Coordinated control system for generator set

Abstract

The invention relates to the technical field of electric power automation control, and discloses a coordinated control system for a generator set, which adopts an intelligent DCS integrated design architecture and performs real-time data interaction with a set decentralized control system through hard wiring or a communication protocol. According to the system, through deep fusion of a model prediction control algorithm and intelligent DCS integrated design, a boiler-steam turbine dynamic coordination model is constructed, and a control instruction is generated in real time in combination with a rolling optimization strategy. By means of the design, the main steam pressure deviation is controlled within + / -0.2 MPa, and the load response rate reaches 1.5% Pe / s and is improved by more than 40% compared with a traditional system. Particularly, under the working condition of deep peak regulation, by means of decoupling control over a fuel-air supply-water supply loop through a neural network algorithm, main steam temperature fluctuation (within + / -5 DEG C) and reheater spray desuperheating water quantity sudden change (the change rate is smaller than or equal to 10 t / min) are effectively restrained, the control quality of a unit under the conditions of variable loads and variable coal quality is remarkably improved, and the control efficiency of the unit is improved. And the high-standard requirement of the power grid on the AGC performance is met.

Description

Technical Field

[0001] This invention relates to the field of power automation control technology, specifically a coordinated control system for generator sets. Background Technology

[0002] With the increasing proportion of new energy sources and the surge in grid regulation demands, traditional generator set coordination and control systems have exposed multiple technical bottlenecks:

[0003] 1. Insufficient control accuracy: Existing PID control strategies are unable to cope with the large lag characteristics of boilers. The dynamic deviation of main steam pressure often reaches ±0.5MPa, and the load response rate is only 0.8%Pe / s, which cannot meet the requirements of power grid AGC assessment.

[0004] 2. Abnormal handling relies on manual intervention: Disturbance alarms account for over 70%, and the diagnosis of critical faults takes more than 30 minutes, leading to frequent unplanned downtime;

[0005] 3. Poor adaptability to operating conditions: During deep peak shaving, the control loop oscillates frequently, and a 10% change in the calorific value of coal can cause combustion instability;

[0006] 4. Low level of automation: Regular equipment switching and parameter adjustment require manual operation, which increases the workload of operators and is prone to human error.

[0007] While existing technologies have attempted to address model mismatch and operational adaptability issues by adding filtering stages or introducing simple feedforward compensation, they have failed to fundamentally resolve these problems. Against this backdrop, this invention, based on an integrated intelligent DCS architecture, integrates advanced technologies such as predictive control and deep reinforcement learning, aiming to construct a coordinated control system for generator sets with self-diagnosis, self-optimization, and self-adaptive capabilities. Summary of the Invention

[0008] The purpose of this invention is to provide a coordinated control system for generator sets to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a coordinated control system for generator sets, wherein the system adopts an intelligent DCS integrated design architecture and interacts with the distributed control system of the generator set in real time through hard wiring or communication protocols. The system includes a boiler-generator coordinated predictive control optimization module, a typical equipment anomaly handling module, and an advanced shift operator decision-making module.

[0010] The coordinated predictive control optimization module constructs a dynamic coordination model of boiler fuel-air supply-feedwater and turbine regulation based on a neural network algorithm;

[0011] The system performance acceptance follows the DL / T 657-2015 standard, and the specific test items include:

[0012] Load variation test: Within the load range of 70% to 100%, the load is increased or decreased at a rate of 3% Pe / min, and the main steam pressure deviation ΔP is recorded. max ≤±0.2MPa, load response deviation ΔN max ≤±2%Pe;

[0013] Load shedding test: When the unit load is reduced from 100% to 50%, ensure that the boiler does not overpressure, the turbine does not overspeed, and the reheater safety valve does not activate.

[0014] Disturbance test: When ±10% step disturbances are applied to the water supply flow rate, fuel quantity, and air supply volume, the settling time T of each main controlled variable is measured. s ≤120 seconds, overshoot σ≤5%;

[0015] All test data is automatically collected by the system's built-in test module and an acceptance report that meets the power grid requirements is generated. The report includes time-domain waveform diagrams, performance index statistics tables, and compliance conclusions.

[0016] The system possesses full lifecycle self-optimization capabilities, and its self-learning mechanism comprises three dimensions:

[0017] Model self-calibration: The boiler dynamic model is updated daily at midnight using the operating data of the previous 24 hours. Cross-validation is used before and after the model update to ensure that the prediction error ε ≤ 3%.

