Double-pumping control strategy design method and device for ultra-supercritical unit

Abstract

The invention provides a double-pumping control strategy design method and device for an ultra-supercritical unit, belongs to the field of ultra-supercritical units, and is used for solving the problem that a control strategy of the ultra-supercritical unit is difficult to quickly and accurately determine in the related technology. The current control strategy vector can be analyzed and predicted based on the target control effect vector and the historical control record, the influence of the multi-dimensional control parameters and the control effect is fully considered, intelligent analysis and prediction are achieved, the control strategy can be rapidly and accurately determined, the control precision is guaranteed, and the supply quality is improved.

Description

Technical Field

[0001] This application relates to the field of ultra-supercritical units, and particularly to a design method and device for the dual-extraction control strategy of ultra-supercritical units. Background Art

[0002] In the field of modern power industry, ultra-supercritical units, as a kind of efficient and energy-saving power generation equipment, have been widely used. With the continuous growth of power demand and the increasing requirements for energy utilization efficiency and power supply stability, the operation optimization of ultra-supercritical units has become a key issue.

[0003] In the dual-extraction system of ultra-supercritical units, the precise control of extraction parameters (such as pressure, temperature, and flow rate, etc.) has a crucial impact on the overall performance and operation reliability of the unit. On the one hand, precise dual-extraction control can ensure the stable supply of external heat supply or industrial steam, meeting the requirements of different users for steam quality and quantity; on the other hand, reasonable extraction control helps to maintain the thermal cycle balance inside the unit, improve power generation efficiency, reduce unit energy consumption, and reduce the fatigue damage and potential failure risks caused by steam parameter fluctuations to the equipment.

[0004] However, there are many limitations in the existing dual-extraction control methods for ultra-supercritical units. Traditional control strategies often rely on simple feedback regulation mechanisms, with certain hysteresis in control, making it difficult to quickly and accurately determine control strategies, having insufficient control accuracy, and being unable to meet the strict requirements of modern power systems for high-quality power and steam supply. Summary of the Invention

[0005] This application provides a design method and device for the dual-extraction control strategy of ultra-supercritical units, which can quickly and accurately determine control strategies, is conducive to ensuring control accuracy, and improving supply quality.

[0006] In a first aspect, this application provides a design method for the dual-extraction control strategy of ultra-supercritical units. The method includes: Obtain a target control effect vector and historical control records. The historical control records carry time stamps, and the historical control records include a historical control strategy vector and a historical control effect vector; Based on the control effect vector, identify the same-target control records in the historical control records. The vector similarity between the historical control effect vector of the same-target control records and the target control effect vector is higher than the effect similarity threshold; Analyze the temporal feature distinctiveness of the historical control strategy vectors of the same-target control records; Judge whether the temporal feature distinctiveness is higher than the temporal distinctiveness threshold; If so, predict the current control strategy vector based on the historical control strategy vectors of all the same-target control records; If not, use the average of the historical control policy vectors of the nearest preset number of the same target control records as the current control policy vector.

[0007] By adopting the above technical solution, it is possible to analyze and predict the current control policy vector based on the target control effect vector and historical control records, fully considering the influence of multi-dimensional control parameters and control effects, realizing intelligent analysis and prediction, which is beneficial to quickly and accurately determine the control policy, ensure the control accuracy, and improve the supply quality.

[0008] Further, the analysis of the temporal feature distinctiveness of the historical control policy vectors of the same target control records includes: Let the historical control policy vector of the i-th same target control record in chronological order of timestamps be , and define the reference control vector as the vector with the most recent timestamp in the associated control vector set ; For the associated control vectors before the reference control vector, calculate the associated vector similarity with the reference control vector respectively , and obtain the associated vector similarity set ; Analyze all the associated vector similarities in the associated vector similarity set to determine the temporal feature distinctiveness, and the temporal feature distinctiveness is associated with the distribution trend distinctiveness of all the associated vector similarities over timestamps.

[0009] Further, the analysis of all the associated vector similarities in the associated vector similarity set to determine the temporal feature distinctiveness includes: Set a time window of size w, slide the window forward from the reference control vector, with a step size of 1 each time. For each time window, calculate the average of the associated control vectors within the window to form an average vector sequence; Analyze the fitting line slope of the average vector sequence based on the linear regression method; Determine the temporal feature distinctiveness according to the fitting line slope, and the temporal feature distinctiveness is positively correlated with the absolute value of the fitting line slope.

