Steam temperature autonomous cruise method based on time sequence classification

Abstract

The invention provides a steam temperature autonomous cruise method based on time sequence classification, which comprises the following steps of: acquiring historical data and preprocessing the historical data to establish a Transform model; collecting current data for preprocessing, and inputting the trained model to obtain a steam temperature state classification result so as to identify an operation state; when the identification result is the abnormal working condition, a preset expert strategy library is called to match a corresponding control strategy, and whether a progressive steam temperature set value compensation strategy is additionally executed or not is detected in real time until the abnormal working condition is finished; meanwhile, the model is updated and the expert strategy library is corrected according to historical correction records. By establishing the time sequence classification model, working conditions under different trends can be classified carefully and effectively, a control strategy is correspondingly selected for correction, and a set value compensation strategy is timely combined under the condition that expectation is not met, so that the response speed and the regulation and control precision of the system can be improved; and a guarantee is provided for safe and stable operation of the thermal generator set.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal power generator set control, and in particular to a steam temperature autonomous cruise method based on Transformer time series classification, which is used to realize autonomous regulation of main / reheat steam temperature. Background Art

[0002] In the energy field, thermal power units are an important part of power supply and play a key role in ensuring stable electricity consumption for social production and life. During the operation of thermal power units, the main / reheat steam temperature is one of the most important operating parameters. Maintaining a stable and appropriate main / reheat steam temperature can not only improve the thermal efficiency of the unit and reduce energy consumption, but also has a vital impact on the safe and stable operation of the unit.

[0003] However, the control of main / reheat steam temperature faces many challenges. It is affected by many complex factors such as the output of each layer of coal mill, the primary air flow of each coal mill, the opening of each layer of air door, the swing angle of the burner, the center position of the flame, the flow rate of cooling water, etc. These factors are interrelated and influence each other, showing a high degree of nonlinearity and coupling.

[0004] The traditional modeling method based on mechanism analysis attempts to establish the main / reheat steam temperature model through in-depth understanding of the physical process and mathematical derivation. However, due to the complexity and uncertainty of various factors in actual operation, it is difficult to accurately describe all influencing factors and their interactions, resulting in limited accuracy and reliability of the model. For example, under different load conditions, the degree and law of the influence of various factors on steam temperature will change, making it difficult for models based on fixed mechanisms to adapt to the diversity of actual operations.

[0005] The data-driven modeling approach relies on a large amount of actual operating data and builds models through data analysis and mining technology. However, there are many factors affecting the main / reheat steam temperature, and data collection, processing and analysis are extremely difficult. In addition, there may be noise, missing values ​​and other problems in the data, which further increases the complexity of modeling. At the same time, data-driven models often lack an in-depth understanding of the physical process and are insufficient in terms of explanatory power and generalization.

[0006] In addition, in actual operation, there are also problems with the adjustment of the main / reheat steam temperature set value. Due to the frequent changes in operating conditions, the set value needs to be adjusted frequently, which not only increases the operation workload of the operator, but is also easily affected by the subjective will of the operator, thus bringing safety hazards. Once the set value is improperly adjusted, the steam temperature may be too high or too low, affecting the efficiency and safety of the unit, and even causing equipment failure.

[0007] In summary, the existing main / reheat steam temperature control technology has problems such as modeling difficulties, low control accuracy and unreasonable set value adjustment when facing complex influencing factors and actual operation requirements. A new method is urgently needed to solve these problems in order to improve the operating efficiency and safety of thermal power units. Summary of the invention

[0008] In view of the problems existing in the prior art, the present invention provides a steam temperature autonomous cruising method based on time series classification which has high processing efficiency and can be automatically operated.

