Artificial intelligence-based double-tower desulfurization slurry circulating pump combination optimization method and system

Abstract

The present application relates to the technical field of desulfurization circulating pump frequency conversion and slurry supply collaborative control, in particular to a double-tower desulfurization slurry circulating pump combination optimization method and system based on artificial intelligence, comprising the following steps: predicting future multi-time-point boiler load values, defining discrete running situations in combination with the start-stop state and running frequency of the circulating pump group, establishing slurry output values corresponding to each situation, matching the slurry amount required by the boiler load at each future time point, and generating a future working condition requirement and running situation mapping set. The present application dynamically defines discrete running situations by predicting the boiler load at multiple future time points and in combination with the start-stop state and running frequency of the circulating pump group, matches each situation with the load requirement, and then calculates the energy consumption transfer value and equipment start-stop wear cost generated by the running frequency change, to realize detailed evaluation of the comprehensive cost of the migration path between running situations.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of desulfurization circulating pump frequency conversion and slurry supply collaborative control, and particularly relates to a double-tower desulfurization slurry circulating pump combination optimization method and system based on artificial intelligence. BACKGROUND

[0002] The technical field of desulfurization circulating pump frequency conversion and slurry supply collaborative control mainly involves dynamically adjusting the operating frequency and combination start-stop strategy of circulating pumps to achieve the goal of precise control of slurry supply in the desulfurization system and efficient use of energy.

[0003] In actual application, the prior art relies on frequency regulation and combination start-stop strategy, and has deficiencies in the prediction accuracy of boiler load changes and slurry demand, and is insufficient in matching the boiler load demand at multiple future time points and the frequency and start-stop of equipment; it lacks sufficient quantitative means in the fine analysis of equipment start-stop loss and energy consumption, leading to frequent switching of operating conditions and an increase in the number of equipment start-stops, which in turn causes increased equipment wear and tear and maintenance costs. Therefore, improvements are needed. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art and to provide a double-tower desulfurization slurry circulating pump combination optimization method and system based on artificial intelligence.

[0005] To achieve the above-mentioned purpose, the present application adopts the following technical solution: a double-tower desulfurization slurry circulating pump combination optimization method based on artificial intelligence, comprising the following steps:

[0006] Predicting future boiler load values at multiple time points, defining discrete operating conditions in combination with the start-stop state and operating frequency of the circulating pump group, establishing slurry output values corresponding to each condition, matching the slurry amount required by the boiler load at each future time point, and generating a future operating condition demand and operating condition mapping set;

[0007] Based on the future operating condition demand and operating condition mapping set, all operating conditions are extracted, the energy consumption transfer value generated by the change in operating frequency between any two conditions is calculated, an energy consumption transfer cost table is established, and based on the energy consumption transfer cost table, the quantified start-stop equipment wear and tear cost is superimposed to generate a comprehensive wear and tear cost of the operating condition migration path;

[0008] Based on the future operating condition demand and operating condition mapping set, all operating conditions that meet the slurry demand at the first time point are filtered to obtain a feasible condition set for the first node, and based on the feasible condition set for the first node, operating conditions that meet the constraints for subsequent time points are matched in sequence and connected to establish a feasible migration network between multiple time nodes;

[0009] Based on the feasible migration network between the multiple time nodes, starting from the current running situation, tracking to the end time node, generating all optional complete path sequences, obtaining the full-quantitative path candidate list, based on the full-quantitative path candidate list, calling the running situation migration path to calculate the total cost of each path and sorting, and establishing the circulating pump combined optimal scheduling instruction path.

[0010] Preferably, the obtaining step of the future working condition demand and running situation mapping set is:

[0011] Based on the boiler operation record, the boiler load value, the inlet sulfur dioxide concentration of the absorption tower, the slurry circulation amount, the liquid-gas ratio, the pH value, the oxygen concentration and the circulating pump operation frequency at each time point in the boiler operation record are extracted to obtain a boiler load prediction input set;

[0012] According to the boiler load prediction input set, the boiler load prediction value at each future time point is calculated;

[0013] Based on the boiler load prediction value, a circulating pump start-stop state and operation frequency defined combined situation set is called, the predicted load is matched with the value of the slurry output corresponding to each situation, and a future working condition demand and running situation mapping set is generated.

[0014] Preferably, the obtaining step of the energy consumption transfer cost table is:

[0015] Based on the future working condition demand and running situation mapping set, the operation frequency values of all circulating pumps in each running situation are extracted in sequence, the frequencies of the corresponding circulating pumps under any two different running situations are paired, and the target operation time of the circulating pump after the situation transfer is synchronously extracted to generate a running frequency transfer and time pairing sequence;

[0016] According to the running frequency transfer and time pairing sequence, the total energy consumption transfer value is calculated;

[0017] Based on the total energy consumption transfer value, the energy consumption changes between each pair of running situations are registered item by item, a matrix structure is constructed according to the running situation number, the energy consumption differences of the transfer paths between all situations are output in row-column mode, and an energy consumption transfer cost table is generated.

[0018] Preferably, the obtaining step of the running situation migration path comprehensive cost is:

[0019] Based on the running situation start-stop record, the running state identifiers of each device before and after the migration of each group of running situations are compared, the devices whose states change from running to stopping or from stopping to running are identified, each device whose state changes is recorded as a start-stop action, and the operation time of the recent operation cycle is extracted to generate a start-stop device and its start-stop pre-operation time pairing table;

[0020] According to the start-stop device and the start-stop time length before starting, the running state trend migration path cost is calculated according to the list and the energy consumption transfer cost table;

[0021] Based on the migration path cost, the migration path cost between all running state trend pairs is sequentially filled into the cost table structure matrix, and is arranged according to the starting state trend and the target state trend index, and a running state trend migration path comprehensive cost is generated.

[0022] Preferably, the step of obtaining the feasible state trend set of the first node is:

[0023] Based on the future working condition demand and the running state trend mapping set, the running frequency and the start-stop state of all circulating pumps of each running state trend in the future working condition demand and the running state trend mapping set are retrieved, the slurry output flow of each circulating pump is calculated one by one, and the slurry output flow of all circulating pumps is summed to form the total slurry output flow of each running state trend;

[0024] Based on the total slurry output flow of each running state trend, the slurry amount demand value required by the boiler load corresponding to the first time point in the future working condition demand and the running state trend mapping set is extracted, the total slurry output flow of each running state trend is compared with the demand value one by one, all running state trends with the total slurry output flow equal to or higher than the demand value are selected, and a preliminary running state trend set satisfying the slurry demand of the boiler load at the first time point is formed;

[0025] Based on the preliminary running state trend set satisfying the slurry demand of the boiler load at the first time point, whether the start-stop state and the frequency of each circulating pump in the running state trend are in the safe running interval allowed by the device is verified item by item, the running state trend with the start-stop state or the frequency exceeding the safe running range is removed, and a feasible state trend set of the first node is formed.

[0026] Preferably, the step of obtaining the feasible migration network between multiple time nodes is:

[0027] Based on the feasible state trend set of the first node, the running frequency, the start-stop state and the slurry output flow of all circulating pumps corresponding to each running state trend in the set are extracted one by one, all estimated running state trends satisfying the slurry amount demand of the second time point are matched with each running state trend according to the slurry amount demand value corresponding to the boiler load demand of the second time point, and a feasible state trend connection set from the first node to the second time node is formed.

[0028] Based on the first node to the second time point of the feasible situation connection set, all the circulating pump operating frequencies, start-stop states and slurry output flow rates of all operating situations at the second time point are retrieved and extracted in sequence, and according to the slurry quantity demand value corresponding to the third time point boiler load demand, it is judged whether each situation can meet the slurry demand at the third time point under the constraints of circulating pump frequency adjustment amplitude and safe frequency interval, and the connected situations meeting the conditions are screened to form the feasible situation connection set from the second time point to the third time point.

[0029] Based on the feasible situation connection set from the second time point to the third time point, the operating situation connection and constraint verification are performed in sequence to all the subsequent remaining time points, the feasible situation connections of all time points are connected in series to form a feasible migration network among multiple time nodes.

[0030] Preferably, the step of obtaining the full-quantitative path candidate list is:

[0031] Based on the feasible migration network among multiple time nodes, the current operating situation in the current time node is selected as the initial starting node, all operating situation nodes connected to the next time node under the initial starting node are extracted one by one, the operating frequencies, start-stop states and slurry output flow rates of all circulating pumps contained in each situation node are recorded in sequence, and the connection relationships between respective nodes are marked to form a situation path set from the initial starting node to the second time node.

[0032] Based on the situation path set from the initial starting node to the second time node, the operating frequencies, start-stop states and slurry output flow rates of all circulating pumps corresponding to each operating situation node in the second time node are retrieved in sequence, all feasible connected operating situations are matched to the third time node, and the connection relationships between each pair of nodes are recorded, and the node connection chain set from the initial starting node to the end time node is formed by progressing to the end time node one by one.

[0033] Based on the node connection chain set from the initial starting node to the end time node, the full-quantitative path candidate list is formed by starting from the initial starting node, traversing all node connection chains one by one, connecting in series to form a continuous and complete path sequence of multiple time nodes, and eliminating the connection chains that are broken or cannot continue to the end time node.

[0034] Preferably, the step of obtaining the circulating pump combination optimal dispatching instruction path is:

[0035] Based on the full quantization path candidate list, each complete path sequence in the list is retrieved one by one, the operation situation transition path comprehensive cost corresponding to each pair of adjacent operation situation nodes between the starting node and the end node of the path is extracted step by step, the operation situation transition path comprehensive cost between each pair of nodes is recorded one by one, and the operation situation transition path comprehensive costs of all adjacent node pairs are accumulated to form the accumulated operation situation transition path comprehensive cost corresponding to each path;

[0036] Based on the accumulated operation situation transition path comprehensive cost corresponding to each path, the accumulated operation situation transition path comprehensive costs of all complete path sequences are sorted from low to high to form a path accumulated path comprehensive cost sorting list;

[0037] Based on the path accumulated path comprehensive cost sorting list, the complete path sequence with the lowest operation situation transition path comprehensive cost in the sorting list is selected, and the combined optimal scheduling instruction path of the circulating pump is formed according to the circulating pump operation frequency and start-stop state recorded by each node in the path sequence.

