A Method and System for Matching Water Service Business Performance Data

Abstract

The present invention relates to the field of water conservancy detection technology, and discloses a method and system for matching water service performance data. A method for matching water service performance data includes the following steps: obtaining basic information; obtaining monitoring information; constructing orderliness information of two structural dimensions; inputting the first feature of the object connection of the orderliness information into the basic information recognition model to output the basic characteristics of the pipe network; inputting the second feature of the object connection of the orderliness information into the monitoring information recognition model to output the monitoring characteristics; splicing the basic characteristics of the pipe network and the monitoring characteristics and inputting them into the generation model to output the current water age data of each object; By analyzing the basic information and monitoring information of the pipe network, extracting the features therein, and calculating the water age of the pipe network based on this, compared with calculating the water age only through pure data, this system combines features such as hydraulics and water quality, making the generated water age statistically realistic while meeting the physical and chemical constraints required for the actual operation of the pipe network.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy detection, and more specifically, it relates to a method and system for matching water service business performance data. Background Art

[0002] The performance assessment of water service business includes KPI assessment and PRI assessment. An important indicator of PRI assessment is the water quality qualification rate, and one of the indicators of the water quality qualification rate is the qualification rate of the water age at the user's end. Since the water age at the user's end is dynamically affected by pipeline blockage, temperature, and water pressure, there will be a situation where the water pressure is temporarily increased during the assessment sampling to reduce the water age at the user's end. To ensure the accuracy of performance assessment, it is necessary to predict the water age through the data monitored daily; however, when calculating the water age using engineering calculation methods, it is based on idealized assumptions, such as pipeline smoothness, valve opening and closing rules, etc. The problems existing in the physical simulation method are also based on some idealized assumptions; but there are many non-linear, sudden, and cross-departmental problems in the real pipe network, such as leakage of old pipelines, accidental water supply interruption, changes in user behavior patterns, etc., which are very likely to exceed the original scope set by the physical simulation model, resulting in deviations between the calculation results of the pipe network hydraulic model and the physical simulation results and the actual data. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for matching water service business performance data in order to solve the above problems.

[0004] The present invention provides a method for matching water service business performance data, including the following steps:

[0005] Obtain basic information, where the basic information includes water source basic information, branch basic information, and user end basic information;

[0006] Obtain monitoring information, where the monitoring information includes water source monitoring information, branch monitoring information, and user end information;

[0007] Based on the basic information and the monitoring information, construct the ordered information of two structural dimensions. The basic structural elements included in the first structural dimension are objects, and the objects include: water source, branch, and user end; the basic structural elements included in the second structural dimension are time series units; the methods for processing to obtain the ordered information include:

[0008] Extract the objects of the ordered information;

[0009] Extract specific features associated with the objects. The specific features include two categories, namely the first feature derived from the basic information and the second feature derived from the monitoring information, and divide the extracted second feature into multiple parts according to the data collection period. One part is mapped to one time series unit, and the order of the time series units is consistent with the time order of the collection period of the second feature;

[0010] Input the first feature of the object connection of the order information into the basic information recognition model. The basic information recognition model includes a first backbone network, and the first backbone network includes a first graph network layer and a first output layer. Among them, the first graph network layer inputs the first feature of the object connection of the order information and outputs a first hidden feature to the first output layer, and the first output layer outputs the pipe network basic feature;

[0011] Input the second feature of the object connection of the order information into the monitoring information recognition model. The monitoring information recognition model includes a second backbone network, and the second backbone network includes a first time series layer, a second graph network layer and a second output layer. Among them, the first time series layer inputs the second feature of the object connection of the order information and outputs a second hidden feature to the second graph network layer, the second graph network layer outputs a third hidden feature to the second output layer, and the second output layer outputs the monitoring feature;

[0012] Concatenate the pipe network basic feature and the monitoring feature and input them into the generation model, and the generation model outputs the current water age data of each object.

[0013] Furthermore, establish a mapping between the extracted object and its associated specific features;

[0014] For any two objects, if there is an association related to the feature task between the associated specific features or there is an association related to the feature task between the objects, an object connection is constructed between these two objects.

