Thermal power generating unit frequency modulation dynamic characteristic prediction system

Abstract

The invention relates to the technical field of frequency modulation prediction, and provides a thermal power generating unit frequency modulation dynamic characteristic prediction system, historical operation data and real-time operation data are comprehensively acquired through a data acquisition module, the integrity and timeliness of the data are ensured, and high-quality input is provided for subsequent feature extraction and model prediction; the feature screening module extracts a frequency modulation capability feature set through preprocessing and a random forest algorithm, data redundancy is effectively reduced, key influence factors are highlighted, and the pertinence of model input is improved; the model prediction module constructs a multi-level driving model based on a deep learning algorithm, can accurately simulate key parameters such as frequency modulation dynamic characteristics, output power adjustment response speed and frequency difference tracking precision under different working conditions and power grid disturbance, and the prediction result better fits the actual operation state and can be directly sent to a unit control system. And data support is provided for real-time optimization of the primary frequency modulation action.

Description

TECHNICAL FIELD

[0001] The application relates to the field of frequency modulation prediction, in particular to a thermal power unit frequency modulation dynamic characteristic prediction system. BACKGROUND

[0003] Foreign researches are mostly focused on the optimization of specific frequency modulation measures, such as improving the frequency modulation capacity of the unit by butterfly valve compensation and heat load-power load signal conversion, and some researches improve the primary frequency modulation success rate of supercritical units by improving the trigger logic or adjusting the steam extraction amount. Domestic researches try to improve the frequency modulation performance of supercritical and ultra-supercritical units by means of coordinated control of steam supply valve, energy storage combined frequency modulation and optimization of control parameters, and some researches combine MATLAB / Simulink and other tools to establish mathematical models to provide data support for frequency modulation optimization.

[0004] Although relevant researches have been carried out at home and abroad, there are still significant technical limitations. The existing technology is not accurate in judging the primary frequency modulation action, the regulation tracking precision is low, and a closed-loop regulation mechanism for the examination index has not been formed, so it is difficult to realize the action analysis with the same precision as the power grid examination. The unit operation data is large but not fully excavated, and a feature set for quantitatively representing the frequency modulation capacity has not been formed, which leads to poor generalization of the frequency modulation capacity evaluation model and cannot accurately reflect the unit dynamic characteristics under different working conditions and power grid disturbances. The optimization research on compensation control algorithm is insufficient in foreign countries, and domestic researches mostly focus on equipment cooperation or strategy design, and the research on accurate prediction of frequency modulation dynamic characteristics and targeted control measures in actual operation is insufficient.

[0005] In view of the problems in the prior art, the technical problem solved by the application is how to accurately predict the frequency modulation dynamic characteristics of the thermal power unit by processing multi-dimensional data. SUMMARY

[0006] In view of the deficiencies of the prior art, the application provides a thermal power unit frequency modulation dynamic characteristic prediction system, which comprises a data acquisition module, a feature screening module and a model prediction module.

[0007] The data acquisition module is used for acquiring historical operation data and real-time operation data of the thermal power unit, wherein the historical operation data comprises unit parameters, control instructions and power grid disturbance data at the time of primary frequency modulation action, and the real-time operation data comprises operating condition parameters, frequency deviation signals and output regulation instructions.

[0008] The feature screening module is used for pre-processing the historical operation data and real-time operation data, and extracting a frequency modulation capacity feature set from the pre-processed historical operation data by means of a random forest algorithm, wherein the frequency modulation capacity feature set comprises boiler heat storage parameters, valve flow characteristic parameters, control characteristic parameters and equipment state parameters.

[0009] The model prediction module is configured to construct a multi-level driving model based on a deep learning algorithm, input the frequency modulation capability feature set into the multi-level driving model, and output a prediction result of a frequency modulation characteristic parameter of the thermal power generating unit under the operating condition parameters and the power grid disturbance data, the prediction result including a power regulation response speed, a frequency difference tracking accuracy, and an output regulation amplitude, and the prediction result is simultaneously displayed and sent to the unit control system to optimize the frequency modulation control.

[0010] As an optional implementation, the unit parameters include a main steam pressure, a steam extraction amount, a steam turbine speed, a load instruction, and a boiler feed water flow.

[0011] As an optional implementation, the data acquisition module is communicatively connected to a DCS system or a SIS system of the thermal power generating unit to acquire real-time operating data of the thermal power generating unit in real time.

[0012] As an optional implementation, the historical operating data and the real-time operating data of the thermal power generating unit are preprocessed, including removing abnormal data by using a Grubbs analysis method and filling missing data by using a KNN algorithm.

[0013] As an optional implementation, when the frequency modulation capability feature set is extracted, the feature screening module determines the frequency modulation capability feature set by calculating a variable projection importance value, and selects a parameter with a variable projection importance value greater than a preset projection threshold value as the frequency modulation capability feature.