[0038] The application also provides a double-tower desulfurization slurry circulating pump combination optimization system, which comprises:

[0039] A prediction module: predicting future boiler load values at multiple time points, defining discrete operation situations in combination with the start-stop state and operation frequency of the circulating pump group, establishing slurry output values corresponding to each situation, matching the slurry amount required by the boiler load at each future time point, and generating a future working condition demand and operation situation mapping set;

[0040] A cost modeling module: based on the future working condition demand and operation situation mapping set, extracting all operation situations, calculating energy consumption transfer values generated due to changes in operation frequency between any two situations, establishing an energy consumption transfer cost table, superimposing the quantified wear and tear cost of the start-stop equipment based on the energy consumption transfer cost table, and generating an operation situation transition path comprehensive cost;

[0041] A network construction module: based on the future working condition demand and operation situation mapping set, filtering all operation situations that meet the slurry amount demand at the first time point, obtaining a feasible situation set of the first node, and based on the feasible situation set of the first node, sequentially matching the operation situations that meet the constraints for each subsequent time point and constructing connections to establish a feasible migration network between multiple time nodes;

[0042] A path optimization module: based on the feasible migration network between multiple time nodes, starting from the current operation situation, tracking to the last time node, generating all selectable complete path sequences, obtaining a full quantization path candidate list, based on the full quantization path candidate list, calling the operation situation transition path comprehensive cost to calculate the total cost of each path and sorting, and establishing a combined optimal scheduling instruction path of the circulating pump.

[0043] Compared with the prior art, the application has the advantages and positive effects that:

[0044] The application dynamically defines discrete operation situations by predicting the boiler load at multiple future time points and combining the start-stop state and operating frequency of the circulating pump group, matches each situation with the load demand, calculates the energy consumption transfer value and equipment start-stop wear and tear cost caused by the change in operating frequency, realizes the detailed evaluation of the comprehensive cost of the migration path between operation situations, further screens the relevance of the slurry demand at each time point and the operation situation combination, forms a feasible migration network and gradually optimizes it, improves the operation accuracy, economy and stability of the desulfurization circulating pump, realizes the real-time matching of the operation state and the boiler load demand, reduces the maintenance cost caused by the operation energy consumption and equipment start-stop wear and tear, improves the smoothness of equipment operation, and avoids the problem of accelerated equipment aging caused by frequent switching of operation situations. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 The application is a step schematic diagram. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical scheme and advantages of the application more clear and explicit, the application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application and do not limit the application.

[0047] Please refer to Figure 1 The application provides a technical scheme, a double-tower desulfurization slurry circulating pump combination optimization method based on artificial intelligence, which comprises the following steps:

[0048] The boiler load values at multiple future time points are predicted, the discrete operation situations are defined in combination with the start-stop state and operating frequency of the circulating pump group, the slurry output values corresponding to each situation are established, the slurry amount required by the boiler load at each future time point is matched, and a future working condition demand and operation situation mapping set is generated;

[0049] Based on the future working condition demand and operation situation mapping set, all operation situations are extracted, the energy consumption transfer value between any two situations caused by the change in operating frequency is calculated, an energy consumption transfer cost table is established, the start-stop equipment wear and tear cost after quantization is superimposed based on the energy consumption transfer cost table, and a comprehensive cost of operation situation migration path is generated;

[0050] Based on the future working condition demand and operation situation mapping set, all operation situations that meet the slurry amount demand at the first time point are screened, a feasible situation set of the first node is obtained, and based on the feasible situation set of the first node, the operation situations that meet the constraints for each subsequent time point are matched in turn and connected, a feasible migration network between multiple time nodes is established;

[0051] Based on the feasible migration network between multiple time nodes, starting from the current running situation, tracking to the end time node, generating all optional complete path sequences, obtaining the full-quantitative path candidate list, based on the full-quantitative path candidate list, calling the running situation migration path to calculate the total cost of each path and sorting, establishing the optimal dispatching instruction path of the circulating pump combination.

[0052] The acquisition step of the future working condition demand and running situation mapping set is:

[0053] Based on the boiler operation record, the boiler load value, the absorption tower inlet sulfur dioxide concentration, the slurry circulation amount, the liquid-gas ratio, the pH value, the oxygen concentration and the circulating pump operation frequency of each time point in the boiler operation record are extracted to obtain the boiler load prediction input set;

[0054] According to the boiler load prediction input set, the boiler load prediction value of each future time point is calculated, and the calculation formula is:

[0055]

[0056] Among them, F t is the boiler load prediction value of the tth time point, h is the total number of data points contained in the current prediction window, C c is the SO2 concentration at the inlet of the absorption tower at the cth time point, Q c is the slurry circulation amount at the cth time point, G c is the liquid-gas ratio at the cth time point, p c is the pH value at the cth time point, Freq c is the circulating pump operation frequency at the cth time point, O2 c is the oxygen concentration at the cth time point, and k0, k1, k2, k3, k4, k5, k6 are fixed regression coefficients determined based on sample training.

[0057] Based on the boiler load prediction value, the circulating pump start-stop state and operation frequency definition combination situation set is called, the predicted load and the slurry output corresponding to each situation are matched in value, and the future working condition demand and running situation mapping set is generated.

[0058] Specifically, based on the historical operation data records of the boiler, which are usually stored in the distributed control system (DCS) or the production information system (SIS) database of the power plant, first, the data is aggregated and aligned according to the preset time interval (for example, every minute or every five minutes), ensuring the consistency of each parameter on the timestamp, then for each time point, the actual load value of the boiler is accurately extracted from the records, which is usually in units of megawatts (MW) or tons / hour (t / h of evaporation), directly reflecting the output state of the boiler, the real-time concentration value of sulfur dioxide (SO2) in the flue gas at the inlet of the absorption tower is synchronously extracted, which is derived from the continuous emission monitoring system (CEMS), and the unit is usually milligrams per standard cubic meter (mg / Nm 3 ), accurately recording the initial amount of SO2 entering the desulfurization system, then the total slurry circulation amount transported by the slurry circulating pump is extracted, which is obtained by the electromagnetic flowmeter or ultrasonic flowmeter installed on the outlet pipeline of the circulating pump, and the unit is cubic meters / hour (m 3 / h), this parameter is crucial to maintain sufficient liquid-solid contact, at the same time, the liquid-gas ratio (L / G) at that time is calculated or extracted, if directly extracted, it is derived from the system preset calculation module, if calculated, it is calculated by the formula L / G = slurry circulation amount / flue gas amount, wherein the flue gas amount is estimated according to the boiler load, fuel characteristics and excess air coefficient, etc. or directly measured by the flue gas flowmeter, with the unit of liters / cubic meter (L / m 3 ) or dimensionless, then the slurry pH value is obtained, which is monitored by the online pH analyzer inserted into the slurry pool or pipeline in real time, reflecting the acidity and alkalinity of the slurry, which is crucial for limestone dissolution and calcium sulfite oxidation, then the oxygen concentration in the flue gas or slurry is extracted, which usually refers to the oxygen content in the flue gas, measured by the oxygen analyzer in the CEMS, with the unit of percentage (%), finally, the running frequency of each slurry circulating pump is recorded, which is obtained from the feedback signal of the frequency converter (VFD), with the unit of hertz (Hz), which directly affects the speed of the pump and the output of the slurry, these boiler load values, SO2 concentration at the inlet of the absorption tower, slurry circulation amount, liquid-gas ratio, pH value, oxygen concentration and running frequency of each circulating pump data collected or calculated at the same time point are integrated to form a structured data set, wherein each record represents a complete snapshot of the working condition at a specific time point, and this data set is the boiler load prediction input set.

[0059] Formula: (k0+k1·C c +k2·Q c +k3·G c +k4·p c +k5·Freq c +k6·O2 cThe advantage of this formula lies in the fact that it uses a multiple linear regression model combined with a moving average method to predict future boiler load. Its benefits are reflected in the following aspects: First, by incorporating multiple operating parameters (C...) closely related to boiler load and desulfurization efficiency... c Q c G c ,p c ,Freq c O2 c The model can more comprehensively capture the complex factors affecting load changes, improving the accuracy and robustness of predictions; secondly, it uses a moving average. The processing can effectively smooth out short-term data fluctuations and noise, making the prediction results more stable and avoiding prediction deviations caused by instantaneous disturbances. Secondly, the linear regression model has a relatively simple structure, high computational efficiency, and is easy to understand and deploy, making it easy for engineering technicians to quickly master and apply. Finally, the regression coefficients (k0 to k6) are determined through training with historical data samples, which can reflect the operating characteristics of specific units and have a certain degree of adaptability.

[0060] Parameter description:

[0061] `h` represents the total number of historical data points included in the current forecast window. This is a parameter used for moving average calculation, determining the length of historical data backtracked during model forecasting. For example, if the data collection frequency is once per minute, setting `h=10` means using data from the past 10 minutes for averaging. The selection of this parameter requires analysis of historical data to balance the smoothness of the forecast and the response speed to load changes. It is obtained by backtesting different `h` values ​​(e.g., 5, 10, 15, 20 data points) on a historical validation set, evaluating the forecast error (e.g., root mean square error RMSE), and selecting the `h` value that minimizes the RMSE. For example, testing a power plant's historical data for one year at minute intervals revealed that `h=12` (i.e., a 12-minute window) provided the best overall performance in terms of forecast accuracy and stability; therefore, `h=12` was chosen.

[0062] C c This represents the SO2 concentration at the inlet of the absorber tower at the c-th time point in the past. This parameter reflects the initial pollutant load entering the desulfurization system. This data is collected in real time by a continuous flue gas monitoring system (CEMS) installed on the inlet flue of the absorber tower. For example, at a historical data point c, the SO2 concentration monitored and recorded by the CEMS system is 1500 mg / Nm³. 3 .

