A method and system for analyzing and optimizing indicators of a wet flue gas desulfurization system

Abstract

The application provides a wet desulfurization system index analysis optimization method and system, the method comprises the following steps: obtaining the operation indexes of a wet desulfurization slurry supply system, the information contribution degrees of each operation index to downstream analysis tasks and the quality scores at the current moment; calculating the effective information evaluation value based on the product of the information contribution degree and the quality score, and dynamically screening and adjusting the weight of the input index set according to the effective information evaluation value; and using the processed index set to execute the downstream analysis task and output the result. The application quantifies the index information value through conditional mutual information and SHAP attribution, establishes an index coupling network with the transmission function parameters as the edges to realize real-time evaluation of sensor quality, realizes adaptive selection and redundancy fault tolerance of the indexes through the effective information evaluation value, and significantly improves the robustness and precision of wet desulfurization prediction and control.

Description

Technical Field

[0001] This invention relates to the field of flue gas desulfurization technology, specifically to a method and system for analyzing and optimizing the indicators of a wet desulfurization system. Background Technology

[0002] Wet limestone-gypsum desulfurization is the most mainstream flue gas desulfurization technology in thermal power plants. The slurry supply system supplies limestone slurry to the absorber, maintaining the circulating slurry pH within a suitable range to ensure SO2 absorption efficiency. The operation of the slurry supply system relies on the real-time acquisition and accurate analysis of a series of operational indicators, including inlet flue gas flow rate, inlet SO2 concentration, circulating slurry pH, outlet SO2 concentration, slurry supply pump frequency, and circulating pump operating status. Monitoring and analyzing these operational indicators is the foundation of all control decisions in the slurry supply system; the quality of indicator analysis directly determines the quality of subsequent prediction and control.

[0003] In recent years, several technical solutions have been proposed for intelligent control of wet desulfurization. For example, one existing intelligent slurry supply control method for wet desulfurization collects operating indicators such as unit load, inlet flue gas flow rate, inlet SO2 concentration, circulating slurry pH, slurry flow rate, slurry density, absorber level, and outlet SO2 concentration. These input characteristics drive a long short-term memory network to establish a circulating slurry pH prediction model and an SO2 outlet concentration prediction model. The model prediction results are then used as feedforward and feedback signals to the controller, achieving closed-loop regulation of the slurry pump frequency and the circulating pump start-stop combination. Another example is an existing integrated intelligent desulfurization method that collects parameters from the entire desulfurization process through a DCS system and intelligent instruments. It extracts derived variables such as inlet SO2 mass flow rate and circulating slurry flow rate, establishes pH and outlet SO2 concentration prediction models respectively, and uses a two-level model predictive controller structure to achieve control.

[0004] However, the aforementioned existing technical solutions all suffer from the following structural defects in indicator analysis: First, the indicator input method is fixed, lacking a quantitative understanding of the information value of the indicators themselves. Existing technologies treat all collected operational indicators as fixed input features of the prediction model, assuming that all indicators are equally important and equally reliable. This leads to low-information-value indicators being included in the model input without distinction, which not only increases computational overhead but also reduces the model's prediction accuracy due to the measurement noise they carry. Second, the indicator quality lacks real-time perception. Sensor degradation leads to silent deterioration of control quality. Existing technologies assume that the collected indicator values ​​are always reliable and lack a mechanism for evaluating and providing feedback on the indicator measurement quality. When sensors experience performance degradation due to high temperature, high humidity, or high corrosion environments, the prediction model will produce systematic biases, leading to incorrect control commands. Third, the coupling relationship between indicators lacks quantitative modeling, and anomaly tracing can only make qualitative inferences. Existing technologies only remain at the level of qualitative description and cannot quantitatively decompose the contribution ratio of each anomaly source. Fourth, information redundancy is regarded as a burden rather than a resource. Existing technologies regard information overlap as a harmful multicollinearity problem and eliminate it, actively abandoning the information backup resources on which the system resilience depends when sensors fail. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies in wet desulfurization system operation index analysis, such as the inability to quantify the value of index information, the lack of real-time perception of index quality, the lack of quantitative modeling of the coupling relationship between indicators, and the ineffective utilization of information redundancy. This invention provides a wet desulfurization system index analysis and optimization method and system that can quantify the value of index information, evaluate index quality in real time, quantitatively model the coupling relationship between indicators, and realize adaptive selection of indicators and dynamic adjustment of weights.