[0018] Rule self-evolution: monthly cluster analysis of the expert rule base, merging processing strategies under similar working conditions, and adding new rules for predicting equipment degradation trends;

[0019] Parameter self-optimization: Global parameter optimization is performed every quarter, with the optimization target defined as:

[0020]

[0021] Where a, b, and c are economic weight coefficients, and T is the optimization period; the optimization process uses a genetic algorithm with a population size of N=50, a mutation probability of Pm=0.1, and the termination condition is that the improvement of the objective function value is <1% for 10 consecutive generations.

[0022] The system has a self-diagnostic function, which can monitor the status of I / O cards, the integrity of communication links and the accuracy of algorithm models in real time. The diagnostic results are stored in the form of logs and trigger a hierarchical alarm mechanism. When the optimization function is abnormal or the DCS side triggers the RB condition, the system immediately freezes the current control command and executes the seamless switching logic, returns control to the DCS and outputs an alarm signal to ensure that the unit smoothly transitions to the basic automatic control mode.

[0023] Preferably, the specific process of constructing a dynamic coordination model for boiler fuel-air supply-feedwater and turbine regulation based on a neural network algorithm includes the following:

[0024] Collect a large amount of historical data during the operation of the boiler and steam turbine, including parameters such as main steam pressure, power, feedwater flow, as well as corresponding fuel quantity, air supply, feedwater quantity and steam turbine regulation parameters;

[0025] The collected data is cleaned to remove outliers and noisy data. At the same time, the data is normalized to convert data of different ranges and magnitudes to the [0,1] or [-1,1] interval in order to improve the efficiency and stability of neural network training.

[0026] According to the problem description, the input layer nodes include the main steam pressure deviation ΔP, power deviation ΔN, and feedwater flow deviation ΔW. Other relevant operating parameters can also be considered as inputs to more comprehensively reflect the system status. The formula is: X(t)=[ΔP(t),ΔN(t),ΔW(t),...], where X(t) is the input vector at time t. Each deviation is collected in real time by sensors and normalized to the interval [0,1] or [-1,1].

[0027] Based on the complexity of the problem and the characteristics of the data, select an appropriate number of hidden layers and nodes. The hidden layers achieve feature extraction through weighted summation and activation functions, using the formula: h(t)=f(W1·x(t)+b1), where W1 is the weight matrix from the input layer to the hidden layer, b1 is the bias vector of the hidden layer, and f(·) is the activation function.

[0028] The number of output layer nodes corresponds to the dimension of the control command. It is used to control the relevant parameters of boiler fuel, air supply, water supply and turbine regulation. The control quantity is generated by linear transformation, and the formula is: u(t)=W2·h(t)+b2, where u(t) is the control command vector at time t, and W2 and b2 are the weights and biases from the hidden layer to the output layer, respectively.

[0029] Preferably, the specific process of constructing a dynamic coordination model for boiler fuel-air supply-feedwater and turbine regulation based on a neural network algorithm further includes:

[0030] The preprocessed data is divided into training set, validation set and test set. The training set is used for learning the model's parameters, the validation set is used to tune the model's hyperparameters, and the test set is used to evaluate the model's generalization ability.

[0031] The backpropagation algorithm or the Levenberg-Marquardt algorithm is used to continuously adjust the weights and thresholds of the neural network to minimize the error between the model's output and the expected output.

[0032] The neural network is trained using the training set data, and the network weights and thresholds are iteratively updated until the preset training stopping conditions are met, such as reaching the maximum number of iterations or the error being less than the set threshold.

[0033] Preferably, the specific process of constructing a dynamic coordination model for boiler fuel-air supply-feedwater and turbine regulation based on a neural network algorithm further includes:

[0034] The trained neural network model is used to predict the future operating state of the system, and parameters such as main steam pressure, power and feedwater flow rate are predicted for a period of time in the future based on the input data at the current moment.

[0035] Within each control cycle, based on the predicted system state and combined with the system constraints u min ≤u(t)≤u max The optimal control command u(t) is solved by optimization algorithms (such as optimization algorithms in model predictive control) so that the system performance indicators (such as minimizing the main steam pressure deviation and power deviation) can be optimized.

[0036] As the system's operating status changes in real time, the input data is continuously updated, and the prediction and optimization process is repeated to achieve dynamic coordinated control.

[0037] The trained model is evaluated using test set data, and performance metrics such as mean squared error (MSE) and mean absolute error (MAE) are calculated to assess the model's prediction accuracy and control performance.

