Method for determining the position of a sensor, device and computer program product thereof
Through simulation of abnormal scenarios and genetic algorithm optimization, the complexity of sensor position arrangement in the water distribution network is solved, and the effect of rapid detection of water quality abnormalities is achieved, reducing the computing resource requirements and economic costs.
Patent Information
- Application Number
- CN201980045698.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2018-10-10
- Filing Date
- 2019-07-08
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2039-07-08
AI Technical Summary
In water distribution networks, it is difficult for the prior art to quickly detect water quality abnormalities with as few sensors as possible, and the optimal arrangement of sensor locations is a complex combination optimization problem, and the inability to effectively utilize limited computing power to achieve the best Detection effect.
By simulating anomaly scenarios, a candidate set of sensor location is determined, and a genetic algorithm is used to iterate the sensor location until the stop standard is reached, ensuring that the sensor can be optimally placed in the network to detect anomaly.
The optimal placement of sensors in the water distribution network is achieved to quickly detect abnormalities, reduce algorithm complexity and computing resource requirements, and ensure the balance of security and economic costs of network users.
Smart Images

Figure CN112424599B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of fluid distribution networks, such as water or gas distribution networks. More specifically, the present invention relates to determining the best positions of sensors allowing the measurement of physicochemical fluid data. Background Art
[0002] The water distribution network may be affected by numerous anomalies that affect the water quality. For example, the water may be polluted by various pollutants introduced intentionally (malicious acts) or accidentally by backflow or connection errors, by particles of matter that are regularly deposited in the pipes and suddenly resume suspension, by reactions with the constituent materials of the network, or by bacteria due to bacterial regrowth phenomena. In order to prevent the damaging effects that may be caused by pollution, or more generally, user discomfort (changes in color, taste, odor, etc.) due to a decrease in water quality, it is important to detect and characterize any events that affect the water quality as quickly and accurately as possible. Problems that affect water quality may also occur in natural water systems, such as waters or waterways of lakes, ponds or rivers or collections of natural waters or waterways. These systems may be affected by, for example, accidental pollution or abnormal growth of algae, which can seriously undermine their condition and hinder their use (drinking water, bathing water, etc.).
[0003] Water distribution networks can also see other anomalies occur such as leaks, over- or under-pressure, incorrect valve positions, etc. Other fluid distribution networks such as gas distribution networks or oil pipeline networks can be affected by anomalies such as over-pressure or leaks requiring corrective action.
[0004] In the case of using the connected sensors, the effectiveness of detecting anomalies in a fluid distribution network can be significantly increased. Such sensors allow one or more physicochemical parameters of the fluid (pressure, velocity, turbidity, pH, etc.) to be measured at one point of the network, and these parameter values are analyzed locally or transmitted to a server for analysis. Thus, anomalies or contamination of the fluid distribution network can be detected automatically, and corrective or preventive actions can be taken immediately. French patent application No. 1763286 filed by the applicant of the present application describes a device for detecting anomalies in a continuous body of water based on the measurement of physicochemical parameters in the continuous body of water.
[0005] Anomalies in a fluid distribution network may appear at various points in the network. For example, contamination in a water distribution network may appear at the entry point to the network in a reservoir, but may also appear at another point in the network, such as a plant, which may trigger a backflow. Contamination generated in this way, by its nature, can only be detected when at least one sensor has been able to detect an abnormal change in a parameter, that is, when the contamination has reached this sensor. Failure to detect contamination is very problematic, especially for drinking water distribution networks, which may affect the health of network users. It may also lead to problems that do not affect the health of users and often require both therapeutic and harmful actions to ensure continuity of service.
[0006] The detection of anomalies can be improved by increasing the number of sensors in the water distribution network. However, these sensors are expensive to purchase and use, especially for their maintenance. It is therefore important to detect anomalies as quickly as possible over the largest possible part of the water distribution network with the smallest possible number of sensors. Therefore, a suitable arrangement of the sensors is essential to achieve satisfactory detection of anomalies with a limited number of sensors.
[0007] The problem of optimal placement of a given number of sensors in a water distribution network is an extremely complex combinatorial problem. In practice, it is not possible to test all solutions on a water distribution network of standard size (e.g., on the scale of a town or region) because exhaustive testing of all possible combinations of sensor locations would require computing power too large to be implemented in practice.
[0008] Cheifetz, N., Sandraz, A.C., Feliers, C., Gilbert, D., Piller, O. and Heim, V. (2017), A greedy algorithm for the positioning of quality sensors in a large water distribution network, Techniques Sciences Méthodes, (11), 55-63, proposes the use of a so-called "greedy" algorithm for the placement of sensors in a water distribution network, that is, by iteratively placing a given number of sensors one after another in positions that are considered to be optimal in combination with the already placed sensors. Although it allows a fairly satisfactory sensor placement to be obtained using limited computing power, this algorithm generally does not allow the optimal placement of sensors to be obtained. In fact, once a sensor is placed, its position is no longer questioned, even if this position (which was considered optimal based on the positions of previously placed sensors) is no longer the case for a larger number of sensors.
[0009] There is therefore a need for a method that allows for the optimal placement of a given number of sensors in a water distribution network in order to detect anomalies in the water distribution network as quickly as possible, regardless of the source of the anomaly.
[0010] More generally, there is a need for a method that allows for optimal placement of a given number of sensors in a fluid distribution network in order to optimize detection of anomalies in the fluid. Summary of the invention
[0011] The invention achieves this object by simulating scenarios in which at least some anomalies are introduced into the fluid; by determining a candidate set of sensor locations; by determining simulated values of parameters of the fluid at each location and in each scenario and the associated anomaly detection; by attributing to each candidate set of sensor locations a score representing the effectiveness of the sensors placed at the locations of each candidate set in being able to detect anomalies. Finally, the candidate set can be modified using a so-called genetic algorithm until a stopping criterion is verified. The genetic algorithm can, for example, include crossover or mutation of the candidate set.
[0012] In this regard, the present invention describes a method for determining a set of locations of multiple sensors for one or more physicochemical parameters of a fluid in a fluid transmission network, the method comprising: obtaining simulated values of physicochemical quantities at a set of points in the network for a set of time steps for at least one set of input parameters; obtaining multiple candidate sets of locations of multiple sensors; iteratively performing the following operations until a stopping criterion is verified: for each candidate set of locations, obtaining a performance score based at least on the simulated values of the physicochemical quantities at the locations of multiple sensors; modifying the multiple candidate sets of sensor locations, the modification comprising at least one operation selected from the following items: retaining at least one candidate set of locations with the most favorable scores; adding at least one candidate set of sensor locations defined by combining the locations of at least two candidate sets; adding at least one candidate set of sensor locations defined by modifying the locations of sensors in the candidate sets; selecting the candidate set of locations with the most favorable scores.
[0013] In this specification, "a plurality of sets" should be understood to mean a set of combinations of positions in a network for a given number of sensors. Thus, the optimization aims to determine the best combination for a given number of sensors to obtain the highest performance score. By means of the step of modifying the plurality of sets, a plurality of combinations to be tested are defined and are evolved in each iteration.
[0014] The stopping criteria are a combination of rules based on the performance level of the population evaluated during the optimization process at each iteration and independent complementary rules: -Rule based on the performance level measured at each iteration: a certain number of consecutive iterations have been performed without the performance level of the best set, for example, 5% of the population with the best performance level, evolving beyond a certain percentage. Reaching a stable level -Rule independent of the performance level: reaching a pre-set maximum number of iterations. This maximum number of iterations depends on the complexity of the problem (it increases when the system to be optimized is more complex and more iterations are needed).
[0015] The "best score" is the value of the objective function calculated for a combination of sensor locations that provides a result that best satisfies the goal assigned to the sensor placement. Depending on the circumstances, the best score is a value greater than another value of the objective function for a less favorable combination of sensor locations. For example, if the score corresponds to the number of nodes protected by the combination of sensors, the best score will be the highest score, corresponding to a greater number of protected sensors. On the other hand, if the score corresponds to the number of nodes for which anomalies have not yet been detected or have been detected too late, the best score corresponds to the lowest value, and therefore to the lowest possible number of nodes for which anomalies have not yet been detected. The score can be calculated based on one or more elements: anomaly detection capability, the number of users who detect anomalies in time, the equivalent consumption of timely detected anomalies, the rapidity with which anomalies are detected, the detection of a set of anomalies that have a significant impact on users, the cost of sensor deployment, etc.
