Data-mechanism dual-driven pump station real-time regulation and control method

Abstract

The invention discloses a data-mechanism dual-driven pump station real-time regulation and control method. The method comprises the steps that key influence factors for determining the number of started pump stations are determined; all feasible regions of real-time regulation and control of the pump station are determined, a real-time regulation and control mechanism driving model of the pump station is constructed and solved, and the distribution condition of the number of started pump stations in all the feasible regions is obtained to serve as a mechanism driving model solution; determining a real-time regulation and control local feasible region of the pump station, constructing and solving a real-time regulation and control data driving model of the pump station, and obtaining the distribution condition of the number of started pump stations in the local feasible region as a data driving model solution; and constructing a fusion data set based on the mechanism-driven model solution and the data-driven model solution, training and verifying a data-mechanism dual-driven model based on the fusion data set, and obtaining a final regulation and control decision result. The method has the advantages that the scheduling decision more conforming to the in-situ regulation and control requirement is generated while high-efficiency operation of the pump station is pursued, and the problem that a unit is frequently started and stopped due to the fact that a mechanism driving model pursues high-efficiency operation of the pump station is avoided to a certain extent.

Description

Technical Field

[0001] The present invention relates to the technical field of pump station optimization scheduling, and in particular to a data-mechanism dual-driven real-time control method for a pump station. Background Art

[0002] In order to complete the task of water transfer in each period, it is necessary to comprehensively consider the real-time water conditions and work conditions information to determine the operation plan of the pump station units. On the basis of known pump performance parameters and under the premise of meeting relevant constraints, it is of great significance to reasonably determine the unit combination mode, flow rate, speed and other parameters of the pump station to achieve the optimal operation of the pump station. At present, water transfer projects with pump stations as the core generally face problems such as low efficiency of pumping devices, energy waste of devices and high operating costs during operation. In order to achieve the goal of energy saving and consumption reduction, domestic and foreign scholars have conducted a lot of research on the optimized operation of pump stations. At this stage, the methods of real-time control of pump stations mainly include the following methods:

[0003] (1) Real-time control of pump stations based on optimization algorithms

[0004] The real-time dispatching method of pumping stations based on optimization algorithms uses optimization algorithms to solve the parameters such as the combination mode of pumping station units, flow rate and speed, under the premise of meeting relevant constraints, so that the objective function of pumping station operation reaches the maximum or minimum. Commonly used optimization algorithms include: genetic algorithm, ant colony algorithm, particle swarm algorithm, artificial neural network, etc. Although a large number of studies have confirmed the significant advantages of this method in improving pumping station efficiency and reducing operating energy consumption, the application of this control method in actual engineering is very limited. In actual applications, this control method may lead to frequent start and stop of units in pursuit of efficient operation of pumping stations, which accelerates the aging of equipment and increases the work pressure of local personnel.

[0005] (2) Real-time control of pump stations based on manual experience

[0006] The real-time control method of pumping stations based on manual experience mainly relies on the experience and intuition of local dispatchers to decide when to start the pumping station units and how many units to start. It cannot be denied that this control method contains the rich wisdom and work experience of local dispatchers, but this method lacks scientificity and precision, and will sacrifice the operating efficiency of the pumping station to a certain extent, causing the pumping station to continue to deviate from the high-efficiency zone. In addition, the experience and intuition of local dispatchers may deviate according to different situations, resulting in uncertainty in the real-time control of the pumping station, making it difficult to adapt to unstable and uncertain environments. Summary of the invention

[0007] The purpose of the present invention is to provide a data-mechanism dual-driven real-time control method for a pump station, thereby solving the above-mentioned problems existing in the prior art.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0009] A data-mechanism dual-driven real-time control method for a pump station comprises the following steps:

[0010] S1. Determination of key impact factors:

[0011] Combined with the actual situation of on-site dispatching of pumping stations and the theoretical research on real-time optimization and control of pumping stations, the corresponding relationship between pumping station water condition data and control strategies is summarized to determine the key influencing factors that determine the number of pumping station start-up units;

