An auxiliary decision-making method for cleaning water supply pipelines based on neural network

Abstract

The present invention relates to the field of water supply pipeline cleaning, and particularly to an auxiliary decision-making method for water supply pipeline cleaning based on a neural network. The present invention provides an auxiliary decision-making method for water supply pipeline cleaning technology based on a neural network. This method can adapt to spatio-temporal changes, be specific to the pipeline level, couple the assessment of environmental impacts and economic benefits, provide scientific decision-making basis and decision-making support for management decision-makers, establish a scientific comprehensive decision-making system for cleaning water supply pipelines with ice slurry and gas-water pulse cleaning, quantitatively evaluate, and effectively improve the comprehensive benefits of water supply pipeline cleaning.

Description

Technical Field

[0001] The present invention relates to the field of water supply pipeline cleaning, and particularly to an auxiliary decision-making method for water supply pipeline cleaning based on a neural network. Background Art

[0002] During the long-term operation of water supply pipelines, sediments, pipe scale, and biofilms are likely to form on the inner wall of the pipes and gradually turn into "growth rings", which significantly affect the water quality of the water supply and are the main cause of "yellow water"; the thickness of the "growth ring" increases year by year, which will also lead to rough inner walls of the pipes, reduce the cross-sectional area of the water flow, and cause a decrease in the water pressure at the user end. To meet the needs of users for high-quality drinking water, water supply companies gradually replace, repair, and clean water supply pipelines. The "Technical Regulations for the Operation, Maintenance, and Safety of Urban Water Supply Pipe Networks" (CJJ 207-2013) stipulates that cleaning and disinfection should be carried out before the use of new pipes, before and after pipe repair, and regular cleaning should be carried out for operating pipes. According to statistics, the length of urban water supply pipelines in China reached 1.103 million kilometers in 2022. In 2020 alone, 2,289 km of water supply pipelines were newly built and renovated in Beijing, indicating that the length of water supply pipeline cleaning is very considerable.

[0003] The commonly used method for cleaning water supply pipelines is one-way flushing. This method is simple to operate and only requires opening and closing specific water supply pipeline valves, but it consumes a large amount of water. For example, the cleaning of the water supply pipelines in the urban area of Beijing in 2015 consumed 13 million m 3 of water, which is equivalent to the water volume of 6.5 Kunming Lakes. At the same time, the electricity required to produce one-way flushing water, the CO 2 emissions, and the costs incurred are very high. The emerging ice slurry cleaning and gas-water pulse technologies have the advantages of saving water consumption, short cleaning time, and good cleaning effect, but new equipment needs to be invested.

[0004] Currently, whether to introduce ice slurry cleaning and gas-water pulse technologies at the level of a certain pipe network / region / country still relies on experience for decision-making, lacking quantitative evaluations of the water consumption, electricity consumption, carbon footprint, and costs of these two technologies. At the same time, the water supply pipe network is a complex system containing pipelines with different diameters and lengths, and the water supply pipe networks in each city are also different. After introducing new technologies, which diameter of pipelines and how long of pipelines should be cleaned using ice slurry cleaning and gas-water pulse cleaning respectively are still blank in the industry. The water supply pipe networks in various cities or regions in China are huge in scale, and the conditions of water supply pipelines in different regions are also different. Therefore, it is difficult to make cleaning decisions one by one. In addition, since the pipeline system needs to be updated and replaced on a large scale over time, the costs of each cleaning plan also vary with different water prices and electricity prices, and the traditional static decision-making model can no longer keep up with the changes. Therefore, it is crucial to develop a dynamic auxiliary decision-making model for water supply pipeline cleaning that adapts to spatio-temporal changes.

[0005] In the context of the booming development of the water supply pipeline cleaning industry, the construction of a water-saving society, and the "dual carbon" goal, it is essential to formulate solutions for ice slurry cleaning and gas-water pulse hybrid cleaning of water supply pipe networks in the future. Quickly determining a pipeline cleaning solution that is technically feasible, low-carbon and environmentally friendly, economically reasonable, and applicable to all pipe networks is an urgent problem that the current industry needs to solve. Summary of the Invention

[0006] To solve the above problems, the present application provides an auxiliary decision-making method for water supply pipeline cleaning based on a neural network, including the following steps:

[0007] S1: Set pipeline characteristic parameters and cleaning decision-making parameters, and construct a virtual pipe network and corresponding cleaning decision-making solutions; collect the water consumption, electricity consumption, required alternative water sources, and costs of cleaning equipment and consumables for one-way flushing, ice slurry cleaning, and gas-water pulse cleaning of 1 km of pipelines in pipelines with different diameters; calculate the water consumption, electricity consumption, costs of all cleaning decision-making solutions, and the carbon footprint reduction compared to traditional one-way flushing.

