Sensor optimization arrangement method and system based on feature importance

Abstract

The invention discloses a sensor optimization arrangement method and system based on feature importance, and the method comprises the steps: carrying out the simulation of the distribution positions of different types of sensors based on a fire-fighting pipe network simulation model, building a data set, and carrying out the calculation of the data set; the data set comprises positions and types of sensors, monitoring data of different distribution positions and corresponding pipe network operation states; training a gradient boosting tree by using the data set to obtain feature importance degrees of the monitoring data at different positions to the operation state of the pipe network; determining point distribution schemes of different types of sensors at different positions according to the feature importance; optimized arrangement of various types of sensors of the fire-fighting pipe network is achieved.

Description

Technical Field

[0001] The present invention relates to the field of machine learning technology, and in particular to a sensor optimization layout method and system based on feature importance. Background Art

[0002] As a crucial power transmission and conversion facility, converter stations require the proper functioning of their fire protection networks, crucial for timely extinguishing fires. Typically, pressure and flow sensors deployed along these networks collect real-time pressure and flow data. This data is then analyzed and feature extracted for network status monitoring and fault diagnosis. Therefore, the optimal placement of pressure and flow sensors in fire protection networks is crucial, as it directly impacts the quality of network status monitoring and fault diagnosis.

[0003] At present, the optimal layout of pressure sensors and flow sensors in fire protection pipe networks often relies on the experience of professionals and simplified theoretical analysis, and is only applicable to the optimal layout of a single type of sensor, and is not applicable to the joint optimal layout of multiple types of sensors. For example, the document "Study on the Optimal Layout and Leakage Location of Pressure Monitoring Points Based on Water Supply Pipeline Network Modeling, Liu Songzi, Master's Thesis" proposes a method for the optimal layout of pressure monitoring points in water supply pipe networks, establishes an optimization model with the maximization of the number of pressure monitoring point nodes as the objective function, and uses the ant lion algorithm to solve the model. The patent application document with publication number CN115048804A proposes a method for the optimal layout of multiple sensors in urban pipe networks, establishes an optimization model with the maximization of the overall coverage of the sensor network on the pipe network as the objective function, and uses a genetic algorithm to solve the model. However, the above studies are only applicable to the optimal layout of a single type of sensor, and are not applicable to the optimal layout of multiple types of pressure and flow sensors.

[0004] However, in actual applications, in order to accurately monitor the operating status of the fire protection pipe network, it is often necessary to simultaneously deploy multiple types of sensors such as pressure sensors and flow sensors, collect pressure data and flow data of the fire protection pipe network, and then perform data analysis and feature extraction to improve the accuracy of fire protection pipe network fault diagnosis. Summary of the Invention

[0005] The technical problem to be solved by the present invention is how to achieve the optimal arrangement of various types of sensors in a fire protection pipe network.

[0006] The present invention solves the above technical problems through the following technical means:

[0007] A sensor optimization layout method based on feature importance is proposed, which includes:

[0008] For the locations of different types of sensors, simulation is performed based on the fire protection pipe network simulation model to construct a data set, which includes the location and type of sensors, as well as the monitoring data of different locations and the corresponding pipe network operation status;

[0009] The data set is used to train a gradient boosting tree to obtain the feature importance of monitoring data at different locations to the operation status of the pipeline network;

[0010] The layout scheme of sensors of different types at different locations is determined according to the feature importance.

[0011] Furthermore, the locations of different types of sensors are simulated based on the fire pipe network simulation model to construct training data, including:

[0012] Based on the sensors already deployed in the fire protection pipe network, other deployment locations are selected to deploy sensors, forming deployment locations for different types of sensors;

[0013] The fire protection pipe network model is used for simulation to obtain monitoring data of different types of sensors at different locations and the corresponding pipe network operation status, and the data set is constructed.

[0014] Furthermore, before simulating the locations of different types of sensors based on the fire protection pipe network simulation model and constructing a data set, the method further includes:

[0015] Use sensors already deployed in the fire protection pipe network to collect real-time monitoring data of the fire protection pipe network;

[0016] Establish a fire protection pipe network simulation model, perform simulation using sensors already deployed in the fire protection pipe network as simulation conditions, and obtain simulation data of the simulation model;

[0017] The simulation data is compared with the real-time monitoring data to verify the fire pipe network simulation model.