[0014] As an optional implementation, when the frequency modulation capability feature set is extracted, the feature screening module further optimizes the features in combination with a thermal stability state under different operating condition parameters of the unit, and the thermal stability state is determined by the deviation of the real-time operating data of the thermal power generating unit being less than an acquisition accuracy in a plurality of continuous sampling periods.

[0015] As an optional implementation, the multi-level driving model includes an input layer, a feature mapping layer, a characteristic prediction layer, and an output layer, the input layer is configured to receive the frequency modulation capability feature set, the feature mapping layer is configured to strengthen a weight of the frequency modulation capability feature by using an attention mechanism, the characteristic prediction layer is configured to simulate a frequency modulation response process under different operating condition parameters, and the output layer is configured to output the prediction result of the frequency modulation characteristic parameter.

[0016] As an optional implementation, the prediction result of the frequency modulation characteristic parameter is evaluated, and a qualified rate of a frequency modulation action and a risk value of the frequency modulation action are output.

[0017] As an optional implementation, the prediction result of the frequency modulation characteristic parameter is displayed in a visual chart form, including a frequency modulation response curve, a parameter sensitivity analysis diagram, and a prediction error distribution diagram.

[0018] As an optional implementation, when the model prediction module outputs the prediction results of the frequency modulation characteristic parameters, it also simultaneously associates the assessment indicators during the frequency modulation process. The assessment indicators include the output response index and the power contribution index, so as to support the optimization of frequency modulation control based on the assessment indicators.

[0019] Compared with existing technologies, the beneficial effects of this application are as follows: the data acquisition module comprehensively acquires historical and real-time operating data, ensuring data integrity and timeliness, and providing high-quality input for subsequent feature extraction and model prediction; the feature selection module extracts frequency regulation capability feature sets through preprocessing and random forest algorithms, effectively reducing data redundancy, highlighting key influencing factors, and improving the targeting of model input; the model prediction module constructs a multi-level driving model based on deep learning, which can accurately simulate the frequency regulation dynamic characteristics under different operating conditions and grid disturbances, outputting key parameters such as power regulation response speed and frequency difference tracking accuracy. The prediction results are more in line with the actual operating state, and the prediction results are directly sent to the unit control system, providing data support for the real-time optimization of primary frequency regulation actions, which helps to improve the action qualification rate and reduce assessment risks. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0021] Figure 1 This is a system flowchart of a frequency regulation dynamic characteristic prediction system for thermal power units provided in an embodiment of this application;

[0022] Figure 2 This is a schematic diagram of the structure of a multi-level drive model for a thermal power unit frequency regulation dynamic characteristic prediction system provided in an embodiment of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application more apparent and understandable, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0024] Example

[0025] like Figure 1 The figure shown is a system flowchart of a thermal power unit frequency regulation dynamic characteristic prediction system provided in this application embodiment. The system includes a data acquisition module, a feature screening module, and a model prediction module.

[0026] The data acquisition module is configured to acquire historical operation data and real-time operation data of the thermal power generating unit, wherein the historical operation data includes unit parameters, control instructions and power grid disturbance data at the time of primary frequency modulation action, and the real-time operation data includes operating condition parameters, frequency deviation signals and output adjustment instructions.

[0027] The operation data of the thermal power generating unit is of various types, and if the acquisition range is not explicitly defined, invalid data redundancy will be caused, the subsequent processing load will be increased, and even the feature selection accuracy will be disturbed. According to the prediction requirement of the primary frequency modulation dynamic characteristics, the acquisition range is divided into two types of data, the historical operation data focuses on the time period when the primary frequency modulation action occurs, and includes unit parameters, control instructions and power grid disturbance data, the real-time operation data covers the whole operating condition of the current unit, and includes operating condition parameters, frequency deviation signals and output adjustment instructions. The start and end points of the primary frequency modulation action are marked by time stamps to ensure the accurate association between the historical data and the action event. Thus, irrelevant data interference is avoided, the pertinence and effectiveness of the acquired data are ensured, and high-quality data are provided for subsequent feature selection and model prediction.

[0028] The operation data of the thermal power generating unit is stored in different systems, wherein the historical data is mostly stored in a database, and the real-time data is obtained from an online control system. The data is fully covered by multi-source acquisition. For the historical operation data, the data is extracted from the historical database of the thermal power generating unit through a data interface, wherein the historical database includes a relational database and a time series database. For the real-time operation data, the real-time data bus of the unit distributed control system or the plant-level monitoring information system is connected through a communication protocol, and the dynamic parameters are acquired in real time. In the acquisition process, a parallel processing mechanism is adopted to ensure that the batch extraction of the historical operation data and the high-frequency acquisition of the real-time operation data do not interfere with each other. Thus, the seamless connection between the historical operation data and the real-time operation data is realized, the integrity and timeliness of the data are ensured, and the demand of the model for full-cycle data is met.