[0008] S2: Use the water consumption, electricity consumption, and costs of the cleaning decision-making solutions as input indicators, and the carbon footprint reduction (compared to traditional one-way flushing) as the output indicator, and use DEA (Data Envelopment Analysis) to calculate the technical benefits, scale benefits, and comprehensive benefits of all solutions.

[0009] S3: Use the pipeline characteristic parameters and cleaning decision-making parameters of all cleaning decision-making solutions as input factors, and use the comprehensive benefits calculated in S2 as output factors, normalize the input factors, and use the K-fold cross-validation method to train the neural network model. Use the ReLU function as the activation function and use the mean squared error (MSE) as the optimization objective of the model.

[0010] S4: When formulating a cleaning decision-making solution for an actual pipe network, obtain the pipeline characteristic parameters of the actual pipe network to be cleaned in the target area, select a series of cleaning decision-making parameters, and use the neural network model trained in S3 to calculate the comprehensive benefits of the cleaning solution in the target area. Decision-makers can compare several solutions as needed and select the solution with the highest comprehensive benefits and the strongest feasibility to clean the water supply pipe network.

[0011] Preferably, the S1 sets pipeline characteristic parameters and cleaning decision-making parameters, constructs a virtual pipe network and corresponding cleaning decision-making solutions; collects the water consumption, electricity consumption, required alternative water sources, and costs of cleaning equipment and consumables for one-way flushing, ice slurry cleaning, and gas-water pulse cleaning of 1 km of pipelines in pipelines with different diameters; calculates the water consumption, electricity consumption, costs of all cleaning decision-making solutions, and the carbon footprint reduction compared to traditional one-way flushing. It includes the following steps:

[0012] Step 1, construct a virtual pipe network and set pipe characteristic parameters. Assume that the total pipe length of the pipe network to be cleaned is 1 km, and assume that there are n types of pipe diameters in the entire pipe network, which are D 1 , D 2 ... D n , and the pipe lengths corresponding to the pipes of each pipe diameter are a 1 , a 2 ... a n (km);

[0013] Step 2, formulate corresponding cleaning decision-making plans and set cleaning decision parameters. For the pipes of the i-th pipe diameter, assume that the pipe length cleaned by ice slurry / the total pipe length of the pipes of this diameter is b i (i = 1, 2, 3... n), and the pipe length of gas-water pulse / the total pipe length of the pipes of this diameter is 1 - b i (i = 1, 2, 3... n);

[0014] Step 3, assume that a n has 5 possible values, and the values of a n are: 0, 0.1, 0.2, 0.3, 0.4, and satisfy

[0015] Step 4, assume that b n has 6 possible values, and the values of b n are: b i = 0, 0.2, 0.4, 0.6, 0.8, 1.0 (i = 1, 2, 3... n).

[0016] Step 5, assume that the water price has 4 possible values, and the water price r w (the cost of each 1 m 3 of water, yuan), and the values of r w are: 1, 5, 10, 20.

[0017] Step 6, assume that the electricity price has 4 possible values, and the electricity price r e (the cost of each 1 kwh of electricity, yuan), and the values of r e are: 0.1, 0.5, 1.0, 2.0.

[0018] Step 7, gather all the cleaning plans for the water supply pipes in this virtual pipe network to form a total of m plans.

[0019] Step 8, collect the water consumption, electricity consumption, required alternative water sources, and the costs of cleaning equipment and consumables for one-way flushing, ice slurry cleaning, and gas-water pulse cleaning of 1 km of pipes in pipes of different diameters.

[0020] Step 9, in the k-th plan, calculate the water consumption W k (m 3), k = 1, 2... m.