[0018] Furthermore, the formula of the gradient boosting tree is expressed as:

[0019]

[0020] Where g m (x i ) is the mth decision tree, M is the number of decision trees, f M (x i ) is a gradient boosting tree model containing M decision trees.

[0021] Furthermore, the gradient boosting tree is trained using the data set to obtain the feature importance of monitoring data at different locations to the operation status of the pipeline network, including:

[0022] In the dataset On the training base classifier f1(x n ), so that the basic classifier f1(x n ) has an error rate of ∈1<0.5, and the formula for the error rate ∈1 is:

[0023]

[0024] Where: is the sample weight, is the model prediction value, x i is the monitoring data, y i Indicates the operation status of the pipe network, n is the number of distribution points, is the judgment function, if If it holds, δ(·)=1, otherwise, δ(·)=0, f1(x n ) is the base classifier f1(·) located at x n The value of

[0025] Update the sample weights of the basic classifier f1(x), which is expressed as follows:

[0026]

[0027] Where, is the correct sample weight after the update classification, is the updated misclassified sample weight, and d is a constant;

[0028] The sample weights of the monitoring data at different distribution points are updated to obtain the characteristic importance of the monitoring data at different distribution points to the operation status of the pipeline network.

[0029] Furthermore, the goal of the gradient boosting tree training is to minimize the loss function:

[0030]

[0031] Where: L(·) is the loss function, f M (x i ) is a gradient boosting tree model containing M decision trees, g m (x i ) is the mth decision tree, and argmin means finding the minimum value of the objective function.

[0032] Furthermore, before using the data set to train the gradient boosting tree to obtain the feature importance of monitoring data at different locations to the operating status of the pipe network, the method further includes:

[0033] For the dataset D = {P i ,F j,S k}, P i is the pressure data at position i, F j is the flow data of location j, S k The operation status of the pipeline network;

[0034] For pressure data P i and flow data F j Perform normalization processing to obtain normalized pressure data and traffic data

[0035] Based on the normalized pressure data and normalized flow data, a data set is constructed y i =S k .

[0036] Furthermore, the determining of the layout scheme of sensors of different types at different locations according to the feature importance includes:

[0037] Sort the features from high to low importance and determine the priority of different types of sensors at different locations;

[0038] For sensors already deployed in the fire protection pipe network, adjust the positions and types of deployed sensors based on the priorities of different types of sensors at different locations;

[0039] For locations where sensors are not installed, the type of sensors to be installed is determined based on the priorities of different types of sensors at different locations.

[0040] Furthermore, the method further comprises:

[0041] Different types of sensors are deployed based on the deployment plan to collect monitoring data of the fire protection pipe network.

[0042] In addition, the present invention also proposes a sensor optimization arrangement system based on feature importance, the system comprising:

[0043] A data set construction module is used to simulate the locations of different types of sensors based on the fire protection pipe network simulation model to construct a data set, which includes the location and type of sensors, as well as monitoring data of different locations and the corresponding pipe network operation status;

[0044] A feature importance calculation module is used to train a gradient boosting tree using the data set to obtain the feature importance of monitoring data at different locations to the operating status of the pipeline network;

[0045] The deployment plan determination module is used to determine the deployment plan of sensors of different types at different locations according to the feature importance.

[0046] The advantages of the present invention are:

[0047] For the locations of different types of sensors, simulation is performed based on the fire pipe network simulation model, a data set is constructed, and the data set is used to train the gradient boosting tree. The characteristic importance of monitoring data at different locations to the operation status of the pipe network is output. Then, based on the characteristic importance of monitoring data at different locations, the priority of different types of sensors at different locations is determined. This not only supports sensor location optimization, but also supports sensor type optimization, which is more in line with the needs of fire pipe network status monitoring.