[0006] According to one objective of the present invention, the present invention provides a method for analyzing and optimizing the performance indicators of a wet desulfurization system, comprising the following steps: Step S1: Obtain the operating indicators of the wet desulfurization slurry supply system. The operating indicators include at least the inlet flue gas flow rate, inlet SO2 concentration, sulfur content of coal, unit load, circulating slurry pH value, outlet SO2 concentration, slurry supply pump operating parameters, and slurry circulation pump operating parameters; obtain the information contribution of each operating indicator to each downstream analysis task; obtain the quality score of each operating indicator at the current time. Step S2: Based on the information contribution and the quality score, calculate the effective information evaluation value of each operating indicator for each downstream analysis task, and dynamically filter and / or adjust the weight of the input indicator set for each downstream analysis task according to the effective information evaluation value; Step S3: Use the index set processed in step S2 to perform the downstream analysis task and output the analysis results.

[0007] Further, in step S1, the information contribution is obtained in the following way: based on the historical operating data of the operating indicators, the independent information increment of each operating indicator to the target downstream analysis task is quantified by the conditional mutual information method and / or SHAP attribution analysis method, given that other indicators are known, and the information contribution of each operating indicator is obtained after normalization processing.

[0008] Further, in step S1, causally related indicator pairs are selected from the operational indicators. The response process of the corresponding downstream indicators when the upstream indicators change significantly in historical operational data is used to estimate the transfer function parameters of each indicator pair through a system identification method. The transfer function parameters include at least gain, time constant, and dead time. An indicator coupling network is established with indicators as nodes and transfer function parameters as edges. Based on the indicator coupling network, the predicted value of each operational indicator at the current moment is calculated from the upstream operational indicators that are causally related to each operational indicator. The quality score is obtained based on the deviation between the current measured value and the predicted value.

[0009] Furthermore, the quality score is quantified as follows: the absolute value of the residual between the current measured value and the predicted value is calculated and divided by the historical statistical standard deviation of the indicator to obtain the normalized residual; the normalized residual is substituted into a monotonically decreasing mapping function to obtain a quality score with a value range of [0,1]. The larger the normalized residual, the lower the quality score; for root node indicators without upstream indicators, the quality score is determined based on the signal range exceeding the limit, the jump amplitude, and the signal freezing rules.

[0010] Further, in step S2, the effective information evaluation value is the product of the information contribution degree and the quality score.

[0011] Further, in step S2, the dynamic screening and / or weight adjustment includes: setting an effective information threshold, removing indicators whose effective information evaluation value is lower than the effective information threshold from the input indicator set of the corresponding downstream analysis task; when the quality score of an indicator is lower than a preset quality threshold, reducing the input weight of the indicator in the corresponding downstream analysis task, and automatically identifying redundant alternative indicators with information contribution close to the indicator, and proportionally increasing the weight of the redundant alternative indicators.

[0012] Furthermore, in step S3, the downstream analysis task includes at least one or more of the following: operating condition identification, prediction of circulating slurry pH change trend, prediction of outlet SO2 concentration change trend, calculation of slurry supply control quantity, and calculation of circulating pump control quantity.

[0013] Furthermore, the information contribution is recalculated using newly added operational data at a preset cycle, and the transfer function parameters of the indicator coupling network are re-identified; when the change in the information contribution of a certain indicator exceeds a preset threshold or the change in the transfer function parameters of a certain indicator pair exceeds a preset threshold, an early warning is triggered.

[0014] A wet desulfurization system index analysis and optimization system based on the above-mentioned wet desulfurization system index analysis and optimization method includes: The acquisition module is used to acquire the operating indicators of the wet desulfurization slurry supply system, the information contribution of each operating indicator to each downstream analysis task, and the quality score of each operating indicator at the current moment. The calculation module, which is communicatively connected to the acquisition module, is used to calculate the effective information evaluation value of each operating indicator for each downstream analysis task based on the information contribution degree and the quality score, and to dynamically filter and / or adjust the weight of the input indicator set of each downstream analysis task according to the effective information evaluation value. The execution module, which is communicatively connected to the calculation module, is used to execute the downstream analysis task with the indicator set processed by the calculation module and output the analysis results.

[0015] Furthermore, the acquisition module has a built-in information contribution calculation unit and a quality score calculation unit; the quality score calculation unit has an embedded index coupling network model, the index coupling network has running indicators as nodes and transfer function parameters as edges, the transfer function parameters include at least gain, time constant and dead time; the acquisition module, calculation module and execution module exchange data through the OPC UA protocol.