[0016] Genetic type algorithms involve modifying the candidate set between iterations by using one or more operations selected from:
[0017] - Retain: A portion of the candidates with the most favorable scores are retained, and the remaining candidates are eliminated;
[0018] - Crossover: combining the positions of sensors of at least two sets. The set between which the crossover is performed can be a previously selected set;
[0019] - Mutation: For example, randomly modifying one or more sensor positions in the set.
[0020] In all cases, modifying the candidate set of sensor locations using a genetic type algorithm allows achieving a placement of the sensors that is close to the optimal placement.
[0021] The method exhibits a relatively low algorithmic complexity (e.g., it is much less greedy than other optimization methods, such as a "brute force" method involving testing all options), and can be executed using general computing power, such as a personal computer or server, with execution times that may vary, but remain reasonable. Thus, a method according to the invention applied to a network of moderate complexity can be executed on a personal computer in minutes to hours, whereas the execution of a "brute force" method would take much longer or, in practice, would even be impossible to execute on such a computer.
[0022] The method is applicable to any type of fluid distribution network.
[0023] The method allows sensors to be placed in the network to optimally detect anomalies at a low cost.
[0024] The method ensures a level of security for network users using a limited number of sensors.
[0025] In an exemplary embodiment, the network is modeled in the form of a graph; each node or arc of the graph is identified by a unique identifier; and the location of the sensor is defined by the identifier of the node or arc of the graph.
[0026] This representation allows simple and efficient identification of the location of the sensor.
[0027] In an exemplary embodiment, at least one set of input parameters includes an introduction of an anomaly in at least one point of the network at at least one time step.
[0028] Advantageously, this allows sensors to be placed to optimize detection of anomalies in the network.
[0029] In an exemplary embodiment, the performance score is calculated based on the following items: the ability of sensors placed in a candidate set of locations to detect at least one anomaly by determining the time step for at least one anomaly to arrive at a set of points of the network; and at least one objective function selected from the following items: the number of points in the set that detected the anomaly before the anomaly arrived; the number of points in the set that detected the anomaly before the anomaly arrived, weighted by the number of users or the consumption of each point; the number of points where the anomaly has not been detected; the number of points where the anomaly has not been detected, weighted by the number of users or the consumption of each point.
[0030] This allows sensors to be optimally placed to prevent anomalies with strong impact (eg, at a water reservoir) from going undetected before affecting network users.
[0031] In an exemplary embodiment, the performance score is calculated based on at least one characteristic selected from the capabilities of the sensors of the candidate set to evaluate a quality indicator of the fluid in the network, and a deployment cost of the sensors.
[0032] This allows the placement of sensors to be optimized for detecting anomalies and other objectives, such as traceability of fluid quality throughout subsections of the network, protection of sensitive locations, or limiting deployment costs.
[0033] In an exemplary embodiment, at least one sensor has a predefined position.
[0034] This allows the presence of pre-existing sensors to be taken into account and the placement of additional sensors to be optimized relative to the pre-existing sensors.
[0035] In an exemplary embodiment, the points at which sensors are placed are restricted to a subset of the points of the network.
[0036] This allows considering certain points of the network where sensor placement is not allowed, and optimizing sensor placement relative to possible placement points. Sites can be selected using site checking based on a priori knowledge of the network's infrastructure, and sites that exhibit too many constraints or too high implementation costs are invalidated.
[0037] In an exemplary embodiment, the definition of a plurality of candidate sets of locations for a plurality of sensors includes a definition of locations at points of interest for the network.
[0038] This allows placing sensors at sensitive points, such as industrial sites, areas interconnected with large water purchases or reservoir outlets, from the first iteration. This allows efficient placement of sensors from the first iteration. As a result, the method converges faster to a solution close to the optimum and requires less computational complexity to be performed.
[0039] In an exemplary embodiment, the definition of the plurality of candidate sets of locations for the plurality of sensors includes definition of locations at nodes connected to a number of arcs greater than or equal to a predefined threshold.
[0040] This allows placing sensors only at nodes that are connected to a number of arcs greater than a threshold in the first iteration (e.g., nodes connected to at least 2, 3, or 4 arcs). Statistically, these nodes are more likely to detect an abnormal arrival or protect more downstream nodes. Thus, starting from the first iteration, this allows for efficient placement of sensors. Thus, the method is likely to converge to a near-optimal solution with less computational complexity to be performed.
[0041] In an exemplary embodiment, the stopping criteria include one or more conditions selected from the following items: a maximum number of iterations; a comparison of the most favorable score in the current iteration and at least one previous iteration, and verification of the stopping criteria when the difference between the most favorable score in the current iteration and at least one previous iteration is less than or equal to a predefined threshold.
[0042] A stopping criterion based on a maximum number of iterations allows the execution time and the computational resources required by the method to be limited.
[0043] A stopping criterion based on a comparison between the most favorable scores between the current iteration and the previous iteration allows the method to be stopped when the optimization gain in the placement of sensors between at least two consecutive iterations becomes negligible. Once the scores have reached a plateau, a smaller number of iterations may be performed.
[0044] In an exemplary embodiment, eliminating at least one candidate set of positions that do not have the most favorable score includes eliminating all candidate sets except a predefined number or a predefined ratio of candidate sets with the most favorable scores; adding at least one candidate set of sensor positions defined by combining the positions of at least two candidate sets, and adding at least one candidate set of sensor positions defined by modifying the positions of sensors in the candidate sets plus a number of candidate sets equal to the number of candidate sets eliminated.
[0045] This allows to keep a constant number of candidate sets between iterations and to keep a given number of candidate sets with the most favorable scores between two iterations. Therefore, the most favorable score in a set in one iteration must be at least equal to the most favorable score in the previous iteration. Furthermore, this makes it possible to keep a pool of candidate sets with a good performance level, which will also be able to generate new candidates with a good performance level by mutation or crossover. This allows the method to converge faster.
[0046] In an exemplary embodiment, the method comprises: definition, iterative modification and selection of a candidate set of sensors for a plurality of predefined numbers of sensors, respectively.
[0047] This allows the performance levels of different numbers of sensors to be compared.
[0048] In an exemplary embodiment, for an integer (n) number of sensors, obtaining a plurality of candidate sets of positions for the plurality of sensors is based on the candidate sets of positions having the most favorable scores for an integer (nm) number of sensors, where 1≤m <n。
[0049] This allows the most efficient placement for multiple (nm) sensors to be used as the basis for an initial placement for (n) sensors, where m is an integer between 1 and n, and this most efficient placement for the (nm) sensors provides a good basis for the efficient placement prior for the (n) sensors.
[0050] In an exemplary embodiment, the method includes selecting one of the candidate sets having the most favorable score for a plurality of predefined numbers of sensors based on the scores of each of the candidate sets and the deployment cost.
[0051] This allows choosing the number of sensors that provides the best trade-off between deployment cost and anomaly detection efficiency.
[0052] The present invention also describes a method for placing multiple sensors for one or more physicochemical parameters of a fluid, the method comprising: determining a set of positions in a fluid transmission network, the determination comprising the steps of the method for determining a set of positions of multiple sensors for one or more physicochemical parameters of a fluid according to the present invention, and placing the sensors in the set of positions.
[0053] The present invention also describes a computer program product comprising program code instructions stored on a computer-readable medium, including a processor for determining a set of positions of multiple sensors for determining one or more physicochemical parameters of a fluid in a fluid transport network, wherein the computer program comprises a computer-readable programming unit for executing the method according to the present invention for determining a set of positions of multiple sensors for determining one or more physicochemical parameters of a fluid in a fluid transport network.
[0054] The present invention also describes a device capable of determining a set of positions of multiple sensors for one or more physicochemical parameters of a fluid in a fluid transport network, the device comprising a processor configured to execute a method according to the present invention for determining a set of positions of multiple sensors for one or more physicochemical parameters of a fluid in a fluid transport network.
