Method for optimizing power station AGC load regulation rate

Abstract

The invention discloses a method for optimizing the AGC load regulation rate of a power station, and the method comprises the steps: building a digital twinborn body of a unit, building a boiler thermal positive and negative balance model, a fuel heat value prediction model and a self-adaptive control parameter library based on the historical operation data of a target unit, and receiving and processing the operation data of the unit in real time; dynamically predicting a boiler theoretical calorific value demand corresponding to the AGC instruction by using a long-short-term memory recurrent neural network, and calculating a deviation between an actual calorific value and a theoretical calorific value; when the deviation exceeds a preset threshold value, multi-mode cooperative compensation control of feed-forward compensation, parameter self-tuning and energy closed-loop calibration is implemented in parallel; the AGC instruction response time is controlled within 50-55 seconds, and model self-updating is achieved through online tracking. According to the method, the defects that traditional thermal power station fuel calorific value calibration is slow in response, a single control model is difficult to deal with complex problems, AGC response time is long and the like are overcome, the power station AGC load regulation rate and thermal power generating unit frequency modulation performance are remarkably improved, and power grid frequency stability is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of power station optimization regulation, and in particular to a method for optimizing the AGC load regulation rate of a power station. Background Art

[0002] In modern power systems, AGC (Automatic Generation Control) load regulation at thermal power plants is a key technology for ensuring grid frequency stability and achieving a dynamic balance between power supply and demand. The AGC system automatically adjusts the generator load by receiving real-time grid dispatch instructions to maintain grid frequency within specified limits. Its performance directly impacts the safe and stable operation of the grid.

[0003] However, traditional AGC load regulation technology for thermal power plants has numerous drawbacks. Traditional fuel calorific value calibration relies on offline testing, with a response delay exceeding five minutes from sample collection to data feedback, making it difficult to adapt to the grid's rapidly changing load demands. The single control model employed cannot simultaneously address complex issues such as coal quality fluctuations, thermal inertia, and parameter mismatches, resulting in poor control accuracy and stability during regulation. Furthermore, the average AGC response time exceeds 90 seconds, far from meeting the grid's requirements for rapid frequency regulation. This results in a low Kp value for the thermal power unit's frequency regulation performance indicator, making it difficult to effectively support grid frequency stability. Therefore, a more efficient and accurate AGC load regulation method for power plants is urgently needed. Summary of the Invention

[0004] (1) Technical issues

[0005] The present invention provides a method for optimizing the AGC load regulation rate of a power plant to solve the problems of long response delay, low control accuracy, poor frequency regulation performance, etc. in the AGC load regulation of a traditional thermal power plant.

[0006] (2) Technical content

[0007] To solve the above technical problems, the technical solution of the present invention is: a method for optimizing the AGC load regulation rate of a power station, comprising the following steps:

[0008] S1. Build a digital twin of the unit: Based on the historical operating data of the target unit, establish a digital twin that includes a boiler thermal forward and reverse balance model, a fuel calorific value prediction model, and an adaptive control parameter library. Receive unit operating data from the on-site DCS / SIS system in real time through a one-way data interface. Use the principal component analysis algorithm to reduce the dimensionality of the real-time data and extract the main characteristic components.

[0009] S2. Dynamically predict fuel calorific value demand: Input the reduced-dimensional data into the long-short-term memory recurrent neural network to predict the boiler's theoretical calorific value demand corresponding to the AGC instruction; simultaneously calculate the absolute deviation between the current actual fuel calorific value and the theoretical demand value;

[0010] S3, multi-modal collaborative compensation control: When the calorific value deviation exceeds the preset threshold, the following control actions are triggered in parallel:

[0011] a) Feedforward compensation: Adjust the coal feed rate in advance based on the prediction model to compensate for boiler thermal inertia;

[0012] b) Parameter self-tuning: Identify operating condition characteristics through fuzzy matching algorithm, dynamically call optimal control parameters, and correct air supply and water supply circuit lag in real time;

[0013] c) Energy closed-loop calibration: The results of the forward and reverse balance models are fed back to the coal quantity control instructions;

[0014] S4. Closed-loop verification of response performance: Control the response time of a single AGC command within 50 to 55 seconds; use the online tracking module to compare the digital twin and the actual unit status in real time, and trigger model self-update when the core parameter deviation exceeds the limit.

[0015] Furthermore, step S1 includes: constructing an incentive virtual DPU simulation environment that supports the full-operation operation of 300-400MW supercritical units; and using a model predictive control algorithm to update parameters with an update cycle of no more than 10 seconds.

[0016] Furthermore, in step S2: the input features of the long short-term memory neural network include fuel carbon content, historical calorific value curve, AGC instruction change gradient and main steam pressure fluctuation; the output is the boiler calorific value demand and coal quantity adjustment within the next 60 seconds.

[0017] Furthermore, the fuzzy matching algorithm of step S3 takes the load change rate, coal quality fluctuation coefficient, and steam temperature drop slope as input variables; and outputs a control parameter adjustment factor, whose proportional adjustment range satisfies: the proportional gain adjustment amount does not exceed 30% of the reference value, the integral time adjustment amount does not exceed 40% of the reference value, and the differential time adjustment amount does not exceed 20% of the reference value.