[0048] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 1 is a flow chart of a method for optimizing sensor placement based on feature importance according to an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of a fire protection pipe network simulation model in one embodiment of the present invention;

[0051] Figure 3 is a schematic diagram of simulation data and experimental data in one embodiment of the present invention;

[0052] Figure 4 is a schematic diagram of the characteristic importance of pressure / flow data at different positions in one embodiment of the present invention;

[0053] Figure 5 1 is a schematic diagram of an installation scheme for optimizing sensor placement based on feature importance in one embodiment of the present invention;

[0054] Figure 6 3 is a structural diagram of a sensor optimization arrangement system based on feature importance proposed in one embodiment of the present invention. DETAILED DESCRIPTION

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0056] like Figure 1 As shown, an embodiment of the present invention proposes a sensor optimization arrangement method based on feature importance, the method comprising the following steps:

[0057] S10. For the locations of different types of sensors, simulation is performed based on the fire protection pipe network simulation model to construct a data set, wherein the data set includes the location and type of the sensors, monitoring data of different locations, and corresponding pipe network operation status;

[0058] It should be noted that in order to monitor the status and diagnose faults of the fire protection pipe network, pressure sensors and flow sensors are generally arranged on the fire protection pipe network to obtain real-time pressure data and flow data of the fire protection pipe network.

[0059] S20, using the data set to train a gradient boosting tree to obtain the feature importance of monitoring data at different locations to the operating status of the pipeline network;

[0060] S30: Determine a layout plan for sensors of different types at different locations based on the feature importance.

[0061] As a further preferred technical solution, step S10: for the locations of different types of sensors, simulation is performed based on the fire pipe network simulation model to construct training data, which specifically includes the following steps:

[0062] S11. Based on the sensors already deployed in the fire protection pipe network, select other deployment locations to deploy sensors, thereby forming deployment locations for different types of sensors;

[0063] S12. Use the fire protection pipe network model to perform simulation to obtain monitoring data of different types of sensors at different locations and the corresponding pipe network operation status, and construct the data set.

[0064] As a further preferred technical solution, before step S10: simulating the locations of different types of sensors based on the fire protection pipe network simulation model to construct a data set, the method further includes the following steps:

[0065] Use sensors already deployed in the fire protection pipe network to collect real-time monitoring data of the fire protection pipe network;

[0066] Establish a fire protection pipe network simulation model, perform simulation using sensors already deployed in the fire protection pipe network as simulation conditions, and obtain simulation data of the simulation model;

[0067] The simulation data is compared with the real-time monitoring data to verify the fire pipe network simulation model.

[0068] It should be noted that in a real fire protection network, due to budget and construction constraints, the number of sensors installed is very limited, and the installation locations are relatively sparse. Computer simulation offers the advantages of flexibility and convenience, providing a new approach for studying the optimal placement of sensors. Based on the actual fire protection network layout, a 1:1 fire protection network simulation model was created using Flomaster network simulation software. To validate the simulation model, sensors were arranged according to the actual fire protection network. The fire protection network was then simulated, and pressure and flow data obtained from the simulation model were compared with experimental data collected by sensors in the actual fire protection network.

[0069] As a further preferred technical solution, the formula of the gradient boosting tree is expressed as:

[0070]

[0071] Where g m (x i ) is the mth decision tree, M is the number of decision trees, f M (x i ) is a gradient boosting tree model containing M decision trees.

[0072] As a further preferred technical solution, step S20: using the data set to train a gradient boosting tree to obtain the feature importance of monitoring data at different locations to the operating status of the pipe network, specifically includes the following steps:

[0073] S21, in the dataset On the training base classifier f1(x n ), so that the basic classifier f1(x n ) has an error rate of ∈1<0.5, and the formula for the error rate ∈1 is:

[0074]

[0075] Where: is the sample weight, is the model prediction value, x i is the monitoring data, y i Indicates the operation status of the pipe network, n is the number of distribution points, is the judgment function, if If it holds, δ(·)=1, otherwise, δ(·)=0, f1(x n ) is the base classifier f1(·) located at x n The value of

[0076] S22, update the sample weight of the basic classifier f1(x), the formula is expressed as:

[0077]

[0078] Where, is the correct sample weight after the update classification, is the updated misclassified sample weight, and d is a constant;

[0079] It should be noted that the specific process of updating the sample weight is to reduce the weight of the sample correctly classified by f1(x), and the weight is divided by a constant d, that is, Increase the weight of samples that are misclassified by f1(x), and multiply the weight by a constant d, that is,

[0080] S23. Update the sample weights of the monitoring data at different distribution points to obtain the characteristic importance of the monitoring data at different distribution points to the operation status of the pipeline network.