[0055] In an exemplary embodiment, the device can be a sensor that can calculate its optimal position and display it to the user through a display unit, and the user will position the sensor at the optimal position by considering all sensors that have been deployed or are currently being deployed. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Other features will become apparent on reading the following detailed description given as a non-limiting example, in light of the accompanying drawings, which show:
[0057] - Figure 1 represents an example of a probe capable of being placed according to the method of the invention, aimed at detecting anomalies in a water distribution network;
[0058] - Figure 2 A flow chart showing a method for placing a sensor according to the present invention;
[0059] - Figure 3 represents an example of a water distribution network to which the present invention can be applied;
[0060] - Figure 4 An example of anomaly detection in a fluid distribution network according to a set of embodiments of the present invention is shown;
[0061] - Figure 5 An example of a definition of optimization constraints representing a set of implementations according to the present invention;
[0062] - Figure 6 Examples of histograms representing different test scenarios for sensor placement, showing the percentage of nodes of the water distribution network affected by each type of contamination;
[0063] - Figures 7a, 7b, 7c and 7d represent four examples of placing 2, 5, 8 and 11 sensors in a water distribution network, respectively, for the sole purpose of detecting contamination, by means of a method for placing sensors in an embodiment of the present invention;
[0064] - Figure 8 representing the trend of the detection of anomalies in percentage of sensors placed in the water distribution network by the method according to the invention as a function of the number of sensors;
[0065] - Fig. 9 An example showing flow distribution in a water distribution network;
[0066] - Fig.10 represents an example of water quality monitoring via sensors located at downstream nodes of a water distribution network;
[0067] - Figures 11a, 11b, 11c and 11d respectively represent four examples of placing 2, 5, 8 and 11 sensors in a water distribution network for the purpose of both detecting pollution and monitoring downstream water quality by a method for placing sensors in an embodiment of the present invention;
[0068] - Fig.12 Shows the trend of pollution detection and water quality monitoring scores according to the combination of the number of sensors placed by the method according to the present invention. DETAILED DESCRIPTION
[0069] In the description below, the method according to the invention is mainly illustrated by examples involving the optimal placement of sensors for detecting pollution and monitoring water quality in a drinking water distribution network. However, the invention is not limited to these examples and can be applied to the placement of sensors for detecting other anomalies in a water network, such as leaks and other purposes. The invention can also be applied to the optimal placement of sensors in other fluid distribution networks such as gas distribution networks or oil pipeline networks.
[0070] Figure 1 An example of a probe intended to detect anomalies in a water distribution network that can be placed by the method according to the invention is represented.
[0071] The probe 100 is intended to detect anomalies in a water distribution network. The probe can be placed at any location in the water distribution network, such as at the outlet of a drinking water production plant, at the outlet of a reservoir, at a consumption point or at any other point. The applicant has filed French patent application number 1763286 relating to a probe such as the probe 100, which enables the determination of anomalies occurring in a continuous body of water.
[0072] The probe 100 is connected to a pipe 130 of a water distribution network, for example through a set of sensor cables 120, or directly to the network (plug-in), and is powered by electricity 110 or batteries.
[0073] The probe 100 can include one or more sensors of physical quantities of the water distribution network. For example, the probe 100 can include one or more sensors selected from the following sensors: chlorine concentration, temperature, TOC (total organic carbon), UV 254 (absorption of 254 nm wavelength ultraviolet light by water), conductivity, pH, color, turbidity, particle count, bacteria count, dissolved oxygen, chlorophyll a, or any sensor that can characterize a physical quantity of water.
[0074] Thus, in one set of embodiments of the invention, the probe 100 allows the measurement of a set of parameters representative of water quality to be performed at one point in a water distribution network.
[0075] In one set of embodiments of the invention, the probe 100 comprises a communication unit for transmitting the measurements of the embedded sensor. For example, the probe can comprise a wired or wireless connection to a server for sending the measurements to a server configured to detect anomalies in the water distribution network. The probe 100 can thus be coupled to a smart water consumption sensor that sends remotely read consumption data by combining the sending of water consumption data and sensor measurements.
[0076] In one set of embodiments of the invention, the probe comprises a processor configured to detect and characterize anomalies in the water distribution network based on the sensor measurements.
[0077] An example of detecting and characterizing anomalies by a processor is provided, for example, in French patent application No. 1 763 286 .
[0078] Although the probe 100 represents an example of a probe in a water distribution network, such a probe can also be deployed in water in a natural environment, such as in a lake, pond, river or in any other aquatic system, and possibly at different depths.
[0079] Probes such as probe 100 allow anomalies in the water distribution network, such as contamination, to be detected. However, these anomalies can only be detected from the measurements performed by the probe. Therefore, contamination generated upstream of the probe 100 can affect the user without being detected. Therefore, it is important to arrange multiple probes in the network to detect contamination in the water distribution network early enough to protect the user. However, probes such as probe 100 have non-negligible production, use and maintenance costs.
[0080] One object of the invention is to provide an optimal placement of a limited number of probes in a water distribution network, in order to obtain anomaly detection that is as effective as possible and, therefore, to provide good protection for the user, while retaining reasonable economic costs. More generally, one object of the invention is to optimally place a limited number of sensors allowing detection of anomalies in the network, in a fluid distribution network, in order to obtain the most effective anomaly detection at limited operating costs.
[0081] Figure 2 A flow chart showing a method for placing a sensor according to the present invention.
[0082] Method 200 is a method for determining a set of locations of a plurality of sensors for determining one or more physicochemical parameters of a fluid in a fluid distribution network. The fluid distribution network can be, for example, a water or gas distribution network or an oil pipeline network.
[0083] The method 200 is based on at least one set of input parameters 210. The input parameters allow the definition of simulation scenarios by defining the initial state of the network and the parameters that modify the state of the network. Thus, the operation of the network can be simulated within a set of time steps. The parameters that modify the state of the network can correspond to the normal operation of the network (for example, water consumption at a certain point in the network), or to the introduction of anomalies for a given time step. The introduction of anomalies can be done in different ways. For example, anomalies can be introduced at predetermined points corresponding to points of interest of the network (for example, introducing contamination at the outlet of a factory or at a fire hydrant). Anomalies can also be introduced randomly at different points of the network. Anomalies can correspond to items of variable severity, such as leaks, overpressure, or accidental or intentional (sabotage) contamination of the network.
[0084] The method 200 comprises obtaining 220 simulated values of physicochemical quantities at a set of points of a network for a set of time steps for at least one set of input parameters 210. Thus, the operation of the network can be simulated at different time steps, and different quantities (temperature, pH, velocity, turbidity, etc.) can be simulated at different points of the network for a set of time steps. The simulation can be performed each time the method is executed. It can also be performed only once, for example at the initialization of the method or at the first iteration, or it can have been pre-calculated. In this case, it is only necessary to recover the simulated values to obtain the behavior of the network in the simulated situation.
[0085] The method 200 then includes obtaining 230 a plurality of candidate sets 240 of positions of the plurality of sensors. In practice, the method 200 aims to determine the best combination of positions of the sensor set. As will be explained below, the method tests different candidate sets of positions and then iteratively modifies them in order to identify the most effective combination iteratively again and again.
[0086] The network and the location of the sensors can be represented in different ways. For example, the network can be modeled in the form of a graph formed by nodes and arcs. The nodes can correspond to places where fluids are produced or consumed (i.e., reservoirs, groups of consumers, factories, etc.), and the arcs can correspond to different infrastructures that allow the fluid to be transported between nodes, such as pipelines. Each node or arc can be identified by a unique identifier, and the location of the sensor can then be defined by the identifier of the node or arc where the sensor is located.
[0087] Obtaining 230 a set of sensor positions enables having a set of initial positions for the set of sensors. These initial positions can be obtained in different ways.
[0088] For example, at least one sensor can have a predefined position. For example, it can be a sensor that is already arranged in the network, whose position remains fixed. Thus, the positioning of the additional sensor can be optimized by taking into account the already existing sensors.
[0089] The location of sensors can also be limited to a subset of points of the network, such as certain nodes and / or arcs. This can avoid placing sensors in locations that are difficult to access, where the deployment cost of the sensors is too high, where there is no space necessary to arrange the sensors, or where a sufficient level of transmission is not allowed, for example.
[0090] Among the possible placement points, the sensors can be arranged randomly initially, for example. Locations can also be defined from points of interest of the network (eg factories, water plants, etc.), which represent good candidates for optimal sensor placement.