[0012] S2. Construction and solution of mechanism-driven model:

[0013] Based on the key influencing factors that determine the number of pumping station start-up units and the constraints of the real-time control operation of the pumping station, the entire feasible domain of the real-time control of the pumping station is determined; by setting the objective function, decision variables, constraints and solution algorithm, the mechanism-driven model of the real-time control of the pumping station is constructed, and the entire feasible domain of the real-time control of the pumping station is input into the mechanism-driven model of the real-time control of the pumping station, and the distribution of the number of pumping stations in operation under the entire feasible domain is obtained as the solution of the mechanism-driven model;

[0014] S3. Construction and solution of data-driven model:

[0015] Based on the historical operating water conditions data and control decisions of the pumping station, the local feasible domain of the real-time control of the pumping station is determined; based on the decision tree classification algorithm, a data-driven model for the real-time control of the pumping station is constructed, and the local feasible domain of the real-time control of the pumping station is input into the data-driven model for the real-time control of the pumping station to obtain the distribution of the number of units in operation of the pumping station under the local feasible domain as the data-driven model solution;

[0016] S4. Construction and solution of data-mechanism dual-driven model:

[0017] A fusion data set is constructed based on the mechanism-driven model solution and the data-driven model solution. The data-mechanism dual-driven model constructed based on the decision tree is trained and verified based on the fusion data set to obtain the final regulatory decision result.

[0018] Preferably, step S1 specifically includes the following contents:

[0019] S101. Consult relevant literature on pump station optimization operation, determine the commonly used objective function for pump station optimization operation, and determine the preliminary influencing factors based on the objective function calculation formula;

[0020] S102. Combined with local needs, consider the water and engineering conditions that affect local regulation and control decisions;

[0021] S103. Based on literature research and local needs, it is determined that the key influencing factors that determine the number of pump station start-up units include pump station flow rate and pump station head.

[0022] Preferably, the determination of the entire feasible domain of real-time control of the pump station in step S2 specifically includes the following contents:

[0023] S201, querying the design parameters of the pump station, including the minimum and maximum values ​​of the key influencing factors that determine the number of units in operation of the pump station;

[0024] S201, discretize the key factors according to a reasonable step size to generate a solution space;

[0025] S201. According to the constraints of the real-time control operation of the pump station, the working conditions that do not meet the startup conditions are eliminated to obtain the entire feasible domain of the real-time control of the pump station.

[0026] Preferably, the constraints for the real-time control of the pump station operation include water pump unit flow constraint, water pump unit head constraint, pump station operation station bottom / top water level constraint, unit operation number constraint, and water pump flow-head characteristic curve constraint.

[0027] Preferably, the objective function of the pump station real-time control mechanism driving model is the highest operating efficiency of the pump station under working conditions; the decision variable of the pump station real-time control mechanism driving model is the number of pump station start-up units; and the solution algorithm of the pump station real-time control mechanism driving model is a particle swarm optimization algorithm.

[0028] Preferably, determining the local feasible region for real-time control of the pump station in step S3 specifically includes the following contents:

[0029] S3011. Collect water condition data and control decision data of the pump station since the year of operation; all data are the dispatching results of the same period;

[0030] S3012. Analyze the distribution characteristics of the historical operation data of the pump station, and obtain the local flow rate and head operable range according to the minimum / maximum values ​​of the operation data flow rate and head according to a certain ratio, and eliminate the working conditions that do not meet the startup conditions according to the operation constraints to determine the local feasible domain for real-time control of the pump station;

[0031] S3013, analyzing the distribution characteristics of the historical operation data of the pump station, and determining the sample size and whether the distribution is balanced. Preferably, the construction and solution of the real-time control data driven model of the pump station in step S3 specifically include the following contents:

[0032] S3021, dividing the pre-processed historical control data of the pump station into a training set and a verification set according to a certain ratio;

[0033] S3022. Construct a data-driven model for real-time control of pump stations based on a decision tree classification algorithm;

[0034] S3023. Train and verify the pump station real-time control data-driven model based on the training set and the verification set, and evaluate the performance of the model based on the evaluation indicators;

[0035] S3024. Input the local feasible domain of the real-time regulation of the pumping station into the trained data-driven model of the real-time regulation of the pumping station, and obtain the distribution of the number of started units under the local feasible domain of the real-time regulation of the pumping station as the solution of the data-driven model.