[0029] The data formats of different sources are different, such as numerical units and time stamp formats, and contain invalid values, such as abnormal jumps caused by sensor failure. The data needs to be preprocessed to meet the compatibility requirements of the subsequent modules. The acquired various types of data are converted into a unified structured format, such as table type data, the time stamp accuracy and physical units are unified, and the data is preliminarily checked through the preset parameter reasonable range, and the abnormal values beyond the range are marked, wherein the parameter reasonable range includes the normal range of the turbine speed. Thus, the data format difference is eliminated, the interference of invalid data on subsequent processing is reduced, and the consistency and reliability of the data are improved.

[0030] The unit parameters include main steam pressure, steam extraction amount, turbine speed, load instruction and boiler feed water flow.

[0031] The main steam pressure refers to the pressure of the main steam at the outlet of the boiler, which is a core parameter reflecting the heat storage capacity of the boiler. The stability of the main steam pressure directly affects the work efficiency of the steam turbine, and further affects the power response speed of the unit during primary frequency modulation. Primary frequency modulation requires rapid adjustment of unit output, while boiler heat storage is the key support for short-term output regulation. The dynamic change of the main steam pressure can directly reflect the state of heat storage release or replenishment. The pressure sensor installed on the main steam pipe of the boiler is used to obtain real-time data. The sampling frequency is synchronized with the unit control system. Thus, the real-time state of the boiler energy reserve is provided to assist in predicting the short-term power regulation potential of the unit during frequency modulation.

[0032] The steam extraction amount refers to the amount of steam extracted from different stages of the steam turbine, which is mainly used for heating or heat recovery systems. Changes in the steam extraction amount will change the effective enthalpy drop of the steam turbine and affect the power generation. For heating units, there is a coupling relationship between steam extraction and electrical load. During primary frequency modulation, the constraints of steam extraction on power regulation need to be considered to avoid deviations in frequency modulation response caused by steam extraction fluctuations. The flow sensor on the steam extraction pipe is used to collect data, which is corrected by the steam extraction pressure and temperature parameters to ensure data accuracy. Thus, the energy distribution state of the cogeneration unit is reflected to improve the prediction accuracy of the model for the frequency modulation characteristics under heating conditions.

[0033] The steam turbine speed refers to the rotational speed of the steam turbine rotor, which is directly related to the grid frequency. The speed and frequency have a fixed proportional relationship, which is the core basis for triggering primary frequency modulation. The grid frequency deviation is reflected through the speed change, and the speed fluctuation amplitude determines the action depth and adjustment direction of primary frequency modulation, which is the original input signal of frequency modulation control. The speed sensor on the steam turbine shaft is used to continuously collect data. The speed sensor includes a magnetic resistance sensor, and the signal is transmitted to the data acquisition module after filtering. Thus, the trigger source information of the frequency modulation action is provided to ensure rapid perception and response prediction of frequency disturbances.

[0034] The load instruction refers to the target power instruction issued by the grid dispatching or unit coordination control system, which is the reference value for unit operation. Primary frequency modulation is a dynamic correction based on the load instruction. The size and trend of the load instruction will affect the amplitude and range of frequency modulation, such as the limited frequency modulation capacity under low load conditions. The instruction value is directly read from the control module of the CCS system, and the issuance time and change rate of the instruction are recorded synchronously. Thus, the basic operation target of the unit is determined to assist the model in judging the coordination relationship between frequency modulation and conventional load regulation.

[0035] The boiler feed water flow is the feed water flow into the boiler economizer, is the key parameter to maintain the normal water circulation and steam production of the boiler, and directly affects the thermal output stability of the boiler; the dynamic change of the feed water flow reflects the load response capability of the boiler, and the boiler needs to quickly adjust the thermal output to support the power change of the steam turbine during primary frequency modulation, and the feed water flow is an important regulation variable; the accuracy of the flow measurement is ensured by collecting the flow through the flow meter on the feed water pipeline, combining the feed water pressure and temperature parameters for density correction; thereby reflecting the dynamic adjustment state of the boiler, providing a basis for model prediction of long-term frequency modulation capability, such as output maintenance for more than 15 seconds.

[0036] The data acquisition module is in communication connection with the DCS system or SIS system of the thermal power generating unit to acquire real-time operation data of the thermal power generating unit in real time.

[0037] The communication protocols adopted by the DCS / SIS system of the thermal power generating unit are different, such as Modbus, OPC UA and PROFINET, and the data acquisition module needs to be compatible with multiple protocols to realize effective connection; an extensible communication interface module is configured, a plurality of industrial communication protocol analyzers are built-in, the communication protocol type of the DCS / SIS system is identified through a protocol scanning tool, and the corresponding interface driver program is automatically matched; thereby realizing seamless connection of the data acquisition module with different types of control systems, and improving the universality and compatibility of the system.