[0063] Q crepresents the slurry circulation rate at the past time point c. This parameter represents the slurry flow rate used for SO2 absorption by the desulfurization system at that time. This data is obtained by real-time monitoring with electromagnetic flow meters or ultrasonic flow meters installed on the outlet header or branch pipes of the slurry circulation pumps. For example, at a certain historical data point c, the flow meter reading is 3000 m3 / h. 3 .

[0064] G c represents the liquid-gas ratio at the past time point c. It reflects the amount of slurry involved in the reaction per unit of flue gas, which is a key operating parameter affecting the desulfurization efficiency. This parameter is calculated by the DCS system based on the real-time slurry circulation rate (Q c ) and the real-time flue gas rate (V gas,c ), with the calculation formula being: G c = Q c / V gas,c . The flue gas rate can be estimated from the boiler load, fuel composition, etc., or directly measured by a flue gas flow meter. For example, if Q c = 3000 m 3 / h and the flue gas rate at that time is V gas,c = 300000 Nm 3 / h, then G c = (3000 x 1000) / 300000 = 10 L / Nm 3 .

[0065] p c represents the slurry pH value at the past time point c. This is a dimensionless parameter that reflects the acidity or alkalinity of the slurry. This data is monitored in real time by an online pH meter installed in the slurry tank or circulation pipeline. The pH value has a significant impact on the dissolution rate of limestone and the progress of the desulfurization reaction. For example, at a certain historical data point c, the online pH meter reading is 5.8.

[0066] Freq c represents the average operating frequency of the main circulation pump or the operating frequency of the circulation pump that contributes most to the total slurry volume at the past time point c. The unit is Hz (Hertz). This parameter reflects the working intensity of the circulation pump and is directly related to the output capacity of the slurry. This data is obtained from the control system or feedback signal of the frequency converter (VFD). If there are multiple pumps running, the average of their frequencies can be taken. For example, if two identical circulation pumps are running at 40 Hz and 45 Hz respectively, then Freq c

[0067] = (40 + 45) / 2 = 42.5 Hz.

[0068] O2 crepresents the oxygen concentration in the flue gas at the past cth time point. The unit is % (volume percentage). The oxygen concentration in the flue gas is not only related to the boiler combustion condition, but also affects the oxidation process of the desulfurization by-product (such as gypsum). This data is monitored in real time by the oxygen analyzer in the CEMS system. For example, at a certain historical data point c, the oxygen analyzer reading is 6.5%.

[0069] k0, k1, k2, k3, k4, k5, k6 are fixed regression coefficients determined by training based on historical operation data samples. These coefficients are the core parameters of the model, which quantify the degree and direction of the influence of each input variable on the boiler load. The way to obtain them is: collect at least three months of continuous historical operation data (including all the above-mentioned input variables and the corresponding actual boiler load), to form a training data set. Use multivariate linear regression analysis method (such as least squares method) to fit the training data, and solve the coefficients that make the residual sum of squares between the predicted value and the actual value minimum. For example, by regression analysis on historical data, k0 = 20.5, k1 = 0.005, k2 = 0.015, k3 = -5.2, k4 = 8.1, k5 = 1.2, k6 = -3.3 are obtained. The units of these coefficients need to match the units of the corresponding variables and the unit of F t to ensure dimensional consistency. For example, if the unit of F t is MW, the unit of C c is mg / Nm 3 , then the unit of k1 is MW / (mg / Nm 3 ).

[0070] Calculation process:

[0071] The calculation process of this formula is divided into two steps: first, calculate the weighted sum of each historical data point c, and then take the average of the h weighted sums.

[0072] Take the prediction of the boiler load F1 at the first future time point t = 1 as an example.

[0073] Set the prediction window h = 2.

[0074] The regression coefficients used (obtained by training on historical data) are:

[0075] k0 = 20.5 (MW);

[0076] k1 = 0.005 (MW / (mg / Nm 3 ));

[0077] k2 = 0.015 (MW / (m 3 / h));

[0078] k3 = -5.2 (MW / (L / Nm 3 ));

[0079] k4 = 8.1 (MW / pH);

[0080] k5 = 1.2 (MW / Hz);

[0081] k6 = -3.3 (MW / %);

[0082] Parameter values for historical data point 1 (c = 1):

[0083] C1 = 1550 mg / Nm 3 ;

[0084] Q1 = 3050 m 3 / h;

[0085] G1 = 10.5 L / Nm 3 ;

[0086] p1 = 5.7;

[0087] Freq1 = 42.0 Hz;

[0088] O21 = 6.2%;

[0089] Calculate load contribution term L1 for historical data point 1:

[0090] L1 = k0 + k1 · C1 + k2 · Q1 + k3 · G1 + k4 · p1 + k5 · Freq1 + k6 · O21;

[0091]

[0092] L1 = 20.5 + 7.75 + 45.75 - 54.6 + 46.17 + 50.4 - 20.46;

[0093] L1 = 95.51 MW;

[0094] Parameter values for historical data point 2 (c = 2):

[0095] C2 = 1500 mg / Nm 3 ;

[0096] Q2 = 3000 m 3 / h;

[0097] G2 = 10.0 L / Nm 3 ;

[0098] p2 = 5.8;

[0099] Freq2 = 42.5 Hz;

[0100] O22 = 6.5%;

[0101] Calculate the load contribution term L2 for historical data point 2:

[0102] L2=k0+k1·C2+k2·Q2+k3·G2+k4·p2+k5·Freq2+k6·O22;

[0103]

[0104] L2=20.5+7.5+45.0-52.0+46.98+51.0-21.45;

[0105] L2 = 97.53 MW;

[0106] Finally, calculate the boiler load forecast value F at future time point t. t (This is F1):

[0107]

[0108] F1 = 96.52 MW;

[0109] The results indicate that, based on the operating parameters from the two most recent historical time points and the pre-set regression model, the predicted boiler load for the next time point is 96.52 MW. This predicted value will serve as the basis for determining the required slurry volume for the future desulfurization system. A higher predicted value means a greater slurry volume is needed to meet desulfurization requirements, and vice versa.

[0110] Based on the boiler load forecast values ​​for each future time point calculated in the preceding steps, for example, F t =96.52MW. First, a pre-built set of combined states and operating frequencies of the circulating pumps needs to be invoked. This set is generated during system initialization or periodic calibration, based on the physical characteristics of all slurry circulating pumps in the dual-tower desulfurization system (e.g., three circulating pumps, A, B, and C), the allowable operating frequency range (e.g., the operating frequency range of a single pump is 25Hz to 50Hz), the minimum start-stop time interval, and the interlocking logic between pumps (e.g., at least one pump A or B must be running before pump C starts). It is generated by combining all permutations and filtering out combinations that do not meet the operating conditions. Each combined state uniquely describes the start-stop (0 represents stop, 1 represents run) and operating frequency of all circulating pumps (discretized within the allowable range, e.g., step size of 1Hz). The total slurry output flow that the system can stably provide under each state is pre-calculated or calibrated. For example, state S1 is described as "Pump A runs @35Hz, Pump B runs @40Hz, Pump C stops," with a corresponding total slurry output of 2800m³. 3 / h, these "situation-slurry output" corresponding relationships are stored as a lookup table or database, then the predicted boiler load F t is converted to the required theoretical slurry amount for the desulfurization system under this load through a preset conversion model or empirical curve (for example, a "boiler load-required slurry amount" relationship curve fitted according to historical data, which also takes into account factors such as inlet SO2 concentration, for example, when the load is 80-100 MW and the SO2 concentration is 1000-1500 mg / Nm 3 , the required slurry amount is Q req = 0.025xF t + 500, unit: m 3 / h), for example, if F t = 96.52 MW, the required slurry amount Q req = 0.025x96.52 + 500 = 2.413 + 500 = 2913 m 3 / h is calculated according to the conversion model, then the calculated required slurry amount Q req (2913 m 3 / h) is compared with the pre-labeled slurry output of each situation in the combined situation set one by one, and all operating situations that can provide not less than Q req and as close as possible (for example, not more than 110% of Q req , this 110% is a preset upper limit coefficient, which is designed to avoid excessive supply, and this coefficient is set according to historical operation data analysis and energy saving goals, such as by analyzing the energy consumption and desulfurization effect under different redundancies, it is found that 10% redundancy is the economic and effective balance point) are screened to form a candidate situation subset, and this matching process is performed for each future predicted time point to finally obtain the future working condition demand and operating situation mapping set.

[0111] The steps for obtaining the energy transfer cost table are as follows:

[0112] Based on the future working condition demand and operating situation mapping set, the running frequency values of all circulating pumps in each operating situation are extracted in turn, the frequencies of the corresponding circulating pumps under any two different operating situations are paired, and the target running time of the circulating pumps after the situation transfer is simultaneously extracted to generate a running frequency transfer and time length pairing sequence;

[0113] According to the running frequency transfer and time length pairing sequence, the total energy transfer value is calculated, and the calculation formula is:

[0114]

[0115] Where E i,j is the total energy transfer value corresponding to the transition from the i-th operating situation to the j-th operating situation, s is the total number of circulating pumps in the operating situation, and ZCu The power proportionality coefficient of the u-th circulating pump (unit: kW / Hz) 3 ), φ i,u Let φ be the operating frequency (in Hz) of the u-th circulating pump in operating state i. j,u Let λ be the operating frequency (in Hz) of the u-th circulating pump in operating state j. u The duration (in hours) during which the u-th circulating pump operates continuously in operating state j;

[0116] Based on the total energy consumption transfer value, the energy consumption changes between each pair of operating states are registered item by item. A matrix structure is constructed according to the operating state number, and the energy consumption differences of all transfer paths between states are output in a row and column manner to generate an energy consumption transfer cost table.