[0036] Preferably, constructing the fused data set in step S4 specifically includes the following contents:

[0037] S4011, connecting the mechanism-driven model solution and the data-driven model solution in series to form a fused data set;

[0038] S4012, normalizing the fused data set;

[0039] S4013. Divide the normalized fused data set into a training set and a validation set.

[0040] Preferably, obtaining the final control decision result in step S4 specifically includes the following contents:

[0041] S4021. Use a decision tree to build a data-mechanism dual-driven model, input the training set into the data-mechanism dual-driven model for training, input the validation set into the trained data-mechanism dual-driven model, and output the model results;

[0042] S4022. Denormalize the model results to obtain the final control decision results.

[0043] Preferably, after step S4022, the method further includes:

[0044] S4023. Evaluate the reliability of the data-mechanism dual-driven model, and use the entire feasible domain of the model as input to calculate the model's global regulatory decision-making results.

[0045] The beneficial effects of the present invention are as follows: 1. The method of the present invention combines the solution results of the mechanism-driven model and the historical operation scheduling decision, and generates a scheduling decision that is more in line with the local control needs while pursuing the efficient operation of the pump station. It can avoid the frequent start and stop problems of the unit caused by the mechanism-driven model pursuing the efficient operation of the pump station to a certain extent, delay the aging of the equipment, and reduce the work pressure of the local control personnel. 2. The method of the present invention takes into account the problem that the actual operating conditions are relatively concentrated and the historical samples are insufficient. The local feasible domain is divided according to the historical operation interval, and the solution of the local feasible domain is solved by the data-driven method, and the local solution based on the data-driven model and the global solution based on the mechanism-driven model are integrated. The characteristics of the global domain and the local characteristics are integrated to enhance the comprehensiveness and insight of the data. 3. The method of the present invention targets the problem of fusion of data-driven models and mechanism-driven models, and performs data fusion of different scales at the data set level to form a fusion data set driven by both mechanism and data. This method is suitable for use when integrating multiple models with similar performance, which greatly improves the reliability and interpretability of control decisions, and can obtain a more comprehensive and optimized control plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a principle flow chart of the control method in an embodiment of the present invention;

[0047] Figure 2 It is a schematic diagram of a decision-making scheme of a mechanism-driven model in an embodiment of the present invention;

[0048] Figure 3 Schematic diagram of a data-driven model decision-making scheme in an embodiment of the present invention;

[0049] Figure 4 It is a schematic diagram of a decision-making scheme of a data-mechanism dual-driven model in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation methods described herein are only used to explain the present invention and are not used to limit the present invention.

[0051] like Figure 1As shown, in this embodiment, in order to solve the problems of frequent start and stop of pump station units and low pump station operation efficiency in existing pump station control methods, a data-mechanism dual-driven real-time control method for pump stations is proposed. This method adopts a particle swarm optimization algorithm, takes the highest pump station operation efficiency as the goal, constructs a pump station mechanism real-time control model, combines the historical scheduling decision plan of pump station operation, learns historical scheduling decisions and the optimal decision of the mechanism-driven model through a decision tree classification model, generates new decision plans, and realizes the reasonable and efficient operation of pump station units. The present invention alleviates the problems of difficulty in applying control based on mechanism-driven models, and can be used as an effective method for real-time control of pump stations. The method of the present invention mainly includes the following four parts:

[0052] 1. Determination of key impact factors

[0053] Combined with the actual situation of on-site dispatching of pumping stations and the theoretical research on real-time optimization and control of pumping stations, the corresponding relationship between pumping station water condition data and control strategies is summarized to determine the key influencing factors that determine the number of pumping station start-up units. The specific contents include the following:

[0054] (1) Consult relevant literature on pump station optimization operation, determine the commonly used objective function for pump station optimization operation (such as pump station efficiency), and determine the preliminary influencing factors based on the objective function calculation formula.