[0038] Based on the evaluation results, analyze the problems and shortcomings of the model, such as unreasonable model structure and insufficient training data, and improve and optimize the model, such as adjusting the network structure, increasing training data, and improving the training algorithm, in order to improve the performance and reliability of the model.

[0039] Preferably, the furnace-machine coordinated predictive control optimization module adopts a model predictive control algorithm, and its objective function is defined as:

[0040]

[0041] Where y is the set of controlled variables including main steam pressure, unit power, and intermediate point temperature; u is the set of control variables including fuel quantity, feedwater flow rate, and turbine valve opening; Q and R are weight matrices; and α and β are adjustment coefficients. The algorithm identifies the boiler dynamic model parameters online using the recursive least squares method, and the model update period ΔT satisfies:

[0042]

[0043] Ensure that the model's prediction accuracy continuously meets the requirements of the DL / T 657-2015 standard regarding load response rate ≥1.5% Pe / s and dynamic pressure deviation ≤±0.3MPa.

[0044] Preferably, the typical equipment anomaly handling module has a built-in expert rule base and fuzzy inference engine. When the temperature of the blower bearing is detected to exceed the threshold T, crit Or the inlet pressure of the water pump is lower than P min When this happens, the system automatically executes the preset protection logic:

[0045] In case of a blower malfunction, the module calculates the cross-limit coefficient K between air and coal. air =f(T) bearing ), dynamically adjust fuel quantity commands;

[0046] In case of water pump malfunction, the module triggers the RB preset function to reduce the target load at a preset rate while maintaining combustion stability.

[0047] During the anomaly handling process, the system compensates for the disturbances in the relevant loops through a multivariate decoupling algorithm to ensure that the main steam temperature fluctuation range is controlled within ±5℃ and the change rate of the reheater spray desuperheating water volume does not exceed 10t / min.

[0048] Preferably, the advanced operator decision-making module integrates a deep reinforcement learning network. The network input state space includes unit load rate, coal calorific value, ambient temperature, and equipment health index, and the output is a suggested operation sequence. The network training adopts a proximal policy optimization algorithm, and the reward function is designed as follows:

[0049]

[0050] Where γ1, γ2, and γ3 are weighting coefficients, C fuel C represents the actual fuel consumption. opt The module generates an operation analysis report every 24 hours, which includes control quality assessment, equipment deterioration warning and parameter optimization suggestions, based on the theoretical optimal fuel quantity.

[0051] Preferably, the system has a parameter self-adjustment function, and its self-learning mechanism includes two levels: the basic level uses a least squares support vector machine to establish a mapping relationship between control parameters and unit operating conditions, and the mapping function satisfies:

[0052]

[0053] Where K(x,x) i ) is the kernel function, α i b are model parameters; the enhancement layer dynamically adjusts the weights of the mapping function based on the reinforcement learning algorithm, and the reward signal is defined as the control loop settling time T. s Linear combination with overshoot σ:

[0054] r = -(w1T) s +w2σ)

[0055] Among them, w1 and w2 are adjustable coefficients; when the unit experiences deep peak shaving conditions or the fuel calorific value changes by more than 10%, the system automatically triggers the parameter resetting process to ensure that the attenuation rate φ of the main steam pressure control loop is maintained between 0.7 and 0.9, and the load response delay time τ ≤ 5 seconds.

[0056] Preferably, the system's interface with the DCS employs a dual-redundant communication protocol, with a data transmission cycle set to 200ms. Key signals such as actual unit load, main steam pressure, and valve opening are handled with a 2-out-of-3 redundancy scheme. When a communication interruption or invalid data is detected, the system immediately switches to the backup link and retains the last valid control command, while simultaneously starting a counter to record the interruption duration t. fault If t fault If the time is ≥3 seconds, a control handover operation is executed; a soft limiting module is set in the interface logic to optimize the fuel quantity command u output by the control system. fuel Dynamic limiting is implemented, and the limiting range is calculated based on the boiler combustion stability:

[0057]

[0058] U0 is the fuel quantity command under the DCS basic automatic control.

[0059] This invention provides a coordinated control system for generator sets. It has the following beneficial effects:

[0060] 1. This system deeply integrates model predictive control algorithms with intelligent DCS integrated design to construct a dynamic coordination model of the boiler-turbine system, and generates control commands in real time using a rolling optimization strategy. This design keeps the main steam pressure deviation within ±0.2MPa and achieves a load response rate of 1.5%Pe / s, an improvement of over 40% compared to traditional systems. Especially under deep peak shaving conditions, the decoupling control of the fuel-feedwater-air circuit through neural network algorithms effectively suppresses main steam temperature fluctuations (within ±5℃) and sudden changes in reheater spray desuperheating water volume (change rate ≤10t / min), significantly improving the control quality of the unit under varying load and coal quality conditions, and meeting the high standards of the power grid for AGC performance.