[0016] The technical solution of this invention quantifies the independent information contribution of each indicator to downstream tasks through conditional mutual information and SHAP attribution analysis, eliminating low-value indicators. A coupled network with the FOPDT transfer function as the edge is established, and downstream indicators are inferred based on the transfer function output. The quality score is obtained by mapping the quality score to the residual of the measured value, enabling real-time detection of sensor degradation, drift, and jamming. An effective information evaluation value is calculated based on the product of contribution and quality score. Indicators are dynamically screened and their weights adjusted at each time step. When the quality of a high-contribution indicator declines, redundant replacement is automatically activated, transforming measurement redundancy into fault-tolerant resources. The contribution matrix and network parameters are updated periodically, and warnings are issued when thresholds are exceeded to ensure the long-term effectiveness of the model. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the method for analyzing and optimizing the performance indicators of a wet desulfurization system according to an embodiment of the present invention. Figure 2 This is a structural block diagram of the wet desulfurization system index analysis and optimization system according to an embodiment of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0021] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0022] Example 1 This embodiment takes a wet desulfurization system of a 2×600 MW thermal power unit as an example. Each unit is equipped with one absorption tower, and each absorption tower is equipped with five slurry circulation pumps and one slurry supply pump. Those skilled in the art will understand that the technical solution of this invention is also applicable to wet desulfurization systems of other capacity levels and equipment configurations.

[0023] like Figure 1 As shown in the figure, this embodiment provides a method for analyzing and optimizing the indicators of a wet desulfurization system, which proceeds in the order of acquisition, calculation, and output.

[0024] Step S1: Acquisition. This step completes three acquisition tasks: acquiring operational metrics, acquiring the information contribution of each operational metric to each downstream analysis task, and acquiring the quality score of each operational metric at the current time.

[0025] (a) Collection of operational indicators.

[0026] The operating parameters of the wet desulfurization slurry supply system are obtained in real time from the following data sources: (1) The inlet flue gas flow rate is obtained in real time from the power plant's distributed control system (DCS), with a collection cycle of 1 second; (2) The SO2 concentration at the inlet is obtained in real time from the Continuous Emission Monitoring System (CEMS) with a collection cycle of 1 second; (3) The sulfur content of coal is obtained from the laboratory's shift report or online elemental analyzer, with the collection cycle being shift or real-time; (4) The unit load is obtained from the DCS in real time, with a collection cycle of 1 second; (5) The pH value of the circulating slurry is measured in real time by the pH meter of the slurry in the absorption tower, with a collection cycle of 10 seconds; (6) The SO2 concentration at the outlet is measured in real time by CEMS with a data acquisition cycle of 1 second; (7) The operating parameters of the slurry pump are obtained from the DCS by the operating frequency fed back by the slurry pump frequency converter and collected in real time; (8) The operating parameters of the slurry circulation pump are obtained from the DCS, including the start-stop status and frequency of each circulation pump.

[0027] In this embodiment, the absorber tower is not equipped with a slurry flow meter, and the slurry supply pump operating parameters are indirectly characterized by the operating frequency fed back from the frequency converter. The slurry circulation pump operating parameters include the start / stop status of each circulation pump, and the total slurry circulation flow rate is characterized by the sum of the rated flow rates of the activated circulation pumps. When a slurry flow meter is installed in the absorber tower, the slurry supply pump operating parameters can be directly replaced by the flow measurement value; when a frequency converter is installed in the circulation pump, the slurry circulation pump operating parameters can be increased by adding the real-time operating frequency of each pump.

[0028] The aforementioned operational metrics are aggregated by the data acquisition server and stored in a unified data asset pool for use in subsequent steps. Two data acquisition servers are deployed with a redundant architecture, serving as hot backups for each other to ensure high availability of data acquisition.

[0029] (ii) Obtaining information contribution.

[0030] Information contribution is used to quantify the independent information value of each operational indicator to each downstream analysis task. In this embodiment, the downstream analysis tasks include three items: operating condition identification, prediction of circulating slurry pH change trend, and prediction of outlet SO2 concentration change trend.

[0031] Method 1: Based on conditional mutual information. Collect historical operational data for operational indicators for more than 12 months, and perform the following calculations for each downstream analysis task: For the target variable Y of the downstream analysis task, calculate each operational indicator X. i Given other indicators X -i Given the conditional mutual information of Y, the formula is:

[0032] Where H(·|·) is the conditional entropy. The greater the conditional mutual information, the stronger the index X. i Even with other indicators known, it can still provide additional independent information for Y. Dividing the conditional mutual information of each indicator by the sum of the conditional mutual information of all indicators yields the normalized information contribution, with a value range of [0,1].

[0033] Method 2: SHAP-based attribution analysis. For the pH prediction model and the outlet SO2 concentration prediction model, the SHAP (SHapley Additive exPlanations) attribution analysis method is applied to the trained prediction models: For each sample in the test set, the SHAP value of each input feature on the model's predicted output is calculated; the mean of the absolute values ​​of the SHAP values ​​of each feature is used as a measure of the feature's information contribution to the prediction task.

[0034] The two methods can be used individually or in combination: using conditional mutual information as the global information contribution and using SHAP value as the local information contribution distribution within different operating condition intervals, the two are complementary.