[0021] Step 10, in the k-th scheme, calculate the electricity consumption E k (kwh) for cleaning the water supply pipeline in the k-th scheme, k = 1, 2... m.

[0022] Step 11, in the k-th scheme, calculate the cost C k (yuan) for cleaning the water supply pipeline in the k-th scheme, k = 1, 2... m.

[0023] Step 12, use Simapro software to calculate the carbon footprint of unidirectional flushing, ice slurry cleaning, and gas-water pulse cleaning for 1 km of pipeline. Assuming that residents use bottled mineral water as an alternative water source, the required water volume is 30% of the daily water supply. When the diameter of the pipeline to be cleaned is greater than or equal to 400 mm, it is assumed that the pipeline is cleaned at night, and the carbon footprint of the alternative water source is not calculated.

[0024]

[0025]

[0026]

[0027] Among them and are respectively the carbon footprints (kgCO 2 ) generated by unidirectional flushing, ice slurry cleaning, and gas-water pulse cleaning of 1 km of pipeline with the i-th pipe diameter in the k-th scheme, f aw , f w , f t and f e are respectively the carbon footprints (kgCO 2 / bottle) generated by each bottle of alternative water source, the carbon footprint (kgCO 3 / m 2 ) generated by producing every 1 m 3 of tap water, the carbon footprint (kgCO 3 / m 2 ) generated by treating every 1 m 3 of cleaning wastewater discharge, and the carbon footprint (kgCO 2 / kwh) generated by using every 1 kwh of electricity. When the diameter of the pipeline to be cleaned is greater than or equal to 400 mm, aw ui , aw pi and aw si are assumed to be 0;

[0028] Step 13, calculate the reduced carbon footprint F k (kgCO 2 ) of the k-th pipeline cleaning scheme compared with traditional unidirectional flushing, k = 1, 2... m;

[0029]

[0030] Preferably, S2 takes the water consumption, power consumption, and cost of the cleaning decision plan as input indicators, and the reduced carbon footprint (compared with traditional one-way flushing) as the output indicator, and uses DEA (Data Envelopment Analysis) to calculate the technical efficiency, scale efficiency, and comprehensive efficiency of all plans. It includes the following steps:

[0031] Step 14, take W k , E k and C k as input indicators, and F k as the output indicator, and establish a DEA data envelopment analysis model.

[0032] Step 15, calculate the technical efficiency, scale efficiency, and comprehensive efficiency (ef k ) of the m plans; among them, the technical efficiency is an indicator to measure whether the performance of the water supply pipeline cleaning plan can be improved in terms of technical level, the scale efficiency is an indicator to measure whether the cleaning scale can be expanded to improve the performance of the water supply pipeline cleaning plan, and the comprehensive efficiency is a comprehensive consideration indicator of technical efficiency and scale efficiency.

[0033] Preferably, S3 takes the pipeline characteristic parameters and decision parameters of all cleaning decision plans as input factors, and the comprehensive efficiency as the output factor, normalizes the input factors, and uses the K-fold cross-validation method to train the neural network model. Use the ReLU function as the activation function and the mean square error (MSE) as the optimization objective of the model. It includes the following steps:

[0034] Step 16, for the k-th plan, take as the input factor and ef k as the output factor to form the k-th neural network training data pair {z k , ef k}, k = 1, 2...m; a total of m neural network training data pairs are formed.

[0035] Step 17, for the convenience of fast convergence in neural network training, perform Min-Max normalization processing on all input factors, and convert the values of the input factors to the range of [0, 1].[[]END]

[0036] Step 18, use the formed m neural network training data pairs to train the neural network model.

[0037] Step 19, this neural network model includes 1 hidden layer, uses the ReLU function as the activation function, and uses the mean square error (MSE) as the optimization objective of the model.

[0038] Step 20: Update the model parameters using stochastic gradient descent optimization with a learning rate of 0.3.

[0039] Step 21: To evaluate the performance of the model and select the best model, the K-fold cross-validation method is adopted, and 10 folds are used in the present invention.

[0040] Step 22: Conduct multiple iterations in each fold, with a total of 100,000 iterations.

[0041] Step 23: During the training process of each fold, calculate and record the loss value on the validation set as a model performance metric. Finally, select the model with the best performance among all folds and use it to predict the test data to obtain the final prediction result.