[0061] 2. A typical equipment anomaly handling module integrating an expert rule base and a fuzzy inference engine enables real-time monitoring and automatic protection of 12 types of critical equipment, such as blower bearing overheating and feedwater pump inlet pressure anomalies. When an anomaly is detected, the system can execute preset protection logic within 200ms, dynamically adjust fuel quantity commands, and trigger RB preset functions, reducing response time by 90% compared to manual intervention. Combined with a dual-redundant communication protocol and a soft-limiting module, even in the event of communication interruption or invalid data, the system can still retain the last valid control command and switch to the backup link, ensuring a smooth transfer of control. This design improves equipment fault diagnosis efficiency by 65%, reduces disturbance alarms by 80%, and reduces unplanned unit shutdowns by more than 50%.

[0062] 3. Through deep reinforcement learning networks and parameter self-adjustment mechanisms, the system achieves continuous evolution of control strategies. Daily updates to the boiler dynamic model predict errors are controlled within 3%, and quarterly global parameter optimization reduces fuel consumption by 2-4%. A specially designed reward function incorporates load response time, pressure deviation, and fuel economy into a unified optimization framework. Combined with acceptance test data from the DL / T 657-2015 standard, the system enables the unit to achieve high-quality regulation during operation in the 70%-100% load range, saving approximately 1.5 million yuan in annual operating costs (taking a 600MW unit as an example). Simultaneously, the self-learning mechanism achieves a 92% accuracy rate in equipment degradation early warning, extending the service life of critical equipment. Attached Figure Description

[0063] Figure 1 This is a schematic diagram of the system module connections of the present invention;

[0064] Figure 2 This is a schematic view of the neural network of the present invention;

[0065] Figure 3 This is a flowchart of the dynamic coordination model of the present invention. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the invention, and should not be construed as limiting the invention.

[0068] A preferred embodiment of the coordinated control system for generator sets provided by the present invention is, for example... Figure 1-3 As shown: A coordinated control system for generator sets. The system adopts an intelligent DCS integrated design architecture and interacts with the distributed control system of the generator set in real time through hard wiring or communication protocols. The system includes a boiler-generator coordinated predictive control optimization module, a typical equipment anomaly handling module, and an advanced shift operator decision module.

[0069] The coordinated predictive control optimization module constructs a dynamic coordination model of boiler fuel-air supply-feedwater and turbine regulation based on neural network algorithms;

[0070] The specific process of constructing a dynamic coordination model for boiler fuel-air supply-feedwater and turbine regulation based on neural network algorithms includes the following:

[0071] Collect a large amount of historical data during the operation of the boiler and steam turbine, including parameters such as main steam pressure, power, feedwater flow, as well as corresponding fuel quantity, air supply, feedwater quantity and steam turbine regulation parameters;

[0072] The collected data is cleaned to remove outliers and noisy data. At the same time, the data is normalized to convert data of different ranges and magnitudes to the [0,1] or [-1,1] interval in order to improve the efficiency and stability of neural network training.

[0073] According to the problem description, the input layer nodes include the main steam pressure deviation ΔP, power deviation ΔN, and feedwater flow deviation ΔW. Other relevant operating parameters can also be considered as inputs to more comprehensively reflect the system status. The formula is: X(t)=[ΔP(t),ΔN(t),ΔW(t),...], where X(t) is the input vector at time t. Each deviation is collected in real time by sensors and normalized to the interval [0,1] or [-1,1].

[0074] Based on the complexity of the problem and the characteristics of the data, select an appropriate number of hidden layers and nodes. The hidden layers achieve feature extraction through weighted summation and activation functions, using the formula: h(t)=f(W1·x(t)+b1), where W1 is the weight matrix from the input layer to the hidden layer, b1 is the bias vector of the hidden layer, and f(·) is the activation function.

[0075] The number of output layer nodes corresponds to the dimension of the control command. It is used to control the relevant parameters of boiler fuel, air supply, water supply and turbine regulation. The control quantity is generated by linear transformation and the formula is: u(t)=W2·h(t)+b2, where u(t) is the control command vector at time t, and W2 and b2 are the weights and biases from the hidden layer to the output layer, respectively.