[0038] The operation data of the thermal power generating unit belongs to key industrial data, and the real-time performance and safety need to be ensured during the transmission process to avoid data delay or leakage; an industrial Ethernet is adopted to establish a special data transmission link, fixed communication ports and data buffer areas are set, symmetric encryption algorithm is adopted to encrypt the transmitted data, and only authorized modules can decrypt; thereby ensuring the real-time performance and safety of data transmission, and preventing data from being tampered with or illegally acquired.

[0039] The communication link will be interrupted due to network fluctuations and equipment failures, etc., resulting in interruption of data acquisition and affecting the continuity of the system; the data packet loss rate, delay time and other indicators of data transmission are monitored in real time, when the indicators exceed the threshold, the fault diagnosis mechanism is triggered, such as checking the interface connection and protocol handshake state, if it is determined that the link is interrupted, the system automatically switches to the backup communication link, such as a redundant Ethernet port, and records the fault information; thereby improving the reliability of data acquisition, reducing data loss caused by communication failure, and ensuring the continuous and stable operation of the system.

[0040] The feature screening module is used for preprocessing the historical operation data and real-time operation data, and extracting a frequency modulation capability feature set from the preprocessed historical operation data through a random forest algorithm, the frequency modulation capability feature set including boiler heat storage parameters, valve flow characteristic parameters, control characteristic parameters and equipment state parameters.

[0041] The feature screening needs to be based on the historical and real-time operation data provided by the data acquisition module. If the data range is not clear, irrelevant data may enter the processing flow, increasing the calculation load and affecting the effectiveness of the features. The structured data transmitted by the data acquisition module is received through a preset interface, the data range is determined as the historical operation data and real-time operation data related to a frequency modulation action, and the two types of data are associated through data identifiers including timestamps and action numbers. Thus, the input data is highly matched with the frequency modulation characteristic prediction requirements, irrelevant data interference is avoided, and a foundation is laid for subsequent preprocessing and feature extraction.

[0042] The original operation data may contain abnormal values and missing values, where the abnormal values include jumps caused by sensor failures, and the missing values include data transmission interruptions. Direct use may cause feature extraction bias and reduce model prediction accuracy. Abnormal data outside the reasonable range is identified and removed through Grubbs analysis, missing data is filled through KNN algorithm, and the data is standardized, such as unifying parameter units and eliminating dimension effects. Thus, the data quality is improved, noise interference is reduced, and the accuracy of subsequent feature extraction is ensured.

[0043] The preprocessed data still contains a large number of parameters, which need to be screened through an algorithm to extract core features related to frequency modulation capability and avoid feature redundancy leading to high model complexity. The historical operation data after preprocessing is trained through a random forest algorithm, the feature importance is calculated, such as the contribution of parameters to frequency modulation response, and a candidate feature set related to frequency modulation capability is preliminarily extracted. Thus, the feature range is narrowed down, key parameters with significant impact on frequency modulation characteristics are retained, and a foundation is provided for subsequent refined screening.

[0044] There may be secondary features in the preliminarily extracted candidate feature set, which need to be further screened through quantitative indicators to ensure the representativeness and effectiveness of the final feature set. The variable projection importance value of the candidate features is calculated, the parameters with variable projection importance value greater than the preset projection threshold are selected as core features, and a frequency modulation capability feature set is formed, including boiler heat storage parameters and valve flow characteristic parameters. Thus, the features with the most significant impact on frequency modulation dynamic characteristics are accurately located, the model input dimension is reduced, and the relevance of features and prediction targets is improved.

[0045] The influence of features on frequency modulation capability differs under different thermal states of the unit, such as stable or fluctuating thermal states. Features extracted based on raw data cannot reflect the true correlation. By determining whether the deviation of real-time operation data of the unit is less than the preset accuracy in consecutive multiple sampling periods, the thermal stability is determined, the features under stable state are weighted and adjusted, and the pertinence of the feature set is optimized. Thus, the feature set is more suitable for the actual operation state of the unit, and the prediction accuracy of the model for frequency modulation characteristics under different working conditions is improved.

[0046] The historical operation data and real-time operation data of the thermal power unit are preprocessed, including removing abnormal data by Grubbs analysis method and filling missing data by KNN algorithm.

[0047] Abnormal data may be generated in unit operation due to sensor failure and electromagnetic interference, such as sudden rise and fall of main steam pressure. If not removed, it will distort the feature distribution and cause feature extraction distortion. Grubbs analysis method is adopted, and a reasonable fluctuation range is set based on the historical distribution of parameters. The extreme values exceeding the range are marked and removed, and the reasons for removal are recorded, such as deviation from mean value by 3 times standard deviation. Thus, the noise in the data is eliminated, and the feature extraction is based on real and reliable operation data.

[0048] Data transmission interruption and storage failure may cause data missing. If directly ignored, it will cause incomplete information and affect the continuity and integrity of the features. Through KNN algorithm, the most similar working condition sample is matched in the historical data according to the working condition parameters before and after the missing data, such as similar load and similar frequency deviation, and the corresponding parameter value in the sample is used to fill the missing data. Thus, the integrity of the data sequence is maintained, and the feature extraction is avoided from appearing breakpoints or deviations due to missing values.

