented by memory 601 for storing instructions, such as computer programs, that can be executed by the processor 602.

[0216] The computer program stored in memory 601 may include one or more modules, each corresponding to a set of instructions.

[0217] Furthermore, the processor 602 is configured to execute instructions to perform the control method of the waste heat power generation system in any of the above embodiments.

[0218] Computer device 600 can operate on an operating system stored in memory 601, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or similar.

[0219] The computer device 600 may also include a power supply component 603 configured to perform power management of the computer device 600, a wired or wireless network interface 604 configured to connect the computer device 600 to a network, and an input / output (I / O) interface 605.

[0220] Wireless methods can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies.

[0221] When the computer program is executed by the processor, it implements a control method for a waste heat power generation system.

[0222] The display unit 607 of the computer device is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device.

[0223] The display screen can be an LCD screen or an e-ink screen. The input device 606 of the computer device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the casing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0224] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0225] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0226] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0227] Any references to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile memory and volatile memory.

[0228] Non-volatile memories can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc.

[0229] Volatile memory may include random access memory (RAM) or external cache memory, etc.

[0230] By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0231] The databases involved in the various embodiments provided in this application may include at least one of relational databases and non-relational databases.

[0232] Non-relational databases can include, but are not limited to, blockchain-based distributed databases.

[0233] The processors involved in the various embodiments provided in this application may be general-purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited thereto.

[0234] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0235] The above-described embodiments are merely examples of several implementation methods of this application. They are described in a relatively specific and detailed manner, but should not be construed as limiting the scope of this application's patent.

[0236] It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of this application, and these all fall within the scope of protection of this application.

[0237] Therefore, the scope of protection of this application shall be determined by the appended claims.

Claims

1. A control method for a waste heat power generation system, characterized in that, include: Acquire the operating status data of the waste heat power generation system at the current moment, including the measured values ​​of steam parameters and the measured values ​​of power generation; The operating status data is input into a pre-built time series prediction model, and the time series prediction model is used to predict the change trend of the operating status data in a predetermined future time domain, and output prediction trend information containing the predicted power generation value at future time. A state space for an intelligent decision-making model is constructed, which includes current control deviation information, deviation change rate information, and future time prediction values ​​in the prediction trend information, wherein the future time prediction values ​​are used as forward-looking features to characterize the future change trend of power generation. The state space is input into the intelligent decision model, and the intelligent decision model outputs the control parameter correction amount for adjusting the system control loop based on the joint constraints of the current control deviation information, the deviation change rate information and the forward-looking feature contained in the state space. The control parameter correction amount is sent to the underlying control system of the waste heat power generation system to proactively and adaptively adjust the control parameters of the system control loop before the actual power generation deviation occurs.

2. The method according to claim 1, characterized in that, The time-series prediction model includes: A convolutional-based temporal feature extraction network is used to extract local features of the running state data in the time dimension. A gated unit based on a recurrent neural network is used to capture the long-term dependencies of the operational state data; and A feature weighting network based on an attention mechanism is used to adaptively assign weights to features at different times.

3. The method according to claim 1, characterized in that, The intelligent decision-making model is a parameter optimization model based on reinforcement learning. The policy network of the reinforcement learning takes the state space as input and the control parameter correction amount as output.

4. The method according to claim 3, characterized in that, The control parameter correction amount is the adjustment increment of the proportional gain, integral time, and derivative time of the proportional-integral-derivative controller in the system control loop.

5. The method according to claim 4, characterized in that, Sending the control parameter correction amount to the underlying control system includes: The control parameter correction amount is divided into multiple consecutive step increments and written into the underlying control system sequentially at preset time intervals. If the running status data exceeds the safety threshold or communication is interrupted during the writing process, the current writing operation is stopped and the operation is rolled back to the previous set of valid control parameters.

6. The method according to claim 1, characterized in that, The predetermined future time domain is a second-level time window. The time series prediction model receives the operating status data in a sliding window manner and outputs the power generation prediction sequence of multiple consecutive sampling times within the second-level time window.

7. The method according to claim 1, characterized in that, Also includes: The cumulative excess power generation is calculated based on the difference between the actual power generation after optimized control and the preset benchmark power. Based on the preset revenue sharing ratio for energy performance contracting, the revenue data for the energy management party and the revenue data for the owner are generated from the cumulative excess power generation.

8. The method according to claim 1, characterized in that, Also includes: When the prediction result of the time series prediction model or the decision result of the intelligent decision model is abnormal, the contribution of each input feature to the prediction result or the decision result is calculated. Output at least one key feature with the highest contribution to the interactive interface for display.

9. A control system for a waste heat power generation system, characterized in that, include: An edge computing gateway, the edge computing gateway being configured to perform the method as described in any one of claims 1-8; The underlying control system is connected to the edge computing gateway via an industrial communication protocol. The underlying control system is used to provide the operating status data to the edge computing gateway and receive the control parameter correction amount sent by the edge computing gateway. The underlying control system includes a distributed control system or a programmable logic controller.

10. A control device for a waste heat power generation system, characterized in that, include: The data acquisition module is used to acquire the operating status data of the waste heat power generation system at the current moment. The operating status data includes the measured values ​​of steam parameters and the measured values ​​of power generation. The prediction module has a built-in time-series prediction model, which is used to predict the change trend of the operating status data in a future predetermined time domain and output prediction trend information containing the predicted power generation value at future time. The decision optimization module has a built-in intelligent decision model. The decision optimization module is used to construct the state space of the intelligent decision model. The state space includes current control deviation information, deviation change rate information, and future time prediction values ​​in the prediction trend information. The future time prediction values ​​are used as forward-looking features to characterize the future change trend of power generation. Based on the joint constraints of the current control deviation information, deviation change rate information, and forward-looking features contained in the state space, the intelligent decision model outputs the control parameter correction amount for adjusting the system control loop. The execution module is used to send the control parameter correction amount to the underlying control system of the waste heat power generation system so as to proactively and adaptively adjust the control parameters of the system control loop before the actual power generation deviation occurs.

Images