System and method for intelligently regulating and controlling circulating water temperature at cold end of thermal power generating unit

Abstract

The invention relates to the field of thermal power generating unit water temperature regulation and control, and provides a thermal power generating unit cold end circulating water temperature intelligent regulation and control system and method, and the system comprises a data collection module which is used for collecting the multi-source operation parameter original data flow of a thermal power generating unit cold end system in real time; the data processing and feature extraction module is used for extracting a time sequence feature vector representing the dynamic characteristics of the cold end system; the intelligent prediction and decision-making module is used for generating an optimization regulation and control instruction set by utilizing the self-adaptive hybrid prediction model; the actuator control module is used for generating and issuing corresponding control signals to the circulating water pump frequency converter and the cooling tower fan speed regulating device to drive the circulating water pump frequency converter and the cooling tower fan speed regulating device to execute corresponding actions; and the system state monitoring and feedback module is used for determining the deviation between the actual operation state after regulation and control and an expected target, and feeding back the deviation to the intelligent prediction and decision-making module so as to carry out online self-correction on the self-adaptive hybrid prediction model. According to the method, the operation mechanism of the thermal power generating unit is deeply fused, and high robustness and self-adaptive capability are achieved.

Description

Technical Field

[0001] This invention relates to the field of thermal power unit water temperature control technology, and in particular to an intelligent control system and method for cold-end circulating water temperature of thermal power units. Background Technology

[0002] With the continuous improvement of operating efficiency and energy conservation and emission reduction requirements for thermal power units, precise control of cold-end circulating water temperature has become a key factor in ensuring the efficient and stable operation of the units. The cold-end system of a thermal power unit removes heat from the turbine exhaust steam through circulating cooling water, and its water temperature control directly affects the condenser vacuum, unit thermal efficiency, and overall energy consumption. Driven by both energy structure transformation and intelligent upgrading, the cold-end circulating water temperature control system for thermal power units faces higher requirements in terms of response speed, control accuracy, and adaptability to operating conditions.

[0003] Intelligent control of circulating water temperature at the cold end of thermal power units involves complex characteristics such as multivariate coupling, strong nonlinearity, and dynamic time-varying. It requires comprehensive consideration of multiple factors, including ambient temperature, load changes, cooling tower performance, and circulating water pump operating status, to achieve real-time optimization and stable control of water temperature. While existing control methods have achieved some success in general cooling systems, they face significant limitations in specific thermal power scenarios. For example, some existing solutions rely on high-precision sensor feedback for closed-loop control, which places extremely high demands on the stability of the measuring equipment. In the complex environment of thermal power units with high temperature, high humidity, and strong electromagnetic interference, these methods are susceptible to noise interference, leading to inaccurate control commands. Other methods use fixed parameters or empirical rules for water temperature regulation, lacking the ability to fuse and analyze multi-source operating data, making it difficult to adapt to nonlinear disturbances caused by rapid load fluctuations or seasonal environmental changes. Furthermore, many existing control strategies have not modeled and optimized the dynamic response characteristics and strong multivariate coupling relationships of the cold end system of thermal power units, resulting in control lag, overshoot, or oscillation. This not only affects the vacuum stability of the unit but may also increase the energy consumption of the circulating water pump, weakening the overall energy-saving benefits. Summary of the Invention

[0004] The present invention aims to solve at least one of the problems existing in the prior art, and provides an intelligent control system and method for the cold end circulating water temperature of thermal power units.

[0005] One aspect of the present invention provides an intelligent control system for the cold-end circulating water temperature of a thermal power unit, the intelligent control system for the cold-end circulating water temperature of the thermal power unit comprising: The data acquisition module is used to collect raw data streams of multi-source operating parameters of the cold end system of thermal power units in real time; The data processing and feature extraction module is used to preprocess, filter and denoise, and normalize the original data stream of the multi-source operating parameters, and extract the time-series feature vectors that characterize the dynamic characteristics of the cold end system based on the sliding time window. The intelligent prediction and decision-making module is used to output the predicted value of circulating water temperature within a future set time period based on the time-series feature vector and a pre-trained adaptive hybrid prediction model. Based on the deviation between the predicted value of circulating water temperature and the preset optimal water temperature range, and combined with the current load of the thermal power unit and the back pressure constraint of the condenser, the module generates an optimized control instruction set containing the target frequency of the circulating water pump and the target speed of the cooling tower fan by solving the corresponding dynamic optimization problem. The actuator control module is used to generate and send corresponding control signals to the circulating water pump frequency converter and the cooling tower fan speed control device according to the optimized control instruction set. According to the corresponding control signals, the circulating water pump frequency converter and the cooling tower fan speed control device are driven to perform corresponding actions to coordinately regulate the circulating water flow and the cooling tower fan speed. The system status monitoring and feedback module is used to monitor the operating status of the cold end system of the thermal power unit after regulation in real time, determine the deviation between the actual operating status of the cold end system of the thermal power unit after regulation and the expected target, and feed the deviation back to the intelligent prediction and decision module to perform online self-correction of the model parameters of the adaptive hybrid prediction model.

[0006] Optionally, the data acquisition module is used to acquire raw data streams of multi-source operating parameters of the cold-end system of the thermal power unit in real time, including: The data acquisition module is used for: The ambient temperature is collected using an ambient temperature sensor installed at the air inlet louver of the cooling tower; The unit load signal is obtained using the Modbus / TCP protocol interface of the distributed control system of the thermal power unit. Measure the vacuum level of the condenser using an absolute pressure transmitter; The operating frequency of the circulating water pump and the speed of the cooling tower fan are collected through the feedback port of the frequency converter.

