CN116011731A - Factory, net and river joint scheduling method based on machine learning algorithm and rainfall flood numerical model - Google Patents
Factory, net and river joint scheduling method based on machine learning algorithm and rainfall flood numerical model Download PDFInfo
- Publication number
- CN116011731A CN116011731A CN202211547299.3A CN202211547299A CN116011731A CN 116011731 A CN116011731 A CN 116011731A CN 202211547299 A CN202211547299 A CN 202211547299A CN 116011731 A CN116011731 A CN 116011731A
- Authority
- CN
- China
- Prior art keywords
- river
- joint scheduling
- rainfall
- model
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Landscapes
- Feedback Control In General (AREA)
Abstract
The invention relates to the field of flood control and waterlogging prevention, and provides a factory-network-river joint scheduling method based on a machine learning algorithm and a rainfall flood numerical model, which comprises the following steps of: 1. establishing an SWMM model of a target area; 2. calibrating SWMM model parameters; 3. prefabricating a plurality of factory network river joint scheduling schemes; 4. constructing an urban rainfall flood scene database; 5. constructing a factory network river joint scheduling effect evaluation system; 6. constructing a urban factory network and river joint scheduling scenario database; 7. constructing a rainfall condition-joint scheduling scheme-benefit index learning model; 8. and acquiring the optimal joint scheduling scheme of the plant, the network and the river under the current rainfall condition by adopting the trained rainfall condition-joint scheduling scheme-benefit index learning model. By adopting the steps, the efficiency of the acquisition scheduling method can be improved.
Description
Technical Field
The invention relates to the field of flood control and waterlogging prevention, in particular to a factory, net and river joint scheduling method based on a machine learning algorithm and a rainfall flood numerical model.
Background
Under the influence of global climate change, hydrologic cycle is accelerated, short-duration extreme rainfall time is increased more frequently, and the probability of urban flood disasters is increased remarkably. At present, the construction of sewage treatment plants, drainage pipe networks and urban river and lake water systems is basically completed in the city of China, but the sewage treatment plants, the drainage pipe networks and the urban river and lake water systems are usually independently operated and managed by different departments, and the support of a joint scheduling data scheme is lacking. Traditional urban water quantity and quality scheduling management carries out subjective scheduling according to actual conditions by means of scheduler experience, and the scheduling lacks a global field of view and has certain hysteresis. Therefore, the traditional scheduling mode is difficult to fully exert the flood control and drainage capacity of the factory network river system.
In recent years, urban flood disaster simulation is carried out by combining a rainfall flood numerical model, and a particle swarm and other strategy optimization algorithm is used for optimizing a scheduling scheme, so that the method becomes a mainstream method in the field of urban water quantity and quality scheduling management. However, when the rainfall flood numerical model simulates a large-scale complex pipe network, the calculation efficiency is low, the addition of an optimization algorithm further improves the calculation cost, the timeliness of the rainfall flood numerical model is often not effectively ensured, and the urban flood disaster problem caused by short-duration extremely-strong rainfall is difficult to effectively solve.
Disclosure of Invention
In order to improve the efficiency of acquiring a scheduling scheme, a plant-network-river joint scheduling method based on a machine learning algorithm and a rainfall flood numerical model is provided.
The invention solves the problems by adopting the following technical scheme:
a factory, net and river joint scheduling method based on a machine learning algorithm and a rainfall flood numerical model comprises the following steps:
step 2, calibrating SWMM model parameters based on target area early-stage monitoring data;
step 3, prefabricating a plurality of factory network river joint scheduling schemes;
step 4, constructing a city rain and flood scene database: based on the storm parameters of the target area and the Chicago rain pattern, constructing short-duration extremely-strong rainfall events with different reproduction periods to construct a city rainfall flood scenario database;
step 5, constructing a factory network river joint scheduling effect evaluation system;
step 6, constructing a urban factory, network and river joint scheduling scene database: combining the prefabricated multiple factory-net-river joint scheduling schemes in the step 3 with the short-duration extremely-strong rainfall events constructed in the step 4, and simulating based on the rated SWMM model to obtain joint scheduling scene data of various short-duration extremely-strong rainfall events under the conditions of different factory-net-river joint scheduling schemes; evaluating the combined schemes by adopting the plant-network-river joint scheduling effect evaluation system constructed in the step 5 to form a city plant-network-river joint scheduling scene database;
step 7, constructing a rainfall condition-joint scheduling scheme-benefit index learning model, wherein the model takes the rainfall condition and the benefit index as input conditions and takes the optimal joint scheduling scheme as an output target; model training is carried out based on the urban factory, network and river joint scheduling scene database constructed in the step 6;
and 8, acquiring an optimal joint scheduling scheme of the plant, the network and the river under the current rainfall condition by adopting a trained rainfall condition-joint scheduling scheme-benefit index learning model.