[0081] The above process is repeated continuously to train the data set, and the sample weights of the pressure / flow data at different locations, that is, the feature importance, are continuously updated until the model training converges, and the feature importance of the pressure / flow data at different locations to the operation status of the pipeline network is output.

[0082] It should be noted that the gradient boosting tree (GBDT) is considered to be one of the best performing methods in statistical learning. Therefore, GBDT is used for model training and feature extraction.

[0083] As a further preferred technical solution, the goal of the gradient boosting tree training is to minimize the loss function:

[0084]

[0085] Where: L(·) is the loss function, f M (x i ) is a gradient boosting tree model containing M decision trees, g m (x i ) is the mth decision tree, and argmin means finding the minimum value of the objective function.

[0086] As a further preferred technical solution, before the step S20 of training the gradient boosting tree using the data set to obtain the feature importance of the monitoring data at different locations to the operation status of the pipe network, the method further includes the following steps:

[0087] S21, for the data set D = {P i ,F j ,S k}, P i is the pressure data at position i, F j is the flow data of location j, S k The operation status of the pipeline network;

[0088] S22, pressure data P i and flow data F j Perform normalization processing to obtain normalized pressure data and traffic data

[0089] S23. Construct a data set based on the normalized pressure data and normalized flow data. y i =S k .

[0090] It should be noted that, after the original data set is normalized in this embodiment, the pressure data and flow data are both located in [0, 1], which is convenient for subsequent model training.

[0091] As a further preferred technical solution, step S30 determines the layout of sensors of different types at different locations based on the feature importance, specifically comprising the following steps:

[0092] S31. Sort the importance of the features from high to low, and determine the priorities of different types of sensors at different locations;

[0093] S32. For sensors already deployed in the fire protection pipe network, adjust the positions and types of the deployed sensors based on the priorities of sensors of different types at different positions;

[0094] S33. For locations to be deployed where no sensors are installed, determine the type of sensors to be installed based on the priorities of different types of sensors at different locations.

[0095] It should be noted that this embodiment prioritizes different sensor types at different locations based on the importance of the pressure / flow data characteristics at each location, from high to low. Specifically, it prioritizes which locations should be equipped with pressure sensors and which should be equipped with flow sensors. The required number of sensors, including both pressure and flow sensors, is determined based on project requirements. If sensors are already installed, their location and type are adjusted. If not, they are installed according to the optimized layout.

[0096] As a further preferred technical solution, the method further comprises the following steps:

[0097] Different types of sensors are deployed based on the deployment plan to collect monitoring data of the fire protection pipe network.

[0098] It should be noted that this embodiment is described in detail through a specific application example:

[0099] (1) The pressure and flow data of the fire protection pipe network are collected in real time through the pressure sensors and flow sensors arranged on the fire protection pipe network.

[0100] To monitor the status and diagnose faults in fire protection pipe networks, pressure and flow sensors are typically deployed throughout the network to collect real-time pressure and flow data. The fire protection pipe network consists of pipes, water tanks, pumps, fire hydrants, and valves. Three pressure sensors and three flow sensors are deployed throughout the network to collect real-time pressure and flow data. Table 1 shows some of the pressure and flow data collected by the pressure and flow sensors. A network status of 0 indicates normal network operation, while a network status of 1 indicates a fault.

[0101] Table 1

[0102]

[0103] (2) Establish a fire protection pipe network simulation model, simulate the fire protection pipe network based on the deployed pressure / flow sensors, obtain the pressure / flow data of the simulation model, and compare and verify the model with the experimental data.