[0015] Furthermore, the first features associated with the water source: water source ID, pipe segment location, pipe diameter, pipe material, valve location, pump location, terrain elevation, water pressure at the initial or design stage;

[0016] The first features associated with the branch: branch ID, pipe segment location, pipe diameter, pipe material, valve location, pump location, terrain elevation, water pressure at the initial or design stage;

[0017] The first features associated with the user: water source ID, pipe segment location, pipe diameter, pipe material, valve location, pump location, terrain elevation, water pressure at the initial or design stage;

[0018] The second features associated with the water source: water source ID, water source location, pressure, water flow rate, water quality index, water temperature;

[0019] The second features associated with the branch: branch ID, branch location, pressure, water flow rate, water quality index, water temperature;

[0020] The second features associated with the user: user ID, user location, pressure, water flow rate, water quality index, water temperature.

[0021] Furthermore, the associations related to the feature task include:

[0022] The branch connected to the water source;

[0023] The branch connected to the user.

[0024] Furthermore, the generative model adopts GAN, which includes a generator and a discriminator. The loss function of the generative model is as follows:

[0025] ;

[0026] Among them, represents the generator, represents the discriminator, represents the game objective function between the discriminator and the generator, represents sampling from the true water age data distribution The expected value of the water age data , the natural logarithm of the correct classification probability of the discriminator for these water age data, represents sampling from the concatenated features The data , through the generator The data generated is correctly identified as fake by the discriminator The natural logarithm of the probability, , represents the basic features of the pipe network, represents the monitoring features, represents concatenating the basic features and monitoring features of the pipe network together.

[0027] Furthermore, use the existing water age simulation data to pre-train the generative model and initialize the parameters θ of the generator and the parameters ϕ of the discriminator.

[0028] Furthermore, the training steps of the discriminator are as follows:

[0029] Step 611: Sample a batch of water age samples from the real data , calculate ;

[0030] Step 612: Sample from the noise or prior distribution , generate , calculate ;

[0031] Step 613: Update the discriminator parameters , so that the discriminator can distinguish real and generated samples;

[0032] The training steps of the generator are as follows:

[0033] Step 621: Generate a new water age sample ;

[0034] Step 622: Calculate the discriminator's discrimination result for the generated sample, and simultaneously call the physical simulation / constraint module to perform a quick verification to obtain the loss value for violating physical laws ;

[0035] ;

[0036] Among them, represents the generated water age data, represents the hydraulic / water quality mechanism model, Penalty(·) represents the metric function for violating physical laws, is the metric weight coefficient, with a default value of 0.9, represents the expected value of applying the physical feasibility penalty term to the sample generated by the generator;

[0037] Step 623: Minimize the loss function of the generator to update the generator parameters , gradually reducing the degree of violation of physical mechanisms, The formula for

[0038] ;

[0039] Among them, represents the data sampled from the concatenated features , and the data generated by the generator with parameters is the natural logarithm of the probability that the discriminator correctly identifies it as fake. λ is the training balance weight, which is used to balance the weights of statistical loss and physical constraint loss, The initial value of is 0.1 and increases as the number of training rounds increases, The upper limit of

[0040] Furthermore, the relationship between and the number of training rounds

[0041] ;

[0042] is the maximum number of training rounds, and its default value is 500.

[0043] The present invention provides a water service performance data matching system, which is used to store computer-readable instructions that can execute the foregoing water service performance data matching method when the computer-readable instructions are read.

[0044] The beneficial effects of the present invention are as follows: By analyzing the basic information and monitoring information of the pipe network, extracting the features therein, and calculating the water age of the pipe network based on this, compared with calculating the water age only through pure data, the system combines features such as hydraulics and water quality, making the generated water age statistically realistic while meeting the physical and chemical constraints required for the actual operation of the pipe network. Brief Description of the Drawings

[0045] Figure 1 It is a flowchart of a water service performance data matching method of the present invention. Detailed Embodiments

[0046] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described relative to some examples can also be combined in other examples.

[0047] In at least one embodiment of the present invention, a water service performance data matching method is disclosed, as Figure 1 shown, including the following steps:

[0048] Step 100, obtaining basic information, where the basic information includes water source basic information, branch basic information, and user end basic information;

[0049] Step 200, obtaining monitoring information, where the monitoring information includes water source monitoring information, branch monitoring information, and user end information;

[0050] Step 300, constructing the ordered information of two structural dimensions based on the basic information and the monitoring information. The basic structural elements included in the first structural dimension are objects, and the objects include: water source, branch, and user end; the basic structural elements included in the second structural dimension are time sequence units;

[0051] In some embodiments of the present invention, the method for processing to obtain the ordered information includes:

[0052] extracting the objects of the ordered information;

[0053] Extract specific features associated with an object. The specific features include two categories, namely the first features derived from basic information and the second features derived from monitoring information. The extracted second features are divided into multiple parts according to the data collection period, and one part is mapped to one time series unit. The order of the time series units is consistent with the time order of the collection period of the second features;

[0054] Establish a mapping between the extracted object and its associated specific features;

[0055] For any two objects, if there is an association related to the feature task between the associated specific features or there is an association related to the feature task between the objects, then an object connection is constructed between these two objects.