[0076] The specific process of constructing a dynamic coordination model for boiler fuel-air supply-feedwater and turbine regulation based on neural network algorithms also includes:

[0077] The preprocessed data is divided into training set, validation set and test set. The training set is used for learning the model's parameters, the validation set is used to tune the model's hyperparameters, and the test set is used to evaluate the model's generalization ability.

[0078] The backpropagation algorithm or the Levenberg-Marquardt algorithm is used to continuously adjust the weights and thresholds of the neural network to minimize the error between the model's output and the expected output.

[0079] The neural network is trained using the training set data, and the weights and thresholds of the network are iteratively updated until the preset training stopping conditions are met, such as reaching the maximum number of iterations or the error being less than the set threshold.

[0080] The specific process of constructing a dynamic coordination model for boiler fuel-air supply-feedwater and turbine regulation based on neural network algorithms also includes:

[0081] The trained neural network model is used to predict the future operating state of the system, and parameters such as main steam pressure, power and feedwater flow rate are predicted for a period of time in the future based on the input data at the current moment.

[0082] Within each control cycle, based on the predicted system state and combined with the system constraints u min ≤u(t)≤u max The optimal control command u(t) is solved by optimization algorithms (such as optimization algorithms in model predictive control) so that the system performance indicators (such as minimizing the main steam pressure deviation and power deviation) can be optimized.

[0083] As the system's operating status changes in real time, the input data is continuously updated, and the prediction and optimization process is repeated to achieve dynamic coordinated control.

[0084] The trained model is evaluated using test set data, and performance metrics such as mean squared error (MSE) and mean absolute error (MAE) are calculated to assess the model's prediction accuracy and control performance.

[0085] Based on the evaluation results, analyze the problems and shortcomings of the model, such as unreasonable model structure and insufficient training data, and improve and optimize the model, such as adjusting the network structure, increasing training data, and improving the training algorithm, in order to improve the performance and reliability of the model.

[0086] The boiler-turbine coordinated predictive control optimization module adopts a model predictive control algorithm, and its objective function is defined as:

[0087]

[0088] Where y is the set of controlled variables including main steam pressure, unit power, and intermediate point temperature; u is the set of control variables including fuel quantity, feedwater flow rate, and turbine valve opening; Q and R are weight matrices; and α and β are adjustment coefficients. The algorithm identifies the boiler dynamic model parameters online using the recursive least squares method, and the model update period ΔT satisfies:

[0089]

[0090] Ensure that the model's prediction accuracy continuously meets the requirements of the DL / T 657-2015 standard regarding load response rate ≥1.5% Pe / s and dynamic pressure deviation ≤±0.3MPa;

[0091] The typical equipment anomaly handling module has a built-in expert rule base and fuzzy inference engine. When the temperature of the blower bearing exceeds the threshold T, it will handle the anomaly. crit Or the inlet pressure of the water pump is lower than P min When this happens, the system automatically executes the preset protection logic:

[0092] In case of a blower malfunction, the module calculates the cross-limit coefficient K between air and coal. air =f(T) bearing ), dynamically adjust fuel quantity commands;

[0093] In case of water pump malfunction, the module triggers the RB preset function to reduce the target load at a preset rate while maintaining combustion stability.

[0094] During the anomaly handling process, the system compensates for the disturbances in the relevant loops through a multivariate decoupling algorithm to ensure that the main steam temperature fluctuation range is controlled within ±5℃ and the change rate of the reheater spray desuperheating water volume does not exceed 10t / min.

[0095] The advanced shift operator decision-making module integrates a deep reinforcement learning network. The network input state space includes unit load rate, coal calorific value, ambient temperature, and equipment health index, and the output is a suggested operation sequence. The network training adopts a proximal policy optimization algorithm, and the reward function is designed as follows:

[0096]

[0097] Where γ1, γ2, and γ3 are weighting coefficients, C fuel C represents the actual fuel consumption. opt The module generates an operational analysis report every 24 hours, which includes control quality assessment, equipment degradation warnings, and parameter optimization suggestions, based on the theoretically optimal fuel quantity.