[0117] Specifically, based on the future operating condition demand and operational status mapping set generated in the previous steps, this mapping set records in detail the boiler load corresponding to each predicted future time point, the calculated required slurry volume, and one or more candidate operational statuses that can meet the slurry demand, along with their specific configurations (including the start / stop status and operating frequency of each circulating pump). First, all unique operational statuses are extracted from this mapping set, and each unique status is assigned an identifier, such as S1, S2, etc. Then, for each unique operational status, such as status Sk, its definition is parsed to obtain the specific operating frequency value φ of each circulating pump (e.g., s pumps numbered u = 1, 2, ..., s) under that status. k,u If a pump is in a stopped state in situation Sk, its operating frequency is recorded as 0Hz. Subsequently, the system systematically considers the transition between any two different operating states, that is, the transition from the initial state Si to the target state Sj (where i is not equal to j). For each such state transition (Si→Sj), the operating frequency φ of each circulating pump u in the initial state Si is extracted. i,u The operating frequency φ of the circulating pump u corresponding to the target situation Sj j,u At the same time, it is necessary to determine the target runtime λ of each circulating pump u running in Sj after the system transitions to the target state Sj. u This duration is typically determined by the system's optimized scheduling cycle or predicted time step. For example, if the system's scheduling instructions are updated every 15 minutes, then λ u The time frame is typically set to 0.25 hours. For a pump that stops in the target state Sj, the corresponding λ is... u This can be interpreted as 0, or as it "running" at 0Hz for 0.25 hours. However, due to subsequent calculations of energy transfer, if φ j,u =0 and φ i,u ≠0, the frequency difference still exists, therefore λ uThe planned duration of the situation Sj shall be unified for each pair (starting situation Si, target situation Sj) and combined with the starting frequency φ i,u of each circulating pump u in the starting situation Si j,u and the target frequency φ Sj of each circulating pump u in the target situation Sj u to form a record unit. All record units of all pumps in all possible situation pairs are finally collected to generate a sequence of frequency shift and duration pairs.

[0118] The formula is: The benefit of the formula is that it quantifies the energy-related cost caused by the circulating pump frequency adjustment (including start / stop) when switching from one operating situation i to another operating situation j. Based on the similarity theory of fluid machinery, i.e. the power consumption of a pump is approximately proportional to the cube of its rotational speed (frequency), the non-linear characteristic of the power impact of frequency change is captured by , and the use of absolute value ensures that both frequency increase and decrease (or start / stop switching) are considered as a "shift behavior" that needs to consume energy or cause disturbance, thus the cost is taken into account. In addition, the power proportionality coefficient ZC u of each pump is introduced in the formula, making the calculation adaptable to circulating pumps of different specifications and efficiencies, enhancing the accuracy and universality of the model. At the same time, by multiplying the duration of continuous operation of the pump in the target situation λ u , the impact of instantaneous power change is accumulated as an energy value within a period of time, making it more practical and directly reflecting the energy transfer cost within a specific scheduling period.

[0119] Parameter introduction:

[0120] s represents the total number of circulating pumps participating in scheduling. It is a fixed integer value determined by the actual hardware configuration of the desulfurization system. For example, a desulfurization system is configured with 3 adjustable frequency circulating pumps, then s = 3.

[0121] ZC u is the power proportionality coefficient of the u-th circulating pump, with the unit of kW / Hz 3 (thousand watts per hertz cube). This coefficient represents the proportional relationship between the input power of the u-th pump and the cube of its operating frequency, reflecting the energy consumption characteristics of the pump. Its acquisition method is: consult the pump's equipment manual to obtain the power-frequency characteristic curve data, or measure and record the actual input power P k and the corresponding frequency f k of the pump at multiple different stable operating frequency points (e.g. at least 5 points covering its commonly used operating range such as 30Hz, 35Hz, 40Hz, 45Hz, 50Hz, etc.) in actual operation.Then, the least squares method is used to analyze the data points (f) k ,P k To fit the data, solve for P = ZC. u ·f 3 +P fixed ZC in u Value (where P) fixed For fixed power losses, which can be ignored if small or determined by fitting, for example, testing of circulation pump No. 1 showed power of 30kW at 30Hz, 48kW at 35Hz, 72kW at 40Hz, 103kW at 45Hz, and 145kW at 50Hz. For example, P... fixed =0, calculate P / f at each point. 3 30 / 30 3 ≈0.001111, 48 / 35 3 ≈0.001119, 72 / 40 3 =0.001125, 103 / 45 3 ≈0.001128, 145 / 50 3 =0.00116. Taking the average value or through regression analysis, ZC1 = 0.001128 kW / Hz. 3 .

[0122] φ i,u This represents the operating frequency of the u-th circulating pump in the initial operating state i, expressed in Hertz (Hz). This value is directly extracted from the "operating frequency transition and duration pairing sequence" generated in the previous step for a specific state i and pump u. If pump u is in a stopped state in state i, then φ i,u = 0Hz. For example, in situation i, the operating frequency of circulation pump 1 is recorded as 38Hz.

[0123] φ j,u This represents the operating frequency of the u-th circulating pump in the target operating situation j, expressed in Hertz (Hz). Similarly, this value is directly extracted from the "operating frequency transition and duration pairing sequence" for a specific situation j and pump u. If pump u is to stop operating in situation j, then φ... j,u =0Hz. For example, in situation j, the operating frequency of circulation pump 1 is planned to be adjusted to 42Hz.

[0124] λ u λ represents the expected duration, in hours (h), for the u-th circulating pump to operate continuously under the target operating condition j. This duration is typically consistent with the time step of the system's optimal scheduling or the time granularity of future operating condition predictions. For example, if the system makes an optimization decision every 30 minutes, then λ... uTypically taken 0.5 hours. This parameter is also taken from the "run frequency shift and time length pairing sequence" and is generally the same for all pumps participating in the run in situation j, i.e. the planned duration of situation j. For example, situation j is planned for 0.5 hours.

[0125] Calculation:

[0126] The system has s = 2 circulating pumps.

[0127] Power scaling factor for pump 1 : ZC1 = 0.001128 kW / Hz 3 .

[0128] Power scaling factor for pump 2 : ZC2 = 0.001250 kW / Hz 3 .

[0129] Frequency configuration of the starting run situation i:

[0130] Pump 1 run frequency φ i,1 = 38 Hz;

[0131] Pump 2 run frequency φ i,2 = 32 Hz;

[0132] Frequency configuration of the target run situation j:

[0133] Pump 1 run frequency φ j,1 = 42 Hz;

[0134] Pump 2 run frequency φ j,2 = 0 Hz (i.e. pump 2 is stopped in situation j);

[0135] The planned duration of the target run situation j is λ = 0.5 h, so λ1 = 0.5 h and λ2 = 0.5 h.

[0136] Calculate the energy consumption shift value contribution E i,j,1 for pump 1 :

[0137]

[0138] E i,j,1 = 0.001128 · |42 3 - 38 3 | · 0.5;

[0139] E i,j,1 = 0.001128 · |74088 - 54872 | · 0.5;

[0140] E i,j,1 = 0.001128 · |19216 | · 0.5;

[0141] E i,j,1= 0.001128 · 19216 · 0.5;

[0142] E i,j,1 = 21.675648 · 0.5;

[0143] E i,j,1 = 10.837824 kWh;

[0144] The energy transfer value contribution E i,j,2 for pump 2 is calculated

[0145]

[0146] E i,j,2 = 0.001250 · |0 3 -32 3 | · 0.5;

[0147] E i,j,2 = 0.001250 · |0 - 32768| · 0.5;

[0148] E i,j,2 = 0.001250 · |-32768| · 0.5;

[0149] E i,j,2 = 0.001250 · 32768 · 0.5;

[0150] E i,j,2 = 40.96 · 0.5;

[0151] E i,j,2 = 20.48 kWh;

[0152] The total energy transfer value E i,j from situation i to situation j is calculated

[0153] E i,j = E i,j,1 + E i,j,2 ;

[0154] E i,j = 10.837824 + 20.48;

[0155] E i,j = 31.317824 kWh;

[0156] This result shows that when the combination of circulating pumps is adjusted from operating situation i (pump 1 @ 38 Hz, pump 2 @ 32 Hz) to operating situation j (pump 1 @ 42 Hz, pump 2 stopped) and situation j is planned to last for 0.5 hours, then the total energy transfer value associated with this frequency adjustment action is 31.317824 kWh.

[0157] Based on each pair of operational states calculated in the previous step (e.g., from state S...), k To the situation S l The total energy transfer value E between ) k,l First, a list containing all unique operational states is compiled. These states have usually been identified and numbered in previous steps. Let there be N unique operational states, numbered S1, S2, ..., S... N Next, an N×N two-dimensional matrix is ​​initialized, where the row and column indices of the matrix correspond to the numbers of these operational states. Each element M in the matrix... k,l Used to store the running status S k Migration to operational status S l Total energy transfer value E k,l Then, iterate through all possible ordered state pairs (S) k ,S l ), where k and l both range from 1 to N. For each situation pair, the corresponding E is found from the calculated result set. k,l (If k≠l), or for diagonal elements (i.e., k=l, indicating that the state is maintained from itself without migration), their energy transfer value E k,k By definition, it is usually set to 0 because there is no frequency change, and the queried or set E k,l The numerical value is filled into the position of the k-th row and l-th column of the matrix. For example, if the E value from situation S1 to S2 is calculated... 1,2 =31.317824kWh, then fill in this value in the 1st row and 2nd column of the matrix, and repeat for all N values. 2 The filling of each element results in a fully filled N×N matrix, which is the energy transfer cost table.

[0158] The steps for obtaining the comprehensive cost of the operational status transition path are as follows:

[0159] Based on the start-stop records of the operation status, compare the operation status identifier of each device before and after each group of operation status migration, identify the devices whose status changes from running to stopping or from stopping to running, record each device with a status change as a start-stop action, and extract the runtime of the most recent operation cycle to generate a list of start-stop devices and their runtime before start-stop.