[0035] In this embodiment, the information contribution matrix of the eight operational indicators obtained by this method to the three downstream analysis tasks is shown in Table 1: Table 1

[0036] As shown in Table 1, for the operating condition identification task, the information contribution of unit load and sulfur content of coal is the highest (total 0.55); for the pH prediction task, the historical value of circulating slurry pH and inlet SO2 concentration contribute the highest (total 0.45); for the outlet SO2 concentration prediction task, the historical values ​​of inlet SO2 concentration and outlet SO2 concentration contribute the highest (total 0.42). This matrix provides a quantitative basis for the dynamic selection and weight allocation of the subsequent step S2.

[0037] (III) Establishment of the index coupling network.

[0038] From the operational metrics, causally related metric pairs are selected to establish a metric coupling network with metrics as nodes and transfer function parameters as edges. The selected metric pairs are shown in Table 2, including: Table 2

[0039] For each pair of indicators, by utilizing the response process of the downstream indicator when the upstream indicator undergoes significant changes in historical operating data, the three parameters of the first-order plus dead-time transfer function are estimated using the system identification method: gain K, time constant τ, and dead-time θ. The transfer function form is as follows: G(s) = K / (τs+1)·e θs .

[0040] The specific identification method is as follows: Significant step events meeting the following conditions are retrieved from historical operating data: The upstream indicator transitions from a pre-steady state to a post-steady state within a 60-minute time window; the absolute value of the difference between the pre- and post-steady state values ​​exceeds twice the standard deviation σ under normal operating conditions; and the standard deviations of fluctuations in the last 5 minutes of both the pre- and post-steady states do not exceed σ / 2. A data window from 5 minutes before the start of the pre-steady state to 10 minutes after the stabilization of the post-steady state is selected as the identification dataset. The least squares method is used to fit the parameters of a first-order model with dead time, minimizing the mean square error between the model output and the measured response.

[0041] Taking the "slurry pump frequency → circulating slurry pH value" index pair as an example, under steady-state conditions, when the slurry pump frequency jumps from 42Hz to 48Hz, the pH value response process is identified as follows: gain K = 0.042 (i.e., for every 1Hz increase in frequency, the pH value eventually rises by approximately 0.042), time constant τ = 180s (it takes approximately 3 minutes for the pH value to reach 63.2% of the new steady state), and dead time θ = 45s (there is no significant pH value response within 45 seconds after the frequency change). These three parameters fully characterize the dynamic process of the slurry supply change being transmitted to the pH value.

[0042] Using this method, the transfer function parameters of all index pairs are identified one by one, and a directed coupled network with the indexes as nodes is established. Each edge of this network carries three parameters: gain K, time constant τ, and dead zone θ. The quality scoring and update mechanism in subsequent steps all depend on this network.

[0043] (iv) Obtaining quality scores.

[0044] For each operational indicator at each sampling time, based on the aforementioned indicator coupling network, the quality score is quantified and obtained in the following manner.

[0045] Let the current time be t, and let the operating index X be... i The retrieval index is coupled to X in the network. i All upstream metrics. Let the set of upstream metrics be {U1, U2, ..., U...}. k The current values ​​of each upstream indicator are {u1(t), u2(t), ..., u}. k (t)}, each upstream indicator to X i The transfer function is {G} 1i (s),G 2i (s),…,G ki (s)}.

[0046] For each upstream indicator U j Set its current value u j (t) as a transfer function G ji Input (s), calculate X i The estimated predicted value:

[0047] Where F Gji [·] denotes the transfer function G ji (s) Calculates the time-domain response to the input.

[0048] Take the weighted average of the predicted values ​​of all upstream indicators as X. i The overall predicted value: In the actual computer implementation, the zero-order hold method is used to convert the continuous transfer function: G(s) = K / (τs+1)·e -θs Discretize into a first-order difference equation: i (t k )=a· i (t k-1 +b·u j t k-d ; Where a=e -Ts / τ b = K·(1-a), Ts is the sampling period, d = θ / Ts The sampling period is an integer multiple of the dead time. The discretized difference equation is iteratively calculated once in each sampling period to obtain the transfer function output of the upstream index.

[0049]

[0050] Wherein, weight w j Take each upstream indicator to the downstream indicator X i The transfer function gain K ji The absolute value, after normalization, is used, i.e., w. j =|Kji| / Σ|Kli|. Compared to using information contribution as weight, using the gain parameter of the coupled network itself as weight more directly reflects the physical strength of the effect of each upstream indicator on the downstream indicator.

[0051] The weighted average method of the above multi-source forecasts is based on the engineering assumption that "the influence of each upstream indicator on the downstream indicator can be approximately linearly superimposed within a small range of change." When multiple upstream indicators change significantly at the same time—that is, when two or more upstream indicators change by more than twice their historical standard deviation—only the estimated forecast value of the upstream indicator with the largest change is taken as the comprehensive forecast value at that time, and the estimated results of the other upstream indicators are discarded, in order to avoid the superposition bias caused by the coupling effect when multiple inputs change significantly at the same time.