[0007] Optionally, the data processing and feature extraction module is used to perform preprocessing, filtering, noise reduction, and normalization operations on the original data stream of the multi-source operating parameters, including: The data processing and feature extraction module is used for: The 3σ criterion is used to identify and remove outliers from the original data stream of the multi-source operating parameters. For missing data points in the original data stream of the multi-source operating parameters, linear interpolation between two valid data points is used to fill in the missing data points, so as to obtain the preprocessed multi-source operating parameters. The db4 wavelet basis function was selected to perform wavelet transform denoising on the preprocessed multi-source operating parameters, and the high-frequency coefficients were filtered by the soft threshold function.

[0008] Optionally, the adaptive hybrid prediction model includes one input layer, two stacked long short-term memory network layers, one multi-head self-attention mechanism layer, and two fully connected output layers; The two long short-term memory network layers contain 128 and 64 units respectively, with a dropout rate of 0.2; the multi-head self-attention mechanism has 8 heads, and each head has a dimension of 16. During the training process of the adaptive hybrid prediction model, the mean squared error loss function and Adam optimizer are used, the initial learning rate is set to 0.001, the exponential decay strategy is adopted, the batch size is 32, and the training period is 500.

[0009] Optionally, the first objective function of the dynamic optimization problem is defined as: ; in, Let the first objective function of the dynamic optimization problem be... This is the predicted value for circulating water temperature. To preset the optimal water temperature, This refers to the real-time power consumption of the circulating water pump. This is the reference power consumption of the circulating water pump. Real-time vacuum level of the condenser. Set the vacuum level for the condenser. , , These are all weighting coefficients, and their value ranges are as follows: , , ; First objective function The constraints include: the upper limit of the frequency of the circulating water pump. Set as Lower limit of circulating water pump frequency Set as Cooling tower fan speed limit Set as Lower limit of cooling tower fan speed Set as .

[0010] Optionally, the second objective function of the dynamic optimization problem Defined as: ; in, For the comprehensive water quality score, For standard water quality scoring, Water quality weighting coefficient and When the overall water quality score exceeds the water quality safety range, the water quality weighting coefficient is increased.

[0011] Optionally, the circulating water pump frequency converter adopts vector control mode when performing corresponding actions; The cooling tower fan speed control device is driven by a permanent magnet synchronous motor when performing corresponding actions.

[0012] Optionally, the actuator control module is further configured to: Based on the corresponding control signals, the dosing pump and the drain valve are driven to perform corresponding actions to control the frequency of the dosing pump and the opening degree of the drain valve. The dosing pump is a diaphragm metering pump, and the drain valve is an electric regulating valve.

[0013] Optionally, the system status monitoring and feedback module is used for: If the deviation for a preset number of consecutive sampling periods exceeds a preset deviation threshold, the online learning process of the adaptive hybrid prediction model in the intelligent prediction and decision-making module is triggered. The online learning process of the adaptive hybrid prediction model in the intelligent prediction and decision-making module includes: The adaptive hybrid prediction model was incrementally trained using the most recent 1000 sets of sample data, with the learning rate adjusted to 0.1 times its initial value.

[0014] Another aspect of the present invention provides a method for intelligent control of the cold-end circulating water temperature of a thermal power unit, applied to the aforementioned intelligent control system for the cold-end circulating water temperature of a thermal power unit, the method comprising: Step S1: Through the sensor array deployed at key nodes of the cold end system of the thermal power unit, the raw data stream of multi-source operating parameters consisting of ambient temperature, unit load, condenser vacuum, circulating water pump operating frequency, cooling tower outlet water temperature and cooling tower fan speed is periodically collected. Step S2: Perform data cleaning, outlier removal, and wavelet transform denoising on the original data stream of the multi-source operating parameters to obtain the processed multi-source operating parameters. Then, perform maximum and minimum value normalization on the processed multi-source operating parameters to map them to the [0,1] interval to form a standardized data sequence. Step S3: Based on the set sliding time window length and sliding step size, the standardized data sequence is cut into multiple data segments, the statistical characteristics of each data segment are calculated, and the amplitude of the main frequency components is extracted using fast Fourier transform to construct a high-dimensional time series feature vector. Step S4: Input the high-dimensional time-series feature vector into the pre-trained adaptive hybrid prediction model to obtain the circulating water temperature prediction value within the future set time period output by the adaptive hybrid prediction model. Based on the deviation between the circulating water temperature prediction value and the preset optimal water temperature range, and combined with the current load of the thermal power unit and the condenser back pressure constraint, generate an optimized control instruction set containing the target frequency of the circulating water pump and the target speed of the cooling tower fan by solving the corresponding dynamic optimization problem. Step S5: Analyze the optimized control instruction set, generate corresponding control signals through the digital-to-analog converter, and drive the circulating water pump frequency converter and the cooling tower fan speed control device to perform corresponding actions according to the corresponding control signals, so as to coordinately adjust the circulating water flow and the cooling tower fan speed. Step S6: Real-time acquisition of response data of the cold end system of the thermal power unit after the execution of the control action, determination of the root mean square error between the response data and the expected target, and input of the root mean square error as a feedback signal into the adaptive hybrid prediction model to trigger the online gradient update of the internal weight parameters of the adaptive hybrid prediction model.

[0015] This invention provides an intelligent control system and method for the cold-end circulating water temperature of thermal power units that deeply integrates the operating mechanism of thermal power units and possesses strong robustness and adaptability. It breaks through the bottlenecks of existing technologies in terms of accuracy, real-time performance, and adaptability to complex operating conditions, and has the following beneficial effects compared with existing technologies: 1. By constructing an adaptive hybrid prediction model that integrates long short-term memory network and attention mechanism, high-precision, multi-step prediction of the temperature change trend of cold end circulating water in thermal power units is achieved. This effectively overcomes the prediction lag and inaccuracy problems caused by the nonlinearity and time-varying nature of the system in traditional methods, and the average absolute error of prediction can be reduced to within 0.3℃.