Further, the step 1 specifically includes:
step 11, simplifying a river channel into a communicated open channel and simplifying a lake into a regulation pool according to river-lake water system information of a target area;
and 12, connecting a discharge port of the pipeline for discharging to the river and the lake, and setting an outflow condition through the water levels of the river and the lake.
Further, the outflow condition in the step 12 is specifically:
when the river water level is smaller than the elevation of the drainage node, the drainage mode is a free outflow mode, and the flow can be expressed as:
wherein A is the cross-sectional area of the exhaust port; h 0 Is the head height; epsilon is the coefficient of lateral contraction; ζ is a local head loss coefficient; g is gravity acceleration;
when the river water level is higher than the discharge port, but the pipeline water head is higher than the river water head, the discharge port discharges the submerged outflow, and the flow can be expressed as:
wherein A is the cross-sectional area of the exhaust port; z is the difference between the pipeline water head and the river water head; epsilon is the coefficient of lateral contraction; ζ is a local head loss coefficient; g is gravity acceleration;
when the river water level is higher than the discharge port, and the pipeline water head is lower than the river water head, the water of the pipeline can not be discharged from the discharge port, and meanwhile, the water of the river can not flow backward into the pipe network.
Further, the short-duration extremely strong rainfall events in the step 4 are specifically 1h, 2h and 3h short-duration extremely strong rainfall events.
Further, in the step 5, the evaluation index is composed of a node accumulated overflow amount, an accumulated overflow node number, accumulated pollutant overflow and discharge amount, and pump station accumulated pumping and regulating water amount, and the above indexes are comprehensively considered by constructing an objective function; the constraint condition is that the maximum power of the pump station cannot exceed the rated power of the pump station, the maximum water inflow of the sewage treatment plant cannot exceed the rated flow of the pump station, the maximum running water level of the regulation and storage lake in the river basin water system cannot be exceeded, and the water level of the flood control and drainage river channel cannot exceed the maximum safe water level.
Further, the specific steps of constructing the rainfall condition-joint scheduling scheme-benefit index learning model are as follows:
step 71, calculating rainfall time sequence characteristic parameters based on rainfall conditions;
step 72, obtaining an optimal scheduling scheme under each rainfall condition in the urban factory, network and river joint scheduling scene database;
step 73, calculating the correlation between rainfall time series characteristic parameters and each optimal scheduling scheme, and screening out parameters with correlation coefficients larger than preset values as training parameters of a machine learning model;
step 74, dividing the training parameters into a training set D and a testing set according to a certain proportion;
step 75, establishing a machine learning model, and performing model training by adopting a training set; the machine learning model takes training set data as input data, and takes the opening degree of each gate, the running power of a pump station and the total inflow water amount of a sewage plant in a joint scheduling scheme as target data to carry out model training.
Further, the machine learning model in the step 75 includes a K-nearest neighbor model, a random forest model, and a limiting random tree model; and step 8, taking the scheduling scheme with the minimum objective function in the three model output schemes as a final optimal scheduling scheme.
Further, the step of training the K nearest neighbor model by adopting the training set comprises the following steps:
the training samples are expressed in a format of (x, f (x)), where x is a characteristic parameter of the sample, and x is represented by (x) 1 ,x 2 ,x 3 ,…,x n ) Constitution, wherein x n An nth attribute value for sample x; for a new input sample x i Calculating x one by using European distance formula i Distance from each sample in training set, and select x from the distance i K samples nearest to; the Euclidean distance is expressed as:
wherein x is i ，x j Two samples respectively;
respectively sample x i And x j Is the first eigenvalue of (c); l (x) j ,x j ) For sample x i And x j Distance between them.
Further, the step of training the random forest model by adopting the training set is as follows:
generating a plurality of parallel training groups through a boost trap resampling method, and independently training a decision tree model: first, the empirical entropy H (D) of the training set D is calculated:
the empirical conditional entropy H (d|a) of feature a on training set D is calculated:
calculating information gain:
g(D,A)＝H(D)-H(D|A)
calculating an information gain ratio:
wherein:
wherein D is the whole data set of the training set, A is the characteristic parameter, K is the total classification quantity, C k For the k-th class, n is the number of values of the characteristic parameter A;
selecting a characteristic parameter A with the maximum information gain ratio g As a node, for characteristic parameter A g Possible values { a } 1 ,a 2 ....,a n Sequentially according to A } g ＝a 1 ,…,A＝a n Dividing the training set D into D i ,D 2 ,…,D n Enter the next layer, with A- { A g And (3) the characteristic parameter set, repeating the steps until all the characteristic parameters are traversed, stopping, and outputting a decision tree model;
all the independently generated decision tree models are combined to construct a random forest model.