[0104] For real fire protection pipe networks, due to engineering budget and construction limitations, the number of sensors installed is very limited and the installation locations are relatively sparse. Computer simulation has the advantages of flexibility and convenience, providing a new way to study the optimal arrangement of sensors. According to the real fire protection pipe network layout, Flomaster pipe network simulation software is used to build a 1:1 fire protection pipe network simulation model, such as Figure 2 To verify the simulation model, the sensors are arranged according to the real fire pipe network, and then the fire pipe network is simulated to obtain the pressure / flow data of the simulation model and compare it with the experimental data collected by the real fire pipe network sensors, as shown in Figure 2. Figure 3 The simulated pressure / flow data are very consistent with the experimental pressure / flow data, which verifies the accuracy of the simulation model and provides a basis for optimizing sensor layout based on the simulation model.

[0105] (3) Based on the simulation model and the deployed pressure / flow sensors, more locations are selected to deploy pressure / flow sensors, the fire protection pipe network is simulated, and the pressure / flow data of each pressure / flow sensor is obtained.

[0106] This embodiment is based on the established simulation model and selects more locations to arrange pressure / flow sensors. The simulation model sensor arrangement is very flexible and is not limited by the project budget and construction. To improve the rationality of the pressure / flow sensor arrangement, the pressure / flow sensors can be arranged densely and at equal intervals. At the same time, to optimize the sensor type, pressure sensors and flow sensors are simultaneously arranged at each sensor position in the simulation model. The fire-fighting pipe network is then simulated to obtain pressure / flow data from each pressure / flow sensor. The simulation model pipe network is 520 meters long, and pressure / flow sensors are evenly arranged at intervals of 20 meters, that is, in the simulation model, 26 pressure / flow sensors are arranged at each location. Since the simulation model is equipped with a large number of pressure / flow sensors, the pressure / flow data of each sensor in the fire-fighting pipe network can be obtained.

[0107] (4) Based on the pressure / flow data and pipeline network operation status at different locations of the simulation model, the sensor location, type and pipeline network operation status are labeled as input parameters, and a machine learning model is established to train the data and output the characteristic importance of the pressure / flow data at different locations to the pipeline network operation status.

[0108] It should be noted that this embodiment continuously updates the sample weights of pressure / flow data at different locations, i.e., feature importance, by training the data set until the model training converges, and outputs the feature importance of pressure / flow data at different locations to the operating status of the pipe network, such as Figure 4 As shown in the figure, the higher the feature importance, the more the data at that location reflects the operational status of the pipe network. Clearly, it's more important to install a sensor at that location. The top 10 feature importances for pressure / flow data at different locations, from high to low, are Flow5, Flow20, Flow11, Pres10, Pres6, Flow2, Flow25, Flow16, Pres11, and Flow9. Clearly, data from different locations and types has varying degrees of influence on the operational status of the pipe network. Therefore, the optimal placement of pressure / flow sensors in fire protection pipe networks is crucial.

[0109] (5) According to the characteristic importance of pressure / flow data at different locations, determine the priority of different types of sensors at different locations from high to low, that is, which locations should be equipped with pressure sensors and which locations should be equipped with flow sensors. In combination with the project requirements, determine the number of sensors that need to be installed, including pressure sensors and flow sensors. If the sensor has been installed, adjust the sensor location and type; if the sensor has not been installed, install the sensor according to the optimized layout plan, such as Figure 5 shown.

[0110] (6) After the pressure / flow sensor is adjusted through the optimized layout plan, it collects the pressure / flow data of the fire protection pipe network and provides data support for the status monitoring and fault diagnosis of the fire protection pipe network.

[0111] In addition, a second embodiment of the present invention provides a sensor optimization arrangement system based on feature importance, the system comprising:

[0112] A data set construction module 10 is used to simulate the locations of different types of sensors based on the fire protection pipe network simulation model to construct a data set, wherein the data set includes the location and type of the sensors, the monitoring data of the different locations, and the corresponding pipe network operation status;

[0113] A feature importance calculation module 20 is used to train a gradient boosting tree using the data set to obtain the feature importance of monitoring data at different locations to the operating status of the pipeline network;

[0114] The deployment plan determination module 30 is configured to determine deployment plans for sensors of different types at different locations based on the feature importance.