[0055] (2) Considering the local dispatching needs, the local dispatching personnel mainly consider factors such as water conditions and working conditions when executing the dispatching decisions, including the flow rate of the pump station, the head of the pump station, the operating status of the unit and the wear of the equipment.

[0056] (3) Based on literature research and local needs, the key influencing factors that determine the number of pump station start-up units are determined to include pump station flow rate and pump station head.

[0057] 2. Construction and solution of mechanism-driven model

[0058] Based on the key influencing factors that determine the number of pumping station start-up units and the constraints of the real-time control operation of the pumping station, the entire feasible domain of the real-time control of the pumping station is determined; by setting the objective function, decision variables, constraints and solution algorithm, a mechanism-driven model of the real-time control of the pumping station is constructed, and the entire feasible domain of the real-time control of the pumping station is input into the mechanism-driven model of the real-time control of the pumping station, and the distribution of the number of start-up units of the pumping station under the entire feasible domain is obtained as the solution of the mechanism-driven model.

[0059] In this embodiment, the entire feasible domain of real-time control of the pump station is determined, which specifically includes the following contents:

[0060] (1) Query the design parameter values ​​of the pump station, including the minimum and maximum values ​​of key influencing factors.

[0061] (2) Discretize the key influencing factors according to a reasonable step size to generate a solution space.

[0062] (3) According to the constraints of the real-time control of the pump station, the working conditions that do not meet the startup conditions are eliminated to obtain the entire feasible domain of the real-time control of the pump station. The real-time control of the pump station must be operated within certain constraints, including: water pump unit flow constraint, water pump unit head constraint, water level constraint below / above the pump station, unit number constraint, water pump flow-head characteristic curve constraint, etc.

[0063] In this embodiment, by taking the key influencing factors as decision variables, a real-time control mechanism driving model of the pump station is constructed based on the particle swarm optimization algorithm, thereby generating the optimal number of units in operation under different flow and head combinations. Specifically, it includes the following contents:

[0064] (1) Establishing the objective function of the pump station real-time control mechanism driving model. In the present invention, the pump station real-time control mechanism driving model aims to maximize the pump station operation efficiency under the calculation conditions.

[0065] (2) Establishing the decision variables of the driving model of the real-time regulation mechanism of the pump station. In the present invention, the driving model of the real-time regulation mechanism of the pump station uses the number of pump stations in operation as the decision variable.

[0066] (3) Establish the constraints of the driving model of the real-time control mechanism of the pump station. In the present invention, when the operating flow of the pump station is determined, the constraints of the pump group flow, head and water level need to be considered.

[0067] (4) Establishing a solution algorithm for the pump station real-time control mechanism driving model. In the present invention, the solution is performed according to the particle swarm optimization algorithm.

[0068] (5) According to the theoretical optimal solution of the real-time control mechanism driving model of the pump station, the distribution of the number of pump stations in operation in the entire feasible domain is determined, and the mechanism driving model solution S1 is generated.

[0069] 3. Construction and solution of data-driven model

[0070] Based on the historical operating water conditions data and control decisions of the pumping station, the local feasible domain of real-time control of the pumping station is determined. Based on the decision tree classification algorithm, a data-driven model for real-time control of the pumping station is constructed, and the local feasible domain of real-time control of the pumping station is input into the data-driven model for real-time control of the pumping station. The distribution of the number of operating units of the pumping station in the local feasible domain is obtained as the solution of the data-driven model.