[0016] 2. Based on dynamic optimization theory, a real-time decision-making mechanism with multiple objectives of water temperature stability, minimum energy consumption and optimal vacuum degree is constructed. By solving the constrained optimization problem, collaborative control instructions are generated, realizing the linkage optimization of circulating water pump and cooling tower fan. Compared with the existing PID control, the circulating water temperature prediction error of the present invention is reduced by 40%, and while ensuring the optimal vacuum of the condenser, the average operating power consumption of the circulating water pump is reduced by about 8%.

[0017] 3. A complete closed-loop control architecture was designed, which includes data preprocessing, feature extraction, intelligent decision-making, precise execution and online feedback. This significantly improves the system's adaptability and robustness under complex conditions such as rapid load fluctuations and drastic changes in ambient temperature. The overshoot of the control system is reduced to below 2%, and the adjustment time is shortened by about 40%.

[0018] 4. An online self-calibration mechanism based on performance deviation triggering is introduced, which can dynamically adjust the prediction model parameters according to the actual control effect, ensuring the maintenance of control accuracy and performance degradation compensation during long-term system operation, and improving the engineering applicability and life cycle of the method. Attached Figure Description

[0019] One or more embodiments are illustrated by way of example with the corresponding pictures in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0020] Figure 1 This is a schematic diagram of the architecture of an intelligent control system for the cold end circulating water temperature of a thermal power unit, provided as an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details are presented in the various embodiments of the present invention to facilitate a better understanding of the invention. However, the technical solutions claimed in the present invention can be implemented even without these technical details and with various variations and modifications based on the following embodiments. The division of the various embodiments below is for ease of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined with and referenced by each other without contradiction.

[0022] One embodiment of the present invention relates to an intelligent control system for the cold-end circulating water temperature of a thermal power unit, the architecture of which is as follows: Figure 1 As shown, the system includes a data acquisition module, a data processing and feature extraction module, an intelligent prediction and decision-making module, an actuator control module, and a system status monitoring and feedback module. The intelligent control system for the cold-end circulating water temperature of thermal power units provided in this embodiment can be applied to the cold-end system of a 600MW ultra-supercritical coal-fired power unit. The following detailed description uses the operating site of a 600MW ultra-supercritical coal-fired power unit's cold-end system as an example to illustrate the intelligent control system for the cold-end circulating water temperature of thermal power units provided in this embodiment.

[0023] The data acquisition module is used to collect raw data streams of multi-source operating parameters of the cold-end system of thermal power units in real time. These multi-source operating parameters include ambient temperature, unit load, condenser vacuum, circulating water pump operating frequency, cooling tower outlet water temperature, and cooling tower fan speed.

[0024] For example, a data acquisition module is used to acquire raw data streams of multi-source operating parameters of the cold-end system of a thermal power unit in real time, including: The data acquisition module is used to: collect ambient temperature using an ambient temperature sensor installed at the air inlet louvers of the cooling tower; acquire unit load signals using the Modbus / TCP protocol interface of the distributed control system of the thermal power unit; measure condenser vacuum using an absolute pressure transmitter; and collect the operating frequency of the circulating water pump and the speed of the cooling tower fan through the frequency converter feedback port.

[0025] For example, in the operation of the cold-end system of a 600MW ultra-supercritical coal-fired power generating unit, the data acquisition module can accurately measure the air temperature at the cooling tower inlet using a PT100 platinum resistance ambient temperature sensor installed at the louvers of the cooling tower inlet. The measurement accuracy reaches ±0.1 degrees Celsius. The installation position of the ambient temperature sensor is strictly selected 1.5 meters behind the louvers of the cooling tower inlet to avoid the influence of direct sunlight and hot air recirculation. The unit load signal is acquired in real time through the distributed control system standard Modbus / TCP protocol interface of the thermal power unit. The communication cycle is fixed at 1 second, the data format follows the IEEE 754 floating-point standard, and CRC-32 checksum is used during transmission to ensure data integrity. The condenser vacuum degree measurement uses an absolute pressure transmitter with a range covering the entire range of 0kPa-101.325kPa and an accuracy class set to 0.1%FS. The absolute pressure transmitter is installed in the center of the condenser throat, and the length of the pressure tap is controlled within 3 meters to reduce measurement lag. The operating frequency of the circulating water pump and the speed of the cooling tower fan are directly collected through the frequency converter feedback port. The frequency resolution reaches 0.01Hz and the speed resolution reaches 1 revolution per minute. The unit load sampling frequency is maintained at 1Hz, and the sampling interval for parameters such as the operating frequency of the circulating water pump and the speed of the cooling tower fan is set to 500 milliseconds.

[0026] In practical implementation, the data acquisition module continuously collects six types of operating parameters at a frequency of 1Hz: ambient temperature, unit load, condenser vacuum, circulating water pump operating frequency, cooling tower outlet water temperature, and cooling tower fan speed. The data stream for each operating parameter undergoes format conversion before transmission, uniformly converting it to IEEE 754 single-precision floating-point format, and appending a timestamp and CRC checksum to the end of the data packet. Data transmission uses the TCP protocol to ensure reliability, establishing a dual-NIC redundant link. When the primary link fails, it automatically switches to the backup link, with a switching time of less than 200 milliseconds.

[0027] The data processing and feature extraction module is connected to the data acquisition module. It is used to receive the raw data stream of multi-source operating parameters transmitted by the data acquisition module, perform preprocessing, filtering and noise reduction and normalization operations on the raw data stream of multi-source operating parameters, and extract the time-series feature vector representing the dynamic characteristics of the cold end system based on the sliding time window.

[0028] For example, the data processing and feature extraction module is used to preprocess, filter, reduce noise, and normalize the raw data stream of multi-source operating parameters, including: The data processing and feature extraction module is used to: identify and remove outliers from the original data stream of multi-source operating parameters using the 3σ criterion; fill in missing data points in the original data stream of multi-source operating parameters by linear interpolation of two valid data points before and after, and obtain preprocessed multi-source operating parameters; perform wavelet transform denoising on the preprocessed multi-source operating parameters using the db4 wavelet basis function; and filter high-frequency coefficients using a soft threshold function.