[0098] The system has a parameter self-adjustment function, and its self-learning mechanism consists of two levels: the basic level uses a least-squares support vector machine to establish a mapping relationship between control parameters and unit operating conditions, and the mapping function satisfies:

[0099]

[0100] Where K(x,x) i ) is the kernel function, α i b are model parameters; the enhancement layer dynamically adjusts the weights of the mapping function based on the reinforcement learning algorithm, and the reward signal is defined as the control loop settling time T. s Linear combination with overshoot σ:

[0101] r = -(w1T) s +w2σ)

[0102] Among them, w1 and w2 are adjustable coefficients; when the unit experiences deep peak shaving conditions or the fuel calorific value changes by more than 10%, the system automatically triggers the parameter resetting process to ensure that the attenuation rate φ of the main steam pressure control loop is maintained between 0.7 and 0.9 and the load response delay time τ ≤ 5 seconds.

[0103] The system's interface with the DCS employs a dual-redundant communication protocol, with a data transmission cycle set to 200ms. Critical signals such as actual unit load, main steam pressure, and valve opening are handled with a 2-out-of-3 redundancy scheme. When a communication interruption or invalid data is detected, the system immediately switches to the backup link and retains the last valid control command, while simultaneously starting a counter to record the interruption duration t. fault If t fault If the time is ≥3 seconds, a control handover operation is executed; a soft limiting module is set in the interface logic to optimize the fuel quantity command u output by the control system. fuel Dynamic limiting is implemented, and the limiting range is calculated based on the boiler combustion stability:

[0104]

[0105] Among them, U0 is the fuel quantity command under the DCS basic automatic control;

[0106] System performance acceptance follows the DL / T 657-2015 standard, and specific test items include:

[0107] Load variation test: Within the load range of 70% to 100%, the load is increased or decreased at a rate of 3% Pe / min, and the main steam pressure deviation ΔP is recorded. max ≤±0.2MPa, load response deviation ΔN max ≤±2%Pe;

[0108] Load shedding test: When the unit load is reduced from 100% to 50%, ensure that the boiler does not overpressure, the turbine does not overspeed, and the reheater safety valve does not activate.

[0109] Disturbance test: When ±10% step disturbances are applied to the water supply flow rate, fuel quantity, and air supply volume, the settling time T of each main controlled variable is measured. s ≤120 seconds, overshoot σ≤5%;

[0110] All test data is automatically collected by the system's built-in testing module, generating an acceptance report that meets power grid requirements. The report includes time-domain waveform diagrams, performance index statistics tables, and compliance conclusions.

[0111] The system possesses full lifecycle self-optimization capabilities, and its self-learning mechanism comprises three dimensions:

[0112] Model self-calibration: The boiler dynamic model is updated daily at midnight using the operating data of the previous 24 hours. Cross-validation is used before and after the model update to ensure that the prediction error ε ≤ 3%.

[0113] Rule self-evolution: monthly cluster analysis of the expert rule base, merging processing strategies under similar working conditions, and adding new rules for predicting equipment degradation trends;

[0114] Parameter self-optimization: Global parameter optimization is performed every quarter, with the optimization target defined as:

[0115]

[0116] Where a, b, and c are economic weight coefficients, and T is the optimization period; the optimization process uses a genetic algorithm with a population size of N=50, a mutation probability of Pm=0.1, and the termination condition is that the improvement of the objective function value is <1% for 10 consecutive generations.

[0117] The system has a self-diagnostic function, which can monitor the status of I / O cards, the integrity of communication links and the accuracy of algorithm models in real time. The diagnostic results are stored in the form of logs and trigger a hierarchical alarm mechanism. When the optimization function is abnormal or the DCS side triggers the RB condition, the system immediately freezes the current control command and executes the seamless switching logic, returns control to the DCS and outputs an alarm signal to ensure that the unit smoothly transitions to the basic automatic control mode.