[0042] Preferably, when formulating a cleaning decision for the actual pipe network, obtain the pipeline characteristic parameters of the actual pipe network to be cleaned in the target area, select a series of cleaning decision parameters, and calculate the comprehensive benefit of the cleaning plan for the target area using the neural network model trained in S3. The decision maker can compare several plans as needed and select the plan with the highest comprehensive benefit and the strongest feasibility to clean the water supply pipe network. It includes the following steps:

[0043] Step 24: When formulating a decision-making plan for the actual water supply pipe network, obtain the pipeline characteristic parameters of the pipe network to be cleaned, find all the pipe diameters, and obtain the number of pipe diameters n (i.e., there are n different pipe diameters in this pipe network).

[0044] Step 25: Obtain the length ai of the pipeline with the i-th pipe diameter i (i = 1, 2, 3…n).

[0045] Step 26: Select the cleaning decision parameters: the ice slurry flushing ratio bi i (i = 1, 2, 3…n), the number of ice slurry cleaning devices qi p , the number of air-water pulse devices qi s , the water price r w , the electricity price r e , and form a cleaning decision plan;

[0046] Step 27: Input z = (a 1 , a 2 , a 3 ...a n , b 1 , b 2 , b 3 ...b n , q p , q s , r w , r e ) into the trained neural network model to obtain the comprehensive benefit ef of this plan.

[0047] Step 28: According to the needs of the decision maker, compare several options. The higher the comprehensive benefit of an option, the higher its feasibility. The decision maker can select the option with the highest comprehensive benefit for the actual cleaning of the water supply pipeline network.

[0048] The present invention provides an auxiliary decision-making method for water supply pipeline cleaning technology based on a neural network. This method can adapt to spatio-temporal changes, be specific to the pipeline level, couple the evaluation of environmental impacts and economic benefits, provide a scientific decision-making basis and decision-making support for management decision makers, establish a scientific comprehensive decision-making system for ice slurry cleaning and air-water pulse cleaning of water supply pipelines, quantitatively evaluate, and effectively improve the comprehensive benefit of water supply pipeline cleaning. Brief Description of the Drawings

[0049] Figure 1 It is a schematic diagram of the method flow described in the embodiment of the present invention.

[0050] Figure 2 It is the mean square error (MSE) of the neural network model classified by pipe diameter in the embodiment of the present invention. Detailed Embodiment

[0051] The following makes a detailed description of the embodiments of the present invention. This embodiment is carried out based on the technical solution of the present invention, and gives a detailed implementation method and specific operation process. Referring to Figure 1 as shown, the technical solution of the present invention is further explained.

[0052] Step 1: Set pipeline characteristic parameters and cleaning decision parameters, and construct a virtual pipe network and corresponding cleaning decision plans. Collect the water consumption, electricity consumption, required alternative water sources, and costs of cleaning equipment and consumables for one-way flushing, ice slurry cleaning, and air-water pulse cleaning of 1 km of pipeline in pipelines with different diameters. Calculate the water consumption, electricity consumption, cost, and carbon footprint reduction compared to traditional one-way flushing for all cleaning decision plans. Specifically, construct a virtual pipe network and formulate corresponding cleaning decision plans (as shown in Table 1). Assume that the total pipeline length of the pipeline network to be cleaned is 1 (dimensionless), assume that there are n types of pipe diameters in the entire pipe network, which are D 1 、D 2 ...D n , and the pipeline lengths corresponding to the pipelines of each pipe diameter are a 1 、a 2 ...a n (dimensionless); for the pipeline of the i-th pipe diameter, assume that the length of ice slurry cleaning / the total length of the pipeline of this pipe diameter is b i (i = 1, 2, 3... n), and the length of air-water pulse / the total length of the pipeline of this pipe diameter is 1 - b i (i = 1, 2, 3... n). Assume that a n has 5 possible values, an The value range of: 0, 0.1, 0.2, 0.3, 0.4, and it satisfies Assume b n has 6 possible values, b n The value range of: b i = 0, 0.2, 0.4, 0.6, 0.8, 1.0 (i = 1, 2, 3…n); Assume the water price has 4 possible values, the water price r w (cost of per 1m 3 of water, yuan), r w The value range of: 1, 5, 10, 20; Assume the electricity price has 4 possible values, the electricity price r e (cost of per 1kwh of electricity, yuan), r e The value range of: 0.1, 0.5, 1.0, 2.0; A total of m solutions are formed.