[0029] Specifically, when using the 3σ criterion to identify and remove outliers from the raw data stream of multi-source operating parameters, if a data point in the raw data stream deviates from its moving average by more than three times the standard deviation, that data point is automatically marked as an outlier and removed. For missing data points in the raw data stream of multi-source operating parameters due to communication interruptions or other reasons, a linear interpolation algorithm using two consecutive valid data points is used to fill in the missing data points. The interpolation weights are dynamically calculated based on the time distance. The linear interpolation formula is: ,in, The parameters to be interpolated are the operating parameters. for The corresponding interpolation results, The running parameters for the previous valid data point, for The corresponding original data values, The running parameters for the next valid data point, for The corresponding original data values, They belong to the same category of operating parameters.

[0030] Wavelet transform denoising uses the db4 wavelet basis function, with a decomposition level of 5. High-frequency coefficients are filtered using a soft thresholding function, and the threshold is adaptively adjusted according to the noise level to ensure that the effective signal component retention rate is greater than 95%.

[0031] After the noise reduction and filtering are completed, the data enters the maximum and minimum value normalization process. This process linearly maps the values ​​of each operating parameter to a closed interval between 0 and 1. The mapping formula is (x-min) / (max-min), where x represents the value of the operating parameter to be normalized, and min and max are the actual minimum and maximum values ​​of the operating parameter corresponding to x in the historical data of the most recent 24 hours, respectively.

[0032] After preprocessing, filtering, denoising, and normalizing the raw data stream of multi-source operating parameters, the resulting data is used as a standardized data sequence for feature extraction to obtain the corresponding time-series feature vector. When extracting time-series feature vectors representing the dynamic characteristics of the cold-end system based on a sliding time window, the data processing and feature extraction module continuously extracts data segments from the standardized data sequence for feature calculation, based on a set sliding time window (e.g., 300 seconds) and a sliding step size (e.g., 10 seconds). Within each window, i.e., each data segment, the statistical features of six operating parameters are simultaneously calculated, including the arithmetic mean, unbiased variance, third-order central moment skewness coefficient, and fourth-order central moment kurtosis coefficient. Simultaneously, the amplitude of the main frequency components is extracted using Fast Fourier Transform, achieving a frequency resolution of 0.0033Hz. The top five main frequency components are selected to form a frequency domain feature subset. Finally, each window, i.e., each data segment, generates a time-series feature vector containing 30 dimensions. This time-series feature vector is then standardized and stored in a circular buffer with a capacity of 1000 feature vectors, managed using a first-in, first-out (FIFO) strategy.

[0033] The data processing and feature extraction module can be deployed on an industrial server equipped with a Xeon E5 processor and 32GB of memory, with data processing latency controlled within 50 milliseconds. Outlier detection employs a dynamic threshold mechanism, with the threshold updated hourly based on the statistical characteristics of the most recent hour's data. The wavelet denoising soft threshold function uses adaptive threshold calculation, with the threshold calculation formula as follows: Where σ is the noise standard deviation and N is the signal length. The actual minimum and maximum values ​​in the normalization process are updated every 6 hours, excluding historical data points with obvious anomalies during the update.

[0034] The intelligent prediction and decision-making module, connected to the data processing and feature extraction module, receives the time-series feature vector sent by the data processing and feature extraction module. Based on the time-series feature vector, it uses a pre-trained adaptive hybrid prediction model to output the predicted circulating water temperature for a future set period. Based on the deviation between the predicted circulating water temperature and the preset optimal water temperature range, and combined with the current load of the thermal power unit and the condenser back pressure constraint, it generates an optimized control instruction set containing the target frequency of the circulating water pump and the target speed of the cooling tower fan by solving the corresponding dynamic optimization problem.

[0035] For example, the core architecture of the adaptive hybrid prediction model includes one input layer, two stacked long short-term memory (LSM) network layers, one multi-head self-attention mechanism layer, and two fully connected output layers. The dimension of the input layer strictly corresponds to the dimension of the temporal feature vector, which is 30 dimensions. The two LSM layers contain 128 and 64 units respectively, with a dropout rate of 0.2. That is, the first LSM layer has 128 units, the second LSM layer has 64 units, and the dropout rate is uniformly set to 0.2. The multi-head self-attention mechanism has 8 heads, each with a dimension of 16. The attention weights are calculated using a scaled dot product attention mechanism. The two fully connected output layers have 32 and 1 neurons respectively. The adaptive hybrid prediction model ultimately outputs the predicted circulating water temperature for a future time period, such as a 5-minute timescale.

[0036] During the training of the adaptive hybrid prediction model, the mean squared error loss function and Adam optimizer were used. The initial learning rate was set to 0.001, and an exponential decay strategy was adopted (the decay rate decreased by 0.95 times every 100 cycles). The batch size was 32, and the training cycle was 500. The training data of the adaptive hybrid prediction model came from the historical operation database of thermal power units, including historical data of multi-source operating parameters such as ambient temperature, unit load, condenser vacuum, circulating water pump operating frequency, cooling tower outlet water temperature, and cooling tower fan speed. This historical data of multi-source operating parameters was preprocessed, filtered and denoised, and normalized (specifically including 3σ criterion outlier removal, linear interpolation to fill missing data, db4 wavelet basis function denoising, and maximum and minimum value normalization). The actual circulating water temperature at the corresponding time was used as the label to form the training data for the adaptive hybrid prediction model.

[0037] For example, the first objective function of a dynamic optimization problem is defined as: .