[0010] Further, the analysis of all the associated vector similarities in the associated vector similarity set to determine the temporal feature distinctiveness further includes: For each associated vector similarity, calculate an expected deviation from the corresponding point on the fitting line; Determine the time trend influence coefficient based on the sum of all the expected deviations, and the time trend influence coefficient is negatively correlated with the sum of all the expected deviations; The temporal feature distinctiveness is positively correlated with the time trend influence coefficient.

[0011] Furthermore, the determining of the temporal feature distinctness by analyzing all the correlation vector similarities in the correlation vector similarity set further includes: Counting the number of correlation vector similarities with the same sign as the slope of the fitting line as the same-trend number; the temporal feature distinctness is positively correlated with the ratio of the same-trend number to the total number of correlation control vectors.

[0012] In a second aspect, the present application provides a design device for the double-extraction control strategy of an ultra-supercritical unit. The device includes: An acquisition module, configured to acquire a target control effect vector and a historical control record, the historical control record carrying a time stamp, and the historical control record including a historical control strategy vector and a historical control effect vector; An identification module, configured to identify a same-target control record in the historical control record based on the control effect vector, and the vector similarity between the historical control effect vector of the same-target control record and the target control effect vector is higher than an effect similarity threshold; An analysis module, configured to analyze the temporal feature distinctness of the historical control strategy vector of the same-target control record; A judgment module, configured to judge whether the temporal feature distinctness is higher than a temporal distinctness threshold; and A determination module, configured to predict a current control strategy vector based on the historical control strategy vectors of all the same-target control records when the temporal feature distinctness is higher than the temporal distinctness threshold, and to use the average value of the historical control strategy vectors of the nearest preset number of same-target control records as the current control strategy vector when the temporal feature distinctness is not higher than the temporal distinctness threshold.

[0013] Furthermore, the analysis module is further configured that the analysis of the temporal feature distinctness of the historical control strategy vector of the same-target control record includes: Let the historical control strategy vector of the i-th same-target control record in the order of time stamps be , and define the reference control vector as the vector with the latest time stamp in the correlation control vector set ; For the correlation control vectors before the reference control vector, calculate the correlation vector similarities with the reference control vector respectively , and obtain a correlation vector similarity set ; Analyze all the correlation vector similarities in the correlation vector similarity set to determine the temporal feature distinctness, and the temporal feature distinctness is related to the distribution trend distinctness of all the correlation vector similarities with respect to the time stamp.

[0014] Further, the analysis module is further configured such that the determination of the temporal feature distinctiveness by analyzing all the association vector similarities in the association vector similarity set includes: Set a time window of size w, and slide the window forward starting from the reference control vector with a step size of 1 each time. For each time window, calculate the average value of the association control vectors within the window to form an average vector sequence; Analyze the fitting line slope of the average vector sequence based on the linear regression method; Determine the temporal feature distinctiveness according to the fitting line slope, and the temporal feature distinctiveness is positively correlated with the absolute value of the fitting line slope.

[0015] Further, the analysis module is further configured such that the determination of the temporal feature distinctiveness by analyzing all the association vector similarities in the association vector similarity set further includes: For each association vector similarity, calculate an expected deviation from the corresponding point on the fitting line; Determine the time trend influence coefficient based on the sum of all the expected deviations, and the time trend influence coefficient is negatively correlated with the sum of all the expected deviations; The temporal feature distinctiveness is positively correlated with the time trend influence coefficient.

[0016] Further, the analysis module is further configured such that the determination of the temporal feature distinctiveness by analyzing all the association vector similarities in the association vector similarity set further includes: Count the number of association vector similarities with the same sign as the fitting line slope as the same-trend number; the temporal feature distinctiveness is positively correlated with the ratio of the same-trend number to the total number of association control vectors.

[0017] In summary, the present application at least includes the following beneficial effects: A design method and device for the double-extraction control strategy of an ultra-supercritical unit are provided, which can intelligently analyze and determine the control strategy based on an algorithm model to achieve fast response and precise control, so as to ensure the supply quality.