[0049] Different parameters have different physical dimensions. Direct calculation may cause the feature importance evaluation to be biased towards parameters with large values, affecting the fairness of screening. Through Min-Max standardization, all parameters are mapped to the same numerical interval, such as 0-1, eliminating the dimensional difference and making each parameter have equal weight in feature evaluation. Thus, the objectivity of feature importance evaluation is ensured, and the screening bias caused by dimensional difference is avoided.

[0050] When the feature selection module extracts the frequency modulation capability feature set, the frequency modulation capability feature set is determined by calculating the variable projection importance value, and the parameters with variable projection importance value greater than the preset projection threshold are selected as the frequency modulation capability features.

[0051] The preprocessed data still contains dozens or even hundreds of parameters, which need to be quickly narrowed down by algorithm to focus on potential features related to frequency modulation capability. A plurality of decision trees are constructed by random forest algorithm, and the contribution of each parameter in the decision tree splitting process is calculated, such as Gini coefficient reduction, to preliminarily select candidate features with high contribution, such as main steam pressure and valve opening rate, etc. Thus, irrelevant parameters, such as auxiliary system parameters not directly related to frequency modulation, are efficiently removed, reducing the subsequent calculation amount.

[0052] The importance of the candidate features needs to be accurately measured by quantitative indicators, and the default weight of the algorithm cannot reflect the actual influence of the parameters on the frequency modulation characteristics; the variable projection importance value of the candidate features is calculated, which comprehensively reflects the explanation ability of the parameters to the frequency modulation capability index, and the higher the variable projection importance value, the more significant the influence of the parameter on the frequency modulation characteristics; thereby providing a quantitative basis for feature selection and avoiding the deviation caused by subjective judgment.

[0053] A clear threshold needs to be set to select the most critical features, and too many features will increase the complexity of the model, and too few will lose important information; the projection threshold is set according to the training results of historical data, for example, selecting the top 30% of the variable projection importance value parameters, and the parameters that meet the conditions are included in the frequency modulation capability feature set, including boiler heat storage parameters and valve flow characteristic parameters, among which the boiler heat storage parameters include the main steam pressure change rate, and the valve flow characteristic parameters include the valve opening degree response time, etc.; thereby forming a simplified and efficient feature set that balances model accuracy and computational efficiency.

[0054] The feature selection module also optimizes features in combination with the thermal stability state of the unit under different operating conditions when extracting the frequency modulation capability feature set, and the thermal stability state is determined by the deviation of the real-time operating data of the thermal power generating unit being less than the acquisition accuracy in continuous multiple sampling periods.

[0055] The operating parameters of the thermal power generating unit under the thermal instability state will be distorted, and the features extracted based on such data cannot truly reflect the frequency modulation capability, so it is necessary to distinguish between stable and unstable states; by monitoring the deviation of real-time operating data, if the deviation is less than the preset accuracy in continuous multiple sampling periods, it is determined to be in a thermal stability state; otherwise, it is in an unstable state; thereby clearly defining the reliable operating condition range for feature extraction, and avoiding the influence of invalid data under the unstable state on the effectiveness of the features.

[0056] The parameters under the thermal stability state can better reflect the real characteristics of the unit, and the weight of the features under such state needs to be strengthened to improve the relevance of the features and the frequency modulation capability; the historical data segments determined to be in a thermal stability state are given a higher weight, for example, increasing the sample weight in random forest training, so that the features under the stable state occupy a larger proportion in the importance evaluation, while the influence of the features under the unstable state is weakened; thereby making the feature set more consistent with the actual frequency modulation capability of the unit, and improving the prediction accuracy of the model for the frequency modulation characteristics under stable operating conditions.

[0057] The unit operating condition changes with the running time, for example, load rise and fall and seasonal adjustment, the parameter distribution of thermal steady state will change, and the feature set needs to be dynamically adjusted to adapt to the new working condition; the current thermal steady state is re-determined regularly, the feature importance distribution under the new working condition is compared, if the variable projection importance value of the core feature changes more than the preset threshold, the feature set is updated, the new key parameters are included or the invalid parameters are excluded; thereby ensuring the timeliness and adaptability of the feature set, so that the model can still maintain high prediction accuracy when the unit operating condition changes.

[0058] The model prediction module is configured to construct a multi-level driving model based on a deep learning algorithm, and input the frequency modulation capability feature set into the multi-level driving model, and output a prediction result of the frequency modulation characteristic parameter of the thermal power generating unit under the operating condition parameter and the grid disturbance data, the prediction result including power regulation response speed, frequency difference tracking accuracy and output regulation amplitude, and the prediction result is displayed and sent to the unit control system to optimize the frequency modulation control.