[0038] in, Let the first objective function of the dynamic optimization problem be... This is the predicted value for circulating water temperature. To preset the optimal water temperature, This refers to the real-time power consumption of the circulating water pump. This is the reference power consumption of the circulating water pump. Real-time vacuum level of the condenser. Set the vacuum level for the condenser. , , These are all weighting coefficients, and their value ranges are as follows: , , , , , The specific values ​​can be dynamically adjusted based on the current operating mode of the thermal power unit, especially the load range. For example, when the load of the thermal power unit is greater than 80%, , , The possible values ​​are respectively , =0.2, However, when the load of the thermal power unit is less than or equal to 80%, , , The possible values ​​are respectively , =0.3, .

[0039] First objective function The constraints include: the upper limit of the frequency of the circulating water pump. Set as Lower limit of circulating water pump frequency Set as Cooling tower fan speed limit Set as Lower limit of cooling tower fan speed Set as .

[0040] With the first objective function To solve the dynamic optimization problem of the objective function, a sequential quadratic programming algorithm can be used: the iterative initialization starts from the current control command, the convergence tolerance is set to 1e-6, and the maximum number of iterations is set to 100.

[0041] The intelligent prediction and decision-making module can run on an inference server equipped with an NVIDIA Tesla T4 graphics card, with a model inference latency of less than 80 milliseconds. The activation function for the Long Short-Term Memory (LSTM) network layer is the tanh function, and the forget gate bias is initialized to 1.0 to mitigate the vanishing gradient problem. The query matrix, key matrix, and value matrix of the multi-head self-attention mechanism layer are all 16×16 in dimension, and the attention score is calculated using scaled dot product attention with a scaling factor of [value missing]. The fully connected output layer uses the ReLU activation function, while the final output layer of the adaptive hybrid prediction model uses a linear activation function.

[0042] The actuator control module, connected to the intelligent prediction and decision-making module, generates and sends corresponding control signals to the circulating water pump frequency converter and the cooling tower fan speed control device according to the optimized control instruction set. Based on the corresponding control signals, it drives the circulating water pump frequency converter and the cooling tower fan speed control device to perform corresponding actions to coordinately regulate the circulating water flow and the cooling tower fan speed.

[0043] Specifically, the actuator control module parses the optimized control instruction set into corresponding control signals. Specifically, it can generate 4mA-20mA analog control signals corresponding to the optimized control instruction set as the corresponding control signals through a digital-to-analog converter.

[0044] For example, the frequency converter of the circulating water pump adopts vector control mode when performing corresponding actions, with a frequency adjustment resolution of 0.01Hz, a response time of less than 100 milliseconds, and an acceleration slope set to 5Hz per second.

[0045] The cooling tower fan speed control device uses a permanent magnet synchronous motor to drive the corresponding actions, with a speed control accuracy of ±1 revolution per minute. The response time to accelerate to the target speed is less than 5 seconds, and the start-up curve adopts an S-shaped acceleration and deceleration mode.

[0046] The actuator control module can use shielded twisted-pair cables to transmit the corresponding control signals to the circulating water pump frequency converter and the cooling tower fan speed control device respectively. The signal grounding is set independently to effectively suppress common-mode interference.

[0047] The actuator control module adopts a dual-CPU redundant architecture, with the main CPU executing the control algorithm and the backup CPU monitoring the main CPU's status in real time. The analog output card uses 16-bit resolution and has an output refresh cycle of 100 milliseconds. Control signals are amplitude-limited before output to ensure they do not exceed the actuator's safe operating range. The frequency variation rate of the circulating water pump is limited to ±2Hz per second, and the speed variation rate of the cooling tower fan is limited to ±50 revolutions per minute.

[0048] The system status monitoring and feedback module is used to monitor the operating status of the cold end system of the thermal power unit after regulation in real time, determine the deviation between the actual operating status of the cold end system of the thermal power unit after regulation and the expected target, and feed the deviation back to the intelligent prediction and decision module to perform online self-correction of the model parameters of the adaptive hybrid prediction model.

[0049] Specifically, when the system status monitoring and feedback module monitors the operating status of the cold end system of the thermal power unit in real time after the control action is executed, it can collect the response data of the cold end system of the thermal power unit in real time, including the actual circulating water temperature, the change in condenser vacuum, and the power consumption of the circulating water pump.

[0050] The deviation between the actual operating state of the cold-end system of the thermal power unit after regulation and the expected target is the root mean square error between the actual operating state of the cold-end system of the thermal power unit after regulation and the expected target. Specifically, it can be selected as the root mean square error between the actual circulating water temperature after regulation and the preset optimal water temperature.

[0051] For example, the system status monitoring and feedback module is used to: trigger the online learning process of the adaptive hybrid prediction model in the intelligent prediction and decision-making module if the deviation for a consecutive preset number of sampling periods exceeds a preset deviation threshold. For instance, when the deviation is the root mean square error between the actual circulating water temperature after regulation and the preset optimal water temperature, the preset deviation threshold can be set to 0.5 degrees Celsius. When the system status monitoring and feedback module monitors the operating status of the cold end system of the thermal power unit after regulation in real time, the sampling period can be set to 10 seconds. When the deviation for a consecutive preset number of sampling periods, such as 3 periods, exceeds the preset deviation threshold, the online learning process of the adaptive hybrid prediction model in the intelligent prediction and decision-making module is triggered.

[0052] For example, the online learning process of the adaptive hybrid prediction model in the intelligent prediction and decision-making module includes: incrementally training the adaptive hybrid prediction model using the most recent 1000 sets of sample data, and adjusting the learning rate to 0.1 times its initial value.

[0053] For example, the intelligent prediction and decision-making module can incrementally train the adaptive hybrid prediction model using the most recent 1000 sets of sample data. The parameters of the adaptive hybrid prediction model are updated using a momentum-driven stochastic gradient descent algorithm, with the momentum coefficient set to 0.9 and the gradient clipping threshold set to 1.0. The sample data are weighted in reverse chronological order, with the weight of the most recent 100 sets of samples set to 0.015 and the weight of the earliest 100 sets of samples set to 0.005. The learning rate is adjusted to 0.1 times its initial value, and the training cycle is shortened to 50 iterations.