[0071] In this embodiment, the local feasible region for real-time control of the pump station is determined, which specifically includes the following contents:

[0072] (1) Collect water condition data of the pump station since its operation, including pump head and flow rate; collect control decisions of the pump station since its operation, including the number of pump stations in operation; all data are the dispatch results of the same period.

[0073] (2) Analyze the distribution characteristics of historical operating data, find the minimum and maximum values ​​of flow rate and head, and fluctuate them up and down according to a certain ratio to obtain the local flow rate and head operable range. According to the operating constraints, eliminate the working conditions that do not meet the startup conditions and establish the local operable domain of the pump station based on historical data.

[0074] (3) Summarize the distribution of the number of pump station startups based on historical operating data and analyze the sample size and distribution.

[0075] In this embodiment, based on the pre-processed historical control data of the pump station, the sampling decision tree classification algorithm generates a real-time control data-driven model for the pump station, and evaluates the performance of the model. Specifically, it includes the following contents:

[0076] (1) The preprocessed historical control data of the pumping station is divided into a training set and a validation set according to a certain ratio.

[0077] (2) A data-driven model for real-time control of pump stations is constructed based on a decision tree classification algorithm. In essence, this model is a classification model for the number of units in operation based on flow rate and head.

[0078] (3) Evaluate the model performance based on accuracy, precision, recall, F1-score and other evaluation indicators.

[0079] (4) The local feasible domain is input into the data-driven model to generate the distribution of the number of units in operation under the local feasible domain of real-time control of the pump station, forming the data-driven model solution S2.

[0080] 4. Construction and solution of data-mechanism dual-driven model

[0081] A fusion data set is constructed based on the mechanism-driven model solution and the data-driven model solution. The data-mechanism dual-driven model constructed based on the decision tree is trained and verified based on the fusion data set to obtain the final regulatory decision result.

[0082] In this embodiment, the fusion data set is constructed, which specifically includes the following contents:

[0083] (1) Connect the mechanism-driven model solution S1 and the data-driven model solution S2 in series to form a fused dataset S.

[0084] (2) Normalize the fused data set.

[0085] (3) Divide the normalized fusion dataset into a training set and a validation set.

[0086] In this embodiment, obtaining the final control decision result specifically includes the following contents:

[0087] (1) Use decision trees to construct a data-mechanism dual-driven model, input the training set into the data-mechanism dual-driven model for training, and input the validation set into the trained data-mechanism dual-driven model to output the model results.

[0088] (2) Denormalize the model results to obtain the final control decision results.

[0089] (3) Evaluate the reliability of the data-mechanism dual-driven model, and use the entire feasible domain of the model as input to calculate the model's global regulatory decision-making results.

[0090] Embodiment 2

[0091] In this embodiment, a pump station of the East Route Project of South-to-North Water Diversion is taken as an example, and the method flow provided by the present invention is used to establish a pump station control scheme. The design scale of the pump station is 120m 3 / s, there are 5 units, one of which is a standby unit, and the flow rate of a single unit is 30m 3 / s, the flow range of the pump station is 14.2~191.4m 3 / s; the pump station has a low operating head, with a design head of 3.73m, an average head of 1.6m, a pump station head range of 0-3.8m, an annual operating time of 5000 hours, and 210 days of operation per year. The basic parameters of the pump unit of the pump station are shown in Table 1.

[0092] Table 1 Basic parameters of pump station water pump unit

[0093] name parameter Water pump type( / ) Rear bulb tubular pump Number of installed units (units) 5 (4 in use and 1 in reserve) Single machine capacity (kW) 2000 Total installed capacity (kW) 10000 Pump impeller diameter (m) 3.2 Pump speed (r / min) 120 Motor speed (r / min) 750 Matching motor rated power (KW) 2000 Total installed capacity of motor (kW) 10000 Transmission mode Gearbox transmission

[0094] 1. Obtaining key impact factors

[0095] Combining literature research and local needs, the key influencing factors that determine the number of pump stations to be opened are determined to include: pump station flow rate and pump station head.