[0115] As a further preferred technical solution, the data set construction module 10 is specifically used to:

[0116] Based on the sensors already deployed in the fire protection pipe network, other deployment locations are selected to deploy sensors, forming deployment locations for different types of sensors;

[0117] The fire protection pipe network model is used for simulation to obtain monitoring data of different types of sensors at different locations and the corresponding pipe network operation status, and the data set is constructed.

[0118] As a further preferred technical solution, the system further includes:

[0119] The acquisition unit is used to collect real-time monitoring data of the fire protection pipe network using sensors arranged in the fire protection pipe network;

[0120] A model building unit is used to build a fire pipe network simulation model, perform simulation using sensors already arranged in the fire pipe network as simulation conditions, and obtain simulation data of the simulation model;

[0121] A verification unit is used to compare the simulation data with the real-time monitoring data to verify the fire pipe network simulation model.

[0122] As a further preferred technical solution, the formula of the gradient boosting tree is expressed as:

[0123]

[0124] Where g m (xi ) is the nth decision tree, M is the number of decision trees, f M (x i ) is a gradient boosting tree model containing M decision trees.

[0125] As a further preferred technical solution, the feature importance calculation module 20 is specifically configured to:

[0126] In the dataset On the training base classifier f1(x n ), so that the basic classifier f1(x n ) has an error rate of ∈1<0.5, and the formula for the error rate ∈1 is:

[0127]

[0128] Where: is the sample weight, is the model prediction value, x i is the monitoring data, y i Indicates the operation status of the pipe network, n is the number of distribution points, is the judgment function, if If it holds, δ(·)=1, otherwise, δ(·)=0, f1(x n ) is the base classifier f1(·) located at x n The value of

[0129] Update the sample weights of the basic classifier f1(x), which is expressed as follows:

[0130]

[0131] Where, is the correct sample weight after the update classification, is the updated misclassified sample weight, and d is a constant;

[0132] The sample weights of the monitoring data at different distribution points are updated to obtain the characteristic importance of the monitoring data at different distribution points to the operation status of the pipeline network.

[0133] As a further preferred technical solution, the system further includes a normalization module, specifically for

[0134] For the dataset D = {P i ,F j ,S k}, P i is the pressure data at position i, F j is the flow data of location j, S k The operation status of the pipeline network;

[0135] For pressure data P i and flow data F j Perform normalization processing to obtain normalized pressure data and traffic data

[0136] Based on the normalized pressure data and normalized flow data, a data set is constructed y i =S k .

[0137] As a further preferred technical solution, the site layout determination module 30 includes:

[0138] A priority determination unit is used to sort the importance of the features from high to low and determine the priority of different types of sensors at different locations;

[0139] An adjustment unit, configured to adjust the positions and types of sensors deployed in the fire protection pipe network based on the priorities of sensors of different types at different positions;

[0140] The type determination unit is used to determine the type of sensor to be installed based on the priorities of different types of sensors at different locations for locations where no sensors are installed.

[0141] It should be noted that other embodiments or specific implementation methods of the sensor optimization arrangement system based on feature importance of the present invention can refer to the above-mentioned method embodiments and will not be described in detail here.

[0142] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0143] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0144] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A sensor optimization layout method based on feature importance, characterized in that: The method comprises: For the locations of different types of sensors, simulation is performed based on the fire protection pipe network simulation model to construct a data set, which includes the location and type of sensors, as well as the monitoring data of different locations and the corresponding pipe network operation status; The data set is used to train a gradient boosting tree to obtain the feature importance of monitoring data at different locations to the operation status of the pipeline network; The layout scheme of sensors of different types at different locations is determined according to the feature importance.