[0053] In this embodiment, set n = 8, the pipe diameter of the virtual pipe network has D 1 = 100mm, D 2 = 150mm, D 3 = 200mm, D 4 = 225mm, D 5 = 300mm, D 6 = 400mm, D 7 = 500mm, D 8 = 600mm. Then a i = (a 1 , a 2 ...a 8 ) has 31 possible combinations, b i = (b 1 , b 2 ...b 8 ) has 1,679,616 possible combinations. Considering the water price and electricity price, there are a total of 833,089,536 solutions (m = 833,089,536).

[0054] Table 1 Constructing a virtual pipe network and formulating corresponding cleaning decision-making solutions

[0055]

[0056] Then collect the water consumption, electricity consumption, required alternative water sources, and the costs of cleaning equipment and consumables for one-way flushing, ice slurry cleaning, and gas-water pulse cleaning of 1 km of pipeline in pipes with different diameters (as shown in Tables 2 and 3).

[0057] Table 2 Water consumption and electricity consumption for one-way flushing, ice slurry cleaning, and gas-water pulse cleaning of 1 km of pipeline

[0058]

[0059] Table 3 Costs of alternative water sources, cleaning equipment, and consumables required for one-way flushing, ice slurry cleaning, and air-water pulse cleaning of 1 km of pipeline

[0060]

[0061] In the k-th scenario, calculate the water demand W for cleaning the water supply pipeline k (m 3 ).

[0062]

[0063] Next, calculate the electricity consumption E for cleaning the water supply pipeline in the k-th scenario k (kwh),

[0064]

[0065] where ei is the energy density for producing 1 m 3 of tap water (kwh / m 3 ), and et is the electricity consumption required for treating 1 m 3 of the cleaning wastewater discharge (kwh / m 3 )

[0066] Calculate the cost C for cleaning the water supply pipeline in the k-th scenario k (yuan),

[0067]

[0068] where is the water price in the k-th scenario (cost per 1 m 3 of water, yuan), is the electricity price in the k-th scenario (cost per 1 kwh of electricity, yuan).

[0069] Use simapro software to calculate the carbon footprint of one-way flushing, ice slurry cleaning, and air-water pulse cleaning of 1 km of pipeline. Assume that residents use bottled mineral water as the alternative water source, and the required water volume is 30% of the daily water supply. When the diameter of the pipeline to be cleaned is greater than or equal to 400 mm, assume that the pipeline is cleaned at night, and the carbon footprint of the alternative water source is not calculated.

[0070]

[0071]

[0072]

[0073] where and They are the carbon footprints generated by one-way flushing, ice slurry cleaning, and air-water pulse cleaning of 1 km of pipelines with the i-th pipe diameter in the k-th scheme, respectively, f aw 、f w 、f t and f e are the carbon footprints generated by each bottle of alternative water source, production of every 1 m 3 of tap water, treatment of every 1 m 3 of cleaning wastewater discharge, and use of every kwh of electricity, respectively. When the diameter of the pipeline to be cleaned is greater than or equal to 400 mm, aw ui 、aw pi and aw si are assumed to be 0.

[0074]

[0075] Calculate the reduced carbon footprint F k (tCO 2 ) of the k-th pipeline cleaning scheme compared with traditional one-way flushing.

[0076] Step 2: Take the water consumption, electricity consumption, and cost of the cleaning decision-making scheme as input indicators, and the reduced carbon footprint (compared with traditional one-way flushing) as the output indicator, and use DEA (data envelopment analysis) to calculate the technical efficiency, scale efficiency, and comprehensive efficiency of all schemes. Specifically, take W k 、E k and C k as input indicators, and F k as the output indicator, establish a DEA data envelopment analysis model, and calculate the technical efficiency, scale efficiency, and comprehensive efficiency (ef k ) of m schemes; among them, the technical efficiency is an indicator to measure whether the performance of the water supply pipeline cleaning scheme can be improved in terms of technical level, the scale efficiency is an indicator to measure whether the cleaning scale can be expanded to improve the performance of the water supply pipeline cleaning scheme, and the comprehensive efficiency is a comprehensive consideration indicator of technical efficiency and scale efficiency. In this embodiment, the ranges of the technical efficiency, scale efficiency, and comprehensive efficiency of 833089536 schemes are shown in Table 4.