[0096] 2. Construction and solution of mechanism-driven model

[0097] 2.1. Determination of all feasible regions for real-time control of pumping stations

[0098] The flow rate of the pump station ranges from 14.2 to 191.4 m 3 / s, the pump station head range is 0~4.73m, and the flow rate and head are calculated as 0.1m 3 / s and 0.1m step size are used for discretization, and the working conditions that do not meet the startup conditions are eliminated to obtain the entire feasible domain of real-time control of the pump station.

[0099] 2.2 Model construction and solution

[0100] The key influencing factors, pump station head and pump station flow, are used as decision variables, and the maximum pump station operation efficiency is used as the objective function. Based on the particle swarm optimization algorithm, a pump station real-time control mechanism driving model is constructed to generate the optimal number of units in operation under different flow and head combinations. In this embodiment, the Sihong Pump Station is equipped with 5 units of the same type. According to the flow and head constraints of the pump station and the calculation results of the mechanism-driven model, there are 1, 2, 3, 4 and 5 units in operation. The pump station control scheme based on the mechanism-driven model is as follows: Figure 1 shown.

[0101] 3. Construction and solution of data-driven model

[0102] 3.1. Determination of local feasible region for real-time control of pumping stations

[0103] Collect historical water condition data and control decisions of the pump station, including: flow rate, head and number of units in operation. Since the head of the pump station cannot be directly monitored by the water level meter, the water level of the inlet and outlet pools of the pump station is collected according to the placement of the water level meter in this embodiment, and the head of the pump station is calculated according to the water level of the inlet and outlet pools. The collected historical operation data of the pump station is shown in Table 2.

[0104] Table 2 Historical operation data

[0105]

[0106]

[0107] According to the operation data, the historical regulation range of the pump station is analyzed: there are 4 situations in the historical samples, namely, 2, 3, 4, and 5 units are in operation, and the flow rate operation range is 37.2~139.5m 3 / s, the lift operating range is 0.05~1.95m, which accounts for a small proportion of the entire feasible domain of the pump station and is concentrated. Based on historical sample data, the local feasible domain of the pump station is set to flow range: 35.0~150.0m 3 / s, head range: 0.0~2.0m, respectively according to 0.1m 3 / s and 0.1m step size, eliminating the working conditions that cannot be started, and obtaining the local operational domain.

[0108] The statistics of the number of pump station startups collected a total of 2,417 pump station control data at historical moments, including 23 with 2 startups and 36 with 3 startups; 2,081 with 4 startups and 277 with 5 startups. The sample size is small and there is an obvious imbalance. In order to solve the problem of small sample size and concentrated distribution of measured data, the next step is to construct a data-driven model solution.

[0109] 3.2 Model construction and solution

[0110] The pre-processed historical control data of the pump station (a total of 2417 items) were randomly divided into a training set (2176 items) and a validation set (241 items) at a ratio of 90% and 10%, and a real-time control data-driven model of the pump station was constructed based on the decision tree classification algorithm. In essence, it is a classification model for the number of units in operation based on the two factors of flow and head. The performance of the data-driven model was evaluated using relevant indicators, and the evaluation results showed that the data-driven model was reliable. The evaluation results are shown in Table 3.

[0111] Table 3 Data-driven model performance

[0112]

[0113] The local feasible domain is input into the data-driven model to generate the distribution of the number of units in operation under the local feasible domain of real-time control of the pump station, forming the data-driven model solution S2, see Figure 3 .

[0114] 4. Construction and solution of data-mechanism dual-driven model

[0115] 4.1. Hybrid Dataset Construction

[0116] The series mechanism-driven model solution S1 (57,391 samples in total) and the data-driven model solution S2 (20,106 samples in total) constitute the fused dataset S (77,497 samples in total). The fused dataset S is normalized and divided into a training set and a validation set in proportion.

[0117] 4.2. Data-Mechanism Dual Drive Model Solution

[0118] The training set of the fused dataset S is input into the data-mechanism dual-driven model for training, and the verification set of the fused dataset S is input into the trained dual-driven model for verification. The model results are denormalized to obtain the final regulatory decision result of the model.