[0009] To achieve the above purpose, the technical solution adopted by the present invention is as follows: The present invention provides a steam temperature autonomous cruising method based on time series classification, which comprises the following steps: Collecting historical data of thermal power unit operating parameters, and preprocessing the data of historical thermal power unit operating parameters to obtain preprocessed historical data; A Transformer model including an encoding component and a decoding component is established, and the time series correlation characteristics of the operating parameters of the thermal power unit are captured through a self-attention mechanism; the pre-processed historical data is input into the Transformer model for training; Collecting data of the current thermal power unit operating parameters for preprocessing to obtain preprocessed real-time data; inputting the preprocessed real-time data into the trained Transformer model, and outputting the classification results of the steam temperature state under all working conditions; Identify the operating state of the thermal power unit according to the full-condition steam temperature state classification result and preset information; When the identification result of the operating state is an abnormal operating condition, a preset expert strategy library is called to match the corresponding control strategy, and the execution result of the control strategy is compared and judged in real time by a threshold comparison method; When the result of the comparison and judgment exceeds the preset range, a progressive steam temperature set point compensation strategy is additionally executed until the abnormal operating condition ends; The parameters of the Transformer model are updated according to the historical correction records, and the priority weights of the control strategies in the expert strategy library are dynamically corrected.

[0010] Optionally, the operating parameters of the thermal power unit include at least coal mill output, primary air flow, burner swing angle, air door opening, flame center position and cooling water flow.

[0011] Optionally, the preprocessing includes the following steps: Divide the rated value range of the operating parameters of thermal power units according to the operating conditions of the thermal power units; Based on the rated value range, outliers are eliminated from the data collected from the operating parameters of the thermal power unit, and the data are supplemented by interpolation to obtain preprocessed data of the operating parameters of the thermal power unit.

[0012] Optionally, the outlier removal further includes the following steps: Taking the sampling step length as the calculation period, and calculating the maximum change value in the calculation period; When the maximum change value is greater than the allowable range, the maximum change value is removed and supplemented by a data smoothing algorithm.

[0013] Optionally, when the identification result of the operating state is an abnormal operating condition, the method further includes the following steps: Perform multi-level attribution analysis on the abnormal operating conditions, including equipment layer attribution factors, operation layer attribution factors and environment layer attribution factors; The device-level attribution factors: detecting actuator deviation and sensor drift; The operating layer attribution factors: detection of coal distribution uniformity and air-coal ratio deviation; The environmental layer attribution factors: calculation of coal calorific value fluctuations and environmental temperature gradients.

[0014] Optionally, the calling strategy of the expert strategy library includes: Establish a multi-level strategy tree based on historical abnormal operating conditions and corresponding correction control strategies; The similarity between the real-time data and the historical data is calculated, and a group of historical data with the highest similarity is selected to retrieve the multi-level strategy tree, and the control strategy is obtained.

[0015] Optionally, the dynamic modification of the expert strategy library includes the following steps: Calculate the strategy execution success rate and dynamically modify the weight of the corresponding control strategy based on the strategy execution success rate.

[0016] Optionally, the dynamic modification of the expert strategy library further includes the following steps: Based on the last-in-first-out system, the control strategies that have not been used or have the lowest utilization rate in the preset time period are eliminated.

[0017] Optionally, in the Transformer model: The number of encoding component layers of the Transformer model is no less than 4; The self-attention mechanism adopts a multi-head attention structure.

[0018] Optionally, the Transformer model update includes the following steps: Keep the encoding component structure and update the fully connected layer parameters; A sliding window mechanism is used to retain valid training data in adjacent time periods.

[0019] Compared with the prior art, the present invention has the following beneficial effects: By establishing a Transformer time series classification model, the present invention can carefully and effectively classify operating conditions under different trends, and select corresponding control strategy corrections for abnormal operating conditions. In addition, when the correction is ineffective or does not meet expectations, the set value correction strategy is timely combined to improve the response speed and control accuracy of the system, effectively prevent the occurrence of overtemperature, and provide a guarantee for the safe and stable operation of thermal power generating units. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] Figure 1 It is a flow chart of the method in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0024] It is worth noting that the methods used in the present invention are all conventional methods unless otherwise specified; the raw materials and devices used are all conventional commercially available products, and their sources are not specifically limited unless otherwise specified.

[0025] It should also be noted that in order to facilitate understanding in the specific embodiments of the present invention, the method steps are described in a certain order, but those skilled in the art can change the order of the steps according to actual needs, and therefore this cannot be used as a limiting condition.

[0026] like Figure 1As shown, this embodiment provides a method for autonomous cruise control of vehicle temperature based on time series classification, comprising the following steps: The implementation process of the present invention is described in detail below in conjunction with a main steam temperature control scenario of a thermal power generating unit: Step 1: Data collection and preprocessing; 1.1 Data collection: Collecting historical data of thermal power unit operating parameters, and preprocessing the historical data of thermal power unit operating parameters to obtain preprocessed historical data; First, data collection is carried out on the key parameters in the thermal power generating unit, including the primary air flow rate of each coal mill, the opening of each layer of air door, the swing angle opening of the burner and other parameters that can characterize the operating status of the steam temperature system, so as to fully understand the operating status of the unit.

[0027] Specifically, parameters such as coal mill output (average current signal of 5 coal mills), primary air flow (ultrasonic measurement value in the air duct), burner swing angle (hydraulic feedback signal), damper opening (DCS command feedback), flame center position (average infrared imaging coordinates), and cooling water flow (accumulated value of flow meter) are collected, with a sampling frequency of 1 Hz.

[0028] 1.2 Data preprocessing: When selecting parameters that affect the main / reheat steam temperature, the main consideration is logic. Before selecting parameters, it is necessary to reasonably analyze the physical process that affects the main / reheat steam temperature during operation, and it needs to be consistent with the actual unit operating conditions. Therefore, the output of each layer of coal mill, the primary air flow of each coal mill, the opening of each layer of air door, the swing angle of the burner, the center position of the flame and the cooling water flow are finally selected as the key factors affecting the main / reheat steam temperature.

[0029] Furthermore, outliers are removed from the data collected from the operating parameters of thermal power units based on the rated value range, and the data are supplemented by interpolation to obtain preprocessed data of the operating parameters of thermal power units. Due to practical problems such as sampling noise and abnormal operating conditions, there are many invalid data in the data stored by thermal power units, which need to be screened out before being used for modeling.

[0030] The specific process is as follows: classify the working conditions, define the rated value range of the parameters according to different working conditions, and filter out the data that exceeds the rated value range of the working conditions; For example, the rated range of parameters is divided according to the unit load (300MW / 600MW / 1000MW) (such as the cooling water flow rate is 50-80t / h at 300MW), and the data points outside this range are eliminated. Then, the missing values ​​are supplemented by interpolation to complete the data set.

[0031] Furthermore, the sampling step is used as the calculation period, and the maximum change value in the calculation period is calculated; when the maximum change value is greater than the allowable range, the maximum change value is eliminated and supplemented by the data smoothing algorithm. Specifically, the maximum change value of the data in each sampling step is specified, and the parameter value exceeding the maximum change range is considered as an abnormal value, and the abnormal value is replaced by the mean of the normal value of the nearby time series.

[0032] For example: take 10 seconds as the sampling step, calculate the maximum change value of the parameter (for example, when the burner swing angle changes by more than ±3° / s, it is judged to be abnormal), and use the sliding window weighted average method to complete it.

[0033] Step 2: Transformer model (deep learning model based on self-attention mechanism) training and classification; A Transformer model including encoding and decoding components is established, and the time series correlation characteristics of the operating parameters of the thermal power unit are captured through the self-attention mechanism; the preprocessed historical data is input into the Transformer model for model training to form a time series model for subsequent use. The number of encoding component layers is set to 4, the number of multi-head attention heads in each layer is 8, and the output dimension of the decoding component is consistent with the number of operating condition categories. The operating conditions include: wall temperature overtemperature, air distribution ratio change, damper swing angle change, start-stop grinding disturbance, rapid load change, etc.

[0034] The model adopts the advanced time series classification model Transformer model, which enables the system to accurately analyze the changing trends of the collected parameter data in different time periods, thereby effectively classifying the unit operating conditions. Subsequently, the trained model can learn the main / reheat steam temperature change rules under all operating conditions, and finally obtain an accurate main / reheat steam temperature condition classification model under all operating conditions.