[0091] Sensors can also be placed only in a subset of points corresponding to good candidates for optimization. This allows obtaining a good level of performance while reducing the complexity of the method. Depending on the goal sought, for example, good candidates can be only points with high performance.
[0092] Sensors can also be placed at network points connected to a large number of arcs, for example, only at nodes connected to a number of arcs greater than or equal to a predefined threshold.
[0093] These different types of initial placement can be combined. Other ways of obtaining initial sensor placement can also be envisioned. Generally, it is desirable that the initial placement of sensors allows the measurements from the sensors to detect phenomena occurring in the network as effectively as possible.
[0094] The method 200 includes iteratively obtaining performance scores for a candidate set of sensors, and modifying the candidate set using a genetic type algorithm.
[0095] The iteration then comprises obtaining 250 for each candidate set of positions at least from simulated values of physicochemical quantities at the positions of the plurality of sensors a performance score. The performance score allows determining the point at which a set of sensor positions is able to fulfill a given task based on sensor measurements.
[0096] In one set of embodiments of the invention, analog values of the physicochemical quantities are obtained directly for the sensor locations and the performance score is calculated from these values.
[0097] The method 200 can also include an intermediate step of obtaining predictions of sensor measurements for each candidate set of the plurality of positions for each set of input parameters. This allows, for each scenario and each candidate set, simulation of the measurements to be performed by the sensor. The measurements can be considered "perfect", i.e., the sensor measures at each time step exactly the value of the parameter at the position of the sensor at that time step. It is also possible to simulate the imperfections of the sensor, for example by using a sensor measurement model. Different types of models can be used, such as using Gaussian noise to represent measurement errors, using sliding windows to represent the time granularity of the measurements, etc. Thus, the placement of the sensor can be optimized using more realistic modeling than what is perceived by the sensor.
[0098] Thus, sensor measurement predictions can be obtained by calculating the measurements, by selecting the measurements from simulated values of the physicochemical parameter on the network, or by recovering pre-calculated measurement predictions.
[0099] Thus, depending on the objective being pursued, the performance score can be determined in different ways. The performance score can be calculated according to an objective function that represents the objective of optimizing the appropriate placement of sensors. The objective function can, for example, reflect the ability of the sensor to detect contaminants before they reach the user; the ability of the sensor to detect anomalies in the network; the ability of the sensor to assess a fluid quality indicator in the network; the cost of deploying the sensor; a combination of factors, such as those previously indicated. In general, the performance score can be defined so as to allow for maximum sensor efficiency at the lowest possible deployment cost.
[0100] At the end of the first step of obtaining a performance score of 250, the method 200 includes modifying the plurality of candidate sets of sensor positions. This modification is performed by a so-called genetic algorithm, that is, by at least one operation selected from the following:
[0101] - retaining 271 at least one candidate set of the positions with the most favorable scores. This so-called selection operation consists in retaining the best candidate sets, that is to say those sets that have obtained the best scores. This makes it possible to ensure that in each iteration the best score among the candidate sets will be at least as good as in the previous iteration. Furthermore, this allows retaining a pool of very good candidates, which will improve as the iterations proceed. In each iteration, the candidates that are not retained will be replaced by new candidates, for example by the following operation;
[0102] - Adding 272 at least one candidate group of sensor positions defined by a combination of positions of at least two candidate sets. This so-called crossover operation consists in combining positions derived from several candidate sets.
[0103] - Adding 273 at least one candidate set of sensor positions defined by a modification of the position of a sensor in the candidate set. This so-called mutation operation consists in randomly modifying the position of at least one sensor in the candidate set.
[0104] Therefore, crossover and mutation can add a number of candidates to the set equal to the number of unretained candidates so that a constant number of candidates is retained.
[0105] Different operations can be combined in different ways to generate a certain percentage of candidates for each iteration. In one set of embodiments of the invention, the 5% of candidates with the best scores are selected, 80% of the candidates with medium scores are crossed, and the remaining 15% are randomly mutated.
[0106] Starting from the second score calculation iteration, i.e. after at least the first modification of the set, the verification of the stopping criterion after the calculation of the score can be verified 260. The verification can be verified in each iteration or only in certain iterations (e.g. once in every two iterations, once in every three iterations, etc.). The verification of the stopping criterion can also be verified only from a given number of iterations (e.g. from 10, 15, 20 iterations).
[0107] The stopping criterion can be based on one or more conditions. For example, the stopping criterion can be based on a maximum number of iterations. The stopping criterion can also be based on a difference between the most favorable score in the current iteration and the most favorable score in at least one previous iteration. In this case, the stopping criterion is verified when the most favorable score no longer changes or only changes slightly after one or more consecutive iterations.
[0108] The stopping criterion can also be based on a combination of these conditions. For example, the stopping criterion can be verified when a maximum number of iterations is reached and / or when the most favorable score does not change any more after one or more iterations.
[0109] As long as the stopping criterion is not verified 260, a new iteration of modification of the candidate set and score calculation is performed.
[0110] When the stopping criteria is verified 260 , the method 200 includes selecting 290 the candidate set of positions having the most favorable scores.
[0111] Thus, method 200 allows obtaining an optimal set of sensor locations in order to achieve the desired functionality (anomaly detection, monitoring fluid quality traceability, etc.) Method 200 has the advantage of being able to be executed in a reasonable time using general purpose computing power while identifying optimal or near-optimal results.
[0112] In one set of embodiments of the invention, method 200 can be performed for a different number of sensors (eg, several predefined numbers of sensors).
[0113] In one set of embodiments of the invention, the method is performed independently for different numbers of sensors.
[0114] In one set of embodiments of the invention, obtaining 230 a candidate set of initial positions is at least partially based on the set selected for a smaller number of sensors. For example, the best set for a number of sensors equal to 12 is used as a basis for an initial set of candidate sensors with 13 or 14 sensors. Indeed, it can be assumed that the best set to which several sensors will be added can provide a good first candidate for a larger number of sensors.
[0115] Then, the sensor that exhibits the best ratio between the performance score and the deployment cost or the number of sensors can be selected from the best candidates for different numbers of sensors.
[0116] Once the position of the sensor is determined, its placement can be done in different ways. Therefore, the invention also relates to a method for placing a sensor at an optimal position. In one set of embodiments of the invention, the sensor is arranged at the optimal position, for example by an operator.
[0117] In one set of embodiments of the invention, the sensor has the ability to place itself. This is the case, for example, with an automatic sensor robot that can move around in the network. Such a sensor can be positioned in different ways, for example using a GPS receiver, using positioning relative to a fixed radio transmitter (positioning relative to 3G, Wi-Fi, etc. antennas whose positions are known). If the robot is equipped with a topographic map of the network and is able to identify, for example, using a camera, where the network is located, such a robot can also be positioned relative to the map of the network. The sensor robot is thus able to autonomously determine its optimal position by taking into account the sensors that are already present and automatically place itself there. This offers the advantage that no operator intervention is required to place the sensor.
[0118] A group of sensor robots can also operate in a network mode by automatically being placed at an optimal position relative to each other. In addition, when the situation in the network changes (e.g., if a new sensor is added, a sensor is moved to another network, a sensor fails, etc.), the sensor can automatically recalculate its optimal position and move to that position without operator intervention.
[0119] Figure 3 An example of a water distribution network to which the present invention can be applied is shown.
[0120] The water distribution network 300 is a town-sized network. Sector 300 is contained in a large network with approximately 25,000 nodes. The sector consists of Figure 3 The hydraulic model represented by the embodiment includes 3305 nodes (e.g., nodes 310, 311, 312, 313 representing consumption points) and 3639 arcs (e.g., arcs 320, 321, 322 representing pipes). The nodes are connected to different numbers of pipes. For example, nodes 311 and 312 are connected to two pipes, while nodes 310 and 313 are connected to three pipes.
[0121] The operation of the network 300 can be simulated through successive time steps by using a hydraulic model. In particular, the water consumption can be combined at different time steps (e.g., according to consumption profile assumptions), and the pressures at the nodes of the network and the velocities on the arcs derived therefrom at each time step. Depending on availability and the wishes of the operator, the hydraulic model selected can represent the median consumption profile of the network or extreme values (such as consumption peaks). Trends in other physicochemical parameters (such as chlorine concentration, temperature, etc.) within the network over the simulation time period can also be derived therefrom.