[0052] Calculate the current measured value x i The absolute value of the residual between (t) and the comprehensive forecast value is divided by the historical statistical standard deviation σi of the index to obtain the normalized residual:

[0053] Where, σ i It is calculated from the time-series data of the indicator's normal operation over the past 30 days.

[0054] Substituting the normalized residual into the monotonically decreasing mapping function, the quality score is obtained: "normal operation" refers to the period during which the corresponding sensor has not experienced a known fault and the desulfurization system is not under maintenance or shutdown. This is automatically filtered by the sensor fault flag bit and the absorption tower operation status flag bit of the DCS.

[0055]

[0056] The quality score ranges from [0,1]: when the normalized residual is 0 (the measured value and the predicted value are in perfect agreement), the quality score is 1; as the normalized residual increases, the quality score decreases monotonically; when the normalized residual reaches 3 (the measured value deviates from the predicted value by 3 standard deviations), the quality score drops to about 0.05, indicating that the current measured value of this indicator deviates significantly from the reasonable range calculated by the coupled network, and the sensor may be malfunctioning.

[0057] The calculation process of quality score is illustrated using the pH value of circulating slurry as an example: The upstream indicators for pH value include at least the inlet SO2 concentration and the slurry pump frequency. Let the current inlet SO2 concentration and slurry pump frequency be C_SO2_in(t) and f_pump(t), respectively. Using the identified "C_SO2_in→pH" and "f_pump→pH" transfer functions, the predicted pH value is calculated separately, and a weighted average is taken to obtain the comprehensive predicted value. The actual pH meter reading is compared with this predicted value, and the normalized residual and mapped mass score are calculated. If the pH mass score remains below 0.3 on a certain day, it indicates a systematic deviation between the pH meter reading and the expected value calculated based on the inlet SO2 load and slurry supply, suggesting that the pH meter may need calibration.

[0058] For root node indicators with an in-degree of 0 in the indicator coupling network (such as unit load L and sulfur content of coal S_coal, which have no upstream indicators in this embodiment), their quality scores are not obtained based on the cross-validation of the coupling network, but are determined based on the signal characteristics of the indicator itself.

[0059] Specific judgment rules include, but are not limited to: a) Whether the signal value exceeds the range of the instrument; b) Whether the jump amplitude between two adjacent sampling points exceeds 3 times the historical standard deviation σ of the indicator; c) If the signal value remains unchanged for 20 consecutive sampling points (approximately 20 seconds), it is determined that the sensor is stuck or communication is interrupted.

[0060] When any of the above rules is triggered, the quality score of the indicator is set to 0; when none are triggered, the quality score is 1. The quality score of the root node indicator is ultimately combined with its information contribution to calculate the effective information evaluation value in step S2.

[0061] Step S2: Calculation.

[0062] Based on the information contribution and quality scores obtained in step S1, this step calculates the effective information evaluation value and dynamically filters and adjusts the weights of the input indicator sets for each downstream analysis task accordingly.

[0063] (a) Calculation of the evaluation value of effective information.

[0064] For downstream analysis task T, the performance index X i The effective information evaluation value E for this task i T(t) represents the information contribution C. i T and the current time quality score Q i The product of (t), i.e. E i T(t)=C i,T×Q i (t).

[0065] Among them, C i T is read from the information contribution matrix in part (II) of step S1, Q i (t) is calculated in real time at each sampling time in part (iv) of step S1.

[0066] The effective information evaluation value integrates two dimensions: how much information the indicator provides for the task value and how reliable it is at this moment. For example, the information contribution of the inlet SO2 concentration to pH prediction is 0.20. If its current quality score is 0.95, the effective information evaluation value is 0.19; if its quality score drops to 0.40 due to sensor failure, the effective information evaluation value drops to 0.08—although this indicator is theoretically important for pH prediction, its unreliability at the current moment significantly reduces its actual usable information.

[0067] All candidate indicators of downstream analysis task T are sorted from high to low according to their effective information evaluation value to obtain a dynamic sorting list.

[0068] In this embodiment, taking the pH value change trend prediction task as an example, the effective information evaluation values ​​of the eight candidate indicators at a certain moment are calculated as shown in Table 3: Table 3

[0069] (ii) Dynamic screening and weight adjustment.

[0070] Based on the evaluation value of effective information, dynamic screening and weight adjustment are performed on the input indicator set of each downstream analysis task.

[0071] Dynamic filtering rules: Set a valid information threshold E_thresh, which is 0.05 in this embodiment. Indicators with valid information evaluation values ​​lower than this threshold are removed from the input indicator set of the current task.