Further, the training set is adopted to train the limit random tree model, which comprises the following steps:
selecting all training sets D for model training, randomly selecting N characteristic parameters from the characteristic parameters A, randomly selecting one characteristic parameter as a splitting node, taking a splitting threshold with the smallest coefficient of the radix as an optimal splitting threshold, generating two child nodes by using data D_left and D_right, and traversing the rest parameters in sequence; the coefficient of the training set D can be expressed as:
the coefficient of the kurting D under the condition of the characteristic parameter a can be expressed as:
wherein D is the whole data set of the training set, A is the characteristic parameter, K is the total classification quantity, C k Class k, D 1 、D 2 Is two subsets divided according to feature a.
Compared with the prior art, the invention has the following beneficial effects: by simulating the scheduling effect of various combined scheduling modes under various storm conditions, a rainfall scene-combined scheduling scheme-benefit index learning model is further built by combining a machine learning algorithm, the rapid adaptation of the combined scheduling scheme to rainfall events is realized, the timeliness and the intelligence of the combined scheduling of the plant, the network and the river are improved, the rapid decision of management personnel in a short period is assisted, and the urban flood disaster loss is reduced.
Drawings
FIG. 1 is a flow chart of a plant-network-river joint scheduling method based on a machine learning algorithm and a rainfall flood numerical model;
FIG. 2 is a flow chart for constructing a joint scheduling context database of urban plants, networks and rivers;
FIG. 3 is a flow chart of the construction of a rainfall condition-joint scheduling scheme-benefit index learning model.
Detailed Description
The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
As shown in fig. 1, the plant-network-river joint scheduling method based on a machine learning algorithm and a rainfall flood numerical model comprises the following steps:
step S1: collecting and arranging basic data of a target area, wherein the basic data comprise topography, weather, drainage facilities and river and lake water system data, and the basic data comprise digital elevation data, land utilization data, typical storm sequences, drainage pipeline data, drainage node data, pump station information, sewage treatment plant information, river and lake water system distribution information and wading engineering facility information.
Step S2: collecting and arranging target area monitoring data including rainfall monitoring data, pipeline flow and water level monitoring data, pipeline discharge flow monitoring data, river water level flow monitoring data, lake water level monitoring data, pump station monitoring data and sewage treatment plant monitoring data.
Step S3: the SWMM model of the target area coupling sewage treatment plant, the urban pipe network and the urban river and lake water system is built, and the specific steps are as follows:
step S3.1: according to river and lake water system information, simplifying a river channel into a communicated open channel, and simplifying a lake into a regulating reservoir;
step S3.2: the discharge port of the pipeline for discharging the pipeline to the river and the lake is connected to the river and the lake, and the outflow condition is set through the water levels of the river and the lake, and the specific outflow rule is as follows:
1) When the river water level is smaller than the elevation of the drainage node, the drainage mode is a free outflow mode, and the flow can be expressed as:
wherein A is the cross-sectional area of the exhaust port; h 0 Is the head height; epsilon is the coefficient of lateral contraction; ζ is a local head loss coefficient; g is gravitational acceleration.
2) When the river water level is higher than the discharge port, but the pipeline water head is higher than the river water head, the discharge port discharges the submerged outflow, and the flow can be expressed as:
wherein A is the cross-sectional area of the exhaust port; z is the difference between the pipeline water head and the river water head; epsilon is the coefficient of lateral contraction; ζ is a local head loss coefficient; g is gravitational acceleration.
3) When the river water level is higher than the discharge port, and the pipeline water head is lower than the river water head, the throttle is opened by default, namely, the water of the pipeline cannot be discharged from the discharge port, and meanwhile, the water of the river cannot flow backward into the pipe network.
Step S4: calibrating SWMM model parameters based on target area early-stage monitoring data: based on the target area monitoring data collected in the step S2, taking rainfall monitoring data, river water level monitoring data, lake water level monitoring data, pump station monitoring data and sewage treatment plant monitoring data as model driving conditions, taking pipeline flow water level monitoring data and pipeline discharge flow monitoring data as verification information, calibrating SWMM model parameters, ensuring that the modeled model has higher reliability, and accurately inverting the storm flood-causing waterlogging process;
step S5: prefabricating a plurality of factory network river joint scheduling schemes: according to the number and positions of pump stations, sewage treatment plants and gates, a plurality of prefabricated scheduling schemes are set in advance, so that the scheduling schemes can be flexibly combined as much as possible to meet the joint scheduling requirements under various rainfall conditions.