[0122] In one set of embodiments of the invention, anomalies or contamination can be introduced into different nodes of the network. These anomalies may represent events such as accidental contamination or malicious behavior.
[0123] Each node or arc can be represented by a unique identifier. The placement of a sensor can be performed by assigning to the sensor a unique identifier of the node or arc on which it is placed.
[0124] Sector 300 is given by way of non-limiting example only. Thus, the present invention can be applied to many other networks, such as networks of different sizes (regional, regional, etc.), or networks other than water distribution networks (oil pipelines, natural gas pipelines, natural gas distribution networks, etc.).
[0125] In the following description, the invention will be illustrated by a use case consisting of optimal placement of sensors in a network 300 with the goal of limiting the impact caused by rapid degradation of water quality. These degradations can be limited by rapid detection. They are called contamination and can occur at any node of the network in different ways:
[0126] - intentional introduction of toxic substances;
[0127] - Return of customers (individuals or industries);
[0128] - penetration and working problems (refilling with water, no prior disinfection, particles, etc.);
[0129] - Excessive release of biofilm due to excessive speed, leading to an increase in bacteria and organic matter;
[0130] -etc.
[0131] Pollutants do not all have the same impact, depending on their location in the network and, in particular, when they occur. Hydraulic modeling allows these aspects to be taken into account by numerically simulating the pollution on a large number of network nodes. Hydraulic modeling therefore allows the effects caused by different pollution to be mapped, in theory, for each scenario. This mapping of the effects makes it possible to determine the combination of nodes that best suits the sensors to be installed therein and to limit or best detect these effects. The invention therefore allows the network to be covered as evenly as possible by a limited number of sensors, by taking into account the hydraulic behavior of the network.
[0132] In one set of embodiments of the invention, contamination is simulated by introducing a contaminant with a given concentration into the numerical model at a given time step. It is also possible to introduce non-reactive substances (tracers) with a dimensionless concentration (e.g. 100) into the numerical model. Different contaminations can be introduced at different points depending on the scenario. In addition to contamination (overpressure, bacterial growth, etc.), anomalies can also be introduced. In this case, it is assumed that the sensor is able to detect contamination above a certain tracer concentration. This is useful for placing, for example, Figure 1 Sensors such as the one shown are particularly effective as they are able to detect many different types of contamination.
[0133] Depending on the scenarios, anomalies can be injected at key points or points of interest in the network (reservoirs, plants, etc.), or even randomly. The placement of sensors can be optimized for a set of scenarios corresponding to the occurrence of anomalies at different points of the network.
[0134] In one set of embodiments of the present invention, anomalies are introduced at nodes randomly selected from nodes strictly connected to more than two pipelines. These nodes can be called highly connected nodes and reflect the structure of the network well by connecting to main branch lines and auxiliary branch lines. Introducing anomalies at nodes randomly selected from nodes strictly connected to more than two pipelines enables a significant reduction in overall computation time while retaining a large number of scenarios to estimate the distribution of impact.
[0135] Thus, the node introducing the anomaly can be randomly selected from a list of nodes corresponding to the chosen criteria (e.g., nodes strictly connected to more than two pipes), listed by its unique identifier, for example, in the following table:
[0136]
[0137]
[0138] Table 1 - Example of a list of highly connected nodes of a network
[0139] In the simulation, the introduced pollutant or substance constituting the digital tracer then diffuses in the network. With successive time steps, the progress of the substance or pollutant is calculated by propagation and dilution via the physical equations, whether a reaction occurs or not.
[0140] In one set of embodiments of the invention, a so-called influence matrix and a so-called detection matrix are used to determine the ability of a sensor to detect pollution. The influence matrix includes scenes in each column and nodes or network points in each row. It provides for each scene (column) and for all points (row) the time step in which pollution reaches the node in the corresponding simulation. It is constructed based on the transmission time, after which the amount of pollution for each node is higher than a predefined value, for example 10 -6 Here, this is a very low threshold, used to model the fact that contamination (even very slight contamination) poses a risk to the user. Therefore, even with an unlimited amount of tracers, the node can be exposed.
[0141] The following table provides an example of an impact matrix, where each time step represents 10 minutes. For example, the value "36" in the cell of the first row and second column means that in the second scenario, the pollution reaches node number 1 after 36×10=360 minutes (6 hours). Here, the value "144" corresponds to the maximum simulation time step (24h in this example), at which time the pollution has not yet reached the node.
[0142]
[0143]
[0144] Table 2 - Example of an impact matrix
[0145] Each column of the impact matrix is specific to each scenario and the impact time remains constant regardless of the placement of the sensor. In a set of embodiments of the invention, the impact matrix is therefore calculated once for all simulated scenarios, or has been pre-calculated and read from a file so that the impact of the pollution at each sensor placement does not have to be recalculated unnecessarily. In one embodiment, once the calculation of the impact matrix has been performed, only a subset of the scenarios of the impact matrix are used. For example, only columns corresponding to pollution scenarios with certain properties can be used: scenarios involving the most nodes or consumption equivalents affected, nodes involving a percentage of affected nodes lying between two thresholds, etc. After simulating the scenarios, this allows to consider only those corresponding to the impacts that are trying to be minimized.
[0146] The detection matrix is constructed in a similar manner to the influence matrix, but indicates the time steps at which contamination is detected for a given node. In one set of embodiments of the invention, contamination is detected when the concentration of the contaminant or tracer at a node equipped with a sensor is above a predefined threshold. For example, the detection matrix can include time steps at which a concentration greater than or equal to 1 has been detected by an upstream node equipped with a sensor. This makes it possible to model the fact that the sensitivity of the sensor is limited and a contaminant may not be detected when its concentration is too low, even though it still has an impact.
[0147] According to different embodiments of the present invention, a number of objective functions can be established based on the comparison between the detection time and the arrival time of the pollution. For example, the objective function can include maximizing the number of nodes for which pollution is detected before the pollution arrives. Conversely, the objective function can include minimizing the number of points for which pollution has not yet been detected. In both cases, the number of points can be weighted by the number of users or the consumption of each point. It can also be divided by the total number of pollution scenes, or just for each node, divided by the scenes that actually affect each node.
[0148] Thus, the comparison of the detection matrix with the impact matrix allows for a given set of sensor locations and for each injection scenario to be constructed, a list of protected nodes. In a set of embodiments of the invention, the convention used is that a node is protected for a scenario if the sensor detects contamination before it reaches it. The advantage of this functionality is the trade-off between the speed of detection and the extent of the coverage area, since, on the one hand, the faster the detection, the smaller the impact; on the other hand, the later and possibly the further downstream in the network the detection is carried out, the greater the number of scenarios detected (i.e., when contamination occurs in the center of the network, it will be detected, for example, by nodes downstream of the network rather than upstream of it; conversely, a sensor located upstream, when detecting upstream contamination, will allow the operation to be performed on almost all nodes of the network). The cost of each scenario not detected is weighted by the node and the consumption affected.
[0149] If the detection time step for a node in a given pollution scenario is strictly smaller than the time step of propagation to the node, the node will therefore be considered protected. In a set of embodiments of the invention, not the node but the consumption of the node is considered protected to take into account distribution inconsistencies and to favor nodes with significant consumption. Therefore, the objective function used is the average consumption that is not protected in all cases or under a set of scenarios that affect each node in different ways, which must be minimized.
[0150] Figure 3An example of a network to which the present invention can be applied is represented, which is modeled in the form of a graph with nodes and arcs. However, the present invention is not limited to this example, and the present invention can be similarly applied to other representations of a network to optimize the ability of a sensor to detect anomalies before they reach different points of the network. For example, a 2D gridding can be defined, in which the location of the sensor is defined by a 2D coordinate (x, y). Similarly, a 3D gridding can be used, in which the location of the sensor is defined by a 3D coordinate (x, y, z).
[0151] Figure 4 An example of anomaly detection in a fluid distribution network according to one set of embodiments of the present invention is shown.
[0152] Figure 4 The example is based on a portion 400 of a water distribution network.