[0119] The model prediction results are compared with the historical control decisions and the mechanism-driven model decisions to analyze the rationality of the results. Some prediction results are shown in Table 4. The verified model is used to predict the number of units started in the entire feasible domain, and the decision results based on the data-mechanism dual-driven model are obtained, such as Figure 4 .

[0120] Table 4 Decision results of the number of pumping stations in different control modes

[0121]

[0122]

[0123] By adopting the above technical solution disclosed in the present invention, the following beneficial effects are obtained:

[0124] The present invention provides a real-time control method for a pump station driven by data and mechanism. The method combines the solution results of the mechanism-driven model and the historical operation scheduling decision. While pursuing the efficient operation of the pump station, it generates a scheduling decision that is more in line with the local control needs. The problem of frequent start and stop of the unit caused by the pursuit of efficient operation of the pump station by the mechanism-driven model can be avoided to a certain extent, delaying the aging of the equipment and reducing the work pressure of the local control personnel. The method of the present invention takes into account the problem that the actual operating conditions are relatively concentrated and the historical samples are insufficient. The local feasible domain is divided according to the historical operation interval, and the solution of the local feasible domain is solved by the data-driven method. The local solution based on the data-driven model and the global solution based on the mechanism-driven model are fused, and the characteristics of the global domain and the local characteristics are integrated to enhance the comprehensiveness and insight of the data. The method of the present invention targets the problem of fusion of the data-driven model and the mechanism-driven model, and performs data fusion of different scales at the data set level to form a fusion data set driven by mechanism and data. The method is suitable for use when integrating multiple models with similar performance, which greatly improves the reliability and interpretability of the control decision, and can obtain a more comprehensive and optimized control scheme.

[0125] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be considered as the scope of protection of the present invention.

Claims

1. A data-mechanism dual-driven real-time control method for a pump station, characterized in that: The following steps are included: S1. Determination of key impact factors: Combined with the actual situation of on-site dispatching of pumping stations and the theoretical research on real-time optimization and control of pumping stations, the corresponding relationship between pumping station water condition data and control strategies is summarized to determine the key influencing factors that determine the number of pumping station start-up units; S2. Construction and solution of mechanism-driven model: Based on the key influencing factors that determine the number of pumping station start-up units and the constraints of the real-time control operation of the pumping station, the entire feasible domain of the real-time control of the pumping station is determined; by setting the objective function, decision variables, constraints and solution algorithm, the mechanism-driven model of the real-time control of the pumping station is constructed, and the entire feasible domain of the real-time control of the pumping station is input into the mechanism-driven model of the real-time control of the pumping station, and the distribution of the number of pumping stations in operation under the entire feasible domain is obtained as the solution of the mechanism-driven model; S3. Construction and solution of data-driven model: Based on the historical operating water conditions data and control decisions of the pumping station, the local feasible domain of the real-time control of the pumping station is determined; based on the decision tree classification algorithm, a data-driven model for the real-time control of the pumping station is constructed, and the local feasible domain of the real-time control of the pumping station is input into the data-driven model for the real-time control of the pumping station to obtain the distribution of the number of units in operation of the pumping station under the local feasible domain as the data-driven model solution; S4. Construction and solution of data-mechanism dual-driven model: A fusion data set is constructed based on the mechanism-driven model solution and the data-driven model solution. The data-mechanism dual-driven model constructed based on the decision tree is trained and verified based on the fusion data set to obtain the final regulatory decision result.

2. The data-mechanism dual-driven real-time control method for a pump station according to claim 1 is characterized in that: Step S1 specifically includes the following contents: S101. Consult relevant literature on pump station optimization operation, determine the commonly used objective function for pump station optimization operation, and determine the preliminary influencing factors based on the objective function calculation formula; S102. Combined with local needs, consider the water and engineering conditions that affect local regulation and control decisions; S103. Based on literature research and local needs, it is determined that the key influencing factors that determine the number of pump station start-up units include pump station flow rate and pump station head.