[0118] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0119] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A coordinated control system for generator sets, characterized in that: The system adopts an intelligent DCS integrated design architecture and interacts with the unit's distributed control system in real time through hard wiring or communication protocols. The system includes a boiler-turbine coordinated predictive control optimization module, a typical equipment anomaly handling module, and an advanced shift operator decision-making module. The coordinated predictive control optimization module constructs a dynamic coordination model of boiler fuel-air supply-feedwater and turbine regulation based on a neural network algorithm; The system performance acceptance follows the DL / T 657-2015 standard, and the specific test items include: Load variation test: Within the load range of 70% to 100%, the load is increased or decreased at a rate of 3% Pe / min, and the main steam pressure deviation ΔP is recorded. max ≤±0.2MPa, load response deviation ΔN max ≤±2%Pe; Load shedding test: When the unit load is reduced from 100% to 50%, ensure that the boiler does not overpressure, the turbine does not overspeed, and the reheater safety valve does not activate. Disturbance test: When ±10% step disturbances are applied to the water supply flow rate, fuel quantity, and air supply volume, the settling time T of each main controlled variable is measured. s ≤120 seconds, overshoot σ≤5%; All test data is automatically collected by the system's built-in test module and an acceptance report that meets the power grid requirements is generated. The report includes time-domain waveform diagrams, performance index statistics tables, and compliance conclusions. The system possesses full lifecycle self-optimization capabilities, and its self-learning mechanism comprises three dimensions: Model self-calibration: The boiler dynamic model is updated daily at midnight using the operating data of the previous 24 hours. Cross-validation is used before and after the model update to ensure that the prediction error ε ≤ 3%. Rule self-evolution: monthly cluster analysis of the expert rule base, merging processing strategies under similar working conditions, and adding new rules for predicting equipment degradation trends; Parameter self-optimization: Global parameter optimization is performed every quarter, with the optimization target defined as: Where a, b, and c are economic weight coefficients, and T is the optimization period; the optimization process uses a genetic algorithm with a population size of N=50, a mutation probability of Pm=0.1, and the termination condition is that the improvement of the objective function value is <1% for 10 consecutive generations. The system has a self-diagnostic function, which can monitor the status of I / O cards, the integrity of communication links and the accuracy of algorithm models in real time. The diagnostic results are stored in the form of logs and trigger a hierarchical alarm mechanism. When the optimization function is abnormal or the DCS side triggers the RB condition, the system immediately freezes the current control command and executes the seamless switching logic, returns control to the DCS and outputs an alarm signal to ensure that the unit smoothly transitions to the basic automatic control mode.

2. The coordinated control system for a generator set according to claim 1, characterized in that: The specific process of constructing a dynamic coordination model for boiler fuel-air supply-feedwater and turbine regulation based on neural network algorithms includes the following: Collect a large amount of historical data during the operation of the boiler and steam turbine, including parameters such as main steam pressure, power, feedwater flow, as well as corresponding fuel quantity, air supply, feedwater quantity and steam turbine regulation parameters; The collected data is cleaned to remove outliers and noisy data. At the same time, the data is normalized to convert data of different ranges and magnitudes to the [0,1] or [-1,1] interval in order to improve the efficiency and stability of neural network training. According to the problem description, the input layer nodes include the main steam pressure deviation ΔP, power deviation ΔN, and feedwater flow deviation ΔW. Other relevant operating parameters can also be considered as inputs to more comprehensively reflect the system status. The formula is: X(t)=[ΔP(t),ΔN(t),ΔW(t),...], where X(t) is the input vector at time t. Each deviation is collected in real time by sensors and normalized to the interval [0,1] or [-1,1]. Based on the complexity of the problem and the characteristics of the data, select an appropriate number of hidden layers and nodes. The hidden layers achieve feature extraction through weighted summation and activation functions, using the formula: h(t)=f(W1·x(t)+b1), where W1 is the weight matrix from the input layer to the hidden layer, b1 is the bias vector of the hidden layer, and f(·) is the activation function. The number of output layer nodes corresponds to the dimension of the control command. It is used to control the relevant parameters of boiler fuel, air supply, water supply and turbine regulation. The control quantity is generated by linear transformation, and the formula is: u(t)=W2·h(t)+b2, where u(t) is the control command vector at time t, and W2 and b2 are the weights and biases from the hidden layer to the output layer, respectively.

3. A coordinated control system for a generator set according to claim 1, characterized in that: The specific process of constructing a dynamic coordination model for boiler fuel-air supply-feedwater and turbine regulation based on neural network algorithms also includes: The preprocessed data is divided into training set, validation set and test set. The training set is used for learning the model's parameters, the validation set is used to tune the model's hyperparameters, and the test set is used to evaluate the model's generalization ability. The backpropagation algorithm or the Levenberg-Marquardt algorithm is used to continuously adjust the weights and thresholds of the neural network to minimize the error between the model's output and the expected output. The neural network is trained using the training set data, and the weights and thresholds of the network are iteratively updated until the preset training stopping conditions are met, such as reaching the maximum number of iterations or the error being less than the set threshold.