[0153] The portion 400 of the network comprises nine nodes 410, 411, 412, 413, 414, 415, 416, 417, 418 linked by arcs forming three branches. The flow direction of water is continuously from node 410 to nodes 411, 412, 413, then from node 413 to nodes 414, 415 and to nodes 416, 417, 418, respectively. This portion 400 of the network is equipped with a single sensor at node 417.
[0154] In the simulation scenario, pollution is injected 420 at a time step at node 411. The pollution gradually propagates to nodes 412, 413, then to nodes 414, 415 on the one hand, and to nodes 416, 417, 418 on the other hand.
[0155] Thus, in the absence of sensors, this contamination is not detected for nodes 411, 412, 413, 414, 415, 416. On the other hand, it is detected by the sensor for node 417. It is also detected for node 418. Indeed, once contamination is detected for node 417, it can be inferred that the contamination will affect downstream nodes, such as node 418, and appropriate corrective measures can be inferred from this. The monitoring tool can also analyze whether the contamination risk spreads to other areas of the network.
[0156] In one set of implementations of the invention, a performance score for sensor placement is determined by evaluating nodes that detected contamination before it arrived and nodes that detected contamination after it arrived or did not detect contamination at all. For example, this can be done by recording for each node the time step at which contamination arrived and the time step at which contamination was detected, and comparing the two. For example, a reference Figure 3 The impact matrix and detection matrix defined.
[0157] exist Figure 4 In the example:
[0158] - No contamination is detected for nodes 411, 412, 413, 414, 415, 416;
[0159] - the pollution is detected after it has arrived at the node 417. Indeed, by its nature, when the measurements indicating the pollution are performed and processed, the pollution has already arrived at the node 417;
[0160] - Pollution is detected before it arrives for node 418. In fact, when pollution is detected based on the measurements of the sensor at node 417, it can be inferred that the pollution will arrive at node 418, even before it arrives, possibly with a delay time.
[0161] The placement of the sensors allows for more or less effective detection of contamination. Figure 4 In the example of , placement of a sensor at node 412 would, for example, allow detection of contamination before it reaches nodes 413 , 414 , 415 , 416 , 417 , 418 .
[0162] The use of a genetic algorithm for the placement of sensors enables the placement of sensors to be defined iteratively, allowing increasing the probability of detecting anomalies before they arrive, wherever they occur, for the greatest possible number of points of the network.
[0163] Typically, applying a genetic algorithm to the placement of sensors associated with this type of anomaly detection has the effect of positioning the sensor so that it detects contamination on average before it touches downstream nodes: that is, the set of nodes that are touched by the contamination front later than the sensor will be protected. If a sensor senses contamination very upstream, an entire branch of the network can be protected.
[0164] According to different embodiments of the present invention, other objective functions can be used. For example, the following functions can be used alone or in combination:
[0165] - Minimize the average length of the polluted network (a pipe is considered polluted if it lies between two polluted nodes);
[0166] - Minimize the subset of contaminated nodes in the node set (this subset can usually contain "critical" or "sensitive" nodes, such as nodes corresponding to sensitive users such as hospitals, schools, etc.);
[0167] - Minimize the average contamination detection time;
[0168] - Minimize the percentage of undetected contaminated scenes.
[0169] This function gives completely different results than the others, because the time when the contamination was introduced and therefore its impact is not important.
[0170] Figure 5 An example of the definition of optimization constraints according to a set of implementations of the present invention is shown.
[0171] The method according to the invention can be parameterized according to a plurality of constraints and objectives. For example, input parameters 210 can be defined. These input parameters can include in particular pollution scenarios 511 and hydraulic models 512. These input parameters allow the operation of the network to be simulated 220 in each scenario.
[0172] Notably, the input parameters 210 allow the determination of the spread of the pollution 521 in order to obtain an influence matrix 522 representing the arrival times of the pollution for each node. It also allows the determination 523 of the flow in each pipe in order to obtain a monitoring matrix 524 for the nodes, indicating which nodes are linked to each other and determining whether a node allows monitoring of water quality in an upstream node. With particular reference to Fig. 9 and Fig.10 The construction of such a matrix is described.
[0173] The method according to the invention also makes it possible to optimize the location of sensors only on a subset 540 of the points of the network. These points are selected from the set of existing points 541 .
[0174] The method according to the invention can be incorporated into equipment already provided and deployed in the network, such as:
[0175] - Hydraulic sensors 542 already deployed;
[0176] - Deployed quality sensors 543;
[0177] - Available measurement chamber 544: a measurement chamber that already includes sensors can be used to cheaply deploy quality sensors;
[0178] - Current solutions 545, including quality sensors already deployed;
[0179] - Pre-selected nodes 546, including the nodes to be equipped, whatever they are (sensitive sites, reservoirs, network entrances, etc.).
[0180] The method according to the invention can also be parameterized with one or more objective functions 530 indicating the objectives to be monitored by the placement of the sensors. These objective functions allow the calculation 250 of a performance score when executing 200 the method according to the invention for placing sensors.
[0181] Thus, the operator can analyze 550 the best placement solution obtained to check whether it is satisfactory. In case the solution is suboptimal, for example if a placement point is not retained because it does not form part of the subset of points authorized for optimization, the operator can detect this problem and thus can check the sensor placement constraints and, if necessary, modify them 560 before making a new placement iteration.
[0182] Figure 6 shows an example of a histogram representing the percentage of nodes of a water distribution network affected by pollution according to 940 test scenarios for placement of sensors on sector 300 .
[0183] Figure 6 express:
[0184] - on the x-axis 610 , the percentage of nodes touched for different scenarios;
[0185] - On the y-axis 620, the number of scenarios corresponding to each percentage of nodes affected.
[0186] therefore, Figure 6 It is shown that the percentage of nodes affected by contamination varies highly depending on the scenario: in more than 250 scenarios, between 0 and 1% of the nodes are affected 632, while in around 20 scenarios, more than 80% of the nodes are affected, and almost zero scenarios affect between 20% and 30% of the nodes 630.
[0187] 7a, 7b, 7c and 7d respectively show four examples of placing 2, 5, 8 and 11 sensors in a water distribution network for the sole purpose of detecting pollution by the method for placing sensors in an embodiment of the present invention.
[0188] In this example, the method according to the invention for optimizing the placement of sensors has been performed with the following parameters.
[0189] In this example, nodes that can host sensors meet various criteria:
[0190] - it is strictly connected to more than two arcs;
[0191] - Its minimum scenario detection is 5. This can be achieved by determining for each scenario a node that can detect pollution. Conversely, the number of pollution scenarios that will be able to be detected can be determined for each node, and nodes that can detect pollution for a number of scenarios greater than a predefined threshold (e.g. 5) can be pre-selected;
[0192] - Eliminate nodes where sensors are already deployed or where it is not conducive to deploy sensors (site constraints requiring too high an investment, constraints on any road works, heavy traffic, etc.).
[0193] Limiting the optimization to certain nodes allows eliminating nodes where sensors cannot be placed and / or simplifying the problem and the execution of the method according to the invention, while basing the optimization on the a priori most favorable nodes. However, these constraints are given purely by way of non-limiting example and other constraints may be used. For example, the minimum number of scene detections may depend on the number of sensors to be tested, in order to adapt to the complexity of the problem.
[0194] Then, the optimization is performed on a number of sensors ranging from 1 to 20 in succession.
[0195] Multiple candidate sets of locations are initially obtained by randomly selecting sensors from pre-selected nodes, and then several optimization iterations through a genetic algorithm are performed.
[0196] In each iteration, a score is calculated for each candidate set based on the percentage of consumption not covered by sensor measurements (i.e., nodes for which no pollution has been detected are identified, weighted by their relative consumption and related to the total number of nodes to obtain the percentage of consumption for which pollution has not been detected. The lower this percentage, the better the candidate set).
[0197] Then, based on the scores, the genetic algorithm is applied to modify multiple candidate sets at each iteration according to the following rules:
[0198] - Keep the best 5% of the candidate set (select the elite);
[0199] - 80% of the intermediate candidate sets are crossed with each other and / or with the “elites” (best candidates) to generate new sets;
[0200] - The 15% of candidate set mutations that obtained the lowest goodness scores, ie they were modified randomly.
[0201] This application of the method 200 according to the invention allows to obtain an optimal placement for each number of sensors. Once the method has been performed for N sensors, the best solution using N sensors is used as a basis for optimizing N+1 sensors.