[0077] Table 4 Proportion of the value ranges of the technical efficiency, scale efficiency, and comprehensive efficiency of 833089536 schemes

[0078] Numerical range Technical benefit Economies of scale Comprehensive benefit 0.8-1.0 51.42％ 77.51％ 25.72％ 0.6-0.8 43.72％ 22.48％ 60.21％ 0.4-0.6 4.85％ 0.00％ 14.06％

[0079] Step 3: Take the pipeline characteristic parameters and decision-making parameters of all cleaning decision-making schemes as input factors, and the comprehensive efficiency as the output factor, normalize the input factors, and use the K-fold cross-validation method to train the neural network model. Use the ReLU function as the activation function and the mean square error (MSE) as the optimization objective of the model. Specifically

[0080] For the first scheme, take as the input factor, and ef 1 as the output factor to form the first neural network training data pair {z 1 , ef 1};

[0081] For the second scheme, take as the input factor, and ef 2 as the output factor to form the second neural network input data pair {z 2 , ef 2};

[0082] And so on. For the k-th scheme, take as the input factor, and ef k as the output factor to form the k-th neural network training data pair {z k , ef k}; A total of m neural network training data pairs are formed.

[0083] For the convenience of fast convergence in neural network training, all input factors are processed by Min-Max normalization to convert the values of the input factors into the range of [0, 1]. The normalization formula is: where y represents the normalized data, x is the original input data, y' is the processed data, x min is the minimum value in the same type of index data, and x max is the maximum value in the same type of index data. The formed m neural network input training data pairs are used to train the neural network model. This neural network model contains 1 hidden layer, uses the ReLU function as the activation function, and uses the mean squared error (MSE) as the optimization objective of the model, where n is the total number of samples, y i is the true value, is the predicted value. Stochastic gradient descent optimization is used to update the model parameters, and the learning rate is 0.3. To evaluate the performance of the model and select the best model, the K-fold cross-validation method is adopted, and 10 folds are used in the present invention. Multiple iterations are performed in each fold, with a total of 100,000 iterations. During the training process of each fold, the loss value on the validation set is calculated and recorded as the model performance index. Finally, the model with the best performance in all folds is selected and used to predict the test data, and the prediction result is finally obtained. In this embodiment, the MSEs of the final 10 folds are respectively: 0.00023, 0.00025, 0.00026, 0.00023, 0.00023, 0.00023, 0.00023, 0.00022, 0.00029, 0.00025, and the error is less than 0.03%. The error classified by pipe diameter is as Figure 2 shown.

[0084] Step 4: When formulating a cleaning decision for an actual pipe network, obtain the pipeline characteristic parameters of the actual pipe network to be cleaned in the target area, select a series of cleaning decision parameters, and use the neural network model trained by S3 to calculate the comprehensive benefits of the cleaning plan for the target area. The decision maker can compare several plans as needed and select the plan with the highest comprehensive benefits and the strongest feasibility to clean the water supply pipe network. Specifically, when formulating a decision-making plan for an actual water supply pipe network, obtain the pipeline characteristic parameters of the pipe network to be cleaned, find all the pipe diameters, and obtain the number n of pipe diameters (i.e., there are n different pipe diameters in this pipe network); obtain the length a of the pipeline with the i-th pipe diameter i (i = 1, 2, 3…n). For a specific cleaning decision plan, select the cleaning decision parameters: the ice slurry flushing ratio b i (i = 1, 2, 3…n), the number q of ice slurry equipment p 、the number q of air-water pulse equipment s 、the water price r w 、the electricity price r e , and substitute z = (a 1 , a 2 , a 3 ...a n , b 1 , b 2 , b 3 ...b n , q p , q s , r w , r e ) into the trained neural network model to obtain the comprehensive benefits ef of this plan. According to the needs of the decision maker, compare several plans. The higher the comprehensive benefits of a plan, the higher its feasibility. The decision maker can select the plan with the highest comprehensive benefits to clean the actual water supply pipe network. In this embodiment, Beijing, Los Angeles, and Perth are selected for case studies. The plan with the highest comprehensive benefits selected can save a large amount of water resources, energy, costs, and reduce the carbon footprint every year. The specific values are shown in Table 5