[0072] In Table 3, the effective information evaluation values ​​for the outlet SO2 concentration and the circulating pump operating parameters are 0.027 and 0.026, respectively, both below 0.05. These two indicators are not included in the input of the pH prediction model at the current time.

[0073] Dynamic weight allocation rule: For indicators retained in the input indicator set, the weight is the proportion of their effective information evaluation value to the total effective information evaluation value.

[0074] For the six indicators selected in Table 3, taking the pH value of the circulating slurry as an example, its weight is 0.240 / (0.240+0.184+0.135+0.132+0.106+0.074)=0.276. The effective information evaluation value of each indicator is directly used as the dynamic weight coefficient of the input features of the prediction model.

[0075] Redundancy Substitution Rule: When the quality score of a high information contribution indicator is lower than the preset threshold Q_min (0.3 in this embodiment), in addition to reducing the weight of the indicator itself, the system automatically identifies other indicators with information contribution closest to the indicator as redundant substitution indicators and increases the weight of the substitution indicator. Specifically, the system retrieves the two indicators with the smallest absolute value of the difference between their information contribution and that of the indicator in the information contribution matrix, amplifies their information contribution proportionally, and then uses them in the calculation of the effective information evaluation value.

[0076] For example, when the quality score of the inlet SO2 concentration drops to 0.20 due to a CEMS analyzer malfunction, its effective information evaluation value plummets to 0.04 (below 0.05) and is automatically discarded. The system finds that the information contribution of coal sulfur content (0.12) is closest to that of the inlet SO2 concentration (0.20), and automatically increases the weight of coal sulfur content by an amplification factor (0.20 / 0.12≈1.67) to partially compensate for the impact of missing inlet SO2 concentration information on the accuracy of pH prediction.

[0077] Step S3: Output.

[0078] This step inputs the indicator set dynamically processed in step S2 into the corresponding downstream analysis model and outputs the analysis results.

[0079] In this embodiment, the downstream analysis tasks include at least three items: operating condition identification, prediction of pH change trend of circulating slurry, and prediction of SO2 concentration change trend at the outlet.

[0080] (a) Operating condition identification.

[0081] The indicator set after dynamic filtering and weighting in step S2 is input into the working condition identification module.

[0082] In this embodiment, the input indicators for the operating condition identification task are mainly the unit load L, the sulfur content of coal S_coal, and the inlet flue gas flow rate Q_gas_in (total information contribution 0.77). Each indicator participates in the operating condition judgment with the effective information evaluation value calculated in step S2 as a weight.

[0083] Calculate the load change rate:

[0084] Δt is set to 60 seconds. A threshold α = 5% / min is set to determine stable / rising / falling operating conditions. Simultaneously, the sulfur content fluctuation rate δS is calculated, and a threshold β = 20% is used to determine whether to switch coal types. When the contribution weight of unit load information decreases due to quality degradation, the weight of inlet flue gas flow rate is automatically increased as a supplementary criterion.

[0085] The analysis results output by the operating condition identification include: the current operating condition category (stable / rising / falling operating condition), load range identifier, and sulfur content range identifier. These results drive the selection of subsequent prediction models and control strategies.

[0086] (II) Prediction of pH value change trend of circulating slurry.

[0087] The index set after dynamic screening and weighting in step S2 is input into the pH prediction model.

[0088] Unlike the fixed feature input in the prior art, the input feature set and its weights of the present invention change dynamically at each sampling time: when all indicators are of good quality, multiple features are weighted according to the effective information evaluation value; when the quality of some indicators declines, the number of features is automatically reduced and redundant replacement is enabled.

[0089] The prediction model employs a gradient boosting decision tree (GBDT) regression model (trained offline according to standard procedures). The model takes a dynamically weighted set of indicators at the current moment as input and outputs a pH prediction sequence {pH_pred(t+k), k=1,...,5} for the next 1-5 minutes. This prediction sequence is used by the slurry supply control system to calculate the feedforward adjustment.

[0090] (III) Forecast of SO2 concentration change trend at exports.

[0091] Similar to pH prediction, the index set dynamically processed in step S2 is input into the SO2 outlet concentration prediction model, which outputs a predicted sequence of outlet SO2 concentrations for the next 1 to 5 minutes. This predicted sequence is used by the circulating pump control system to calculate the feedforward adjustment.

[0092] (iv) Calculation and execution of downstream control quantities.

[0093] The above prediction results are further used to calculate the control quantities of the slurry supply control system and the circulating pump control system: the slurry supply control quantity is calculated based on the deviation between the pH value prediction sequence and the target set value, and the slurry supply pump frequency feedforward adjustment quantity is calculated and sent to the DCS system for execution with a control cycle of 10 seconds; the circulating pump control quantity is calculated based on the deviation between the outlet SO2 concentration prediction sequence and the target set value, and the circulating pump start-stop combination and frequency conversion adjustment quantity are calculated and sent to the DCS system for execution with a control cycle of 30 seconds.