[0202] Thus, an optimal placement is obtained for each of a plurality of between 1 and 20 sensors.
[0203] These examples show that, in general, the method according to the invention produces a relatively uniform placement of sensors on the network, thereby allowing detection of anomalies occurring at any point. One advantage of the invention is that after adding a sensor, the position of a previously placed sensor is retained only if that position remains optimal. This is the case with Figures 7a to 7d:
[0204] - When the number of sensors increases 710b, 720b, 710c, 720c, 710d, 720d, two sensors 710a, 720a are retained;
[0205] - When the number of sensors is changed to 8 (730c, 740c, 750c) or 11 (730d, 740d, 750d), the additional three sensors 730c, 740b, 750b of the solution with 5 sensors will remain.
[0206] However, in contrast to other algorithms such as "greedy" algorithms, in the present invention, if the addition of a new sensor makes the position of a previously placed sensor non-optimal, the position will be modified.
[0207] Figure 8 Figure 2 shows the trend of the percentage of anomaly detections of sensors placed in a water distribution network by the method according to the invention as a function of the number of sensors.
[0208] The x-axis 810 represents the number of sensors placed in the previously described example between 1 and 20. The y-axis 820 represents the percentage of nodes that are considered covered, that is, the percentage of nodes where contamination is detected before it arrives.
[0209] Figure 8 It is shown that an increase in the number of sensors initially allows for significant protection gains. As the number of deployed sensors increases, the marginal protection gains associated with adding new sensors become less and less significant. Thus, the present invention allows for the selection of the optimal number of sensors based on a target coverage or based on a percentage optimization of coverage / deployment costs.
[0210] Referring to Figures 7a to 7d and Figure 8 The optimizations presented are given purely by way of non-limiting examples. The invention is not limited to these examples and the method according to the invention can optimize the placement of sensors according to many different constraints.
[0211] For example, in some sectors only a subset of points can be optimized in order to best utilize the existing infrastructure.
[0212] The sectors with large-sized reservoirs can be optimized by introducing measurement points downstream of the reservoirs in order to monitor the contamination that may specifically contact them. These points are considered fixed and are included in the tested sensor combination group. They have the function of detecting contamination at the reservoir. If the reservoir is not used as an intrusion point in the tested scenario, the input from these points in the performance level is zero.
[0213] The sector including the hospital enables the generation of an additional constraint that equips the latter with an entry point, which is in fact one of the consumption nodes of the hydraulic model. This is taken into account for the optimization when enforcing the solution. Therefore, the point of the hospital appears in the set of combinations evaluated during the optimization process. It helps to detect the performance level by detecting pollution scenarios that are located upstream of the hospital and can contact other points.
[0214] Another sector with several sensitive installations (i.e. military, industrial and other such installations) can be optimized by weighting the nodes of the model corresponding to these sites in order to prioritize them in the optimization process. Thus, the objective function allows placing sensors by protecting sensitive locations from contamination or even protecting them by setting their weight to 1 and all other weights to 0.
[0215] The fact of being able to choose to optimize the network as a whole or in a differentiated way according to sectors therefore allows greater flexibility and adaptation to site constraints and the expertise of drinking water distribution professionals. Thus, optimizing by sector in all sectors enables a reduction in computational time compared to a global optimization.
[0216] Thus, the method according to the invention can be adapted to very different operating contexts and allows to achieve optimal sensor placement for very different objectives.The user can configure the method according to the invention based on his or her knowledge of the applied network.
[0217] Fig. 9 Example representing flow distribution in a water distribution network.
[0218] The method according to the invention can be used to optimize the placement of sensors for detecting anomalies or contamination. It can also be used for other purposes, such as optimizing water quality monitoring. To this end, the score of each candidate set assigned to the sensor location can be calculated based on a function including determining the number of nodes that the water passes through when measuring. This type of monitoring (referred to as downstream monitoring) can rely on the generation of a flow score matrix. For a set of nodes, the percentage of the flow passing in each adjacent pipe downstream is calculated at different time steps.
[0219] Fig. 9 Examples of flow scores are provided. Fig. 9 The diagram shows the direction of water flow, and the fraction of flow that flows in each pipe. In this example, flow in the network is generally from left to right. The numbers indicated on each pipe indicate the fraction of flow that goes from one node to another. For example:
[0220] - For one unit, the flow input at node 910 is divided into 0.8 (80%) in pipe 911 and 0.2 (20%) in pipe 912;
[0221] - For one unit, the flow input at node 920 is divided into 1 (100%) in pipe 921;
[0222] - For one unit, the flow input at node 930 is divided into 0.3 (30%) in pipe 931 and 0.7 (70%) in pipe 932.
[0223] This relationship at directly adjacent nodes is then propagated throughout the network. In one set of embodiments of the invention, a downstream monitoring relationship is defined for a pair of nodes if one of the two nodes receives the majority of the traffic (>50%) flowing through the other node. Links shown in bold show nodes that are considered to be linked to each other. For example, node 910 is considered to be linked to node 920, and node 920 is linked to location 930. If a 50% traffic threshold provides a good indication of a link between two nodes, other thresholds (40%, 60%, etc.) can also be used.
[0224] Thus, for each node of the graph, it is possible to determine the number of upstream nodes to which it is linked, that is, the number of nodes through which the water is received. It is also possible to assign nodes to populations or associated consumptions. Thus, it is possible to infer for each node which populations or water consumptions will be covered by the quality monitoring performed at this node.
[0225] Fig.10 Represents an example of monitoring water quality through sensors located at downstream nodes of a water distribution network.
[0226] Fig.10 Shown is how a node surrounded by a diamond is considered to be traced by a node surrounded by a star. Flows are represented symbolically, highlighting the structure and hydraulic behavior of the network at that time step.
[0227] Fig.10 1 shows a sector 300, a portion 1000 of which is shown enlarged below the image. Fig.10 In the example of , nodes are iteratively linked when more than 50% of the traffic entering a node passes through the next node. For example, node 1011 is linked to node 1012, which itself is linked to node 1013, and so on, until node 1010. Therefore, node 1011 is linked to node 1014 to node 1010. Fig.10 All nodes that are linked to each other are represented by diamond shapes, and links between nodes are represented by thick grey lines, such as by line 1014 . Fig.10 It is shown that most of the nodes of the network converge to downstream node 1010 or downstream node 1020. Therefore, this representation allows a better understanding of the structure of the flow and the hydraulic behavior of the network at a given time step.
[0228] Thus, sensors placed at nodes 1010 and 1020 allow water quality to be checked over a large portion of the network.
[0229] The method according to the invention allows placing sensors to optimally monitor water quality by calculating a performance score using an objective function that defines the number of nodes to which a node is linked, that is, the number of nodes for which the node can perform downstream monitoring. Such an objective function allows placing sensors so as to be able to monitor water quality originating from different sources, in particular from different areas where these sources are mixed. Typically, sensors placed using this objective function are also able to detect contamination occurring downstream of the network.
[0230] This objective function can be used alone or in combination with an objective function related to anomaly or contamination detection. Thus, the presence of sensors for downstream monitoring can complement sensors placed with the objective function of contamination detection. This combination is also well suited to subdividing the network into small pieces of water corresponding to sources or mixtures of sources.
[0231] 11a , 11b , 11c and 11d respectively represent four examples of placing 2, 5, 8 and 11 sensors in a water distribution network in order to detect pollution and monitor water quality by the method for placing sensors in an embodiment of the present invention.
[0232] The placement of the sensors is performed in the same principle as the example of Figures 7a to 7d, except for the following points.
[0233] As mentioned above, in order to simplify the problem while achieving a good level of performance, it is possible to place sensors only in a subset of the network nodes. Fig. 9 and Fig.10 As explained, the nodes that allow the best placement of sensors for monitoring pollution downstream are nodes that are linked to a large number of nodes of the network located upstream. Therefore, the subset of nodes on which sensors can be placed can include the nodes according to the reference Fig. 9 The defined rules are connected to nodes that are a number of nodes upstream greater than or equal to a predefined threshold. The consumption of each node can also be considered, and the subset of nodes on which the optimization is performed can include nodes that are connected to a number of nodes upstream that correspond to consumption greater than the predefined threshold.