4. A coordinated control system for a generator set according to claim 1, characterized in that: The specific process of constructing a dynamic coordination model for boiler fuel-air supply-feedwater and turbine regulation based on neural network algorithms also includes: The trained neural network model is used to predict the future operating state of the system, and parameters such as main steam pressure, power and feedwater flow rate are predicted for a period of time in the future based on the input data at the current moment. Within each control cycle, based on the predicted system state and combined with the system constraints u min ≤u(t)≤u max The optimal control command u(t) is obtained by optimizing the algorithm, so that the system performance index reaches the optimal level. As the system's operating status changes in real time, the input data is continuously updated, and the prediction and optimization process is repeated to achieve dynamic coordinated control. The trained model is evaluated using test set data. Performance metrics such as mean squared error and mean absolute error are calculated to assess the model's prediction accuracy and control performance. Based on the evaluation results, analyze the problems and shortcomings of the model, such as unreasonable model structure and insufficient training data, and improve and optimize the model, such as adjusting the network structure, increasing training data, and improving the training algorithm, in order to improve the performance and reliability of the model.

5. A coordinated control system for a generator set according to claim 1, characterized in that: The furnace-machine coordination predictive control optimization module adopts a model predictive control algorithm, and its objective function is defined as: Where y is the set of controlled variables including main steam pressure, unit power, and intermediate point temperature; u is the set of control variables including fuel quantity, feedwater flow rate, and turbine valve opening; Q and R are weight matrices; and α and β are adjustment coefficients. The algorithm identifies the boiler dynamic model parameters online using the recursive least squares method, and the model update period ΔT satisfies: Ensure that the model's prediction accuracy continuously meets the requirements of the DL / T 657-2015 standard regarding load response rate ≥1.5% Pe / s and dynamic pressure deviation ≤±0.3MPa.

6. A coordinated control system for a generator set according to claim 1, characterized in that: The typical equipment anomaly handling module incorporates an expert rule base and a fuzzy inference engine. When the temperature of the blower bearing exceeds the threshold T, crit Or the inlet pressure of the water pump is lower than P min When this happens, the system automatically executes the preset protection logic: In case of a blower malfunction, the module calculates the cross-limit coefficient K between air and coal. air =f(T) bearing ), dynamically adjust fuel quantity commands; In case of water pump malfunction, the module triggers the RB preset function to reduce the target load at a preset rate while maintaining combustion stability. During the anomaly handling process, the system compensates for the disturbances in the relevant loops through a multivariate decoupling algorithm to ensure that the main steam temperature fluctuation range is controlled within ±5℃ and the change rate of the reheater spray desuperheating water volume does not exceed 10t / min.

7. A coordinated control system for a generator set according to claim 1, characterized in that: The advanced operator decision-making module integrates a deep reinforcement learning network. The network input state space includes unit load rate, coal calorific value, ambient temperature, and equipment health index, and the output is a suggested operation sequence. The network training adopts a proximal policy optimization algorithm, and the reward function is designed as follows: Where γ1, γ2, and γ3 are weighting coefficients, and C fuel C represents the actual fuel consumption. opt The module generates an operation analysis report every 24 hours, which includes control quality assessment, equipment deterioration warning and parameter optimization suggestions, based on the theoretical optimal fuel quantity.

8. A coordinated control system for a generator set according to claim 1, characterized in that: The system has a parameter self-adjustment function, and its self-learning mechanism consists of two levels: the basic level uses a least-squares support vector machine to establish a mapping relationship between control parameters and unit operating conditions, and the mapping function satisfies: Where K(x,x) i ) is the kernel function, α i b are model parameters; the enhancement layer dynamically adjusts the weights of the mapping function based on the reinforcement learning algorithm, and the reward signal is defined as the control loop settling time T. s Linear combination with overshoot σ: r=-(w1T s +w2σ) Among them, w1 and w2 are adjustable coefficients; when the unit experiences deep peak shaving conditions or the fuel calorific value changes by more than 10%, the system automatically triggers the parameter resetting process to ensure that the attenuation rate φ of the main steam pressure control loop is maintained between 0.7 and 0.9, and the load response delay time τ ≤ 5 seconds.

9. A coordinated control system for a generator set according to claim 1, characterized in that: The system's interface with the DCS employs a dual-redundant communication protocol, with a data transmission cycle set to 200ms. Critical signals such as actual unit load, main steam pressure, and valve opening are handled with a 2-out-of-3 redundancy scheme. When a communication interruption or invalid data is detected, the system immediately switches to the backup link and retains the last valid control command, while simultaneously starting a counter to record the interruption duration t. fault If t fault If the time is ≥3 seconds, a control handover operation is executed; a soft limiting module is set in the interface logic to optimize the fuel quantity command u output by the control system. fuel Dynamic limiting is implemented, and the limiting range is calculated based on the boiler combustion stability: U0 is the fuel quantity command under the DCS basic automatic control.

Images