[0160] Based on the list of start-up and shutdown equipment and their pre-start-up / shutdown running time, and the energy consumption transfer cost table, calculate the operational status migration path cost using the following formula:

[0161]

[0162] Among them, R i,j E represents the migration path cost from the i-th operational state to the j-th operational state.i,j is the total energy consumption shift value corresponding to the running situation, x is the total number of devices that start and stop during the migration process, ω d is the start-stop cost coefficient of the dth device that starts and stops (unit: kWh), τ d is the time length of the dth device continuously running before this migration (unit: h), represents the equivalent wear intensity of a single start-stop of the device after experiencing the running cycle;

[0163] Based on the migration path cost, the migration path costs between all pairs of running situations are sequentially filled into the cost table structure matrix, and are arranged according to the starting situation and target situation indexes, to generate a comprehensive migration path cost of the running situation.

[0164] Specifically, based on the start-stop records of the running situation, which are usually derived from the detailed logs of the system for each situation change and contain the running (1) or stopping (0) state identification of each circulating pump under each situation, for each group of running situation migration, for example, from the starting situation S i to the target situation S j , the system will compare the running state identification of each circulating pump (device number d = 1, 2, …, s) in S i and S j one by one. If the state of a pump d in S i is 1 (running) and the state in S j is 0 (stopping), or in S i is 0 (stopping) and in S j is 1 (running), it is determined that the device d has undergone a start-stop action, and the number of the device is recorded. At the same time, for each device d identified as having undergone a start-stop action, the time length τ d of the most recent continuous uninterrupted running of the device before the start-stop action occurs needs to be queried and extracted from its historical running data (usually maintained by the DCS or device management system). The time length is in hours. If the device is started from a stopped state, its “running time before start-stop” can be understood as 0. All identified start-stop devices and their corresponding τ d values are organized into a list, and each element in the list is a two-tuple (device number d, τ d ), forming a start-stop device and its start-stop running time pair list.

[0165] Formula: The benefit of the formula is that it comprehensively considers two different types of costs to evaluate the total cost of running situation migration, i.e., the energy consumption shift cost E i,jAnd the equivalent wear and tear costs caused by equipment start-up and shutdown operations. This comprehensive assessment allows for more holistic decision-making, not only pursuing the lowest short-term energy consumption but also considering the long-term operational reliability and maintenance costs of the equipment. The wear and tear costs associated with start-up and shutdown are also considered. The duration τ of continuous operation of the equipment before start-up and shutdown was introduced. d As an influencing factor, through the logarithmic function To quantify wear intensity, frequent start-ups and shutdowns, or start-ups and shutdowns after long periods of continuous operation, cause greater impact and wear on the equipment. Wear intensity increases with τ. d The increase is accompanied by an increase in [something], but the growth rate gradually slows down, which is consistent with the general pattern of equipment aging and fatigue accumulation, and also avoids [something]. d Excessive wear and tear leads to an infinitely linear increase in costs. This can be addressed by setting an independent start-stop cost coefficient ω for each device. d Different weights can be assigned to different equipment based on factors such as their importance, purchase cost, and ease of maintenance, making the calculation of wear and tear costs more targeted and reasonable.

[0166] Parameter description:

[0167] E i,j This is the total energy consumption transfer value corresponding to the migration from operating state i to the j-th operating state, in kWh. This parameter is directly obtained from the "Energy Consumption Transfer Cost Table" calculated in the previous step, based on the indices of the initial state i and the target state j. For example, in the aforementioned example, the calculated E i,j = 31.317824 kWh.

[0168] x represents the total number of devices (circulating pumps) that start or stop during the transition from situation i to situation j. This value is obtained from the previous step of "generating a list of devices that start or stop and their running time before starting or stopping". For example, if a situation transition results in one pump starting and one pump stopping, then x = 2.

[0169] ω d This is the start-stop cost coefficient for the d-th device that initiates a start-stop operation, expressed in kWh. This coefficient represents the cost of a single "standard" start-stop operation of this device (i.e., ...). The equivalent energy loss or maintenance cost conversion corresponding to a certain benchmark value) is converted. The setting of this coefficient includes: the purchase cost of the equipment, the expected service life, the mean time between failures (MTBF), the average cost of a single maintenance (including spare parts and labor), the known impact of start-stop on key components of the equipment (such as motor winding, bearing, seal), etc. The way to obtain it is: first, collect the historical failure data and maintenance records of the specific type of circulating pump, analyze the correlation between the number of start-stop and the failure rate, maintenance cost. Second, refer to the guidance provided by the equipment manufacturer on start-stop limits and life impact. Third, expert experience evaluation can be combined, for example, for a key circulating pump with a value of 100,000 yuan and an expected upper limit of 10,000 times of start-stop, if the cumulative depreciation caused by start-stop is calculated as 20% of the purchase cost, then the base cost of a single start-stop can be estimated as 100,000 yuan x 20% / 10,000 times = 2 yuan / time. Then convert this monetary cost into equivalent energy through the local average industrial electricity price (for example, 0.6 yuan / kWh): 2 yuan / 0.6 yuan / kWh ≈ 3.33 kWh. This value can be used as a benchmark for ω d , and adjusted according to the importance and maintenance sensitivity of the equipment. For example, for pump No. 1 (key pump), set ω1 = 4.0 kWh; for pump No. 2 (general pump), set ω2 = 2.5 kWh.

[0170] τ d is the time length of the last continuous operation of the dth start-stop device before the current migration (start-stop) action occurs, with the unit being hours (h). This data is extracted from the "start-stop device and its running time before start-stop list" generated in the previous step. For example, if pump No. 1 has been continuously running for 150.5 hours before being stopped this time, then τ1 = 150.5 h.

[0171] represents the equivalent wear intensity of a single start-stop operation of the device after it has experienced a running period of τ d . This is a dimensionless amplification factor that converts the base start-stop cost coefficient ω d according to the recent "labor" level of the device. The use of natural logarithm makes the wear intensity increase with the increase of τ d , but the growth rate decreases, for example, if τ d = 150.5

[0172] h, then ln(1+150.5 2 ) = ln(1+22650.25) = ln(22651.25) ≈ 10.028.

[0173] Calculation process:

[0174] Consider migrating from running situation i to running situation j.

[0175] From the "Energy Transfer Cost Table" we find E i,j = 31.317824 kWh.

[0176] In this migration, for example, x = 2 devices have been started and stopped:

[0177] Device 1 (Pump 1): A stop action has occurred. Its start-stop cost coefficient ω1= 4.0 kWh. Before this stop, it had been running continuously for τ1= 150.5 h.

[0178] Device 2 (Pump 2): A start action has occurred. Its start-stop cost coefficient ω2= 2.5 kWh. For a device that has been stopped to running, its "previous continuous running duration" is 0. So, τ2= 0 h.

[0179] Calculate the start-stop wear cost M1 of Device 1:

[0180]

[0181] M1= 4.0 · ln(1 + 150.5 2 ) ;

[0182] M1= 4.0 · ln(1 + 22650.25) ;

[0183] M1= 4.0 · ln(22651.25) ;

[0184] M1= 4.0 · 10.02806 ;

[0185] M1= 40.11224 kWh ;

[0186] Calculate the start-stop wear cost M2 of Device 2:

[0187]

[0188] M2= 2.5 · ln(1 + 0 2 ) ;

[0189] M2= 2.5 · ln(1) ;

[0190] M2= 2.5 · 0 ;

[0191] M2= 0 kWh ;

[0192] Calculate the total start-stop wear cost ∑M d :

[0193]

[0194] Calculate the total running situation migration path cost R i,j :

[0195]

[0196] R i,j =31.317824 + 40.11224;

[0197] R i,j = 71.430064 kWh;

[0198] The results show that the overall cost of migrating from operating state i to operating state j is 71.430064 kWh. This value combines the energy consumption impact of frequency adjustment (31.317824 kWh) and the equivalent wear cost (40.11224 kWh) caused by equipment start-up and shutdown (mainly the shutdown of pump 1 after long-term operation). The higher this overall cost, the greater the total "cost" of this state migration. Therefore, when optimizing the path, it is advisable to avoid choosing paths with high Rw. i,j Value migration steps.

[0199] Based on all operational situation pairs calculated in the previous step (e.g., from situation S...), k To the situation S l The migration path cost R between ) k,l First, it is necessary to reconfirm all the unique operating states that have been identified and numbered in the previous steps. For example, there are still N states, numbered S1, S2, ..., S... N Next, a new N×N two-dimensional matrix is ​​created. This matrix is ​​specifically used to store the integrated migration path costs, and its row and column indices also correspond to the numbers of these N operational states. Each element C in the matrix... k,l Stored from running status S k Migration to operational status S l The overall cost R k,l Then, iterate through all possible ordered state pairs (S) k ,S l ), k, l∈[1,N], for each situation pair, from the result set calculated by the previous formula (which contains all R k,l (Value) to find the corresponding R k,l Numerical value: If k = l, meaning the situation remains unchanged, then the migration path cost R k,k It is usually set to 0 because there is neither energy transfer nor start-stop wear (or E). k,k =0 and the start / stop wear item is also 0), find or set R k,l The value is accurately filled into the k-th row and l-th column of the cost table structure matrix. For example, if the comprehensive migration path cost R from situation S1 to S2 is calculated... 1,2 =71.430064kWh, then enter this value in the first row, second column of the cost table, and so on for all N items.2 The assignment of values ​​to each element results in a fully filled N×N matrix, which is the comprehensive cost table for the operational state migration path. It clearly categorizes and arranges the comprehensive costs of all direct migration steps according to the indices of the initial state (row) and the target state (column).

[0200] The steps to obtain the set of feasible states for the first node are as follows:

[0201] Based on the mapping set of future working conditions and operating status, the operating frequency and start / stop status of all circulating pumps in each operating status of the mapping set are retrieved one by one. The slurry output flow rate corresponding to each circulating pump is calculated one by one, and the slurry output flow rates of all circulating pumps are summed to form the total slurry output flow rate of each operating status.