Step S6: constructing a city rainfall flood scenario database, constructing short-duration extreme rainfall events with different reproduction periods based on the storm parameters of a research area and Chicago rainfall so as to fully cover various possible short-duration extreme rainfall events of a target area, and comprehensively considering the conditions of each rainfall; this example builds a 1h, 2h, 3h short duration extreme rainfall event.
Step S7: the method comprises the steps of constructing a plant-network-river joint scheduling effect evaluation system, wherein in the embodiment, benefit evaluation indexes of a plant-network-river joint scheduling scheme consist of node accumulated overflow quantity, accumulated overflow node quantity, accumulated pollutant overflow and discharge quantity and pump station accumulated pumping water quantity, and comprehensively considering the indexes by constructing an objective function; the constraint condition is that the maximum power of the pump station cannot exceed the rated power of the pump station, the maximum water inflow of the sewage treatment plant cannot exceed the rated flow of the pump station, the maximum running water level of the regulation and storage lake in the river basin water system cannot exceed the maximum safe water level, and further the effect of the scheduling scheme is comprehensively evaluated, and the scheduling scheme is optimized. Other indexes can be selected for evaluation according to actual needs, and the evaluation is not limited herein.
The benefit evaluation index of the factory network river scheduling scheme is specifically as follows:
1) Node accumulated overflow quantity: the accumulated overflow quantity of the nodes is used for measuring the surface submerged water quantity and is expressed by the following formula:
q in it The overflow quantity of the ith node in the unit time of T time is represented by delta T, N is the number of accumulated nodes, and T is the accumulated simulation time.
2) Accumulating the number of overflow nodes: the number of the accumulated overflow nodes is used for measuring the flooding range of the earth surface and is expressed by the following formula:
q in it The overflow quantity of the ith node in the unit time of T time is represented by delta T, N is the number of accumulated nodes, and T is the accumulated simulation time.
3) Accumulating pollutant overflow and discharge amount: the accumulated pollutant overflow and discharge amount is used for measuring the control degree of the pollutant, and when the accumulated pollutant overflow and discharge amount is high, the control effect of the pollutant is poor, and the accumulated pollutant overflow and discharge amount is expressed by the following formula:
q in it Is the overflow quantity of the ith node in the unit time of t time, out jt C is the overflow quantity of the jth outlet in unit time at t time it And c jt The pollutant concentrations of the node i and the discharge port j at the moment T are respectively, deltat is a time step, N and M are respectively the accumulated node number and the accumulated discharge port number, and T is accumulated simulation time.
4) Accumulated pumping and regulating water quantity of pump station: the accumulated pumping capacity of the pump station is used for considering the main energy consumption in the scheduling process and is expressed by the following formula:
q in kt The overflow quantity of the ith node in the unit time of T time is represented by delta T, N is the number of accumulated nodes, and T is the accumulated simulation time.
The optimal objective function is set as:
wherein:
w 1 +w 2 +w 3 +w 4 ＝1
in the middle of
Respectively representing the normalized values of the node accumulated overflow quantity, the accumulated overflow node quantity, the accumulated pollutant overflow quantity and the discharge quantity and the pump station accumulated pumping and regulating water quantity, and w 1 、w 2 、w 3 、w 4 For their respective pairsWeight parameters for the application.
The constraint conditions are that the maximum power of the pump station cannot exceed the rated power of the pump station, the maximum inflow flow of the sewage treatment plant cannot exceed the rated flow of the pump station, the maximum running water level of the regulation and storage lake in the river basin water system cannot be exceeded, the water level of the flood control and drainage river channel cannot exceed the maximum safe water level, and the method can be expressed as follows:
wherein Z is j For the water level of the j-th river in the flood control zone,
for the maximum safe water level determined according to the requirements of flood control and drainage in urban areas, the water level is->
The maximum safe water level is determined according to flood control and drainage requirements.
Step S8: constructing a city factory network river joint scheduling scene database: combining the prefabricated multiple factory, net and river joint scheduling schemes proposed in the step S5 and the short-duration extremely-strong rainfall events proposed in the step S6, simulating by the SWMM model rated and completed in the step S4 to obtain joint scheduling scene data of various extremely-strong rainfall events under different scheduling scheme situations, calculating joint scheduling effect evaluation indexes proposed in the step S7, and arranging the joint scheduling effect evaluation indexes to form a city factory, net and river joint scheduling scene database; the flow chart is shown in fig. 2, where N is the number of rainfall events and M is the number of joint scheduling schemes.