[0085] Table 5 Water resources, energy, costs saved and carbon footprint reduced annually by the plans with the highest comprehensive benefits in Beijing, Los Angeles, and Perth

[0086]

Claims

1. A water supply pipeline cleaning technology auxiliary decision-making method based on a neural network, comprising the following steps: S1: Set pipeline characteristic parameters and cleaning decision parameters, build a virtual pipeline network and corresponding cleaning decision schemes; collect water demand, electricity consumption, required alternative water sources, and the cost of cleaning equipment and consumables for one-way flushing, ice slurry cleaning, and air-water pulse cleaning of 1 km of pipelines with different pipe diameters; calculate the water consumption, electricity consumption, cost, and carbon footprint reduction of all cleaning decision schemes compared with traditional one-way flushing; S2: Taking the water consumption, electricity consumption and cost of the cleaning decision-making scheme as input indicators and the reduced carbon footprint as output indicators, DEA is used to calculate the technical benefits, scale benefits and comprehensive benefits of all schemes; S3: Take the pipeline characteristic parameters and cleaning decision parameters of all cleaning decision schemes as input factors, take the comprehensive benefits calculated by S2 as output factors, normalize the input factors, and use the K-fold cross-validation method to train the neural network model; S4: When making a cleaning decision plan for the actual pipe network, obtain the pipeline characteristic parameters of the actual pipe network to be cleaned in the target area, select a series of cleaning decision parameters, and use the neural network model trained in S3 to calculate the comprehensive benefits of the cleaning plan for the target area; decision makers can compare several plans as needed and select the plan with the highest comprehensive benefits and the strongest feasibility to clean the water supply network.

2. The water supply pipeline cleaning technology auxiliary decision-making method according to claim 1 is characterized by: The S1 comprises the following steps: Step 1: Build a virtual pipe network and set the pipe characteristic parameters. Assume that the total length of the pipe network to be cleaned is 1 km and that there are n pipe diameters in the entire pipe network, namely D1, D2…D n , the corresponding pipe lengths of each pipe diameter are a1, a2…a n (km); Step 2: Formulate a corresponding cleaning decision plan and set the cleaning decision parameters. For the i-th pipe diameter, assume that the length of the pipe cleaned by ice slurry / the total length of the pipe of this diameter is b i , where i = 1, 2, 3 ... n, the length of the pipeline of the gas-water pulse / the total length of the pipeline of this diameter is 1-b i , where i = 1, 2, 3…n; Step 3, assuming a n There are 5 possible values, a n The value of: 0, 0.1, 0.2, 0.3, 0.4, and satisfy Step 4, assuming b n There are 6 possible values, b n Value: b i =0, 0.2, 0.4, 0.6, 0.8, 1.0, where i = 1, 2, 3…n; Step 5: Assume that there are four possible water price values, water price r w , r w The value of: 1,5,10,20; Step 6: Assume that there are four possible values ​​for the electricity price, the electricity price r e , r e The value of: 0.1, 0.5, 1.0, 2.0; Step 7, collect all water supply pipe cleaning plans in the virtual pipe network to form m plans in total; Step 8: Collect the water demand, electricity consumption, required alternative water sources, and costs of cleaning equipment and consumables for one-way flushing, ice slurry cleaning, and air-water pulse cleaning of 1 km of pipelines with different pipe diameters; Step 9: In the kth solution, calculate the water demand W for cleaning the water supply pipeline in the kth solution. k (m 3 ), k=1, 2, ..., m; Step 10: In the kth solution, calculate the electricity consumption E for cleaning the water supply pipe in the kth solution. k (kwh), k=1,2...m; Step 11: In the kth solution, calculate the cost C of cleaning the water supply pipeline in the kth solution. k (yuan), k = 1, 2...m; Step 12, using Simapro software to calculate the carbon footprint of one-way flushing, ice slurry cleaning and air-water pulse cleaning of 1 km of pipeline, assuming that residents use bottled mineral water as an alternative water source, the required water volume is 30% of the daily water supply, and when the diameter of the pipeline to be cleaned is greater than or equal to 400 mm, it is assumed that the pipeline is cleaned at night, and the carbon footprint of the alternative water source is not calculated; in and are the carbon footprints (kgCO2) generated by one-way flushing, ice slurry cleaning, and gas-water pulse cleaning of 1 km of pipeline with the i-th diameter in the k-th scheme, respectively, and f aw 、f w 、f t and f e They are the carbon footprint of each bottle of alternative water source (kgCO2 / bottle), the carbon footprint of each 1m 3 Carbon footprint of tap water (kgCO2 / m 3 ), process every 1m 3 Carbon footprint of cleaning wastewater (kgCO2 / m 3 ) and the carbon footprint of each 1kwh of electricity used (kgCO2 / kwh); when the diameter of the pipe to be cleaned is greater than or equal to 400mm, aw ui ,aw pi and aw si Assume it is 0; Step 13, calculate the carbon footprint F reduced by the kth pipeline cleaning solution compared to the traditional one-way flushing k (kgCO2), k = 1, 2...m; 3. The water supply pipeline cleaning technology auxiliary decision-making method according to claim 1 is characterized by: The S2 comprises the following steps: Step 14: W k 、E k and C k As an input indicator, F k As an output indicator, a DEA data envelopment analysis model is established; Step 15: Calculate the technical benefits, scale benefits and comprehensive benefits ef of m solutions k Among them, technical benefit is an indicator to measure whether the performance of the water supply pipe cleaning solution can be improved in terms of technical level, scale benefit is an indicator to measure whether the cleaning scale can be expanded to improve the performance of the water supply pipe cleaning solution, and comprehensive benefit is a comprehensive consideration indicator of technical benefit and scale benefit.