[0202] Based on the total slurry output flow rate of each operating condition, the slurry quantity requirement value of the boiler load corresponding to the first time point in the mapping set of future operating conditions and operating conditions is extracted. The total slurry output flow rate of each operating condition is compared with the requirement value one by one. All operating conditions with a total slurry output flow rate equal to or higher than the requirement value are selected to form a preliminary set of operating conditions that meet the slurry requirement of the boiler load at the first time point.

[0203] Based on the initial set of operating conditions that meet the slurry demand of the boiler load at the first time point, the start-stop status and frequency of each circulating pump within the operating condition are verified one by one to see if they are within the safe operating range allowed by the equipment. Operating conditions that exceed the safe operating range in terms of start-stop status or frequency are eliminated to form the set of feasible conditions for the first node.

[0204] Specifically, based on the future operating condition demand and operational status mapping set obtained from the previous steps, this mapping set includes the boiler load, required slurry volume, and candidate operational statuses that meet the conditions at each predicted time point, along with their detailed configurations. First, it iterates through each unique operational status recorded in this mapping set. For each operational status, such as status S... k Read the operating frequency φ of each circulating pump defined therein (e.g., there are s pumps in total, numbered u = 1, 2, ..., s). k,u And start / stop status (1 for running, 0 for stopped). Then, based on the characteristic curve of each circulating pump u (this curve is usually provided by the pump manufacturer or obtained through field calibration, describing the pump's output flow rate at different operating frequencies, for example...), Where a u ,b u ,c u It is a specific coefficient of pump u. If pump u is in state S k If the pump is in a stopped state, its slurry output flow rate is 0. Calculate the pump's current state S. k The slurry output flow rate Q k,uThen, the situation S k The calculated slurry output flow rate Q k,u is accumulated to obtain the running situation S k The total slurry output flow rate of the system This calculation is performed for all unique running situations in the mapping set, and finally each running situation is marked with its total slurry output flow rate.

[0205] Based on the total slurry output flow rate Q total,k calculated for each running situation in the previous step, the first predicted time point (e.g., t = 1) is located from the future working condition requirement and the running situation mapping set, and the theoretical required slurry amount demand value corresponding to the predicted boiler load at this time point is extracted from the record, denoted as Q demand,t＝1 , for example, Q demand,t＝1

[0206] = 2913 m 3 / h, then, the total slurry output flow rate Q total,k of all unique running situations (e.g., a total of N situations, S1, S2, …, S N ) in the system (where k = 1, …, N) is compared with the slurry amount demand value Q demand,t＝1 at this first time point, and the selection condition is that the total slurry output flow rate Q total,k of the running situation must be equal to or slightly higher than the demand value Q demand,t＝1 , and the specific standard can be set as Q total,k ≥ Q demand,t＝1 and Q total,k ≤ α · Q demand,t＝1 , where α is a coefficient slightly larger than 1, for example, 1.1, representing the maximum allowed redundancy, and this coefficient is set according to actual running experience and energy saving requirements, for example, if historical data shows that a 10% flow redundancy can effectively respond to short-term fluctuations and the energy consumption increase is within an acceptable range, then α = 1.1, and all running situations that meet this condition (for example, if the total slurry output of situation S5 is 2950 m 3 / h, and 2950 ≥ 2913 and 2950 ≤ 1.1 × 2913 = 3204.3, then S5 is selected) are collected to form a preliminary running situation set that meets the slurry demand of the boiler load at the first time point.

[0207] Based on the preliminary running situation set that meets the slurry demand of the boiler load at the first time point obtained in the previous step, the safety check for each running situation in this set needs to be performed, and for any running situation in the preliminary set, for example, situation S p , the start-stop state and running frequency φ of each circulating pump u defined in it are checked one by onep,u The verification of the start-stop state is mainly to ensure that the minimum start-stop time interval requirement is met, for example, if a pump has just been stopped for less than 5 minutes (the 5 minutes is the minimum shutdown restart interval recommended by the equipment manufacturer or a protective threshold set according to historical operation experience, aiming to avoid motor overheating or mechanical impact), the situation S p will be considered unsafe, and for the verification of the operating frequency, it is necessary to ensure that φ p,u is strictly within the allowed safe operating frequency interval in the pump equipment manual or field calibration, for example, the allowed operating frequency range of a certain pump is 25 Hz to 50 Hz, if the frequency of the pump in situation S p is set to 20 Hz or 55 Hz, then the situation is unsafe, in addition, it is also necessary to check whether there are interlocking conditions that are not met, for example, the start of some pumps depends on the fact that other specific pumps are already in operation, or the total number of operating pumps does not exceed the system power supply or pipeline limit, all operating situations in the initial operating situation set that have any circulating pump whose start-stop state (such as violating the minimum start-stop interval) or operating frequency (such as exceeding the safe upper and lower limits) does not meet the pre-set safety specifications are excluded from the set, after this round of safety filtering, the final feasible situation set of the first node (i.e. the first predicted time point) is obtained.

[0208] The steps for obtaining the feasible migration network between multiple time nodes are as follows:

[0209] Based on the feasible situation set of the first node, the operating frequency, start-stop state and slurry output flow of all circulating pumps corresponding to each operating situation in the set are extracted one by one, and all estimated operating situations that can meet the slurry quantity demand value corresponding to the boiler load demand at the second time point are matched in turn for each operating situation to form a feasible situation connection set from the first node to the second time point;

[0210] Based on the feasible situation connection set from the first node to the second time point, the operating frequency, start-stop state and slurry output flow of all circulating pumps of all operating situations at the second time point in the connection set are retrieved and extracted in turn, and it is judged whether each situation can meet the slurry demand value corresponding to the boiler load demand at the third time point under the constraints of circulating pump frequency adjustment amplitude and safe frequency interval, and the connection situations that meet the conditions are selected to form a feasible situation connection set from the second node to the third node;

[0211] Based on the feasible situation connection set from the second node to the third node, the operating situation connection and constraint verification are performed in turn to all subsequent remaining time nodes, and the feasible situation connections of all time nodes are connected in series to form a feasible migration network between multiple time nodes.

[0212] Specifically, based on the feasible scenario set of the first node (i.e. the first predicted time point) determined in the previous step, the system will run each scenario in the set (e.g. scenario S 1,k , where the first subscript represents time node 1 and the second subscript k represents the kth feasible scenario under the time node) is processed, first, for each S 1,k , its detailed configuration information is extracted, including the current operating frequency, start-stop state of all circulating pumps and the total slurry output flow under the scenario, which has been calculated or explicitly in the previous step, and at the same time, the slurry demand value corresponding to the predicted boiler load at the second time point (t = 2) is obtained from the future operating condition demand and operating scenario mapping set, denoted as Q demand,t＝2 , then, for the currently processed S 1,k , the system will traverse all pre-defined or dynamically generated potential target operating scenarios (these target scenarios are for the second time point, which can be referred to as estimated operating scenarios S 2,m ), for each estimated operating scenario S 2,m , its total slurry output flow Q is calculated or looked up, and it is determined whether the flow meets the demand at the second time point, i.e. and (where the setting of a is the same as before, e.g. 1.1), if it meets, it is considered that there is a potential effective connection from S 1,k to S 2,m , and this pair of connections (S 1,k → S 2,m ) is recorded, and this matching process is performed for all S 1,k in the feasible scenario set of the first node, and finally all successful connection pairs are summarized to form the feasible scenario connection set of the first node to the second time point.

[0213] Based on the feasible scenario connection set of the first node to the second time point generated in the previous step, which contains all pairs of feasible scenarios S 1,k successfully connected from the first time point to the feasible scenario S 2,m at the second time point, now the system will process all operating scenarios S 2,m at the second time point in this connection set in turn, for each S 2,m , its complete circulating pump operating frequency, start-stop state and total slurry output flow information is extracted, and at the same time, the slurry demand value corresponding to the boiler load demand at the third time point (t = 3) is obtained from the future operating condition demand and operating scenario mapping set, denoted as Q demand,t＝3 , then, for the current S 2,m, the system needs to consider all possible next operating scenarios (i.e. the estimated operating scenario S 3,n at the third time point) and judge each S 3,n : first, check whether the frequency adjustment range of each circulating pump is within the allowed range when converting from S 2,m to S 3,n , for example, the upper limit of single frequency adjustment is 5 Hz (this value is set according to the characteristics of the equipment and the process requirements, for example, to protect the frequency converter and motor and avoid excessive current impact, the manufacturer or operating regulations stipulate that the frequency change between adjacent scheduling periods should not exceed 5 Hz), i.e. for each pump u, Hz, secondly, confirm that the operating frequency of all pumps in S 3,n is within their respective safe frequency range (for example, 25 Hz to 50 Hz), thirdly, ensure that the total slurry output flow rate of S 3,n meets the demand at the third time point, i.e. and Finally, it also checks the start-stop constraints, such as whether the minimum running or stopping time meets the requirements (for example, the pump has not been running for less than 10 minutes, which is an empirical value to ensure that the pump is fully effective or to avoid frequent switching), only when all these constraints are met, it is considered that the connection from S 2,m to S 3,n is feasible, record this pair of connections (S 2,m → S 3,n ), perform this screening process for all scenarios S 2,m at the second time point, and collect all connections that meet the conditions to form a feasible scenario connection set from the second time point to the third time point.