[0018] (3) Technical effects

[0019] Compared with the existing technology, the advantages of the present invention are: by constructing a digital twin of the unit, the present invention utilizes real-time data dimensionality reduction and multi-model collaboration to accurately predict the fuel calorific value demand, greatly shortening the response time compared to traditional offline detection; the multi-modal collaborative compensation control strategy implements feedforward compensation, parameter self-tuning and energy closed-loop calibration in parallel, effectively overcoming the problems of coal quality fluctuations, thermal inertia and parameter mismatch; the AGC command response time is controlled within the range of 50-55 seconds, significantly improving the AGC response speed, thereby improving the frequency regulation performance index Kp value of the thermal power unit, ensuring the stability of the power grid frequency, and improving the safety and reliability of the power system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of a method for optimizing the AGC load regulation rate of a power station according to the present invention. DETAILED DESCRIPTION

[0021] The present invention will be described in further detail below with reference to the accompanying drawings.

[0022] Combined with attachment Figure 1 A method for optimizing the AGC load regulation rate of a power station comprises the following steps:

[0023] S1. Build a digital twin of the unit: Based on the historical operating data of the target unit, establish a digital twin that includes a boiler thermal forward and reverse balance model, a fuel calorific value prediction model, and an adaptive control parameter library. Receive unit operating data from the on-site DCS / SIS system in real time through a one-way data interface. Use the principal component analysis algorithm to reduce the dimensionality of the real-time data and extract the main characteristic components.

[0024] S2. Dynamically predict fuel calorific value demand: Input the reduced-dimensional data into the long-short-term memory recurrent neural network to predict the boiler's theoretical calorific value demand corresponding to the AGC instruction; simultaneously calculate the absolute deviation between the current actual fuel calorific value and the theoretical demand value;

[0025] S3, multi-modal collaborative compensation control: When the calorific value deviation exceeds the preset threshold, the following control actions are triggered in parallel:

[0026] a) Feedforward compensation: Adjust the coal feed rate in advance based on the prediction model to compensate for boiler thermal inertia;

[0027] b) Parameter self-tuning: Identify operating condition characteristics through fuzzy matching algorithm, dynamically call optimal control parameters, and correct air supply and water supply circuit lag in real time;

[0028] c) Energy closed-loop calibration: The results of the forward and reverse balance models are fed back to the coal quantity control instructions;

[0029] S4. Closed-loop verification of response performance: Control the response time of a single AGC command within 50 to 55 seconds; use the online tracking module to compare the digital twin and the actual unit status in real time, and trigger model self-update when the core parameter deviation exceeds the limit.

[0030] The step S1 includes: building an incentive virtual DPU simulation environment that supports the full-operation operation of 300-400MW supercritical units; and using a model predictive control algorithm to update parameters with an update cycle of no more than 10 seconds.

[0031] In step S2, the input features of the long short-term memory neural network include the carbon content of the fuel, the historical calorific value curve, the AGC instruction change gradient and the main steam pressure fluctuation; the output is the boiler calorific value demand and coal quantity adjustment within the next 60 seconds.

[0032] The fuzzy matching algorithm of step S3 takes the load change rate, coal quality fluctuation coefficient, and steam temperature drop slope as input variables; and outputs a control parameter adjustment factor, whose proportional adjustment range satisfies: the proportional gain adjustment amount does not exceed 30% of the reference value, the integral time adjustment amount does not exceed 40% of the reference value, and the differential time adjustment amount does not exceed 20% of the reference value.

[0033] This embodiment takes a 330MW supercritical unit as an application object and describes in detail the specific working process of the method for optimizing the AGC load regulation rate of a power plant.

[0034] S1. Build a digital twin of the unit

[0035] An incentive-based virtual DPU simulation environment was built to support the full-scale operation of a 330MW supercritical unit. Connected to the on-site DCS / SIS system via a one-way data interface, it receives real-time unit operating data, covering key parameters such as main steam pressure, temperature, coal feed rate, air volume, and unit load. Principal component analysis (PCA) is used to reduce the dimensionality of this massive amount of data. For example, if data containing 20 feature dimensions is received at a given moment, the algorithm extracts five components that reflect the key characteristics of the unit's operating status, effectively reducing data redundancy and improving data processing efficiency. Simultaneously, based on the unit's historical operating data over the past year, a digital twin was established, including a boiler thermal balance model, a fuel calorific value prediction model, and an adaptive control parameter library. A model predictive control algorithm is used to update the parameters in the digital twin, with an 8-second update cycle, to ensure that the digital twin reflects the actual unit operating status in a timely manner.