4. The water supply pipeline cleaning technology auxiliary decision-making method according to claim 1 is characterized by: The S3 uses the ReLU function as the activation function and the mean square error MSE as the optimization target of the model.

5. The water supply pipeline cleaning technology auxiliary decision-making method according to claim 4 is characterized by: The S3 comprises the following steps: Step 16: For the kth solution, As input factor, ef k As the output factor, it constitutes the kth neural network training data pair {z k ,ef k }, k = 1, 2...m; a total of m neural network training data pairs are formed; Step 17: To facilitate rapid convergence in neural network training, all input factors are normalized using Min-Max to convert the values ​​of the input factors into the range of [0,1]. Step 18, using the formed m neural network training data pairs to train the neural network model; Step 19, this neural network model contains 1 hidden layer, uses ReLU function as activation function, and uses mean square error as the optimization target of the model; Step 20, use stochastic gradient descent optimization to update the model parameters with a learning rate of 0.3; Step 21, in order to evaluate the performance of the model and select the best model, the K-fold cross validation method was used with 10 folds; Step 22, perform multiple iterations in each compromise, for a total of 100,000 iterations; Step 23, during the training process of each fold, calculate and record the loss value on the validation set as the model performance indicator; finally, select the model with the best performance among all folds and use it to predict the test data to finally obtain the prediction results.

6. The water supply pipeline cleaning technology auxiliary decision-making method according to claim 1 is characterized by: The S4 comprises the following steps: Step 24, when making a decision plan for the actual water supply network, obtain the pipeline characteristic parameters of the pipeline network to be cleaned, find all the pipeline diameters, and obtain the number of pipeline diameters n; Step 25, obtain the length a of the pipe of the i-th pipe diameter i , where i = 1, 2, 3…n; Step 26, select the cleaning decision parameter: ice slurry flushing ratio b i , where i = 1, 2, 3 ... n, the number of ice slurry cleaning equipment q p , Number of air-water pulse equipment q s 、Water price w 、Electricity price e , forming a cleaning decision plan; Step 27, z=(a1,a2,a3…a n ,b1,b2,b3…b n ,q p ,q s ,r w ,r e ) Input the trained neural network model to obtain the comprehensive benefit ef of the scheme; Step 28, according to the needs of decision makers, several plans are compared and selected. The plan with higher comprehensive benefits has higher feasibility. The decision maker can choose the plan with the highest comprehensive benefits to clean the actual water supply network.

Images