[0035] After that, the data of the current thermal power unit operating parameters are collected and preprocessed to obtain preprocessed real-time data; the preprocessed real-time data is input into the trained Transformer model, and the classification results of the steam temperature state under all working conditions are output. Finally, the operating state of the thermal power unit is identified based on the classification results and preset information.

[0036] The model can accurately determine the impact of each important parameter of the unit on the steam temperature system. According to the impact of each important parameter on the steam temperature system, the specific operating conditions are divided, and the trend of steam temperature changes is monitored at the same time. When the important parameters are monitored to deteriorate, the system will take corresponding intervention measures to effectively avoid the continuation or deterioration of abnormal operating conditions.

[0037] Step 3: Abnormal operating condition attribution and strategy matching; First, establish an expert strategy library; specifically, first collect and classify the abnormal operating conditions that have occurred in the past, and then record the scene information corresponding to each abnormal operating condition one by one based on the past control strategy and the corresponding thermal power unit parameters. On this basis, the system enters the optimal control strategy corresponding to each abnormal operating condition to form a comprehensive control strategy library (expert strategy library), that is, a multi-level strategy tree is established based on historical abnormal operating conditions and corresponding correction control strategies, so as to respond quickly and accurately in actual operation.

[0038] Among them, the specific control strategy is that the system controls the change of the actual value of the steam temperature by adjusting the steam temperature set value. It is not entirely a fault condition, but also includes specific scenarios, such as the initial startup and the rapid load increase period. The operating parameters corresponding to these conditions are also different from the parameters under normal stable state conditions, so they are also classified as abnormal conditions. Therefore, different control schemes are selected according to different conditions. When the disturbance occurs faster, the adjustment of the set value also needs to be faster. When the duration of the disturbance is short, the duration of the set value adjustment needs to match the actual disturbance time. The following are examples of control strategies under some conditions: Grinding disturbance: During the grinding disturbance change, due to special circumstances such as powder accumulation, the set value needs to be lowered by 3-5℃ in a short period of time during the grinding disturbance, and it should be maintained for a short period of time to avoid over-temperature and other situations.

[0039] Rapid load increase: When the load is increased rapidly, over-temperature and other conditions are very likely to occur due to the rapid increase in coal volume. The set value needs to be adjusted down to avoid over-temperature. The reduction range of the set value is adapted to the load variation range, and the duration is the same as the load variation duration.

[0040] 3.1 Multi-level attribution analysis: When the identification result of the operating status is an abnormal condition, it also includes the steps of: performing a multi-level attribution analysis on the abnormal condition, including equipment layer attribution factors, operation layer attribution factors and environment layer attribution factors.

[0041] Among them, the attribution factors of the equipment layer: detection of actuator deviation and sensor drift; the attribution factors of the operation layer: detection of coal distribution uniformity and air-coal ratio deviation; the attribution factors of the environment layer: calculation of coal calorific value fluctuation and ambient temperature gradient. For example: Equipment level: Detection found that the deviation between the #3 burner swing angle feedback signal and the command was greater than 5%; Operation layer: calculate the uniformity of coal distribution (standard deviation > 1.5t / h), and the air-coal ratio deviates from the design value by 10%; Environmental layer: Fluctuation in coal calorific value (current value is 500kcal / kg lower than the benchmark).

[0042] Through multi-level attribution analysis, more abundant technical information under abnormal working conditions can be recorded, thus providing data support for subsequent tracing.

[0043] 3.2 Strategy matching: When the operating status is identified as an abnormal condition, in order to more effectively respond to various abnormal conditions, the Transformer main / reheat steam temperature condition classification model also has the ability to trace the source of each abnormal condition. Through in-depth analysis of historical data and real-time data, the system can accurately find the root cause of the abnormal condition, such as excessive air volume and uneven distribution of coal in each layer. Once the root cause of the abnormality is determined, the system will intelligently select the most appropriate control strategy from the control strategy library to respond based on the preset logic and rules, thereby achieving rapid response and accurate processing of abnormal conditions and ensuring stable operation of the unit.