Step S9: building a rainfall condition-joint scheduling scheme-benefit index learning model: in this embodiment, a model is built based on a machine learning algorithm, as shown in fig. 3, 3 machine learning algorithms based on K nearest neighbors, random forests and extreme random trees are used to build a model between rainfall condition-joint scheduling scheme-benefit index, the model uses rainfall condition as input condition, and the optimal joint scheduling scheme corresponding to the rainfall condition as output target, and other algorithms and combinations thereof can be selected according to actual needs, without limitation. The specific steps in this embodiment include:
step S9.1: constructing a python program get_rain_parameter, and calculating rainfall time sequence characteristic parameters based on rainfall conditions, wherein the rainfall time sequence characteristic parameters comprise accumulated rainfall duration, accumulated rainfall before peak value, maximum 1-hour rainfall, maximum 2-hour rainfall and maximum 4-hour rainfall;
step S9.2: calculating the benefit evaluation index proposed in the step S7 in a statistics way, and reserving an optimal scheduling scheme under each rainfall condition;
step S9.3: analyzing the correlation between the rainfall characteristic parameters in the step S9.1 and the joint scheduling scheme by a pearson correlation coefficient method, and screening out parameters with correlation coefficients larger than 0.4 as training parameters of a machine learning model, wherein the pearson correlation coefficients can be calculated by the following formula:
wherein ρ is xy The pearson correlation coefficient between the parameter x and the parameter y; e (x), E (y), E (xy) are mathematical expectations of the parameters x, y and xy respectively; sigma (sigma) x ，σ y The variance of the parameter x and the parameter y; cov (x, y) is the covariance between parameter x and parameter y.
Step S9.4: after the feature parameter data screened in the step S9.3 is processed by parameter normalization, the data set is divided into a training set D and a test set according to the ratio of 9:1, and the normalization processing mode in this embodiment selects linear normalization, which can be represented by the following formula:
wherein x' is a parameter value after normalization processing, x is a parameter value before normalization processing, min (x) is a minimum value in the parameter x, and max (x) is a maximum value in the parameter x;
step S9.5: the training set obtained in the step S9.4 is used for K neighbor model training, the normalization parameters obtained in the step S9.3 are used as input data, and the opening degree of each gate, the running power of a pump station and the total water inflow amount of a sewage plant in a joint scheduling scheme are used as target data for model training, and the specific steps are as follows:
representing training samples in a format of (x, f (x)); where x is a characteristic parameter of the sample, x is defined by (x 1 ,x 2 ,x 3 ,…,x n ) Constitution, wherein x n For the nth attribute value of sample x, i.e. the number of characteristic parameters is equal to the dimension of the vector composition, for a new input sample x i Calculating x one by using European distance formula i Distance from each sample in training set, and select x from the distance i The K samples closest to the sample. The Euclidean distance can be expressed as:
wherein x is i ，x j Two samples respectively;
respectively sample x i And x j Is the first eigenvalue of (c); l (x) j ,x j ) For sample x i And x j Distance between them.
Step S9.6: inputting the test set data obtained in the step S9.4 into a K nearest neighbor model, and outputting an optimal scheme predicted by the K nearest neighbor model;
step S9.7: the training set obtained in the step S9.4 is used for random forest model training, the normalization parameters obtained in the step S9.3 are used as input data, and the opening degree of each gate, the running power of a pump station and the total water inflow amount of a sewage plant in a joint scheduling scheme are used as target data for model training, and the specific steps are as follows:
and generating a plurality of parallel training groups by a boost trap resampling method, and independently training a decision tree model. First, the empirical entropy H (D) of the training set D is calculated:
the empirical conditional entropy H (d|a) of feature a on training set D is calculated:
calculating information gain:
g(D,A)＝H(D)-H(D|A)
calculating an information gain ratio:
wherein:
wherein D is the whole data set of the training set, A is the characteristic parameter, K is the total classification number, ck is the kth class, and n is the number of the characteristic parameter A.
Selecting a characteristic parameter A with the maximum information gain ratio g As a node, for characteristic parameter A g Possible values { a } 1 ,a 2 ....,a n Sequentially according to A } g ＝a 1 ,…,A＝a n Dividing the training set D into D i ,D 2 ,…,D n Enter the next layer, with A- { A g And (3) the characteristic parameter set, repeating the steps until all the characteristic parameters are traversed, stopping, and outputting the decision tree model. And further combining all the independently generated decision tree models to construct a random forest model, and outputting a final joint scheduling optimization scheme in a voting mode by considering the result of each decision tree when a new sample needs to be predicted.