[0059] The input data of the model prediction is derived from the frequency modulation capability feature set output by the feature screening module and the real-time operating data of the data acquisition module, and if the data has format error or is missing, it will directly lead to distortion of the prediction result; the frequency modulation capability feature set and the real-time operating data are received through a preset interface, the data format is checked, the data format including parameter dimension and data type, and the data integrity is checked, if there is missing key parameter, the backup data preprocessed by the feature screening module is triggered or used; thereby ensuring the standardization and integrity of the input data, and laying a foundation for the accuracy of the model prediction.

[0060] The model prediction needs to be based on the pre-trained multi-level driving model, and the loading model and the execution prediction are the core functional links; the stored multi-level driving model is called, the model including an input layer, a feature mapping layer, a characteristic prediction layer and an output layer, the feature set and the real-time data after verification are input into the model, and the deep learning algorithm is used for operation, the deep learning algorithm including a long short-term memory network based on an attention mechanism, and the prediction result of the frequency modulation characteristic parameter is output, the prediction result including power regulation response speed, frequency difference tracking accuracy and output regulation amplitude; thereby realizing the core prediction function, and converting the abstract feature data into specific parameters which can directly guide the frequency modulation control.

[0061] The model prediction result will have errors, which need to be evaluated to determine its reliability, and if necessary, to be corrected to improve its practicability; based on the deviation law of the predicted value and the actual value in the historical data, the current prediction result is calculated for the qualified rate and evaluated for the risk value, wherein the qualified rate includes the proportion meeting the frequency modulation requirement, the risk value evaluation includes the probability of deviating from the qualified range, if the risk value exceeds the preset risk threshold, a correction algorithm is called to fine-tune the prediction result, wherein the correction algorithm includes a compensation function based on historical deviation; thereby improving the credibility of the prediction result and reducing the control decision errors caused by model errors.

[0062] The prediction result needs to be presented to the operator in an intuitive way and synchronized to the unit control system to achieve optimal control; the prediction result is converted into a visual chart, such as a frequency modulation response curve and a parameter sensitivity analysis chart, which is displayed through a human-computer interaction interface, and at the same time, the structured prediction data, including the response speed and the adjustment amplitude in digital form, is converted into a format according to the communication protocol of the unit control system and sent to the DCS or CCS system; thereby improving the readability and operability of the prediction result and realizing seamless connection with the unit control link.

[0063] The frequency modulation control needs to meet the examination requirements of two grid rules, and the prediction result needs to be associated with the examination index to determine the optimization direction; the predicted frequency modulation characteristic parameters are mapped and analyzed with the examination index, the examination compliance probability corresponding to the prediction result is calculated, and the targeted optimization suggestions are output, such as adjusting the valve response speed to improve the output response index; thereby making the prediction result directly serve the optimization of the examination index and enhancing the adaptability of the system to the actual operation examination demand.

[0064] As shown in Figure 2 The multi-level driving model includes an input layer, a feature mapping layer, a characteristic prediction layer, and an output layer, the input layer is used to receive the frequency modulation capability feature set, the feature mapping layer is used to strengthen the weight of the frequency modulation capability feature through the attention mechanism, the characteristic prediction layer is used to simulate the frequency modulation response process under different operating condition parameters, and the output layer is used to output the prediction result of the frequency modulation characteristic parameters.

[0065] The frequency modulation process of thermal power generating units involves multiple links, such as boiler heat storage and valve adjustment, which need to be simulated respectively through a multi-level structure to ensure the pertinence of the model; the model is divided into an input layer, a feature mapping layer, a characteristic prediction layer, and an output layer, the input layer is responsible for receiving the frequency modulation capability feature set, the feature mapping layer strengthens the weight of the key features through the attention mechanism, wherein the key features include the main steam pressure change rate, the characteristic prediction layer simulates the frequency modulation response process under different operating conditions, such as the adjustment difference under high load and low load, and the output layer outputs the final prediction result of the frequency modulation characteristic parameters; thereby making the model structure match the physical process of unit frequency modulation and improving the accuracy of prediction.

[0066] The deep learning model needs a large amount of high-quality data for training to ensure performance, and needs to reduce the risk of overfitting through parameter optimization; Select the frequency modulation data under different operating parameters in the historical operation as the training sample, where the input is the frequency modulation capability feature set, and the output is the actual frequency modulation characteristic parameter; Use the back propagation algorithm to iteratively optimize the model parameters, such as the weights of each layer of neurons, and adjust the network depth and node number through cross-validation to avoid overfitting; Thus, the model has generalization ability and can adapt to the frequency modulation characteristic prediction under different operating conditions.

[0067] Different features have different degrees of influence on frequency modulation characteristics, for example, valve flow characteristics are more important than secondary equipment parameters, and attention mechanism is needed to highlight the role of key features; Set the attention weight matrix in the feature mapping layer, and based on the correlation between features and frequency modulation results in historical data, assign dynamic weights to each feature, for example, the weight of the valve flow characteristic parameter is higher than that of the auxiliary system parameter, so that the model focuses on high-weight features during operation; Thus, the sensitivity of the model to key influencing factors is improved, and the interference of secondary features on the prediction result is reduced.