[0214] Based on the feasible scenario connection set from the second time point to the third time point constructed in the previous step, this process will continue to iterate forward, covering all subsequent predicted time points until the last time point (for example, if a total of T time points are predicted, then proceed from the T-1 time point to the T time point), in each iteration, for example, when constructing connections from time point t to time point t+1, the system will use the known feasible scenario set at time point t (these scenarios are obtained by t-1 step connection screening and constraint verification), and for each feasible scenario S t,p at time point t, extract its configuration information, then obtain the slurry demand Q demand,t+1 at time point t+1, and then traverse all potential estimated operating scenarios S t+1,q at time point t+1, and perform comprehensive constraint verification for each pair (S t,p → S t+1,q ), these constraints include: S t+1,qwhether the total slurry output flow of the system meets Q demand,t+1 whether the upper and lower limit requirements of Q t,p are met, S t+1,q whether the pump frequency adjustment amplitude of each pump is within the maximum single-step adjustment value (e.g., 5 Hz) allowed, S t+1,q whether the operating frequency of each pump in S is within its safe operating range, and whether all start-stop operations meet the minimum running / stop time interval requirements for device protection and process stability, only when all these dynamic (related to the previous situation) and static (the situation itself) constraints are met, the connection (S t,p → S t+1,q ) is considered valid and is added to the feasible situation connection set from time point t to time point t+1. Through this way of forward recursion, screening and connection of each time node, all feasible situation connections between all time points are finally connected in series to form a network structure composed of nodes (feasible operating situations) and directed edges (feasible transitions), i.e., a multi-time node feasible transition network, covering the entire prediction time domain.

[0215] The acquisition step of the full-quantitative path alternative list is:

[0216] Based on the multi-time node feasible transition network, the current operating situation in the current time node is selected as the initial starting node, all operating situation nodes connected to the next time node under the initial starting node are extracted one by one, the operating frequency, start-stop state and slurry output flow of all circulating pumps contained in each situation node are recorded in turn, and the connection relationship between each node is marked to form a situation path set from the initial starting node to the second time node;

[0217] Based on the situation path set from the initial starting node to the second time node, the operating frequency, start-stop state and slurry output flow of all circulating pumps corresponding to each operating situation node in the second time node are retrieved in turn, all feasible connection operating situations are matched to the third time node, and the connection relationship between each pair of nodes is recorded, and the node connection chain set from the initial starting node to the end time node is formed by progressing one by one to the end time node;

[0218] Based on the node connection chain set from the initial starting node to the end time node, all node connection chains are traversed one by one from the initial starting node, and a continuous and complete path sequence of multiple time nodes is formed in series, the connection chains with discontinuity or unable to continue to the end time node are eliminated, and a full-quantitative path alternative list is formed.

[0219] Specifically, based on the multi-time node feasible migration network constructed in the previous step, which stores all the running states (nodes) and all feasible migration paths (directed edges) between them at all prediction time points that meet the constraint conditions in the form of a graph, first, the current actual running circulating pump combination state (including the start-stop and frequency of each pump) is mapped to a specific running state in the first time node (i.e., t = 1) in the network, which serves as the initial starting node for path search. Then, starting from the initial starting node, the feasible migration network is queried to find all running state nodes directly connected to the next time node (t = 2). For each such next node, the system will record in detail the running frequency, start-stop state of all circulating pumps, and total slurry output flow under this state, which are part of the network node attributes. At the same time, the connection relationship between the initial starting node and this next node is explicitly marked, for example, if the initial node is S 1,current , the states connected to the second time node are S 2,a , S 2,b , and S 2,c , then three preliminary path segments will be generated: (S 1,current → S 2,a ), (S 1,current → S 2,b ), and (S 1,current → S 2,c ), and these path segments and the node details (frequency, start-stop, flow) they contain will be stored to form the state path set from the initial starting node to the second time node.

[0220] Based on the state path set from the initial starting node to the second time node obtained in the previous step, each path segment in this set ends with a running state at the second time node. The system will expand these path segments, processing each end node of the path segment in turn. For example, for the path segment (S 1,current → S 2,a ), the current end node is S 2,a , the system will retrieve the corresponding running frequency, start-stop state, and slurry output flow of all circulating pumps for S 2,a . Then, from the multi-time node feasible migration network, all feasible running state nodes (e.g., S 3,x , S 3,y ) connected from S 2,a to the third time node (t = 3) are found. For each such feasible connection (e.g., S 2,a → S 3,x and S 2,a → S 3,y ), the original path segment is expanded to form a new longer path segment, such as (S 1,current → S2,a → S 3,x ) and (S 1,current → S 2,a → S 3,y ), and record the new connection relationship and the detailed information of the new node. This process is recursively (or iteratively) performed, i.e. for all the currently constructed path segments, the end nodes are processed to find and connect to the next possible situation of the time node, until all path segments are extended to the predetermined end time node (for example, if the total prediction time length is T time units, then it is extended to the Tth time node), and each step of extension is strictly based on the pre-verified connection relationship in the feasible migration network. Finally, a chain set consisting of a series of ordered connected running situation nodes from the initial starting node to the end time node is obtained, which is the node connection chain set from the initial starting node to the end time node.

[0221] Based on the node connection chain set from the initial starting node to the end time node generated in the previous step, each chain in this set represents a potential scheduling path from the current time (initial starting node) to the end time node, but these chains are not complete and effective, therefore, a final screening and sorting is needed. Each node connection chain in the set is traversed from the initial starting node, for each chain, it is checked whether it really constitutes a complete path sequence continuously and uninterruptedly extending from the initial time node to the end time node, which means that each intermediate node in the chain must have an effective connection to the next node, and the last node of the chain must be located at the predetermined end time node. If it is found that a chain is broken at a certain time point (i.e. the current node has no outgoing edge pointing to the next time node in the feasible migration network), or the chain fails to continue to the end time node (for example, it only extends to the second last time point and cannot continue), such chain will be regarded as an invalid path and removed. After this screening, all the remaining chains are complete and continuous from beginning to end, and each node records the detailed information of the running situation at this time point (such as pump frequency, start-stop state, total slurry flow, etc.). These verified complete path sequences together constitute the full-quantitative path candidate list.

[0222] The acquisition step of the optimal scheduling instruction path of the circulating pump combination is as follows:

[0223] Based on the full-quantization path candidate list, each complete path sequence in the list is retrieved one by one, and the running state transition path comprehensive cost corresponding to each pair of adjacent running state nodes between the starting node and the end node of the path is extracted step by step, the running state transition path comprehensive cost between each pair of nodes is recorded one by one, and the running state transition path comprehensive cost of all adjacent node pairs is accumulated to form the accumulated running state transition path comprehensive cost corresponding to each path;

[0224] Based on the accumulated running state transition path comprehensive cost corresponding to each path, the accumulated running state transition path comprehensive costs of all complete path sequences are sorted from low to high to form a path accumulated path comprehensive cost sorting list;

[0225] Based on the path accumulated path comprehensive cost sorting list, the complete path sequence with the lowest running state transition path comprehensive cost in the sorting list is selected, and the combined optimal dispatching instruction path of the circulating pump is formed according to the circulating pump running frequency and start-stop state recorded by each node in the path sequence.

[0226] Specifically, based on the full-quantization path candidate list generated in the previous step, the list contains multiple complete and continuous running state path sequences starting from the current running state to the end time node, and the system will evaluate the cost of each complete path sequence in the list. Specifically, for any selected path sequence, for example, path P k =(S 1,a →S 2,b →S 3,c →…→S T,z ), the system will start from the starting node S 1,a of the path, and extract each pair of adjacent running state nodes in turn, for example, the first pair is (S 1,a ,S 2,b ), the second pair is (S 2,b ,S 3,c ), and so on, until the last pair (S T-1,y ,S T,z ). For each pair of adjacent nodes (S t,curr ,S t+1,next ), the system will call the previously calculated and stored "running state transition path comprehensive cost table" to find the running state transition path comprehensive cost R curr,next corresponding to the transition from state S t,curr to state S t+1,next . The cost R curr,next has integrated the energy consumption transition and device start-stop wear costs. The comprehensive cost R curr,next between each pair of adjacent nodes is recorded one by one, and then the path P kThe running state transition path comprehensive cost of all these adjacent node pairs is accumulated and summed up, and through this accumulation process, the complete path sequence P is obtained k The corresponding total accumulated running state transition path comprehensive cost.

[0227] Based on the accumulated running state transition path comprehensive cost of each complete path sequence in the full-quantitative path candidate list calculated in the previous step, for example, the total cost of path P1 is C(P1), the total cost of path P2 is C(P2), and so on, for example, there are M such complete path sequences and their corresponding total costs in the list, the system will then sort these M paths according to their accumulated running state transition path comprehensive cost, the sorting rule is from low to high, that is, the path with the smallest total cost is placed at the front, and the path with the largest total cost is placed at the back, if there are two or more paths with the same total cost, the secondary sorting rule (for example, prefer to select the path with fewer start-stop times, or prefer to select the path with a smoother frequency fluctuation, these secondary rules can be introduced by introducing a small bias term in the comprehensive cost calculation, or comparing these indicators when the cost is the same, here for example, only sort by total cost, if the same, the order is arbitrary or in the dictionary order of path number) can be used to determine, through standard sorting algorithms (such as quicksort, mergesort, etc.) to process these M (path sequence, total cost) pairs, finally generate an ordered list, each item in this list is a complete path sequence, and these path sequences are strictly arranged from smallest to largest according to their accumulated running state transition path comprehensive cost, this ordered list is the path accumulated path comprehensive cost sorting list.

[0228] Based on the path accumulated path comprehensive cost sorting list generated in the previous step, which arranges all feasible complete scheduling paths from the current state to the end of the future time node according to their total cost from low to high, the system will directly select the complete path sequence at the top of this sorting list (i.e. the first one), because this path has the lowest accumulated running state transition path comprehensive cost among all candidate paths according to the aforementioned comprehensive cost calculation model, which represents the optimal economic operation strategy under the current prediction and constraint conditions, once this optimal path sequence is selected, for example, path P optimal =(S 1,opt →S 2,opt →…→S T,opt ), the system will follow the running state S t,optThe detailed configuration information of the desulfurization system is inputted into the computer, and the operation frequency and start-stop state of each slurry circulating pump at each time point t are extracted. The time-sequenced specific operation instructions (for example, at t=1, pump A operates at 35 Hz, pump B operates at 40 Hz, and pump C stops; at t=2, pump A operates at 38 Hz, pump B operates at 42 Hz, and pump C stops; and at t=T, …) are organized to form a group of clear scheduling instruction sequences which can be directly issued to the desulfurization control system. The group of instruction sequences is the final combined optimal scheduling instruction path of the circulating pumps.