[0054] The system status monitoring and feedback module employs a sliding window mechanism when calculating the root mean square error, with a window length set to 10 sampling periods. The feedback signal, which triggers the online learning process of the adaptive hybrid prediction model in the intelligent prediction and decision module, undergoes low-pass filtering before transmission to the intelligent prediction and decision module, with a cutoff frequency set to 0.1Hz to avoid frequent model updates caused by high-frequency disturbances. During the online learning process of the adaptive hybrid prediction model, training data uses time-weighted sampling, and the weight of recent data can be set to twice the weight of long-term data. Under operating conditions of 50%-100% unit load and ambient temperature of 15℃-35℃, the average operating power consumption of the circulating water pump is reduced by approximately 8%.

[0055] Another embodiment of the present invention relates to an intelligent control method for the cold end circulating water temperature of a thermal power unit, which is applied to the intelligent control system for the cold end circulating water temperature of the thermal power unit described in the above embodiment. The intelligent control method for the cold end circulating water temperature of the thermal power unit includes the following steps S1 to S6.

[0056] Step S1: Through the sensor array deployed at key nodes of the cold end system of the thermal power unit, the raw data stream of multi-source operating parameters consisting of ambient temperature, unit load, condenser vacuum, circulating water pump operating frequency, cooling tower outlet water temperature and cooling tower fan speed is periodically collected. Step S2 involves cleaning the original data stream of multi-source operating parameters, removing outliers, and performing wavelet transform denoising to obtain the processed multi-source operating parameters. The processed multi-source operating parameters are then normalized to their maximum and minimum values ​​and mapped to the [0,1] interval to form a standardized data sequence. Step S3: Based on the set sliding time window length and sliding step size, the standardized data sequence is cut into multiple data segments, the statistical characteristics of each data segment are calculated, and the amplitude of the main frequency components is extracted using fast Fourier transform to construct a high-dimensional time series feature vector. Step S4: Input the high-dimensional time series feature vector into the pre-trained adaptive hybrid prediction model to obtain the circulating water temperature prediction value within the future set time period output by the adaptive hybrid prediction model. Based on the deviation between the circulating water temperature prediction value and the preset optimal water temperature range, and combined with the current load of the thermal power unit and the condenser back pressure constraint, generate an optimized control instruction set containing the target frequency of the circulating water pump and the target speed of the cooling tower fan by solving the corresponding dynamic optimization problem. Step S5: Analyze and optimize the control instruction set, generate corresponding control signals through digital-to-analog converter, and drive the circulating water pump frequency converter and cooling tower fan speed control device to perform corresponding actions according to the corresponding control signals, so as to coordinate the adjustment of circulating water flow and cooling tower fan speed. Step S6: Real-time acquisition of response data of the cold end system of the thermal power unit after the control action is executed, determination of the root mean square error between the response data and the expected target, and input of the root mean square error as a feedback signal into the adaptive hybrid prediction model to trigger the online gradient update of the internal weight parameters of the adaptive hybrid prediction model.

[0057] This invention provides an intelligent control system and method for the cold-end circulating water temperature of thermal power units, deeply integrating the operating mechanism of thermal power units and possessing strong robustness and adaptability. It overcomes the bottlenecks of existing technologies in terms of accuracy, real-time performance, and adaptability to complex operating conditions. By constructing an adaptive hybrid prediction model integrating a long short-term memory network and an attention mechanism, it achieves high-precision, multi-step prediction of the cold-end circulating water temperature change trend of thermal power units, effectively overcoming the prediction lag and inaccuracy problems caused by the nonlinearity and time-varying nature of traditional methods. The average absolute error of the prediction can be reduced to within 0.3℃. Based on dynamic optimization theory, a real-time decision-making mechanism with multiple objectives of water temperature stability, minimum energy consumption, and optimal vacuum degree is constructed. By solving a constrained optimization problem, collaborative control commands are generated, realizing the control of the circulating water pump and... Compared with existing PID control, the linkage optimization of cooling tower fans in this invention reduces the circulating water temperature prediction error by 40%, and reduces the average operating power consumption of the circulating water pump by about 8% while ensuring the optimal vacuum of the condenser. A complete closed-loop control architecture including data preprocessing, feature extraction, intelligent decision-making, precise execution, and online feedback is designed, which significantly improves the system's adaptability and robustness under complex conditions such as rapid load fluctuations and drastic changes in ambient temperature. The overshoot of the control system is reduced to below 2%, and the settling time is shortened by about 40%. An online self-correction mechanism based on performance deviation triggering is introduced, which can dynamically adjust the prediction model parameters according to the actual control effect, ensuring the maintenance of control accuracy and performance degradation compensation during long-term system operation, and improving the engineering practicality and life cycle of the method.

[0058] The following section optimizes the intelligent control system for cold-end circulating water temperature of thermal power units provided in the above implementation method, taking into account the application scenario of a 300MW subcritical coal-fired power generating unit.

[0059] The data acquisition module is equipped with circulating water quality monitoring sensors, including a pH sensor, a conductivity sensor, and a turbidity sensor, for real-time measurement of water pH, conductivity, and turbidity parameters. The pH sensor uses a glass electrode, with a measurement range of 2-12 and an accuracy of 0.1 pH. The conductivity sensor uses a four-electrode design, with a range of 0-2000 μS / cm and an accuracy of 1% FS. The turbidity sensor uses a diffused light design, with a range of 0 NTU-100 NTU and an accuracy of 2% FS. These sensors sample at a frequency of 0.5 Hz and are connected to the data acquisition system via a PROFIBUS-DP bus.