[0018] It should be understood that the content described in the summary of the invention section is not intended to limit the key or important features of the embodiments of the present application, nor is it used to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In combination with the accompanying drawings and with reference to the following detailed description, the above and other features, advantages, and aspects of the embodiments of the present application will become more apparent. In the drawings, the same or similar reference numerals denote the same or similar elements, where: Figure 1 The flowchart of a design method for the double-extraction control strategy of an ultra-supercritical unit in an embodiment of the present application is shown.

[0020] Figure 2 The block diagram of a design device for the double-extraction control strategy of an ultra-supercritical unit in an embodiment of the present application is shown. Detailed implementation manners

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0022] In addition, the term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the front and rear associated objects.

[0023] The present application provides a design method and device for the double-extraction control strategy of an ultra-supercritical unit, which can realize the intelligent, fast, and accurate determination of the double-extraction control strategy of the ultra-supercritical unit.

[0024] In a first aspect, an embodiment of the present application discloses a design method for the double-extraction control strategy of an ultra-supercritical unit. This method can be specifically executed by a controller in the ultra-supercritical unit.

[0025] Figure 1 The flowchart of a design method for the double-extraction control strategy of an ultra-supercritical unit in an embodiment of the present application is shown.

[0026] Referring to Figure 1 , this method specifically includes the following steps: S110: Obtain a target control effect vector and historical control records.

[0027] In the method of this step, the historical control records carry time stamps, and the historical control records include a historical control strategy vector and a historical control effect vector.

[0028] The control strategy of an ultra-supercritical unit essentially determines the parameter values of multiple control parameters to achieve a series of control results. In the method of this step, the target control effect vector reflects a series of control results that the ultra-supercritical unit hopes to achieve. The target control effect is generally pre-obtained, which can be directly input by humans with a series of precise effect limitations, or a series of control results determined based on the work of other parts of the controller, or determined after intelligent analysis of human fuzzy input, etc. The acquisition of the target control effect vector does not belong to the improvement of this method, so it will not be elaborated.

[0029] The historical control record is the record of the historical control process of the ultra-supercritical unit, which carries a time stamp, the historical control effect vector that was hoped to be achieved at that time, and the historical control strategy vector that was selected at that time.

[0030] The key content of the historical control strategy vector (and the subsequent current control strategy vector) is the selection of control parameters and vector construction. For specific reference, please refer to the following content: The vector similarity between the historical control effect vector of the same target control record and the target control effect vector is higher than the effect similarity threshold; the dual-extraction control strategy of the ultra-supercritical unit involves multiple key parameters, such as the steam inlet volume of the steam turbine , the opening of the medium-pressure extraction control valve , the opening of the low-pressure extraction control valve , the boiler combustion rate , the feed water flow rate etc. Suppose there are n parameters in total. At time the values of these parameters are respectively , then the control parameter vector , where represents the value of the nth control parameter at time . For some parameters, normalization processing may be required to make their value ranges within a reasonable interval. For example, for the steam inlet volume , the linear normalization formula can be used: , where are the minimum and maximum values of the steam inlet volume respectively. This can avoid the influence of large differences in dimension and value range of different parameters on subsequent calculations and improve the stability and accuracy of the model.

[0031] The key content of the target control effect vector and the historical control effect vector is the selection of control effect indicators and vector construction. For specific reference, please refer to the following content: The control effect can be measured from multiple aspects, such as the power generation stability index (for example, it can be represented by the sum of the squares of the deviations between the actual power generation and the planned power generation: , where is the actual power generation in the th time interval, is the planned power generation, and N is the number of statistical time intervals), the extraction steam pressure stability index (which can be measured by the fluctuation variances of the medium-pressure extraction steam pressure and the low-pressure extraction steam pressure: , where represents the variance operation, are the medium-pressure and low-pressure extraction steam pressures respectively), the unit thermal efficiency index (which is reflected by the ratio of the actual heat consumption rate to the designed heat consumption rate: , where is the actual heat consumption rate, is the designed heat consumption rate), etc. Suppose there are m control effect indicators in total, then at time the control effect vector . Similarly, for some control effect indicators, appropriate transformation or normalization processing can also be carried out to make them more in line with the calculation requirements and practical significance of the model. For example, for the thermal efficiency index, if its value range is small and high precision is required, logarithmic transformation can be adopted: , which can magnify the differences between the indicators and facilitate subsequent similarity calculation and analysis.

[0032] Continuously collect the double-extraction control strategy parameters and the corresponding control effect data of the ultra-supercritical unit at different operation time periods to form the control parameter vector set and the control effect vector set , and each vector is accompanied by the corresponding accurate time stamp , and the precision of the time stamp should be determined according to the operation characteristics and control requirements of the unit. For example, it can be accurate to the minute or second level to better capture the change trend and timeliness of the control strategy.

[0033] Regarding the specific determination of the target control effect vector, please refer to the following content for details: According to the current operation tasks and requirements of the unit, determine the target control effect vector , for example, during a certain heating period, it is required that the power generation be stable within a certain range (set ), the extraction steam pressure fluctuation be as small as possible (set ), and at the same time ensure that the unit thermal efficiency is not lower than a certain threshold (set ). These target values can be manually set by the operation personnel according to experience and actual situation, or can be automatically obtained and updated through interaction with external systems such as the power grid dispatching system and the heating demand system. When constructing the target control effect vector, the same preprocessing and normalization operations as the historical data need to be carried out on each indicator to ensure comparability and consistency with the historical control effect vector.

[0034] S120: Identify the same target control records in the historical control records based on the control effect vector.

[0035] In the method of this step, the vector similarity between the historical control effect vector of the same target control record and the target control effect vector is higher than the effect similarity threshold.

[0036] The method of this step first needs to select a suitable vector similarity calculation method, such as the commonly used Euclidean distance formula: , or the cosine similarity formula: , where are two control effect vectors, are respectively the j-th element of the vector respectively represent the norms of the vectors. The smaller the Euclidean distance or the closer the cosine similarity is to 1, the more similar the two vectors are.

[0037] According to the operating characteristics of the unit and the distribution of historical data, set the effect similarity threshold . It is possible to determine a suitable threshold by statistical analysis of historical data, such as calculating the reusability of control strategies and the stability of control effects at different similarity levels. Initially, a relatively loose threshold can be set, and then dynamically adjusted and optimized according to the actual effects during the operation of the model. For example, the threshold can be initially set to a certain percentile (such as percentile) of the similarity distribution in the historical data, and then observe the quantity and quality of the historical control records matched under this threshold. If the quantity is too small or the quality is not high, the threshold can be appropriately reduced; if the quantity is too large and contains many invalid or low-quality records, the threshold can be appropriately increased.

[0038] Secondly, the method of this step needs to perform screening and determination of associated control vectors. Specifically, calculate the similarity between the target control effect vector and each vector in the set of historical control effect vectors , find all historical control records that satisfy , and the corresponding control parameter vector is the associated control vector, forming a set of associated control vectors , where p is the number of associated control vectors.

[0039] S130: Analyze the temporal feature distinctness of the historical control strategy vectors of the same target control records.

[0040] The method of this step specifically includes: Let the historical control strategy vector of the i-th same target control record in chronological order of timestamps be , and define the reference control vector as the vector with the most recent timestamp in the set of associated control vectors ; for the associated control vectors before the reference control vector, calculate the associated vector similarity with the reference control vector respectively , an associated vector similarity set is obtained ; analyze all the associated vector similarities in the associated vector similarity set to determine the temporal feature distinctiveness, where the temporal feature distinctiveness is associated with the distribution trend distinctiveness of all the associated vector similarities over time stamps.

[0041] In one example, the analyzing all the associated vector similarities in the associated vector similarity set to determine the temporal feature distinctiveness includes: setting a time window of size w, starting from the reference control vector and sliding the window forward, with a step size of 1 each time. For each time window, calculate the average value of the associated control vectors within the window to form an average vector sequence; analyze the fitting line slope of the average vector sequence based on the linear regression method; determine the temporal feature distinctiveness according to the fitting line slope, where the temporal feature distinctiveness is positively correlated with the absolute value of the fitting line slope.

[0042] Regarding the specific calculation process of the fitting line slope, for example: the first time window contains the w associated control vectors before the reference control vector, that is , and its average vector is calculated as follows: ; The second time window contains the w associated control vectors before the vector one before the reference control vector, that is , and its average vector is: ; And so on, until sliding to the starting position of the historical data or reaching the preset upper limit of the sliding times, to obtain the average vector sequence , where q is the number of sliding windows.

[0043] After determining the average vector sequence, perform slope calculation and trend analysis. For the average vector sequence, also use the linear regression method to calculate the slope. Assume that the time stamps corresponding to the average vector sequence are numerically processed as (for example, ), calculate the similarity sequence between the average vectors, where , and (the similarity of a vector to itself is 1).

[0044] Use the linear regression formula to calculate the slope : ; Among them, .

[0045] In another example, it is also possible to determine the slope of the fitted line based on the similarity of correlation vectors within a single time window. For example, a time window size w is determined, and w correlation vector similarities are selected forward from the reference control vector to form a subset. The selection of the time window size w needs to comprehensively consider factors such as the operating inertia of the unit, the adjustment frequency of the control strategy, and the richness of historical data. If the operating state of the unit changes relatively slowly and the control strategy is not adjusted frequently, a larger time window can be selected to better capture the long-term time series trend; if the operating conditions of the unit are relatively complex and changeable and the control strategy needs to respond quickly, a smaller time window should be selected to highlight the timeliness and importance of recent data. Generally, a comparative experiment can be conducted on the model prediction effects under different time window sizes, and the time window size that minimizes the prediction error is selected as the initial value and fine-tuned according to the actual situation during subsequent operation. The process of calculating the slope of the fitted line based on the subset can refer to the aforementioned process of calculating the slope and will not be described repeatedly.

[0046] In another example, determining the obviousness of the time series characteristics by analyzing all the correlation vector similarities in the correlation vector similarity set may further include: for each correlation vector similarity, calculating an expected deviation from the corresponding point on the fitted line; determining a time trend influence coefficient based on the sum of all the expected deviations, where the time trend influence coefficient is negatively correlated with the sum of all the expected deviations; and the obviousness of the time series characteristics is positively correlated with the time trend influence coefficient.

[0047] Specifically, for each average vector , calculate its deviation from the corresponding point on the fitted line. Assume that the time stamp corresponding to the average vector is , then the similarity prediction value of the corresponding point on the fitted line is . Calculate the actual similarity value , and the deviation between and the predicted value. Regarding the calculation of the time trend influence coefficient, it specifically includes: calculating the sum of all the deviations and using this to define the time trend influence degree . To make the time trend influence degree negatively correlated with the sum of the deviations, the following formula can be used: ; where is an adjustment coefficient used to adjust the influence degree of the sum of the deviations on the time trend influence degree. When the sum of the deviations is larger, is smaller, indicating that the reliability of the time trend is lower; conversely, is larger, and the time trend is more reliable.

[0048] In addition, in other examples, determining the temporal feature distinctiveness by analyzing all the correlation vector similarities in the set of correlation vector similarities may further include: counting the number of correlation vector similarities with the same sign as the slope of the fitting line as the co-trend number; the temporal feature distinctiveness is positively correlated with the ratio of the co-trend number to the total number of correlation control vectors.

[0049] Based on the foregoing, regarding the calculation of the temporal feature distinctiveness T, specifically, the following formula can be referred to: ; In the formula, the slope of the fitting line is , the co-trend number is , the total number of correlation control vectors is , the time trend influence coefficient is , are respectively preset weight coefficients.

[0050] By introducing the time trend influence , the deviation between the average vector and the fitting line is comprehensively considered, making the evaluation of the temporal feature distinctiveness more accurate and reasonable. When the deviation between the correlation vector similarity and the fitting line is small, the time trend influence is large, and the contribution to the temporal feature distinctiveness is also large; conversely, when the deviation is large, the time trend influence decreases, thereby reducing the influence of unreliable time trends on the distinctiveness. In practical applications, it is necessary to determine the appropriate values of the adjustment coefficient , weight coefficient through experiments and optimization according to the historical data characteristics of the unit to achieve the best control strategy decision-making effect.

[0051] S140: Determine whether the temporal feature distinctiveness is higher than the temporal distinctiveness threshold.

[0052] The temporal distinctiveness threshold is preset in the controller. Since both the temporal feature distinctiveness and the temporal distinctiveness threshold are quantitative data, the comparison result between the two can be directly obtained by the controller.

[0053] S150: If the temporal feature distinctiveness is higher than the temporal distinctiveness threshold, then predict the current control strategy vector based on the historical control strategy vectors of all the same-target control records.

[0054] In the method of this step, if the temporal feature distinctiveness is higher than the set temporal distinctiveness threshold , it is considered that the historical correlation control vector has an obvious time series trend, and the current control strategy can be predicted based on all correlation control vectors. Multiple prediction methods can be used, such as simple linear regression prediction, polynomial regression prediction, or more complex machine learning methods (such as neural networks, support vector machines, etc.).

[0055] Taking linear regression prediction as an example, assume that the control parameter vector and the timestamp have a linear relationship: , where is the coefficient vector to be determined. By performing linear regression fitting on all correlation control vectors and their corresponding timestamps, the estimated value of the coefficient vector can be obtained. Then, according to the current time , predict the current control strategy vector . When using machine learning methods, the correlation control vectors need to be used as input features, and the corresponding timestamps can be used as additional features or not used (relying on the model to automatically learn the time series relationship in the data), and the current control strategy vector is predicted by training the model. Techniques such as cross-validation can be used during the training process to avoid overfitting and improve the generalization ability of the model. At the same time, according to the prediction error and performance metrics of the model, the structure and parameters of the model are adjusted and optimized. For example, parameters such as the number of layers, the number of nodes, and the learning rate of the neural network are adjusted, or different kernel functions and penalty parameters are selected to optimize the support vector machine model.

[0056] S160: If the time series feature significance is not higher than the time series significance threshold, then use the average value of the historical control strategy vectors of the same target control records in the recent preset number as the current control strategy vector.

[0057] In the method of this step, if the time series feature significance is not higher than the time series significance threshold , then use the average value of the recent correlation control vectors as the current control strategy vector . The specific calculation formula is: ; where, The value of needs to be determined according to the stability of the historical data and the operation requirements of the unit.

[0058] If the historical data fluctuates little and the control strategy is relatively stable, a larger value can be selected to make full use of more historical information and smooth the change of the control strategy; if the historical data changes greatly, or the unit has a high requirement for the timeliness of the current control strategy, a smaller value should be selected, focusing on the recent correlation control vectors. It can be achieved by different Test and evaluate the control effect under the value, and select the value with the best control effect as the initial setting, and adjust it according to the situation during actual operation.

[0059] Based on the above, through the above detailed model design and optimization methods, the historical operation data of ultra-supercritical units can be better utilized to adaptively determine the double-extraction control strategy, improve the operation efficiency, stability and control accuracy of the units, and at the same time provide strong support for the intelligent operation and optimization management of the units.

[0060] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present application is not limited by the described action sequence, because according to the embodiments of the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present application.

[0061] In a second aspect, an embodiment of the present application discloses a device for designing a double-extraction control strategy for an ultra-supercritical unit. The device can be implemented as a controller of an ultra-supercritical unit or be included in the controller of an ultra-supercritical unit.

[0062] Figure 2 The block diagram of a device for designing a double-extraction control strategy for an ultra-supercritical unit in an embodiment of the present application is shown.

[0063] Referring to Figure 2 , the device includes: An acquisition module 110, configured to acquire a target control effect vector and a historical control record, where the historical control record carries a time stamp, and the historical control record includes a historical control strategy vector and a historical control effect vector; An identification module 120, configured to identify a same-target control record in the historical control record based on the control effect vector, where the vector similarity between the historical control effect vector of the same-target control record and the target control effect vector is higher than an effect similarity threshold; An analysis module 130, configured to analyze the temporal feature distinctness of the historical control strategy vector of the same-target control record; A judgment module 140, configured to judge whether the temporal feature distinctness is higher than a temporal distinctness threshold; and A determination module 150, configured to predict a current control policy vector based on historical control policy vectors of all same-target control records when the temporal feature distinctiveness is higher than a temporal distinctiveness threshold, and to use the average value of historical control policy vectors of the nearest preset number of same-target control records as the current control policy vector when the temporal feature distinctiveness is not higher than the temporal distinctiveness threshold.

[0064] Further, the analysis module 130 is further configured such that the analysis of the temporal feature distinctiveness of the historical control policy vectors of the same-target control records includes: Let the historical control policy vector of the i-th same-target control record in chronological order of timestamps be , and define the reference control vector as the vector with the most recent timestamp in the associated control vector set ; For the associated control vectors before the reference control vector, calculate the associated vector similarity with the reference control vector respectively , to obtain an associated vector similarity set ; Analyze all the associated vector similarities in the associated vector similarity set to determine the temporal feature distinctiveness, and the temporal feature distinctiveness is associated with the distribution distinctiveness of all the associated vector similarities over the timestamps.

[0065] Further, the analysis module 130 is further configured such that the analysis of all the associated vector similarities in the associated vector similarity set to determine the temporal feature distinctiveness includes: Set a time window of size w, slide the window forward starting from the reference control vector, with a step size of 1 each time. For each time window, calculate the average value of the associated control vectors within the window to form an average vector sequence; Analyze the fitting line slope of the average vector sequence based on the linear regression method; Determine the temporal feature distinctiveness according to the fitting line slope, and the temporal feature distinctiveness is positively correlated with the absolute value of the fitting line slope.

[0066] Further, the analysis module 130 is further configured such that the analysis of all the associated vector similarities in the associated vector similarity set to determine the temporal feature distinctiveness further includes: For each associated vector similarity, calculate an expected deviation from the corresponding point on the fitting line; Determine a time trend influence coefficient based on the sum of all the expected deviations, and the time trend influence coefficient is negatively correlated with the sum of all the expected deviations; The temporal feature distinctiveness is positively correlated with the time trend influence coefficient.

[0067] Further, the analysis module 130 is further configured that determining the temporal feature distinctiveness by analyzing all the correlation vector similarities in the correlation vector similarity set further includes: Counting the number of correlation vector similarities with the same sign as the slope of the fitting line as the co-trend number; the temporal feature distinctiveness is positively correlated with the ratio of the co-trend number to the total number of correlation control vectors.

[0068] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the described device can refer to the corresponding process in the foregoing method embodiment, which will not be elaborated here.

[0069] In summary, the present application at least includes the following beneficial effects: A dual-extraction control strategy design method and device for an ultra-supercritical unit are provided, which can intelligently analyze and determine the control strategy based on an algorithm model, achieve fast response and precise control, so as to ensure the supply quality.

[0070] The above description is only a preferred embodiment of the present application and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the foregoing disclosure concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the present application.

Claims

1. A method for designing a double extraction control strategy for an ultra-supercritical unit, characterized in that: include: Obtaining a target control effect vector and a historical control record, wherein the historical control record carries a timestamp and includes a historical control strategy vector and a historical control effect vector; identifying a same-target control record in historical control records based on the control effect vector, wherein a vector similarity between the historical control effect vector of the same-target control record and the target control effect vector is higher than an effect similarity threshold; Analyze the temporal characteristics of the historical control strategy vectors with the same target control records; Determine whether the temporal feature prominence is higher than the temporal prominence threshold; If so, the current control strategy vector is predicted based on the historical control strategy vectors of all control records with the same target; If not, the average value of the historical control strategy vectors of the most recent preset number of control records with the same target is used as the current control strategy vector.

2. The method according to claim 1, characterized in that The analysis of the time series feature significance of the historical control strategy vector recorded with the target control includes: Suppose the historical control strategy vector of the i-th control record with the same target in the order of timestamp is , define the reference control vector as the vector with the closest timestamp in the set of associated control vectors ; For the associated control vector before the reference control vector, calculate the similarity of the associated vector with the reference control vector , get the association vector similarity set ; The temporal feature prominence is determined by analyzing the similarities of all associated vectors in the associated vector similarity set, where the temporal feature prominence is associated with the distribution trend prominence of all associated vector similarities with time stamps.

3. The method according to claim 2, characterized in that The analyzing the similarities of all associated vectors in the associated vector similarity set to determine the temporal feature prominence comprises: Set a time window of size w, slide the window forward from the reference control vector, and the step length of each slide is 1. For each time window, calculate the average value of the associated control vector in the window to form an average vector sequence; Analyze the slope of the fitted line of the average vector sequence based on the linear regression method; The temporal feature significance is determined according to the slope of the fitting straight line, and the temporal feature significance is positively correlated to the absolute value of the slope of the fitting straight line.

4. The method according to claim 3, characterized in that The analyzing of all the association vector similarities in the association vector similarity set to determine the temporal feature significance is also For each correlation vector similarity, a deviation from the corresponding point on the fitting line is calculated as the expected deviation; Determine a time trend impact coefficient based on the sum of all expected deviations, wherein the time trend impact coefficient is negatively correlated to the sum of all expected deviations; The temporal feature significance is positively correlated with the temporal trend influence coefficient.

5. The method according to claim 3 or 4, characterized in that: The analyzing of all the association vector similarities in the association vector similarity set to determine the temporal feature significance is also The number of similarities of associated vectors with the same slope sign as the fitted straight line is counted as the number of same trends; the time series feature significance is positively correlated to the ratio of the number of same trends to the total number of associated control vectors.

6. A device for designing double extraction control strategy for ultra-supercritical units, characterized in that: include: An acquisition module (110) is used to acquire a target control effect vector and a historical control record, wherein the historical control record carries a timestamp and includes a historical control strategy vector and a historical control effect vector; An identification module (120) is used to identify a target control record in the historical control record based on the control effect vector, wherein the vector similarity between the historical control effect vector of the target control record and the target control effect vector is higher than an effect similarity threshold; An analysis module (130) is used to analyze the time series feature significance of the historical control strategy vectors of the same target control record; A judgment module (140), used to judge whether the temporal feature significance is higher than a temporal significance significance threshold; as well as A determination module (150) is used to predict the current control strategy vector based on the historical control strategy vectors of all control records with the same target when the time series feature significance is higher than the time series significance threshold, and to use the average value of the historical control strategy vectors of the most recent preset number of control records with the same target as the current control strategy vector when the time series feature significance is not higher than the time series significance threshold.

7. The device according to claim 6, characterized in that The analysis module (130) is further configured to analyze the temporal feature significance of the historical control strategy vector of the target control record, including: Suppose the historical control strategy vector of the i-th control record with the same target in the order of timestamp is , define the reference control vector as the vector with the closest timestamp in the set of associated control vectors ; For the associated control vector before the reference control vector, calculate the similarity of the associated vector with the reference control vector , get the association vector similarity set ; The temporal feature significance is determined by analyzing the similarities of all associated vectors in the associated vector similarity set, wherein the temporal feature significance is associated with the significance of the distribution trend of all associated vector similarities with the time stamp.

8. The device according to claim 7, characterized in that The analysis module (130) is further configured to analyze the similarities of all associated vectors in the associated vector similarity set to determine the temporal feature significance Set a time window of size w, slide the window forward from the reference control vector, and the step length of each slide is 1. For each time window, calculate the average value of the associated control vector in the window to form an average vector sequence; Analyze the slope of the fitted line of the average vector sequence based on the linear regression method; The temporal feature significance is determined according to the slope of the fitting straight line, and the temporal feature significance is positively correlated to the absolute value of the slope of the fitting straight line.

9. The device according to claim 8, characterized in that The analysis module (130) is further configured to analyze the similarities of all associated vectors in the associated vector similarity set to determine the temporal feature significance. For each correlation vector similarity, a deviation from the corresponding point on the fitting line is calculated as the expected deviation; Determine a time trend impact coefficient based on the sum of all expected deviations, wherein the time trend impact coefficient is negatively correlated to the sum of all expected deviations; The temporal feature significance is positively correlated with the temporal trend influence coefficient.

10. The device according to claim 8 or 9, characterized in that The analysis module (130) is further configured to analyze the similarities of all associated vectors in the associated vector similarity set to determine the temporal feature significance. The number of similarities of correlation vectors with the same slope sign as the fitted straight line is counted as the number of the same trend; The time series feature significance is positively correlated to the ratio of the number of the same trend to the total number of associated control vectors.

Images