[0234] To jointly optimize the placement of sensors for anomaly detection and downstream monitoring involves preselecting nodes that form good candidates for anomaly detection and nodes that form good candidates for downstream water quality monitoring. For example, the subset of nodes on which the optimization is performed can include nodes that are connected to a large number of nodes upstream (forming good candidates for downstream monitoring), and nodes that are connected to a number of arcs greater than a predefined threshold (forming good candidates for anomaly detection).
[0235] During optimization, a score assigned to each candidate set of locations is calculated based on a combined objective function, equal to the average of the target anomaly detection function and the target water quality monitoring function, so as to favor candidate sets of locations that allow good anomaly detection and good water quality monitoring.
[0236] Then, based on the scores, the genetic algorithm is applied to modify multiple candidate sets for each iteration according to the following rules:
[0237] - Keep the best 5% of the candidate set (select the elite);
[0238] - 80% of the intermediate candidate sets are crossed with each other, and / or with the “elite” (the best candidates that are kept) to generate new sets;
[0239] - The 15% of candidate set mutations that obtained the lowest goodness scores, ie they were modified randomly.
[0240] This application of the method 200 according to the invention allows to obtain an optimal placement for each number of sensors. Once the method has been performed for N sensors, the best solution for N sensors is used as a basis for optimizing N+1 sensors.
[0241] Thus, an optimal placement is obtained for each number of sensors between 1 and 20.
[0242] The placement of nodes is indicated by circles such as circle 1110a for 2, 5, 8 and 11, respectively. The placement here is different from the placement performed for Figures 7a to 7d for the same number of sensors. For example, in the solution with 5 sensors, sensor 1120b is placed in an area where no sensor is placed in Figure 7b (the solution with 5 sensors is only used for pollution detection). This sensor 1120b allows water quality detection in the entire northern part of the network and therefore allows the solution with 5 sensors shown in Figure 11b to obtain a good score for the combination of "detection of anomalies + water quality monitoring".
[0243] These examples demonstrate the ability of the present invention to provide optimal sensor placement according to user defined objectives.Thus, the present invention is applicable to a wide range of use cases and allows optimal placement of sensors to be obtained according to a wide range of different criteria.
[0244] Fig.12 Shows the trend of pollution detection and water quality monitoring scores according to the combination of the number of sensors placed by the method according to the present invention.
[0245] Fig.12 The curve is based on Figure 8The curves are created using the same principle: the x-axis 1210 represents the number of placed sensors, the vertical axis 1220 represents the performance score of the best solution, expressed as a percentage, and in the example of Figures 11a to 11d, the curve 1230 represents the trend of the score according to the number of sensors.
[0246] As described above, in the combined optimization example of Figures 11a to 11d, the performance score is calculated as the average of the water quality monitoring score and the anomaly detection score. The top curve 1240 represents the water quality monitoring score, while the bottom curve 1250 represents the anomaly detection score. When a new sensor is added, the total score can be increased by increasing the water quality monitoring score and / or the anomaly detection score. For example, here the third and fourth sensors are added, which mainly improve the water quality monitoring scores 1241, 1242, while the second, fifth and sixth sensors are added, which mainly improve the anomaly detection scores 1251, 1252, 1253.
[0247] This example demonstrates the ability of the present invention to optimally arrange sensors according to a combination of objectives by improving one or the other objective to obtain the best overall score.
[0248] The above examples demonstrate the ability of the present invention to optimize the placement of sensors according to different target criteria. However, they are given by way of example only and in no way limit the scope of the present invention as defined by the appended claims.
Claims
1. A method for determining a plurality of sensor locations of one or more physicochemical parameters of a fluid in a fluid transport network, the method comprising: include: - obtaining, for at least one set of input parameters, simulated values of physicochemical quantities at a set of points of said network for a set of time steps, wherein at least one of the sets of input parameters comprises the introduction of an anomaly in at least one point of said network for at least one time step; - obtaining a plurality of candidate sets of positions of the plurality of sensors; - Iteratively do the following until the stopping criteria is verified: o for each candidate set of locations, obtaining a performance score based at least on the simulated values of the physicochemical quantity at the locations of the plurality of sensors; o modifying the plurality of candidate sets of sensor locations, the modifying comprising at least one operation selected from the following: retaining at least one candidate set of positions with the most favorable scores; adding at least one candidate set of positions of said sensor defined by a combination of positions of at least two candidate sets; adding at least one candidate set of sensor positions defined by modifications to the positions of sensors in the candidate set; - selecting said candidate set of positions having said most favourable scores; - placing the sensor in the selected candidate set of locations having the most favorable score; wherein the performance score is calculated based on: the ability of the sensor to be placed in the candidate set of locations to detect the at least one anomaly by determining a time step for the at least one anomaly to arrive at the set of points of the network; and at least one objective function selected from: - the number of points in the set for which the anomaly was detected before the anomaly arrived; - the number of points in the set for which the anomaly was detected before its arrival, weighted by the number of users or the consumption of each point; - the number of points for which said anomaly has not yet been detected; - The number of points for which said anomaly has not yet been detected, weighted by the number of users or consumption per point.
2. The method according to claim 1, in: - the network is modeled in graphical form; - Each node or arc of the graph is identified by a unique identifier; - The location of the sensor is defined by an identifier of a node or arc of the graph.
3. The method according to one of claims 1 to 2, in, The performance score is calculated based on at least one characteristic selected from capabilities of the sensors in the candidate set to evaluate a quality indicator of the fluid in the network and a deployment cost of the sensors.
4. The method according to claim 1, in, At least one sensor has a predefined position.
5. The method according to one of claims 1 to 2, in, The points at which the sensors are placed are restricted to a subset of the points of the network.
6. The method according to one of claims 1 to 2, in, The definitions of the plurality of candidate sets of locations for the plurality of sensors include definitions of locations at points of interest for the network.
7. The method according to claim 2, in, The definitions of the plurality of candidate sets of locations for the plurality of sensors include definitions of locations at nodes that are connected to a number of arcs greater than or equal to a predefined threshold.
8. The method according to one of claims 1 to 2, in, The stopping criteria include one or more conditions selected from the following: - maximum number of iterations; - a comparison of the most favorable scores in a current iteration and at least one previous iteration, and said verification of said stopping criterion if a difference between said most favorable scores at said current iteration and at said at least one previous iteration is less than or equal to a predefined threshold.
9. The method according to one of claims 1 to 2, in: - eliminating at least one candidate set of positions not having said most favorable score comprises eliminating all candidate sets except a predefined number or a predefined ratio of candidate sets having said most favorable scores; - adding at least one candidate set of sensor positions defined by combining the positions of at least two candidate sets, and adding at least one candidate set of sensor positions defined by modifying the positions of sensors in the candidate sets comprises: adding a number of candidate sets equal to the number of candidate sets eliminated.
10. The method according to one of claims 1 to 2, include: A candidate set of sensors is defined, iteratively modified, and selected for a plurality of predefined numbers of sensors, respectively.
11. The method according to claim 10, in, The plurality of candidate sets of positions for the plurality of sensors obtained for an integer number n of sensors is based on the candidate sets of positions having the most favorable scores for an integer number nm of sensors, where 1≤m <n。 12. The method according to claim 10, include: A selection of one of the candidate sets having the most favorable score for the plurality of predefined numbers of sensors is performed based on the score of each of the candidate sets and the deployment cost.
13. The method of claim 1, wherein a constant number of candidate sets is retained between the iterations, and a given number of candidate sets with the most favorable scores is retained between second iterations.
14. A method for placing a plurality of sensors for one or more physicochemical parameters of a fluid, the method include: determining a set of locations in a fluid transport network, said determining comprising the steps of a method according to one of claims 1 to 13, and • Placing the sensors in the set of locations.
15. A computer program product comprising program code instructions stored on a computer readable medium, comprising a processor for determining a set of positions of a plurality of sensors for determining one or more physicochemical parameters of a fluid in a fluid transport network, the computer program comprising a computer readable programming unit for executing the method according to one of claims 1 to 13.
16. A device comprising a set of positions of a plurality of sensors capable of determining one or more physicochemical parameters of a fluid in a fluid transport network, the device comprising a processor configured to perform the method according to one of claims 1 to 13.