[0060] The data processing and feature extraction module adds water quality-related feature extraction to the existing feature extraction, including pH value change gradient, conductivity moving average, and turbidity peak count. The sliding time window length is adjusted to 600 seconds to adapt to the slower dynamic characteristics of subcritical units, while the sliding step size remains unchanged at 10 seconds. The feature vector dimension is expanded to 36 dimensions, and new features include the one-hour pH value change rate, the three-hour moving average of conductivity, and the statistics of abnormal turbidity fluctuations.

[0061] The network structure of the adaptive hybrid prediction model in the intelligent prediction and decision-making module has been adjusted to a 36-dimensional input layer. The number of units in the first long short-term memory network layer has been reduced to 96 to reduce computational complexity, while the number of units in the second long short-term memory network layer remains unchanged at 64. The number of heads in the multi-head attention mechanism has been adjusted to 4, and the dimension of each head has been increased to 24. The number of neurons in the two fully connected output layers has been adjusted to 24 and 1, respectively, to adapt to changes in feature dimensions. The batch size for training the adaptive hybrid prediction model has been increased to 64, and the training cycle has been extended to 800 to accommodate the more complex dynamic characteristics of subcritical units.

[0062] For example, in the intelligent prediction and decision-making module, the second objective function of the dynamic optimization problem Defined as: ; in, For the comprehensive water quality score, For standard water quality scoring, Water quality weighting coefficient and When the comprehensive water quality score exceeds the safe water quality range, the water quality weight coefficient is increased. In the constraints of the second objective function, the lower limit of the circulating water pump frequency is adjusted to 25Hz, and the lower limit of the cooling tower fan speed is adjusted to 250 rpm, to adapt to the lower load operation requirements of the subcritical unit.

[0063] The actuator control module adds water quality regulation device control functions, including dosing pump frequency control and drain valve opening control. Specifically, the actuator control module is also used to drive the dosing pump and drain valve to perform corresponding actions based on the corresponding control signals, thereby controlling the dosing pump frequency and the drain valve opening. The dosing pump is a diaphragm metering pump with a control signal of 4 mA-20 mA, a frequency adjustment range of 0 Hz-100 Hz, and a resolution of 0.1 Hz. The drain valve is an electrically adjustable valve with an opening control accuracy of 1% and a full stroke time of 30 seconds. A water quality priority judgment is added to the control logic; when water quality parameters exceed the safe range, the water quality weight coefficient is automatically increased.

[0064] The system status monitoring and feedback module expands its monitoring indicators, adding scaling trend prediction and corrosion risk assessment. Scaling trend is calculated in real-time based on the Langerile index, while corrosion risk is assessed based on the Larssen index. Water quality control effectiveness assessment is added to the feedback signal; when water quality parameters continuously deviate from the set value for more than a preset time (e.g., 2 hours), a chemical dosing strategy optimization process is triggered. The online learning dataset for the adaptive hybrid prediction model now includes correlation data between water quality parameters and control effects, and the weights of water quality-related features during training are increased to 1.5 times the original weights.

[0065] The intelligent control system for cold-end circulating water temperature in thermal power units can also be equipped with a safety interlock protection mechanism. When the condenser vacuum level is below 85 kPa, it automatically switches to manual control mode; when the circulating water temperature exceeds 42 degrees Celsius, it automatically increases the cooling tower fan speed to its maximum value. Detailed logs of protection actions are recorded, including trigger time, parameter values, action type, and recovery time, and the logs are retained for at least 90 days. The intelligent control system for cold-end circulating water temperature in thermal power units can be configured with dual redundant power supplies. When the main power supply fails, the backup power supply automatically switches on, with a switching time of less than 10 milliseconds.

[0066] Those skilled in the art will understand that the above embodiments are specific implementations of the present invention, and in practical applications, various changes can be made in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A smart temperature control system for cold-end circulating water in a thermal power unit, characterized in that, The intelligent control system for cold-end circulating water temperature of the thermal power unit includes: The data acquisition module is used to collect raw data streams of multi-source operating parameters of the cold end system of thermal power units in real time; The data processing and feature extraction module is used to preprocess, filter and denoise, and normalize the original data stream of the multi-source operating parameters, and extract the time-series feature vectors that characterize the dynamic characteristics of the cold end system based on the sliding time window. The intelligent prediction and decision-making module is used to output the predicted value of circulating water temperature within a future set time period based on the time-series feature vector and a pre-trained adaptive hybrid prediction model. Based on the deviation between the predicted value of circulating water temperature and the preset optimal water temperature range, and combined with the current load of the thermal power unit and the back pressure constraint of the condenser, the module generates an optimized control instruction set containing the target frequency of the circulating water pump and the target speed of the cooling tower fan by solving the corresponding dynamic optimization problem. The actuator control module is used to generate and send corresponding control signals to the circulating water pump frequency converter and the cooling tower fan speed control device according to the optimized control instruction set. According to the corresponding control signals, the circulating water pump frequency converter and the cooling tower fan speed control device are driven to perform corresponding actions to coordinately regulate the circulating water flow and the cooling tower fan speed. The system status monitoring and feedback module is used to monitor the operating status of the cold end system of the thermal power unit after regulation in real time, determine the deviation between the actual operating status of the cold end system of the thermal power unit after regulation and the expected target, and feed the deviation back to the intelligent prediction and decision module to perform online self-correction of the model parameters of the adaptive hybrid prediction model.

2. The intelligent temperature control system for cold-end circulating water of thermal power units according to claim 1, characterized in that, The data acquisition module is used to acquire raw data streams of multi-source operating parameters of the cold-end system of thermal power units in real time, including: The data acquisition module is used for: The ambient temperature is collected using an ambient temperature sensor installed at the air inlet louver of the cooling tower; The load signal of the thermal power unit is obtained by using the Modbus / TCP protocol interface of the distributed control system of the thermal power unit. Measure the vacuum level of the condenser using an absolute pressure transmitter; The operating frequency of the circulating water pump and the speed of the cooling tower fan are collected through the feedback port of the frequency converter.