[0229] The above is only a preferred embodiment of the present application, and does not limit the present application in other forms. Any person skilled in the art can use the disclosed technical content to make changes or modifications into equivalent embodiments applied to other fields. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present application without departing from the technical solution content of the present application still falls within the protection scope of the present application.

Claims

1. A method for optimizing a dual-tower desulfurization slurry circulating pump combination based on artificial intelligence, characterized in that, The method comprises the following steps: predicting future multi-time point boiler load values, defining discrete running situations combining the start-stop state and running frequency of the circulating pump group, establishing slurry output values corresponding to each situation, matching the slurry output required by the boiler load at each future time point, and generating a future working condition requirement and running situation mapping set; based on the future working condition requirement and running situation mapping set, extracting all running situations, calculating the energy consumption transfer value generated by the change in running frequency between any two situations, establishing an energy consumption transfer cost table, superimposing the quantified wear and tear cost of the start-stop equipment based on the energy consumption transfer cost table, and generating a running situation migration path comprehensive cost; based on the future working condition requirement and running situation mapping set, filtering all running situations that meet the slurry output requirement at the first time point, obtaining a feasible situation set for the first node, matching the running situation that meets the constraint for each subsequent time point based on the feasible situation set for the first node and constructing a connection, and establishing a feasible migration network between multiple time nodes; based on the feasible migration network between multiple time nodes, starting from the current running situation and tracking to the last time node, generating all selectable complete path sequences, obtaining a full-quantized path candidate list, calculating the total cost of each path based on the full-quantized path candidate list and sorting the total cost, and establishing a circulating pump combination optimal dispatching instruction path.

2. The AI-based dual-tower desulfurization slurry circulating pump set combination optimization method according to claim 1, characterized in that, The step of obtaining the future working condition requirement and running situation mapping set is: based on the boiler operation record, extracting the boiler load value, absorption tower inlet sulfur dioxide concentration, slurry circulation amount, liquid-gas ratio, pH value, oxygen concentration, and circulating pump running frequency at each time point in the boiler operation record to obtain a boiler load prediction input set; calculating the boiler load prediction value at each future time point according to the boiler load prediction input set; based on the boiler load prediction value, calling the circulating pump start-stop state and running frequency defined combination situation set, matching the predicted load with the slurry output corresponding to each situation, and generating a future working condition requirement and running situation mapping set.

3. The AI-based dual-tower desulfurization slurry circulating pump combination optimization method of claim 1, wherein, The step of obtaining the energy consumption transfer cost table is: based on the future working condition requirement and running situation mapping set, extracting the running frequency value of all circulating pumps in each running situation in turn, pairing the frequencies of the corresponding circulating pumps under any two different running situations, and simultaneously extracting the target running time of the circulating pumps after the situation transfer, generating a running frequency transfer and time pairing sequence; calculating the total energy consumption transfer value according to the running frequency transfer and time pairing sequence; based on the total energy consumption transfer value, recording the energy consumption change between each pair of running situations item by item, constructing a matrix structure according to the running situation number, outputting the energy consumption difference of the transfer path between all situations in row-column mode, and generating an energy consumption transfer cost table.

4. The AI-based dual-tower desulfurization slurry circulating pump combination optimization method of claim 1, wherein, The step of obtaining the running situation migration path comprehensive cost is: Based on the running situation start-stop record, the running state identifiers of each device before and after the migration of each group of running situations are compared, the devices whose state changes from running to stopping or from stopping to running are identified, each device that has a state change is recorded as a start-stop action, the running duration of each device in the recent running cycle is extracted, and a start-stop device and its running duration before start-stop are generated to form a list; According to the start-stop device and its running duration before start-stop list and the energy consumption transfer cost table, the running situation migration path cost is calculated; Based on the migration path cost, the migration path costs between all running situation pairs are sequentially filled into the cost table structure matrix, and the migration path costs are arranged according to the starting situation and target situation indexes to generate a comprehensive migration path cost of the running situation.

5. The AI-based dual-tower desulfurization slurry circulating pump combination optimization method according to claim 1, characterized in that, The acquisition step of the feasible situation set of the first node is: Based on the future working condition demand and running situation mapping set, the running frequency and start-stop state of all circulating pumps in each running situation in the future working condition demand and running situation mapping set are retrieved, the slurry output flow of each circulating pump is calculated, and the slurry output flows of all circulating pumps are summed to form the total slurry output flow of each running situation; Based on the total slurry output flow of each running situation, the slurry demand value required by the boiler load corresponding to the first time point in the future working condition demand and running situation mapping set is extracted, the total slurry output flow of each running situation is compared with the demand value one by one, all running situations whose total slurry output flow is equal to or higher than the demand value are selected to form a preliminary running situation set that meets the slurry demand of the boiler load at the first time point; Based on the preliminary running situation set that meets the slurry demand of the boiler load at the first time point, whether the start-stop state and frequency of each circulating pump in the running situation are within the safe running interval allowed by the device is verified, and the running situation whose start-stop state or frequency is out of the safe running range is removed to form the feasible situation set of the first node.

6. The AI-based dual-tower desulfurization slurry circulating pump combination optimization method of claim 1, wherein, The acquisition step of the feasible migration network between multiple time nodes is: Based on the feasible situation set of the first node, the running frequency, start-stop state, and slurry output flow of all circulating pumps corresponding to each running situation in the set are extracted one by one, and all estimated running situations that can meet the slurry demand at the second time point are matched with each running situation according to the slurry demand value corresponding to the boiler load demand at the second time point to form a feasible situation connection set from the first node to the second time point; Based on the feasible situation connection set from the first node to the second time point, the running frequency, start-stop state, and slurry output flow of all circulating pumps of all running situations at the second time point in the connection set are retrieved and extracted, and whether each situation can meet the slurry demand at the third time point under the constraints of the circulating pump frequency adjustment amplitude and safe frequency interval is judged according to the slurry demand value corresponding to the boiler load demand at the third time point, and the connection situation that meets the condition is selected to form a feasible situation connection set from the second time node to the third time node; Based on the feasible state connection set from the second time point to the third time point, the running state connection and constraint verification are sequentially progressed to all subsequent remaining time points, the feasible state connection of all time points is connected in series to form a feasible migration network among multiple time nodes.

7. The AI-based dual-tower desulfurization slurry circulating pump combination optimization method according to claim 1, characterized in that, The acquisition step of the full-quantitative path alternative list is: Based on the feasible migration network among multiple time nodes, the current running state in the current time node is selected as an initial starting node, all running state nodes connected to the next time node under the initial starting node are extracted one by one, the running frequency, start-stop state and slurry output flow of all circulating pumps contained in each state node are recorded in sequence, and the connection relationship between respective nodes is marked to form a state path set from the initial starting node to the second time node; Based on the state path set from the initial starting node to the second time node, the running frequency, start-stop state and slurry output flow of all circulating pumps corresponding to each running state node in the second time node are sequentially retrieved, all feasible connected running states are matched to the third time node, and the connection relationship between each pair of nodes is recorded, and the node connection chain set from the initial starting node to the end time node is sequentially progressed to the end time node to form a node connection chain set from the initial starting node to the end time node; Based on the node connection chain set from the initial starting node to the end time node, the full-quantitative path alternative list is formed by starting from the initial starting node, sequentially traversing all node connection chains, connecting in series to form a continuous and complete path sequence of multiple time nodes, eliminating the connection chains that are broken or cannot be continuously continued to the end time node.

8. The method of claim 1, wherein the method is based on an artificial intelligence-based dual-tower desulfurization slurry circulating pump combination optimization method. The acquisition step of the circulating pump combination optimal scheduling instruction path is: Based on the full-quantitative path alternative list, each complete path sequence in the list is retrieved one by one, the running state migration path comprehensive cost corresponding to each pair of adjacent running state nodes between the starting node and the end node of the path is extracted step by step, the running state migration path comprehensive cost between each pair of nodes is recorded, and the running state migration path comprehensive cost of all adjacent node pairs is accumulated to form the cumulative running state migration path comprehensive cost corresponding to each path; Based on the cumulative running state migration path comprehensive cost corresponding to each path, the cumulative running state migration path comprehensive costs of all complete path sequences are sorted from low to high to form a path cumulative path comprehensive cost sorting list; Based on the path cumulative path comprehensive cost sorting list, the complete path sequence with the lowest running state migration path comprehensive cost in the sorting list is selected, and the circulating pump running frequency and start-stop state recorded in the path sequence are used to form a circulating pump combination optimal scheduling instruction path.

9. The dual-tower desulfurization slurry circulating pump combination optimization system based on artificial intelligence according to any one of claims 1-8, characterized in that, It comprises: A prediction module: predicting the future boiler load values of multiple time points, defining discrete running states in combination with the start-stop state and running frequency of the circulating pump group, establishing the slurry output values corresponding to each state, matching the slurry amount required by the boiler load of each time point in the future, and generating a future working condition demand and running state mapping set; A cost modeling module: based on the future working condition demand and the running situation mapping set, all running situations are extracted, the energy consumption transfer value generated by the change of the running frequency between any two situations is calculated, the energy consumption transfer cost table is established, based on the energy consumption transfer cost table, the wear and tear cost of the start-stop equipment after quantification is superimposed, and the running situation migration path comprehensive cost is generated; A network construction module: based on the future working condition demand and the running situation mapping set, all running situations meeting the pulp quantity demand at the first time point are screened, the feasible situation set of the first node is obtained, based on the feasible situation set of the first node, the running situations meeting the constraints for subsequent time points are matched and connected in sequence, and the feasible migration network between multiple time nodes is established; A path optimization module: based on the feasible migration network between multiple time nodes, starting from the current running situation, tracking to the last time node, generating all selectable complete path sequences, obtaining the full-quantized path alternative list, based on the full-quantized path alternative list, calling the running situation migration path comprehensive cost to calculate the total cost of each path and sorting, and establishing the optimal dispatching instruction path of the circulating pump combination.

Images