[0056] In some embodiments of the present invention, the first features associated with the water source: water source ID, pipe section location, pipe diameter, pipe material, valve location, pump location, terrain elevation, water pressure at the initial or design stage;

[0057] The first features associated with the branch: branch ID, pipe section location, pipe diameter, pipe material, valve location, pump location, terrain elevation, water pressure at the initial or design stage;

[0058] The first features associated with the user: water source ID, pipe section location, pipe diameter, pipe material, valve location, pump location, terrain elevation, water pressure at the initial or design stage;

[0059] In some embodiments of the present invention, the second features associated with the water source: water source ID, water source location, pressure, water flow rate, water quality indicators (residual chlorine, turbidity, dissolved oxygen, etc.), water temperature;

[0060] The second features associated with the branch: branch ID, branch location, pressure, water flow rate, water quality indicators (residual chlorine, turbidity, dissolved oxygen, etc.), water temperature;

[0061] The second features associated with the user: user ID, user location, pressure, water flow rate, water quality indicators (residual chlorine, turbidity, dissolved oxygen, etc.), water temperature;

[0062] In some embodiments of the present invention, the associations related to the feature task include:

[0063] The water source and the branch connected to it;

[0064] The user and the branch connected to it.

[0065] Step 400: Input the first feature of the object connection of the orderliness information into the basic information recognition model. The basic information recognition model includes a first backbone network, and the first backbone network includes a first graph network layer and a first output layer. Among them, the first graph network layer inputs the first feature of the object connection of the orderliness information and outputs a first hidden feature to the first output layer, and the first output layer outputs the pipeline network basic feature.

[0066] Step 500: Input the second feature of the object connection of the orderliness information into the monitoring information recognition model. The monitoring information recognition model includes a second backbone network, and the second backbone network includes a first time series layer, a second graph network layer and a second output layer. Among them, the first time series layer inputs the second feature of the object connection of the orderliness information and outputs a second hidden feature to the second graph network layer, the second graph network layer outputs a third hidden feature to the second output layer, and the second output layer outputs the monitoring feature.

[0067] In an embodiment of the present invention, the first time series layer uses a time series neural network, such as RNN or LSTM or GRU.

[0068] An expression of one form of the first time series layer is as follows:

[0069]

[0070]

[0071]

[0072]

[0073] and : Activation vectors of the reset gate and update gate at the th step;

[0074] : Candidate state generated at the th step;

[0075] : Second hidden feature at the th step;

[0076] : First, second, third, fourth, fifth, and sixth transformation matrices (trainable parameters);

[0077] : First, second, and third biases (trainable parameters);

[0078] represents the th time series unit;

[0079] , When

[0080] : Sigmoid activation function;

[0081] : tanh activation function.

[0082] Both the first network layer and the second graph network layer adopt a multi-layer structure. The calculation formula for the th layer is as follows:

[0083]

[0084] represents the object hidden feature of the th object in the th layer, represents the object hidden feature of the th object in the th layer, and respectively represent the sets of objects that have object connections with object and object , represents the cardinality of the set, represents the planar recognition weight matrix of the th layer of the second graph network layer or the fourth graph network layer, , represents the total number of layers. In the first graph network, When , represents the specific feature associated with the u-th object. In the second graph network, When , , represents the last hidden feature output when the specific feature inputs of all the timing units associated with the th object enter the first timing layer, is the total number of timing units, When is equal to the second or third hidden feature of object (the first graph network layer outputs the second hidden feature, and the second graph network layer outputs the third hidden feature, but the trainable parameters of the first graph network layer and the second graph network layer are different), is the sigmoid function.

[0085] The expression of the first output layer is as follows:

[0086]

[0087] wherein represents a basic feature of the pipe network, represents the th second hidden feature of the represents the set of all objects, represents the total number of layers, and FC represents fully connected.

[0088] The expression of the second output layer is as follows:

[0089]

[0090] wherein represents a monitoring feature, represents the th third hidden feature of the represents the set of all objects, represents the total number of layers, and FC represents fully connected.