[0036] S2. Dynamically predict fuel calorific value requirements

[0037] The reduced data is fed into a long-short-term memory (LSTM) recurrent neural network (RNN). Input features include the fuel's carbon content, the historical calorific value curve for the past two hours, the gradient of the AGC command, and main steam pressure fluctuations. When the grid dispatcher issues a new AGC command, requiring the unit to increase its load from 280 MW to 320 MW within 10 minutes, the LSTM RNN, based on these input features, outputs the boiler's calorific value requirement and coal quantity adjustment for the next 60 seconds. It also calculates the absolute deviation between the actual fuel calorific value and the theoretical demand.

[0038] S3. Multi-modal collaborative compensation control

[0039] When the calculated calorific value deviation exceeds the preset threshold, the following control actions are triggered in parallel:

[0040] a) Feedforward compensation

[0041] Based on the fuel calorific value prediction model, the coal feed is adjusted in advance to compensate for boiler thermal inertia. Based on the calculated coal feed adjustment for the next 60 seconds, the coal feed is immediately increased by a certain compensation factor (0.8 in this example). This reduces the impact of boiler thermal inertia on load regulation and enables the unit to respond more quickly to load changes.

[0042] b) Parameter self-tuning

[0043] A fuzzy matching algorithm identifies operating condition characteristics, using the current load change rate (the rate of change from 280MW to 320MW), the coal quality fluctuation coefficient (calculated based on parameters such as fuel carbon content), and the steam temperature drop slope as input variables to output control parameter adjustment factors. Calculated adjustments are made to the proportional gain by 25% of the baseline value, the integral time by 30%, and the differential time by 15%. This dynamically calls for optimal control parameters, correcting air and water supply circuit hysteresis in real time to better adapt the control system to changing operating conditions.

[0044] c) Energy closed-loop calibration

[0045] The results of the boiler's thermal balance model calculations are fed back into the coal quantity control instructions. For example, if the forward and reverse balance model calculations reveal that after adjusting the current coal feed, the actual heat generated still deviates from the theoretical demand, the coal quantity control instructions are further adjusted based on this deviation to ensure a balance between the unit's energy input and output.

[0046] S4. Closed-loop verification of response performance

[0047] The response time for a single AGC command is strictly controlled within 50 to 55 seconds. During the unit load adjustment process from 280MW to 320MW, the online tracking module compared the digital twin with the actual unit status in real time, monitoring core parameters such as main steam pressure, temperature, and unit load. If a core parameter deviates beyond the limit, a model self-update is immediately triggered, and the model and parameters in the digital twin are corrected using the latest operating data to ensure consistency between the digital twin and the actual unit and continuously optimize AGC load adjustment performance.

[0048] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A method for optimizing the AGC load regulation rate of a power station, characterized in that: The following steps are involved: S1. Build a digital twin of the unit: Based on the historical operating data of the target unit, establish a digital twin that includes a boiler thermal forward and reverse balance model, a fuel calorific value prediction model, and an adaptive control parameter library. Receive unit operating data from the on-site DCS / SIS system in real time through a one-way data interface. Use the principal component analysis algorithm to reduce the dimensionality of the real-time data and extract the main characteristic components. S2. Dynamically predict fuel calorific value demand: Input the reduced-dimensional data into the long-short-term memory recurrent neural network to predict the boiler's theoretical calorific value demand corresponding to the AGC instruction; simultaneously calculate the absolute deviation between the current actual fuel calorific value and the theoretical demand value; S3, multi-modal collaborative compensation control: When the calorific value deviation exceeds the preset threshold, the following control actions are triggered in parallel: a) Feedforward compensation: Adjust the coal feed rate in advance based on the prediction model to compensate for boiler thermal inertia; b) Parameter self-tuning: Identify operating condition characteristics through fuzzy matching algorithm, dynamically call optimal control parameters, and correct air supply and water supply circuit lag in real time; c) Energy closed-loop calibration: The results of the forward and reverse balance models are fed back to the coal quantity control instructions; S4. Closed-loop verification of response performance: Control the response time of a single AGC command within 50 to 55 seconds; use the online tracking module to compare the digital twin and the actual unit status in real time, and trigger model self-update when the core parameter deviation exceeds the limit.

2. The method for optimizing the AGC load regulation rate of a power station according to claim 1, characterized in that: The step S1 includes: building an incentive virtual DPU simulation environment that supports the full-operation operation of 300-400MW supercritical units; and using a model predictive control algorithm to update parameters with an update cycle of no more than 10 seconds.

3. The method for optimizing the AGC load regulation rate of a power station according to claim 1, characterized in that: In step S2, the input features of the long short-term memory neural network include the carbon content of the fuel, the historical calorific value curve, the AGC instruction change gradient and the main steam pressure fluctuation; the output is the boiler calorific value demand and coal quantity adjustment within the next 60 seconds.

4. The method for optimizing the AGC load regulation rate of a power station according to claim 1, characterized in that: The fuzzy matching algorithm of step S3 takes the load change rate, coal quality fluctuation coefficient, and steam temperature drop slope as input variables; and outputs a control parameter adjustment factor, whose proportional adjustment range satisfies: the proportional gain adjustment amount does not exceed 30% of the reference value, the integral time adjustment amount does not exceed 40% of the reference value, and the differential time adjustment amount does not exceed 20% of the reference value.

Images