[0068] The unit operating condition changes over time, such as equipment aging and load range adjustment, and the model needs to have adaptability to ensure long-term effectiveness; Regularly use newly collected operating data to verify the model, if the prediction error exceeds the threshold, start incremental training, that is, only fine-tune the model parameters with new data, update the weight configuration of the feature mapping layer for new characteristics after unit modification, such as valve replacement, to ensure that the model matches the actual equipment characteristics; Thus, the effective life cycle of the model is extended, and the dynamic changes of the unit operating state are adapted.

[0069] The prediction results of the frequency modulation characteristic parameters are evaluated, and the qualified rate of the frequency modulation action and the risk value of the frequency modulation action are output.

[0070] The evaluation standard needs to be clear to quantify the reliability of the prediction results and avoid subjective judgment; Define the frequency modulation action qualified rate and the frequency modulation action risk value, the frequency modulation action qualified rate refers to the proportion of the prediction results falling within the qualified range, and the qualified range is set based on the grid frequency modulation requirements, the frequency modulation action risk value refers to the probability of the prediction results deviating from the qualified range, which is calculated through the historical error distribution, forming a complete evaluation index system; Thus, an objective basis is provided for the prediction result evaluation, ensuring the consistency and comparability of the evaluation results.

[0071] The quality of the prediction result intuitively reflected by specific numerical values provides a reference for decision-making; the predicted frequency modulation characteristic parameters are compared with the preset qualified interval, and the qualified rate is counted, wherein the predicted frequency modulation characteristic parameters include the power regulation response speed, the preset qualified interval includes the response speed range required by the power grid, and the probability that the current prediction value exceeds the qualified interval, i.e., the risk value, is calculated based on the historical prediction error distribution of the parameter, such as normal distribution; thereby quantifying the reliability of the prediction result to help the operator quickly judge whether the result can be used for control adjustment.

[0072] When the prediction result risk is high, a correction mechanism needs to be triggered to avoid misleading control decisions; a risk threshold is set, and if the current risk value exceeds the risk threshold, it is determined that the result is not reliable enough, and a correction module is automatically called to compensate for the current result based on the prediction deviation under the same working condition in the near future; thereby reducing the negative impact of high-risk prediction results on control optimization and improving the robustness of the system.

[0073] The prediction result of the frequency modulation characteristic parameter is displayed in the form of a visual chart, including a frequency modulation response curve, a parameter sensitivity analysis graph, and a prediction error distribution graph.

[0074] The prediction result in the form of pure numbers is difficult to intuitively reflect the trend and relevance, and needs to be enhanced in readability through charts; three types of core charts are designed, wherein the horizontal axis of the frequency modulation response curve is time, and the vertical axis is the predicted power regulation value, reflecting the dynamic change process, the parameter sensitivity analysis graph is a bar chart showing the influence degree of each characteristic parameter on the prediction result, and the prediction error distribution graph is a histogram showing the distribution of historical prediction errors; thereby making the prediction result and the underlying influencing factors easier to understand, and assisting the operator to quickly grasp the key information.

[0075] The running state of the thermal power generating unit changes in real time, and the prediction result needs to be dynamically updated to reflect the latest state; a visualization tool is used to automatically generate charts, such as a WebGL-based drawing library, when new prediction results are generated, a chart refreshing mechanism is triggered, historical curves are retained to compare trends, and chart interaction is supported, such as zooming and local magnification, to facilitate the operator to view details; thereby ensuring the timeliness of the displayed content and improving the operator's intuitive perception of the frequency modulation dynamic process.

[0076] The model prediction module also synchronously associates the examination index in the primary frequency modulation action process when outputting the prediction result of the frequency modulation characteristic parameter, and the examination index includes the output response index and the power contribution index, to support the frequency modulation control optimization based on the examination index.

[0077] The corresponding relationship between the predicted frequency modulation characteristic parameters and the grid evaluation indicators needs to be determined in order to associate the prediction results with the evaluation requirements; the calculation rules of the evaluation indicators in the two regulations, such as the 15-second output response index and the power contribution index, are analyzed, and the mathematical mapping relationship between the frequency modulation characteristic parameters and the evaluation indicators is established, for example, the faster the response speed, the higher the 15-second output response index; thereby providing a basis for the association analysis of the prediction results and the evaluation indicators, and directly linking the technical parameters with the management requirements.

[0078] The corresponding evaluation performance of the current prediction results needs to be determined through association analysis to indicate the direction of optimization control; the predicted frequency modulation characteristic parameters are substituted into the mapping relationship to calculate the corresponding evaluation index prediction value, for example, the 15-second output response index is calculated based on the predicted response speed, which is compared with the evaluation standard to determine whether it meets the standard, and the influencing factors of the unqualified indicators are analyzed, for example, the insufficient output regulation amplitude leads to the low power contribution index; thereby converting the abstract prediction results into specific evaluation standard expectations, and making the control optimization more targeted.