[0091] Step 600: Concatenate the basic feature of the pipe network and the monitoring feature and input them into the generation model, and the generation model outputs the current water age data of each object.

[0092] In an embodiment of the present invention, the generation model adopts GAN, and GAN includes a generator and a discriminator. The loss function of the generation model is as follows:

[0093]

[0094] wherein, represents the generator, represents the discriminator, represents the game objective function between the discriminator and the generator, represents the expected value of the water age data sampled from the true water age data distribution the natural logarithm of the correct classification probability of the discriminator for these water age data, represents the data sampled from the concatenated features the data generated by the generator is correctly identified as false by the discriminator the natural logarithm of the probability, , represents the basic feature of the pipe network, represents the monitoring feature, represents concatenating the basic feature of the pipe network and the monitoring feature together.

[0095] In one embodiment of the present invention, a pre-trained generation model is generated using existing water age simulation data, and the parameters of the generator are initialized. and the parameters of the discriminator .

[0096] In one embodiment of the present invention, the training steps of the discriminator are as follows:

[0097] Step 611: Sample a batch of water age samples from the real data , and calculate ;

[0098] Step 612: Sample from the noise or prior distribution , generate , and calculate ;

[0099] Step 613: Update the discriminator parameters so that the discriminator can distinguish real and generated samples;

[0100] The training steps of the generator are as follows:

[0101] Step 621: Generate new water age samples ;

[0102] Step 622: Calculate the discrimination result of the discriminator on the generated samples, and at the same time call the physical simulation / constraint module to perform a quick check, and obtain the loss value for violating physical laws ;

[0103]

[0104] Among them, represents the generated water age data, represents the hydraulic / water quality mechanism model, Penalty(·) represents the measurement function for violating physical laws (such as pressure gradient, water quality attenuation), is the measurement weight coefficient, and the default value is 0.9, represents the expected value of applying the physical feasibility penalty term to the samples generated by the generator;

[0105] Step 623: Minimize the loss function of the generator to update the generator parameters , and gradually reduce the degree of violation of physical mechanisms, The formula of

[0106]

[0107] Among them, represents the concatenated features Sampled data , by using a parameter generator Generated data is the natural logarithm of the probability that the discriminator correctly identifies as false, λ is the training balance weight, the weight used to balance the statistical loss and the physical constraint loss, The initial value of is 0.1 and increases as the number of training rounds increases, The upper limit of is 0.6. In the initial stage of training, it is pursued that the generation of the generator conforms to the distribution of the training data, and the constraint of physical laws is improved as the number of training rounds increases.

[0108] has a linear or non-linear growth relationship with the number of training rounds , for example , where is the maximum number of training rounds, and the default value is 500.

[0109] In one embodiment of the present invention, steps 611-623 are repeated until certain convergence conditions are met or the maximum number of training rounds is reached. The convergence conditions include that the loss function of the generator is less than a set threshold (for example, 0.01), and the default value of the maximum number of training rounds is 500.

[0110] In at least one embodiment of the present invention, a water service performance data matching system is provided, which is used to store computer-readable instructions that can execute the foregoing water service performance data matching method when the computer-readable instructions are read.

[0111] The above describes the embodiments of the present invention, but these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. A water service performance data matching method, characterized in that: The following steps are involved: Obtain basic information, including water source basic information, branch basic information and user end basic information; Obtain monitoring information, including water source monitoring information, branch monitoring information and user end information; Based on basic information and monitoring information, order information of two structural dimensions is constructed. The basic structural elements included in the first structural dimension are objects, including: water source, branch, user end; the basic structural elements included in the second structural dimension are time sequence units; the method for processing and obtaining order information includes: Objects for extracting orderliness information; Extract specific features associated with the object. The specific features include two types, namely, first features derived from basic information and second features derived from monitoring information. The extracted second features are divided into multiple parts according to the data collection cycle. One part is mapped to a time sequence unit. The order of the time sequence units is consistent with the time order of the collection cycle of the second features. Inputting the first feature of the object connection of the ordered information into the basic information recognition model, the basic information recognition model includes a first backbone network, the first backbone network includes a first graph network layer and a first output layer, wherein the first graph network layer inputs the first feature of the object connection of the ordered information, outputs the first hidden feature to the first output layer, and the first output layer outputs the basic feature of the pipeline network; Inputting the second feature of the object connection of the ordered information into the monitoring information recognition model, the monitoring information recognition model includes a second backbone network, the second backbone network includes a first time sequence layer, a second graph network layer and a second output layer, wherein the first time sequence layer inputs the second feature of the object connection of the ordered information, outputs the second hidden feature to the second graph network layer, the second graph network layer outputs the third hidden feature to the second output layer, and the second output layer outputs the monitoring feature; The basic characteristics of the pipeline network and the monitoring characteristics are spliced ​​and input into the generation model, which then outputs the current water age data of each object.