2. The sensor optimization layout method based on feature importance according to claim 1, characterized in that: The locations of different types of sensors are simulated based on the fire pipe network simulation model to construct training data, including: Based on the sensors already deployed in the fire protection pipe network, other deployment locations are selected to deploy sensors, forming deployment locations for different types of sensors; The fire protection pipe network model is used for simulation to obtain monitoring data of different types of sensors at different locations and the corresponding pipe network operation status, and the data set is constructed.

3. The sensor optimization layout method based on feature importance according to claim 1, characterized in that: Before simulating the locations of different types of sensors based on the fire protection pipe network simulation model and constructing a data set, the method further includes: Use sensors already deployed in the fire protection pipe network to collect real-time monitoring data of the fire protection pipe network; Establish a fire protection pipe network simulation model, perform simulation using sensors already deployed in the fire protection pipe network as simulation conditions, and obtain simulation data of the simulation model; The simulation data is compared with the real-time monitoring data to verify the fire pipe network simulation model.

4. The sensor optimization layout method based on feature importance according to claim 1, characterized in that: The formula of the gradient boosting tree is expressed as: Where g m (x i ) is the mth decision tree, M is the number of decision trees, f M (x i ) is a gradient boosting tree model containing M decision trees.

5. The sensor optimization layout method based on feature importance according to claim 4, characterized in that: The gradient boosting tree is trained using the data set to obtain the feature importance of monitoring data at different locations to the operation status of the pipeline network, including: In the dataset On the training base classifier f1(x n ), so that the basic classifier f1(x n ) has an error rate of ∈1<0.5, and the formula for the error rate ∈1 is: Where: is the sample weight, is the model prediction value, x i is the monitoring data, y i Indicates the operation status of the pipe network, n is the number of distribution points, is the judgment function, if If it holds, δ(·)=1, otherwise, δ(·)=0, f1(x n ) is the base classifier f1(·) located at x n The value of Update the sample weights of the basic classifier f1(x), which is expressed as follows: Where, is the correct sample weight after the update classification, is the updated misclassified sample weight, and d is a constant; The sample weights of the monitoring data at different distribution points are updated to obtain the characteristic importance of the monitoring data at different distribution points to the operation status of the pipeline network.

6. The sensor optimization layout method based on feature importance according to claim 5, characterized in that: The goal of the gradient boosting tree training is to minimize the loss function: Where: L(·) is the loss function, f M (x i ) is a gradient boosting tree model containing M decision trees, g m (x i ) is the mth decision tree, and argmin means finding the minimum value of the objective function.

7. The sensor optimization layout method based on feature importance according to claim 5, characterized in that: Before using the data set to train the gradient boosting tree to obtain the feature importance of monitoring data at different locations to the operating status of the pipeline network, the method further includes: For the dataset D = {P i ,F j ,S k }, P i is the pressure data at position i, F j is the flow data of location j, S k The operation status of the pipeline network; For pressure data P i and flow data F j Perform normalization processing to obtain normalized pressure data and traffic data Based on the normalized pressure data and normalized flow data, a data set is constructed y i =S k .

8. The sensor optimization layout method based on feature importance according to claim 1, characterized in that: The method of determining the placement of sensors of different types at different locations according to the feature importance includes: Sort the features from high to low importance and determine the priority of different types of sensors at different locations; For sensors already deployed in the fire protection pipe network, adjust the positions and types of deployed sensors based on the priorities of different types of sensors at different locations; For locations where sensors are not installed, the type of sensors to be installed is determined based on the priorities of different types of sensors at different locations.

9. The sensor optimization layout method based on feature importance according to claim 1, characterized in that: The method further comprises: Different types of sensors are deployed based on the deployment plan to collect monitoring data of the fire protection pipe network.

10. A sensor optimization placement system based on feature importance, characterized in that: The system comprises: A data set construction module is used to simulate the locations of different types of sensors based on the fire protection pipe network simulation model to construct a data set, which includes the location and type of sensors, as well as monitoring data of different locations and the corresponding pipe network operation status; A feature importance calculation module is used to train a gradient boosting tree using the data set to obtain the feature importance of monitoring data at different locations to the operating status of the pipeline network; The deployment plan determination module is used to determine the deployment plan of sensors of different types at different locations according to the feature importance.

Images