3. The intelligent temperature control system for cold-end circulating water of thermal power units according to claim 1, characterized in that, The data processing and feature extraction module is used to preprocess, filter, reduce noise, and normalize the original data stream of the multi-source operating parameters, including: The data processing and feature extraction module is used for: The 3σ criterion is used to identify and remove outliers from the original data stream of the multi-source operating parameters. For missing data points in the original data stream of the multi-source operating parameters, linear interpolation between two valid data points is used to fill in the missing data points, so as to obtain the preprocessed multi-source operating parameters. The db4 wavelet basis function was selected to perform wavelet transform denoising on the preprocessed multi-source operating parameters, and the high-frequency coefficients were filtered by the soft threshold function.

4. The intelligent temperature control system for cold-end circulating water of thermal power units according to claim 1, characterized in that, The adaptive hybrid prediction model comprises one input layer, two stacked long short-term memory network layers, one multi-head self-attention mechanism layer, and two fully connected output layers. The two long short-term memory network layers contain 128 and 64 units respectively, with a dropout rate of 0.2; the multi-head self-attention mechanism has 8 heads, and each head has a dimension of 16. During the training process of the adaptive hybrid prediction model, the mean squared error loss function and Adam optimizer are used, the initial learning rate is set to 0.001, the exponential decay strategy is adopted, the batch size is 32, and the training period is 500.

5. The intelligent temperature control system for cold-end circulating water of thermal power units according to claim 1, characterized in that, The first objective function of a dynamic optimization problem is defined as: ； in, Let the first objective function of the dynamic optimization problem be... This is the predicted value for circulating water temperature. To preset the optimal water temperature, This refers to the real-time power consumption of the circulating water pump. This is the reference power consumption of the circulating water pump. Real-time vacuum level of the condenser. Set the vacuum level for the condenser. , , These are all weighting coefficients, and their value ranges are as follows: , , ; First objective function The constraints include: the upper limit of the frequency of the circulating water pump. Set as Lower limit of circulating water pump frequency Set as Cooling tower fan speed limit Set as Lower limit of cooling tower fan speed Set as .

6. The intelligent control system for cold-end circulating water temperature of thermal power units according to claim 5, characterized in that, The second objective function of the dynamic optimization problem Defined as: ； in, For the comprehensive water quality score, For standard water quality scoring, Water quality weighting coefficient and When the overall water quality score exceeds the water quality safety range, the water quality weighting coefficient is increased.

7. The intelligent control system for cold-end circulating water temperature of thermal power units according to claim 1, characterized in that, The frequency converter of the circulating water pump adopts vector control mode when performing corresponding actions; The cooling tower fan speed control device is driven by a permanent magnet synchronous motor when performing corresponding actions.

8. The intelligent control system for cold-end circulating water temperature of thermal power units according to claim 1, characterized in that, The actuator control module is also used for: Based on the corresponding control signals, the dosing pump and the drain valve are driven to perform corresponding actions to control the frequency of the dosing pump and the opening degree of the drain valve. The dosing pump is a diaphragm metering pump, and the drain valve is an electric regulating valve.

9. The intelligent control system for cold-end circulating water temperature of thermal power units according to claim 1, characterized in that, The system status monitoring and feedback module is used for: If the deviation for a preset number of consecutive sampling periods exceeds a preset deviation threshold, the online learning process of the adaptive hybrid prediction model in the intelligent prediction and decision-making module is triggered. The online learning process of the adaptive hybrid prediction model in the intelligent prediction and decision-making module includes: The adaptive hybrid prediction model was incrementally trained using the most recent 1000 sets of sample data, with the learning rate adjusted to 0.1 times its initial value.

10. A method for intelligent control of cold-end circulating water temperature in a thermal power unit, characterized in that, The intelligent control system for cold-end circulating water temperature of thermal power units, applicable to any one of claims 1-9, comprises the following methods: Step S1: Through the sensor array deployed at key nodes of the cold end system of the thermal power unit, the raw data stream of multi-source operating parameters consisting of ambient temperature, unit load, condenser vacuum, circulating water pump operating frequency, cooling tower outlet water temperature and cooling tower fan speed is periodically collected. Step S2: Perform data cleaning, outlier removal, and wavelet transform denoising on the original data stream of the multi-source operating parameters to obtain the processed multi-source operating parameters. Then, perform maximum and minimum value normalization on the processed multi-source operating parameters to map them to the [0,1] interval to form a standardized data sequence. Step S3: Based on the set sliding time window length and sliding step size, the standardized data sequence is cut into multiple data segments, the statistical characteristics of each data segment are calculated, and the amplitude of the main frequency components is extracted using fast Fourier transform to construct a high-dimensional time series feature vector. Step S4: Input the high-dimensional time-series feature vector into the pre-trained adaptive hybrid prediction model to obtain the circulating water temperature prediction value within the future set time period output by the adaptive hybrid prediction model. Based on the deviation between the circulating water temperature prediction value and the preset optimal water temperature range, and combined with the current load of the thermal power unit and the condenser back pressure constraint, generate an optimized control instruction set containing the target frequency of the circulating water pump and the target speed of the cooling tower fan by solving the corresponding dynamic optimization problem. Step S5: Analyze the optimized control instruction set, generate corresponding control signals through the digital-to-analog converter, and drive the circulating water pump frequency converter and the cooling tower fan speed control device to perform corresponding actions according to the corresponding control signals, so as to coordinately adjust the circulating water flow and the cooling tower fan speed. Step S6: Real-time acquisition of response data of the cold end system of the thermal power unit after the execution of the control action, determination of the root mean square error between the response data and the expected target, and input of the root mean square error as a feedback signal into the adaptive hybrid prediction model to trigger the online gradient update of the internal weight parameters of the adaptive hybrid prediction model.

Images