3. The data-mechanism dual-driven real-time control method for a pump station according to claim 1 is characterized in that: The entire feasible domain for real-time control of the pump station determined in step S2 specifically includes the following contents: S201, querying the design parameters of the pump station, including the minimum and maximum values ​​of the key influencing factors that determine the number of units in operation of the pump station; S201, discretize the key factors according to a reasonable step size to generate a solution space; S201. According to the constraints of the real-time control operation of the pump station, the working conditions that do not meet the startup conditions are eliminated to obtain the entire feasible domain of the real-time control of the pump station.

4. The data-mechanism dual-driven real-time control method for a pump station according to claim 3 is characterized by: The constraints for the real-time control of the pump station operation include water pump unit flow constraint, water pump unit head constraint, pump station operation station bottom / top water level constraint, unit operation unit number constraint, and water pump flow-head characteristic curve constraint.

5. The data-mechanism dual-driven real-time control method for a pump station according to claim 1 is characterized in that: The objective function of the model driven by the real-time control mechanism of the pump station is to achieve the highest operating efficiency of the pump station under the working conditions; The decision variable of the model driven by the real-time control mechanism of the pump station is the number of pump stations in operation; The solution algorithm for the pump station real-time control mechanism driving model is the particle swarm optimization algorithm.

6. The data-mechanism dual-driven real-time control method for a pump station according to claim 1 is characterized in that: The determination of the local feasible region for real-time control of the pump station in step S3 specifically includes the following contents: S3011. Collect water condition data and control decision data of the pump station since the year of operation; all data are the dispatching results of the same period; S3012. Analyze the distribution characteristics of the historical operation data of the pump station, and obtain the local flow rate and head operable range according to the minimum / maximum values ​​of the operation data flow rate and head according to a certain ratio, and eliminate the working conditions that do not meet the startup conditions according to the operation constraints to determine the local feasible domain for real-time control of the pump station; S3013. Analyze the distribution characteristics of the historical operation data of the pump station to determine the sample size and whether the distribution is balanced.

7. The data-mechanism dual-driven real-time control method for a pump station according to claim 1 is characterized in that: The construction and solution of the real-time control data-driven model of the pump station in step S3 specifically include the following contents: S3021, dividing the pre-processed historical control data of the pump station into a training set and a verification set according to a certain ratio; S3022. Construct a data-driven model for real-time control of pump stations based on a decision tree classification algorithm; S3023. Train and verify the pump station real-time control data-driven model based on the training set and the verification set, and evaluate the performance of the model based on the evaluation indicators; S3024. Input the local feasible domain of the real-time regulation of the pumping station into the trained data-driven model of the real-time regulation of the pumping station, and obtain the distribution of the number of started units under the local feasible domain of the real-time regulation of the pumping station as the solution of the data-driven model.

8. The data-mechanism dual-driven real-time control method for a pump station according to claim 1 is characterized by: The construction of the fusion data set in step S4 specifically includes the following contents: S4011, connecting the mechanism-driven model solution and the data-driven model solution in series to form a fused data set; S4012, normalizing the fused data set; S4013. Divide the normalized fused data set into a training set and a validation set.

9. The data-mechanism dual-driven real-time control method for a pump station according to claim 1 is characterized in that: The final control decision result obtained in step S4 specifically includes the following contents: S4021. Use a decision tree to build a data-mechanism dual-driven model, input the training set into the data-mechanism dual-driven model for training, input the validation set into the trained data-mechanism dual-driven model, and output the model results; S4022. Denormalize the model results to obtain the final control decision results.

10. The data-mechanism dual-driven real-time control method for a pump station according to claim 9 is characterized in that: After step S4022, the method further includes: S4023. Evaluate the reliability of the data-mechanism dual-driven model, and use the entire feasible domain of the model as input to calculate the model's global regulatory decision-making results.

Images