[0044] For input data x and model parameters θ , working conditions i The classification probability It can be expressed as: ; In the formula, The model is for the working condition i The output score of , C is the total number of operating condition categories.

[0045] In another embodiment, in order to increase the accuracy of the control strategy selection, the similarity between the real-time data and the historical data can be calculated, and a group of historical data with the highest similarity can be selected to retrieve the multi-level strategy tree, and the control strategy can be obtained, and it can be superimposed, fitted or jointly judged with the control strategy corresponding to the model classification result. For example, for cases with matching similarity greater than 90%, the "burner swing angle deviation compensation + air-coal ratio correction" combination strategy in the strategy tree is called.

[0046] Step 4: Control strategy execution and progressive compensation; 4.1 Matching control strategy adjustment: Execute the control strategy finally output by the Transformer model and the expert strategy library, and correct the steam temperature set value.

[0047] 4.2 Progressive compensation: However, due to the complexity of the industrial site, this correction may not achieve the expected effect. It is also necessary to compare and judge the execution results of the control strategy in real time through the threshold comparison method; when the result of the comparison and judgment exceeds the preset range, the progressive steam temperature set value compensation strategy is additionally executed until the abnormal condition ends. Specifically, after the correction and compensation in 4.1, the main / reheat steam temperature is monitored in real time to determine whether the correction is effective and whether it achieves the expected effect. If it does not achieve the expected effect, continue to make corrections according to the actual operating conditions, such as: After the strategy is executed for the first time, the main steam temperature deviation is still +8°C (the preset threshold is ±5°C), triggering progressive compensation: First compensation: The steam temperature set point is reduced by 2°C, but the deviation still exceeds the threshold range; Second compensation: additional reduction of 1.5℃, deviation ≤5℃, end compensation.

[0048] Step 5: Update the strategy library; The dynamic modification of the expert strategy library includes the following steps: Based on the last-in-first-out system, the control strategies that have not been used and / or have the lowest utilization rate in a preset time period are eliminated.

[0049] Furthermore, the success rate of each control strategy is counted. For example, if the success rate of "burner swing angle deviation compensation" is 20% (does not meet the preset success rate threshold of 60%) and is ranked the lowest, it will be removed from the expert strategy library, or the weight of the corresponding control strategy will be dynamically corrected according to the strategy execution success rate. For example, the success rate of the control strategy is proportional to the weight (weight refers to the numerical value corresponding to each node, feature or branch in the multi-level strategy tree, which is used to measure their relative importance in the decision-making process).

[0050] Furthermore, optional amendments also include the elimination of the "coal mill output sudden change suppression strategy" which has not been used for nearly 6 months.

[0051] Step 6: Model parameter update; Update the parameters of the Transformer model based on historical correction records: The continuous optimization system continuously optimizes and improves its monitoring and adjustment capabilities through a cycle of continuous monitoring, precise analysis, and set value compensation. The application effect is corrected based on actual working conditions. After several corrections, the method will correct it according to the application of the strategy and model, making the Transformer model more in line with actual working conditions and enhancing the performance of the algorithm when the system deteriorates.

[0052] Among them, the model fusion and ensemble learning formula is: ; In the formula, is the output result after the control strategy is updated, is the model weighting coefficient, Output result for each actual control under the current model, N A collection of working conditions.

[0053] Furthermore, the update process also includes retaining the main structure and some parameters of the encoding component, updating the parameters (weights) of the fully connected layer, and retaining the data of the last 30 days for training data (sliding window mechanism).

[0054] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions of the technical solution of the present invention by ordinary technicians in this field do not deviate from the essence and scope of the technical solution of the present invention.