2. A water service performance data matching method according to claim 1, characterized in that: Establishing a mapping between the extracted objects and their associated specific features; For any two objects, if there is a correlation between the associated specific features related to the feature task or there is a correlation between the objects related to the feature task, an object relationship is established between the two objects.

3. A water service performance data matching method according to claim 2, characterized in that: The first characteristics associated with water sources: water source ID, pipe segment location, pipe diameter, pipe material, valve location, water pump location, terrain elevation, and water pressure at the initial or design stage; The first characteristics of branch association: branch ID, pipe segment location, pipe diameter, pipe material, valve location, water pump location, terrain elevation, water pressure at the initial or design stage; The first features associated by the user: water source ID, pipe segment location, pipe diameter, pipe material, valve location, water pump location, terrain elevation, and water pressure at the initial or design stage; The second characteristics associated with water sources: water source ID, water source location, pressure, water flow, water quality index, and water temperature; Secondary features associated with branches: branch ID, branch location, pressure, water flow, water quality index, and water temperature; Secondary features associated with the user: user ID, user location, pressure, water flow, water quality index, and water temperature.

4. A water service performance data matching method according to claim 2, characterized in that: The correlations related to the feature tasks include: The branch to which the water source is connected; The branch the user is connected to.

5. A water service performance data matching method according to claim 1, characterized in that: The generative model uses GAN, which includes a generator and a discriminator. The loss function of the generative model is as follows: ； in, represents a generator, represents the discriminator, represents the game objective function between the discriminator and the generator, Indicates the distribution of real water age data Water age data sampled in The expected value of the discriminator is the logarithm of the probability of correctly classifying these water age data with the natural constant e as the base, Indicates that from the splicing feature The data sampled in , through the generator Generated data Discriminator The logarithm of the probability of correctly identifying a false positive to the base of the natural constant e, , Indicates the basic characteristics of the pipe network. Indicates monitoring characteristics, It means to splice the basic characteristics of the pipeline network and the monitoring characteristics together.

6. A water service performance data matching method according to claim 5, characterized in that: Use the existing water age simulation data to pre-train the generative model and initialize the parameters θ of the generator and ϕ of the discriminator.

7. A water service performance data matching method according to claim 6, characterized in that: The training steps of the discriminator are as follows: Step 611: Sampling a batch of water age samples from real data ,calculate ; Step 612: Sampling from noise or prior distribution ,generate ,calculate ; Step 613: Update the discriminator parameters , so that the discriminator can distinguish between real and generated samples; The training steps of the generator are as follows: Step 621: Generate a new water age sample ; Step 622: Calculate the discriminator's discrimination result for the generated sample and call the physical simulation / constraint module at the same time Do a quick check to get the loss value that violates the laws of physics ; ； in, represents the generated water age data, represents the hydraulic / water quality mechanism model, Penalty(·) represents the metric function for violating physical laws, is the measurement weight coefficient, the default value is 0.9, Represents the samples generated by the generator The expected value of the applied physical feasibility penalty; Step 623: Minimize the generator’s loss function To update the generator parameters , gradually reducing the degree of violation of physical mechanisms, The formula is as follows: ； in, Indicates that from the splicing feature The data sampled in , by using the parameter Generator Generated data Discriminator The logarithm of the probability of correctly identifying a false positive with the natural constant e as the base, λ is the training balance weight, which is used to balance the weight of statistical loss and physical constraint loss, The initial value of is 0.1, and it increases with the increase of training rounds. The upper limit is 0.

6. In the early stage of training, the generator is required to generate data that conforms to the distribution of training data. As the number of training rounds increases, the constraints of physical laws are improved.

8. A water service performance data matching method according to claim 7, characterized in that: And training rounds The relationship is: ； The maximum number of training rounds is 500 by default.

9. A water service performance data matching system, characterized in that: It is used to store computer-readable instructions, which, when read, can execute a water service performance data matching method as described in any one of claims 1-8.

Images