Step S9.8: performing limit random tree model training by using the training set obtained in the step S9.4, performing model training by using the normalized parameters obtained in the step S9.3 as input data and using the opening degree of each gate, the running power of a pump station and the total water inflow amount of a sewage plant in a joint scheduling scheme as target data, wherein the specific steps are as follows:
and selecting all training sets D for model training, randomly selecting N characteristic parameters from the characteristic parameters A, randomly selecting one characteristic parameter as a splitting node, taking a splitting threshold with the smallest coefficient of the radix as an optimal splitting threshold, generating two child nodes by using the data D_left and D_right, and traversing the rest parameters in sequence.
The coefficient of the training set D can be expressed as:
the coefficient of the kurting D under the condition of the characteristic parameter a can be expressed as:
wherein D is the whole data set of the training set, A is the characteristic parameter, K is the total classification quantity, C k Class k, D 1 、D 2 Is two subsets divided according to feature a.
Step S10: and (3) verifying the effect of the modeled model, verifying the effect of the model constructed in the step S9 through test set data, if the accuracy requirement is met, storing the model, otherwise, returning to the step 9, and carrying out model training again.
Step S11, carrying out factory-net-river joint scheduling prediction: and driving the model stored in the step S10 according to the real-time rainfall meteorological data, and taking a scheduling scheme with the minimum objective function in the three model output schemes as a final optimal scheduling scheme.
Claims (10)
1. The factory, net and river joint scheduling method based on the machine learning algorithm and the rainfall flood numerical model is characterized by comprising the following steps of:
step 1, establishing an SWMM model of a target area coupling sewage treatment plant, an urban pipe network and an urban river and lake water system;
step 2, calibrating SWMM model parameters based on target area early-stage monitoring data;
step 3, prefabricating a plurality of factory network river joint scheduling schemes;
step 4, constructing a city rain and flood scene database: based on the storm parameters of the target area and the Chicago rain pattern, constructing short-duration extremely-strong rainfall events with different reproduction periods to construct a city rainfall flood scenario database;
step 5, constructing a factory network river joint scheduling effect evaluation system;
step 6, constructing a urban factory, network and river joint scheduling scene database: combining the prefabricated multiple factory-net-river joint scheduling schemes in the step 3 with the short-duration extremely-strong rainfall events constructed in the step 4, and simulating based on the rated SWMM model to obtain joint scheduling scene data of various short-duration extremely-strong rainfall events under the conditions of different factory-net-river joint scheduling schemes; evaluating the combined schemes by adopting the plant-network-river joint scheduling effect evaluation system constructed in the step 5 to form a city plant-network-river joint scheduling scene database;
step 7, constructing a rainfall condition-joint scheduling scheme-benefit index learning model, wherein the model takes the rainfall condition and the benefit index as input conditions and takes the optimal joint scheduling scheme as an output target; model training is carried out based on the urban factory, network and river joint scheduling scene database constructed in the step 6;
and 8, acquiring an optimal joint scheduling scheme of the plant, the network and the river under the current rainfall condition by adopting a trained rainfall condition-joint scheduling scheme-benefit index learning model.
2. The method for joint scheduling of plant and river based on machine learning algorithm and rainfall flood numerical model according to claim 1, wherein the step 1 specifically comprises:
step 11, simplifying a river channel into a communicated open channel and simplifying a lake into a regulation pool according to river-lake water system information of a target area;
and 12, connecting a discharge port of the pipeline for discharging to the river and the lake, and setting an outflow condition through the water levels of the river and the lake.
3. The method for joint scheduling of the plant and the river based on the machine learning algorithm and the rainfall flood numerical model according to claim 2, wherein the outflow condition in the step 12 is specifically:
when the river water level is smaller than the elevation of the drainage node, the drainage mode is a free outflow mode, and the flow can be expressed as:
wherein A is the cross-sectional area of the exhaust port; h 0 Is the head height; epsilon is the coefficient of lateral contraction; ζ is a local head loss coefficient; g is gravity acceleration;
when the river water level is higher than the discharge port, but the pipeline water head is higher than the river water head, the discharge port discharges the submerged outflow, and the flow can be expressed as:
wherein A is the cross-sectional area of the exhaust port; z is the difference between the pipeline water head and the river water head; epsilon is the coefficient of lateral contraction; ζ is a local head loss coefficient; g is gravity acceleration;
when the river water level is higher than the discharge port, and the pipeline water head is lower than the river water head, the water of the pipeline can not be discharged from the discharge port, and meanwhile, the water of the river can not flow backward into the pipe network.