Claims

1. A steam temperature autonomous cruising method based on time series classification, characterized by: The steps include: Collecting historical data of thermal power unit operating parameters, and preprocessing the data of historical thermal power unit operating parameters to obtain preprocessed historical data; A Transformer model consisting of encoding and decoding components is established, and the temporal correlation characteristics of the operating parameters of thermal power units are captured through the self-attention mechanism; Inputting the preprocessed historical data into the Transformer model for training; Collect the data of the current thermal power unit operating parameters for preprocessing to obtain preprocessed real-time data; Inputting the preprocessed real-time data into the trained Transformer model, and outputting the classification results of the steam temperature status under all working conditions; Identify the operating state of the thermal power unit according to the full-condition steam temperature state classification result and preset information; When the identification result of the operating state is an abnormal operating condition, a preset expert strategy library is called to match the corresponding control strategy, and the execution result of the control strategy is compared and judged in real time through a threshold comparison method; When the result of the comparison and judgment exceeds the preset range, a progressive steam temperature set point compensation strategy is additionally executed until the abnormal operating condition ends; The parameters of the Transformer model are updated according to the historical correction records, and the priority weights of the control strategies in the expert strategy library are dynamically corrected.

2. The method for autonomous cruise control of vehicle temperature based on time series classification according to claim 1 is characterized in that: The operating parameters of a thermal power unit include at least the coal mill output, primary air flow, burner swing angle, air door opening, flame center position and cooling water flow.

3. The method for autonomous cruise control of vehicle temperature based on time series classification according to claim 1 is characterized in that: The pre-processing comprises the following steps: Divide the rated value range of the operating parameters of thermal power units according to the operating conditions of the thermal power units; Based on the rated value range, outliers are eliminated from the data of the operating parameters of the thermal power unit, and the data are supplemented by interpolation to obtain preprocessed data of the operating parameters of the thermal power unit.

4. The method for autonomous cruise control of vehicle temperature based on time series classification according to claim 3 is characterized in that: The outlier elimination also includes the following steps: Taking the sampling step length as the calculation period, and calculating the maximum change value in the calculation period; When the maximum change value is greater than the allowable range, the maximum change value is removed and supplemented by a data smoothing algorithm.

5. The method for autonomous cruise control of vehicle temperature based on time series classification according to claim 1 is characterized in that: When the identification result of the operating state is an abnormal operating condition, the method further includes the following steps: Perform multi-level attribution analysis on the abnormal operating conditions, including equipment layer attribution factors, operation layer attribution factors and environment layer attribution factors; The device-level attribution factors: detecting actuator deviation and sensor drift; The operating layer attribution factors: detection of coal distribution uniformity and air-coal ratio deviation; The environmental layer attribution factors: calculation of coal calorific value fluctuations and environmental temperature gradients.

6. The method for autonomous cruise control of vehicle temperature based on time series classification according to claim 1 is characterized in that: The calling strategies of the expert strategy library include: Establish a multi-level strategy tree based on historical abnormal operating conditions and corresponding correction control strategies; The similarity between the real-time data and the historical data is calculated, and a group of historical data with the highest similarity is selected to retrieve the multi-level strategy tree, and the control strategy is obtained.

7. The method for autonomous cruise control of vehicle temperature based on time series classification according to claim 6 is characterized by: The dynamic modification of the expert strategy library includes the following steps: Calculate the strategy execution success rate and dynamically modify the weight of the corresponding control strategy based on the strategy execution success rate.

8. The method for autonomous cruise control of vehicle temperature based on time series classification according to claim 7 is characterized in that: The dynamic modification of the expert strategy library also includes the following steps: Based on the last-in-first-out system, the control strategies that have not been used or have the lowest utilization rate in the preset time period are eliminated.

9. The method for autonomous cruise control of vehicle temperature based on time series classification according to claim 1 is characterized by: In the Transformer model: The number of encoding component layers of the Transformer model is no less than 4; The self-attention mechanism adopts a multi-head attention structure.

10. The method for autonomous cruise control of vehicle temperature based on time series classification according to claim 9 is characterized in that: The Transformer model update includes the following steps: Keep the encoding component structure and update the fully connected layer parameters; A sliding window mechanism is used to retain valid training data in adjacent time periods.

Images