[0079] Specific optimization suggestions need to be given according to the results of the association analysis in order to effectively support control decisions; for the unqualified evaluation indicators, the frequency modulation characteristic parameters that need to be adjusted are deduced in reverse, for example, the power contribution index does not meet the standard and the output regulation amplitude needs to be increased, combined with the feature importance analysis of the feature screening module, the device parameters that can be optimized are prompted, for example, increasing the valve opening to improve the regulation amplitude; thereby converting the evaluation requirements into specific device control suggestions, and shortening the decision chain from prediction to optimization control.

Claims

1. A system for predicting dynamic characteristics of frequency modulation of a thermal power unit, characterized in that, The utility model relates to a kind of power plant unit's primary frequency modulation characteristic prediction method and system, including: Data acquisition module, feature screening module, model prediction module; The data acquisition module is used to obtain the historical operation data and real-time operation data of thermal power generating unit, and the historical operation data includes unit parameters, control instructions and power grid disturbance data when primary frequency modulation is operated, and the real-time operation data includes operating condition parameters, frequency deviation signal and output regulation instruction; The feature screening module is used to preprocess the historical operation data and real-time operation data, and extract the frequency modulation capability feature set from the preprocessed historical operation data through random forest algorithm, and the frequency modulation capability feature set includes boiler heat storage parameter, valve flow characteristic parameter, control characteristic parameter and equipment state parameter; The model prediction module is used to construct a multi-level drive model based on deep learning algorithm, and input the frequency modulation capability feature set into the multi-level drive model, output the prediction result of frequency modulation characteristic parameter of thermal power generating unit under operating condition parameter and power grid disturbance data, and the prediction result includes power regulation response speed, frequency difference tracking accuracy and output regulation amplitude, and is simultaneously displayed and sent to unit control system to optimize frequency modulation control.

2. The system for predicting dynamic characteristics of frequency modulation of a thermal power generating unit according to claim 1, wherein The unit parameters include main steam pressure, steam extraction amount, turbine speed, load instruction and boiler feed water flow.

3. The system for predicting the dynamic characteristics of frequency modulation of a thermal power generating unit according to claim 2, wherein The data acquisition module is communicatively connected with the DCS system or SIS system of the thermal power generating unit to obtain the real-time operation data of the thermal power generating unit in real time.

4. The system for predicting the dynamic characteristics of frequency modulation of a thermal power generating unit according to claim 3, wherein The preprocessing of the historical operation data and real-time operation data of the thermal power generating unit includes eliminating abnormal data by Grubbs analysis method and filling missing data by KNN algorithm.

5. The system for predicting the dynamic characteristics of a frequency modulation of a thermal power generating unit according to claim 4, wherein When extracting the frequency modulation capability feature set, the feature screening module determines the frequency modulation capability feature set by calculating variable projection importance value, and selects the parameters with variable projection importance value greater than the preset projection threshold as the frequency modulation capability feature.

6. The system for predicting the dynamic characteristics of frequency modulation of a thermal power generating unit according to claim 5, wherein When extracting the frequency modulation capability feature set, the feature screening module also optimizes the features in combination with the thermal stability state under different operating condition parameters of the unit, and the thermal stability state is determined by the deviation of the real-time operation data of the thermal power generating unit being less than the acquisition accuracy in continuous multiple sampling periods.

7. The system for predicting the dynamic characteristics of a frequency modulation of a thermal power generating unit according to claim 6, wherein The multi-level drive model includes input layer, feature mapping layer, characteristic prediction layer and output layer, the input layer is used to receive the frequency modulation capability feature set, the feature mapping layer is used to strengthen the weight of the frequency modulation capability feature through attention mechanism, the characteristic prediction layer is used to simulate the frequency modulation response process under different operating condition parameters, and the output layer is used to output the prediction result of frequency modulation characteristic parameter.

8. The system for predicting the dynamic characteristics of a frequency modulation of a thermal power generating unit according to claim 7, wherein The prediction result of frequency modulation characteristic parameter is evaluated, and the qualified rate of frequency modulation action and the risk value of frequency modulation action are output.

9. The system for predicting the dynamic characteristics of a frequency modulation of a thermal power generating unit according to claim 8, wherein, The prediction result of frequency modulation characteristic parameter is displayed in the form of visual chart, including frequency modulation response curve, parameter sensitivity analysis graph and prediction error distribution graph.

10. The system for predicting the dynamic characteristics of a frequency modulation of a thermal power generating unit according to claim 9, wherein, When outputting the prediction result of frequency modulation characteristic parameter, the model prediction module also synchronously associates the evaluation index in the process of primary frequency modulation, and the evaluation index includes output response index and power contribution index, to support the optimization of frequency modulation control based on evaluation index.

Images