4. The method for joint scheduling of the plant and river based on the machine learning algorithm and the rainfall flood numerical model according to claim 1, wherein the short duration extremely strong rainfall events in the step 4 are specifically 1h, 2h and 3h short duration extremely strong rainfall events.
5. The method for joint scheduling of plant, network and river based on machine learning algorithm and rainfall flood numerical model according to claim 1, wherein in step 5, the evaluation index consists of node accumulated overflow amount, accumulated overflow node number, accumulated pollutant overflow and discharge amount, pump station accumulated water pumping amount, and the above index is comprehensively considered by constructing objective function; the constraint condition is that the maximum power of the pump station cannot exceed the rated power of the pump station, the maximum water inflow of the sewage treatment plant cannot exceed the rated flow of the pump station, the maximum running water level of the regulation and storage lake in the river basin water system cannot be exceeded, and the water level of the flood control and drainage river channel cannot exceed the maximum safe water level.
6. The method for joint scheduling of plant, network and river based on machine learning algorithm and rainfall flood numerical model according to claim 5, wherein the specific steps of constructing the rainfall condition-joint scheduling scheme-benefit index learning model are as follows:
step 71, calculating rainfall time sequence characteristic parameters based on rainfall conditions;
step 72, obtaining an optimal scheduling scheme under each rainfall condition in the urban factory, network and river joint scheduling scene database;
step 73, calculating the correlation between rainfall time series characteristic parameters and each optimal scheduling scheme, and screening out parameters with correlation coefficients larger than preset values as training parameters of a machine learning model;
step 74, dividing the training parameters into a training set D and a testing set according to a certain proportion;
step 75, establishing a machine learning model, and performing model training by adopting a training set; the machine learning model takes training set data as input data, and takes the opening degree of each gate, the running power of a pump station and the total inflow water amount of a sewage plant in a joint scheduling scheme as target data to carry out model training.
7. The method for joint scheduling of plant and river based on machine learning algorithm and rainfall flood numerical model according to claim 6, wherein the machine learning model in step 75 comprises a K-nearest neighbor model, a random forest model and a limit random tree model; and step 8, taking the scheduling scheme with the minimum objective function in the three model output schemes as a final optimal scheduling scheme.
8. The method for joint scheduling of a plant and a river based on a machine learning algorithm and a rainfall flood numerical model according to claim 7, wherein the step of training the K-nearest neighbor model by using a training set is as follows:
the training samples are expressed in a format of (x, f (x)), where x is a characteristic parameter of the sample, and x is represented by (x) 1 ,x 2 ,x 3 ,…,x n ) Constitution, wherein x n An nth attribute value for sample x; for a new input sample x i Calculating x one by using European distance formula i Distance from each sample in training set, and select x from the distance i K samples nearest to; the Euclidean distance is expressed as:
wherein x is i ，x j Two samples respectively;
respectively sample x i And x j Is the first eigenvalue of (c); l (x) j ,x j ) For sample x i And x j Distance between them.
9. The method for joint scheduling of plant, network and river based on machine learning algorithm and rainfall flood numerical model according to claim 7, wherein the step of training the random forest model by using training set is:
generating a plurality of parallel training groups through a boost trap resampling method, and independently training a decision tree model: first, the empirical entropy H (D) of the training set D is calculated:
the empirical conditional entropy H (d|a) of feature a on training set D is calculated:
calculating information gain:
g(D,A)＝H(D)-H(D|A)
calculating an information gain ratio:
wherein:
wherein D is the whole data set of the training set, A is the characteristic parameter, K is the total classification quantity, C k For the k-th class, n is the number of values of the characteristic parameter A;
selecting a characteristic parameter A with the maximum information gain ratio g As a node, for characteristic parameter A g Possible values { a } 1 ,a 2 ....,a n Sequentially according to A } g ＝a 1 ,…,A＝a n Dividing the training set D into D i ,D 2 ,…,D n Enter the next layer, with A- { A g And (3) the characteristic parameter set, repeating the steps until all the characteristic parameters are traversed, stopping, and outputting a decision tree model;
all the independently generated decision tree models are combined to construct a random forest model.