[0094] Regular update mechanism: The information contribution matrix and indicator coupling network are updated periodically using newly added operational data. Every 30 days, the information contribution matrix is ​​updated by recalculating the conditional mutual information and SHAP values ​​using newly added operational data. An alert is triggered when the information contribution of a particular indicator decreases by more than 20% compared to the previous period. Every 30 days, the indicator coupling network re-identifies the transfer function parameters of each indicator pair using newly added operational data. An alert is also triggered when the gain K of an indicator pair changes by more than 30% or the dead time θ changes by more than 60 seconds.

[0095] Example 2 like Figure 2 As shown in the figure, this embodiment provides a wet desulfurization system index analysis and optimization system, including an acquisition module, a calculation module and an execution module.

[0096] The acquisition module is deployed on the data acquisition server to acquire the operating indicators of the wet desulfurization slurry supply system from DCS, CEMS, and online analytical instruments. The acquisition module has built-in information contribution calculation unit and quality score calculation unit: the information contribution calculation unit stores and periodically updates the information contribution matrix of each operating indicator to each downstream analysis task; the quality score calculation unit embeds an indicator coupling network model, calculates the predicted value based on the coupling network at each sampling time and compares it with the measured value, and outputs the [0,1] quality score of each indicator.

[0097] The calculation module is deployed on an advanced control server and communicates with the acquisition module via an industrial Ethernet network. In each control cycle, the calculation module reads the information contribution and quality score of the current moment from the acquisition module, calculates the effective information evaluation value of each operating indicator for each downstream analysis task, and generates a dynamically filtered set of indicators and corresponding weight vectors.

[0098] The execution module is deployed on an advanced control server. It receives the dynamic weighted index set output by the calculation module, calls the corresponding downstream analysis models, including the operating condition identification model, pH value prediction model, and outlet SO2 concentration prediction model, performs inference, outputs analysis results and control commands, and sends them to the DCS system for execution through the safety assurance layer.

[0099] The acquisition module, calculation module, and execution module exchange data via the OPC UA protocol to ensure the real-time performance and reliability of data transmission.

[0100] In the above embodiments: Once an online chloride ion analyzer is installed in the absorption tower, the chloride ion concentration can be added as a new operating indicator to the indicator list in step S1, and participate in the information contribution calculation and coupled network modeling to achieve quality monitoring of the slurry quality dimension.

[0101] The effective information threshold E_thresh and the quality score lower limit threshold Q_min in step S2 can be set separately according to the sensitivity of each downstream analysis task to the amount of input information, rather than taking a uniform value for all tasks. For example, the working condition identification task has a high requirement for information completeness, so E_thresh can be set to 0.08; the pH value prediction task can be set to E_thresh = 0.04.

[0102] In step S1, in addition to cross-validation based on coupled networks, the quality score can be further integrated with the scores of the signal-to-noise ratio dimension or the data integrity dimension, and the final quality score can be obtained by weighted summation, thereby enhancing the sensitivity of the quality score in the early stage of sensor failure.

[0103] When the unit is configured as a supercritical or ultra-supercritical unit, the dynamic range of operating indicators is wider. The calculation of information contribution and the identification of coupling networks should use historical data covering the entire operating condition range to ensure the comprehensiveness and representativeness of the evaluation benchmark.

[0104] This invention quantifies the independent information contribution of each operational indicator to downstream tasks through conditional mutual information and SHAP attribution analysis, eliminating the negative impact of low-information-value indicators on prediction accuracy. It establishes an indicator coupling network with FOPDT transfer function parameters as edges, calculates the predicted values ​​of downstream indicators based on the upstream indicator transfer function output, and obtains a quality score by mapping the normalized residuals of the upstream and downstream indicators to achieve real-time detection of sensor degradation, drift, and jamming. Based on the product of information contribution and quality score, it calculates an effective information evaluation value, dynamically filters and adjusts the weights of the downstream task input indicator set time-by-time, and automatically activates redundant replacement indicators when the quality of high-contribution indicators declines, transforming inherent measurement redundancy into fault-tolerant resources in case of sensor failure. The information contribution matrix and coupling network parameters are updated periodically, and an early warning is triggered when the indicator information value or transfer function parameter changes exceed a threshold, ensuring the long-term effectiveness of the analysis model.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing and optimizing the performance indicators of a wet desulfurization system, characterized in that, Includes the following steps: Step S1: Obtain the operating indicators of the wet desulfurization slurry supply system. The operating indicators include at least the inlet flue gas flow rate, inlet SO2 concentration, sulfur content of coal, unit load, circulating slurry pH value, outlet SO2 concentration, slurry supply pump operating parameters, and slurry circulation pump operating parameters; obtain the information contribution of each operating indicator to each downstream analysis task; obtain the quality score of each operating indicator at the current time. Step S2: Based on the information contribution and the quality score, calculate the effective information evaluation value of each operating indicator for each downstream analysis task, and dynamically filter and / or adjust the weight of the input indicator set for each downstream analysis task according to the effective information evaluation value; Step S3: Use the index set processed in step S2 to perform the downstream analysis task and output the analysis results.

2. The method for analyzing and optimizing the performance indicators of a wet desulfurization system according to claim 1, characterized in that, In step S1, the information contribution is obtained in the following way: based on the historical operating data of the operating indicators, the independent information increment of each operating indicator to the target downstream analysis task is quantified by the conditional mutual information method and / or SHAP attribution analysis method, given that other indicators are known. After normalization processing, the information contribution of each operating indicator is obtained.

3. The method for analyzing and optimizing the performance indicators of a wet desulfurization system according to claim 1, characterized in that, In step S1, select causally related indicator pairs from the operational indicators, and use the response process of the corresponding downstream indicators when the upstream indicators change significantly in historical operational data to estimate the transfer function parameters of each indicator pair through the system identification method. The transfer function parameters include at least gain, time constant and dead time, and establish an indicator coupling network with indicators as nodes and transfer function parameters as edges. Based on the index coupling network, the predicted value of each operational index at the current moment is calculated from the upstream operational indexes that are causally related to each operational index, and the quality score is obtained based on the deviation between the current measured value and the predicted value.

4. The method for analyzing and optimizing the performance indicators of a wet desulfurization system according to claim 3, characterized in that, The quality score is quantified as follows: the absolute value of the residual between the current measured value and the predicted value is calculated and divided by the historical statistical standard deviation of the indicator to obtain the normalized residual; the normalized residual is substituted into a monotonically decreasing mapping function to obtain a quality score with a value range of [0,1]. The larger the normalized residual, the lower the quality score; for root node indicators without upstream indicators, the quality score is determined based on the signal range exceeding the limit, the jump amplitude, and the signal freezing rules.

5. The method for analyzing and optimizing the performance indicators of a wet desulfurization system according to claim 1, characterized in that, In step S2, the effective information evaluation value is the product of the information contribution degree and the quality score.

6. The method for analyzing and optimizing the performance indicators of a wet desulfurization system according to claim 1, characterized in that, In step S2, the dynamic screening and / or weight adjustment includes: setting an effective information threshold, removing indicators with effective information evaluation values ​​lower than the effective information threshold from the input indicator set of the corresponding downstream analysis task; when the quality score of an indicator is lower than a preset quality threshold, reducing the input weight of the indicator in the corresponding downstream analysis task, and automatically identifying redundant alternative indicators with information contribution close to the indicator, and proportionally increasing the weight of the redundant alternative indicators.

7. The method for analyzing and optimizing the performance indicators of a wet desulfurization system according to claim 1, characterized in that, In step S3, the downstream analysis task includes at least one or more of the following: operating condition identification, prediction of circulating slurry pH change trend, prediction of outlet SO2 concentration change trend, calculation of slurry supply control quantity, and calculation of circulating pump control quantity.

8. The method for analyzing and optimizing the performance indicators of a wet desulfurization system according to claim 3, characterized in that, The information contribution is recalculated using newly added operational data at a preset cycle, and the transfer function parameters of the indicator coupling network are re-identified. An early warning is triggered when the change in the information contribution of a certain indicator exceeds a preset threshold or the change in the transfer function parameters of a certain indicator pair exceeds a preset threshold.

9. A wet desulfurization system index analysis and optimization system based on the wet desulfurization system index analysis and optimization method according to any one of claims 1-8, characterized in that, include: The acquisition module is used to acquire the operating indicators of the wet desulfurization slurry supply system, the information contribution of each operating indicator to each downstream analysis task, and the quality score of each operating indicator at the current time. The calculation module, which is communicatively connected to the acquisition module, is used to calculate the effective information evaluation value of each operating indicator for each downstream analysis task based on the information contribution degree and the quality score, and to dynamically filter and / or adjust the weight of the input indicator set of each downstream analysis task according to the effective information evaluation value. The execution module is communicatively connected to the calculation module and is used to execute the downstream analysis task with the indicator set processed by the calculation module and output the analysis results.

10. The wet desulfurization system index analysis and optimization system according to claim 9, characterized in that, The acquisition module has a built-in information contribution calculation unit and a quality score calculation unit; the quality score calculation unit has an embedded index coupling network model, the index coupling network has running indicators as nodes and transfer function parameters as edges, the transfer function parameters include at least gain, time constant and dead time; the acquisition module, calculation module and execution module exchange data through the OPC UA protocol.

Images