10. The method for joint scheduling of plant and river based on machine learning algorithm and rainfall flood numerical model according to claim 7, wherein the step of training the limit random tree model by using training set is as follows:
selecting all training sets D for model training, randomly selecting N characteristic parameters from the characteristic parameters A, randomly selecting one characteristic parameter as a splitting node, taking a splitting threshold with the smallest coefficient of the radix as an optimal splitting threshold, generating two child nodes by using data D_left and D_right, and traversing the rest parameters in sequence; the coefficient of the training set D can be expressed as:
the coefficient of the kurting D under the condition of the characteristic parameter a can be expressed as:
wherein D is the whole data set of the training set, A is the characteristic parameter, K is the total classification quantity, C k Class k, D 1 、D 2 Is two subsets divided according to feature a.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211547299.3A CN116011731A (en) | 2022-12-05 | 2022-12-05 | Factory, net and river joint scheduling method based on machine learning algorithm and rainfall flood numerical model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211547299.3A CN116011731A (en) | 2022-12-05 | 2022-12-05 | Factory, net and river joint scheduling method based on machine learning algorithm and rainfall flood numerical model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116011731A true CN116011731A (en) | 2023-04-25 |
Family
ID=86030803
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211547299.3A Pending CN116011731A (en) | 2022-12-05 | 2022-12-05 | Factory, net and river joint scheduling method based on machine learning algorithm and rainfall flood numerical model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116011731A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116738874A (en) * | 2023-05-12 | 2023-09-12 | 珠江水利委员会珠江水利科学研究院 | Gate pump group joint optimization scheduling method based on Multi-Agent PPO reinforcement learning |
| CN117828489A (en) * | 2024-03-05 | 2024-04-05 | 河钢国际科技(北京)有限公司 | Intelligent ship remote dynamic control system |
-
2022
- 2022-12-05 CN CN202211547299.3A patent/CN116011731A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116738874A (en) * | 2023-05-12 | 2023-09-12 | 珠江水利委员会珠江水利科学研究院 | Gate pump group joint optimization scheduling method based on Multi-Agent PPO reinforcement learning |
| CN116738874B (en) * | 2023-05-12 | 2024-01-23 | 珠江水利委员会珠江水利科学研究院 | Gate pump group joint optimization scheduling method based on Multi-Agent PPO reinforcement learning |
| CN117828489A (en) * | 2024-03-05 | 2024-04-05 | 河钢国际科技(北京)有限公司 | Intelligent ship remote dynamic control system |
| CN117828489B (en) * | 2024-03-05 | 2024-05-14 | 河钢国际科技(北京)有限公司 | Intelligent ship remote dynamic control system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116011731A (en) | Factory, net and river joint scheduling method based on machine learning algorithm and rainfall flood numerical model | |
| Mehta et al. | Optimal operation of a multi-purpose reservoir using neuro-fuzzy technique | |
| CN113887787B (en) | Flood forecast model parameter multi-objective optimization method based on long-short-term memory network and NSGA-II algorithm | |
| CN117035201B (en) | Multi-target scheduling rule making method and system for plain river network water engineering cluster | |
| CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
| CN103729550A (en) | Multi-model integrated flood forecasting method based on propagation time clustering analysis | |
| CN115758886A (en) | Optimal distribution method of storage tanks based on rain flood pipe network model and decision tree algorithm | |
| CN111737853A (en) | Low-impact development multi-target interval optimization configuration method based on SWMM model | |
| CN113705877A (en) | Real-time monthly runoff forecasting method based on deep learning model | |
| CN114777192B (en) | Secondary network heat supply autonomous optimization regulation and control method based on data association and deep learning | |
| CN104732300A (en) | Neural network wind power short-term forecasting method based on fuzzy partition theory | |
| CN114429053A (en) | Basin scale WEFE system adaptability simulation optimization method | |
| CN109242265A (en) | Based on the smallest Urban Water Demand combination forecasting method of error sum of squares | |
| CN116050628A (en) | Reservoir flood control optimization scheduling method considering forecast uncertainty based on digital twinning | |
| Nourani et al. | A new hybrid algorithm for rainfall–runoff process modeling based on the wavelet transform and genetic fuzzy system | |
| CN107766694A (en) | Hydro-Model Parameter Calibration Technology method of estimation based on FCM NSGA II | |
| CN115186857A (en) | Neural network reservoir water level prediction method based on ensemble learning | |
| CN110276477A (en) | A kind of Flood Forecasting Method based on layering Bayesian network and incremental learning | |
| CN114936505B (en) | Method for rapidly forecasting multi-point water depth of urban rainwater well | |
| CN113435630A (en) | Basin hydrological forecasting method and system with self-adaptive runoff yield mode | |
| CN114580762A (en) | Hydrological forecast error correction method based on XGboost | |
| CN117575289B (en) | Reservoir optimal scheduling method with water quantity-water quality multi-objective cooperation | |
| CN114862073A (en) | Method for forecasting medium and long term runoff by four-dimensional coupling of reservoir water of air and land | |
| CN116485584A (en) | Method and system for cooperative regulation and control of WEE of river with large bottom slope in alpine region | |
| Shuofeng et al. | LSTM based hybrid method for basin water level prediction by using precipitation data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |