Wireless sensor network maintenance method and related equipment

Abstract

The embodiment of the present invention discloses a wireless sensor network maintenance method and related equipment, firstly establishes a Bayesian network according to expert prior knowledge; then determines the initial condition probability table of each node in the Bayesian network; optimizes the initial condition probability table according to the discrete characteristic data of the wireless sensor network to obtain the optimized condition probability table of each node; then determines the abnormal node that meets the first preset abnormal condition in the top-level characteristic variable, and infers the lowest-level influencing variable that meets the second preset abnormal condition from the abnormal node according to the optimized condition probability table; finally, adjusts the lowest-level influencing variable that meets the second preset abnormal condition, and determines the first maintenance plan that makes the abnormal node meet the preset normal condition. The method of the embodiment of the present invention, based on the representation, learning, and reasoning capabilities of the Bayesian network, can not only locate the fault, but also provide a maintenance strategy, reduce subjectivity, and reduce maintenance costs.

Description

Technical Field

[0001] The present invention relates to the field of wireless sensor technology, and in particular to a wireless sensor network maintenance method and related equipment. Background Art

[0002] Wireless Sensor Networks (WSN) is a distributed sensor network in which sensor data is uploaded to a gateway wirelessly. For small-scale wireless sensor networks, the probability of failure is very low and the maintenance cost is also low. However, for large-scale wireless sensor networks, there are tens of thousands of sensor nodes. Even if the probability of sensor node failure or channel quality deterioration is very small, once a failure occurs, the network maintenance requires a large cost.

[0003] The existing maintenance method for large-scale wireless sensor networks relies on human experts to conduct on-site troubleshooting based on the collected high-quality information and expert experience to maintain the normal operation of the system. This maintenance method is too costly and it is difficult to provide an effective maintenance improvement strategy for the current system. It is limited to finding faults. Even if experts can provide maintenance strategies, the maintenance strategies are too subjective and cannot effectively solve the problems in wireless sensor networks. Summary of the invention

[0004] The embodiment of the present invention provides a wireless sensor network maintenance method and related equipment, which can reduce the maintenance cost of the wireless sensor network.

[0005] In a first aspect, an embodiment of the present invention provides a wireless sensor network maintenance method, including:

[0006] According to the expert prior knowledge, a Bayesian network is established between the observable top-level characteristic variables of the wireless sensor network and the influencing variables affecting the top-level characteristic variables, wherein the top-level characteristic variables are used to characterize the performance of the wireless sensor network, and the influencing variables include unobservable variables;

[0007] Determining an initial condition probability table for each node in the Bayesian network;

[0008] Optimizing the initial conditional probability table according to the discrete feature data of the wireless sensor network to obtain an optimized conditional probability table for each node, wherein the discrete feature data is observation data of observable variables in the Bayesian network;

[0009] Determine an abnormal node in the top-level feature variable that satisfies a first preset abnormal condition, and infer the lowest-level influencing variable that satisfies a second preset abnormal condition from the abnormal node according to the optimized condition probability table;

[0010] The bottom-level influencing variables are adjusted to determine a first maintenance plan that enables the abnormal node to meet a preset normal condition.

[0011] Optionally, determining an initial condition probability table for each node in the Bayesian network includes:

[0012] The initial condition probability table of each node is determined according to the expert prior knowledge and / or historical statistical data.

[0013] Optionally, optimizing the initial condition probability table according to the discrete characteristic data of the wireless sensor network to obtain an optimized condition probability table for each node includes:

[0014] The initial conditional probability table is optimized according to the discrete feature data and the expected maximum algorithm to obtain the optimized conditional probability table.

[0015] Optionally, the step of inferring downward from the abnormal node to obtain the lowest-level influencing variable that satisfies the second preset abnormal condition includes:

[0016] According to the abnormal nodes and the Bayesian network reasoning algorithm, the bottom-level influencing variables are inferred downwards.

[0017] Optionally, the Bayesian network reasoning algorithm includes an elimination reasoning algorithm or a clique tree reasoning algorithm.

[0018] Optionally, adjusting the bottom-level influencing variables to determine a first maintenance plan that enables the abnormal node to meet a preset normal condition includes:

[0019] Try to adjust the bottom-level influencing variables to determine at least one second maintenance plan that enables the abnormal node to meet the preset normal conditions;

[0020] Determining a cost function value corresponding to the second maintenance solution;

[0021] The second maintenance plan with the smallest cost function value is used as the first maintenance plan.

[0022] Optionally, the influencing variable corresponding to the top-level characteristic variable is an influencing variable set including at least one level, and the one level includes at least one influencing variable.

[0023] In a second aspect, an embodiment of the present invention provides a wireless sensor network maintenance device, including:

[0024] An establishment module is used to establish a Bayesian network between observable top-level characteristic variables of the wireless sensor network and influencing variables affecting the top-level characteristic variables according to expert prior knowledge, wherein the top-level characteristic variables are used to characterize the performance of the wireless sensor network, and the influencing variables include unobservable variables;

[0025] A determination module, used to determine an initial condition probability table for each node in the Bayesian network;

[0026] An optimization module, configured to optimize the initial condition probability table according to discrete feature data of the wireless sensor network to obtain an optimized condition probability table for each node, wherein the discrete feature data is observation data of observable variables in the Bayesian network;

[0027] An inference module, used to determine an abnormal node in the top-level feature variable that satisfies a first preset abnormal condition, and infer the lowest-level influencing variable that satisfies a second preset abnormal condition from the abnormal node downward according to the optimized condition probability table;

[0028] The adjustment module is used to adjust the bottom-level influencing variables to determine a first maintenance plan that enables the abnormal node to meet a preset normal condition.

[0029] In a third aspect, an embodiment of the present invention provides a wireless sensor network maintenance device, including: a processor and a memory;

[0030] The processor is connected to a memory, wherein the memory is used to store program codes, and the processor is used to call the program codes to execute the wireless sensor network maintenance method as described in the first aspect.

[0031] In a fourth aspect, an embodiment of the present invention provides a computer storage medium, wherein the computer storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the wireless sensor network maintenance method as described in the first aspect is executed.

[0032] In the embodiment of the present invention, firstly, according to the expert prior knowledge, a Bayesian network is established between the observable top-level characteristic variables of the wireless sensor network and the influencing variables affecting the top-level characteristic variables, the top-level characteristic variables are used to characterize the performance of the wireless sensor network, and the influencing variables include unobservable variables; then, the initial condition probability table of each node in the Bayesian network is determined; then, according to the discrete characteristic data of the wireless sensor network, the initial condition probability table is optimized to obtain the optimized condition probability table of each node, and the discrete characteristic data is the observed data of the observable variables in the Bayesian network; then, the abnormal node that meets the first preset abnormal condition in the top-level characteristic variables is determined, and according to the optimized condition probability table, the lowest level influencing variables that meet the second preset abnormal condition are inferred from the abnormal node; finally, the lowest level influencing variables that meet the second preset abnormal condition are adjusted to determine the first maintenance plan that makes the abnormal node meet the preset normal condition. The method of the embodiment of the present invention, based on the representation, learning, and reasoning capabilities of the Bayesian network, can not only detect the lowest level influencing variables in a degraded state (i.e., a fault state) and locate the fault, but also provide a maintenance strategy, reduce subjectivity, improve maintenance efficiency, and reduce maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0034] Figure 1 It is a flow chart of a wireless sensor network maintenance method provided by an embodiment of the present invention;

[0035] Figure 2 is a schematic diagram of the structure of a Bayesian network provided by an embodiment of the present invention;

[0036] Figure 3 It is a schematic diagram of quantization and discretization of a variable provided by an embodiment of the present invention;

[0037] Figure 4 yes Figure 3 The initial condition probability indicates the intention;

[0038] Figure 5 It is a structural schematic diagram of a wireless sensor network maintenance device provided by an embodiment of the present invention;

[0039] Figure 6 It is a structural diagram of a wireless sensor network maintenance device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0041] It should be understood that the terms "first", "second", etc. in the specification, claims and drawings of the present application are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or devices.

[0042] Reference to "embodiments" in the present invention means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described in the present invention may be combined with other embodiments.

[0043] A Bayesian network is a graphical representation that describes the uncertain relationship between variables. It consists of two parts: a structural model and a conditional probability distribution set. The structural model is a directed acyclic graph (DAG), where nodes represent random variables and directed edges represent dependencies between variables. The degree of dependency between two variables is described by the probability distribution attached to each node. The conditional probability distribution set, or conditional probability table, is a collection of local probability distributions associated with each node.

[0044] See also Figure 1 , Figure 1 1 is a flow chart of a wireless sensor network maintenance method provided by an embodiment of the present invention; the wireless sensor network maintenance method comprises:

[0045] Step 101, based on the expert prior knowledge, a Bayesian network (causal relationship diagram) is established between the observable top-level characteristic variables of the wireless sensor network and the influencing variables that affect the top-level characteristic variables, wherein the top-level characteristic variables are used to characterize the performance of the wireless sensor network, and the influencing variables include unobservable variables;

[0046] Specifically, the top-level characteristic variables are defined as quantifiable parameters (can be one or more than two) that characterize the performance of the wireless sensor network, and the parameters that cause the top-level characteristic variables to change are defined as influencing variables (can be one or more than two), such as end-to-end transmission delay. The influencing variables corresponding to each top-level characteristic variable are an influencing variable set including at least one level, and each level includes at least one influencing variable. The number of levels of the influencing variable set is determined according to demand and is not particularly limited. For example, taking the influencing variable set including two levels as an example, refer to Figure 2 , Figure 2 : is a schematic diagram of the structure of a Bayesian network provided by an embodiment of the present invention; assuming that node A is a top-level feature variable, and nodes B, C, D, E and F are influencing variables of node A, wherein nodes B and C are at the same level, and nodes D, E and F are at the same level. In particular, the influencing variable set includes unobservable variables, and the influencing variable set may also include observable variables.

[0047] In addition, in the embodiment of the present invention, the fault base is also defined as the leaf node variable at the bottom layer of the Bayesian network, which can improve the top-level characteristic variables through maintenance. The fault base is generally difficult to observe when the wireless sensor network is running. The middle layer variable between the fault base and the top-level characteristic variable is the intermediate quantity. The degradation of the state of the fault base causes the change of the intermediate quantity, which in turn causes the change of the top-level network performance parameter. The listening network attached to the wireless sensor network can be used to collect data to obtain the data of the variables in the middle layer. For example, the listening network can monitor the data transmission rate of the wireless sensor network, the type of data packet and other related parameters.

[0048] To build a Bayesian network for a wireless sensor network, experts in the field of wireless sensor networks can be consulted to obtain prior knowledge in the field. The expert prior knowledge can provide the specific types of the top-level characteristic variables of the wireless sensor network and the various influencing variables that affect it. Based on the influence relationship between the top-level characteristic variables and the influencing variables, and between the influencing variables, a directed acyclic graph, namely the Bayesian network, can be established.

[0049] Among them, building a Bayesian network based on expert prior knowledge can reduce the requirements for high-quality data.

[0050] Step 102, determining the initial conditional probability table (CPT) of each node in the Bayesian network;

[0051] Specifically, for a domain expert, it is easy and reliable to determine the network structure of the Bayesian network, but it is difficult to give specific parameters. Therefore, other methods are needed to determine the initial condition probability table of each node in the Bayesian network.

[0052] Step 103, optimizing the initial conditional probability table according to the discrete characteristic data of the wireless sensor network to obtain an optimized conditional probability table for each node, wherein the discrete characteristic data is the observed data of the observable variables in the Bayesian network;

[0053] Specifically, a large amount of field observation data of observable variables (including top-level characteristic variables and some observable influencing variables) in the wireless sensor network is obtained as discrete characteristic data, and based on the discrete characteristic data, the initial conditional probability table in step 102 is optimized to obtain an optimized conditional probability table for each node.

[0054] Step 104, determining an abnormal node in the top-level feature variable that satisfies a first preset abnormal condition, and inferring the lowest-level influencing variable that satisfies a second preset abnormal condition from the abnormal node downward according to the optimized condition probability table;

[0055] Specifically, the first preset abnormal condition defines the conditions that the discrete top-level characteristic variable value (i.e., the field observation data of the top-level characteristic variable) needs to satisfy when the top-level characteristic variable is in an abnormal state, and the second preset abnormal condition defines the conditions that the conditional probability of the bottom-level influencing variable needs to satisfy when it is in a degraded state (i.e., a faulty state); the first preset abnormal condition and the second preset abnormal condition can be specifically set according to actual needs. For example, taking the top-level characteristic variable as the end-to-end delay as an example, assuming that the first preset abnormal condition is that the delay is greater than 50s, when the end-to-end delay observed on site is 60s, it can be judged that the top-level characteristic variable of the end-to-end delay is in an abnormal state, that is, it is an abnormal node. Taking the bottom-level influencing variable as temperature as an example, when the conditional probability that the temperature is very high is greater than or equal to 0.9, it is judged that the temperature variable is in a degraded state.

[0056] In fact, firstly, the top-level feature variables with abnormal performance in the Bayesian network can be determined according to the discrete feature data obtained in step 103, and then the state of the corresponding fault base can be inferred downward from the top-level feature variables according to the optimized conditional probability table.

[0057] Step 105 , adjusting the lowest-level influencing variables that satisfy the second preset abnormal condition, and determining a first maintenance plan that enables the abnormal node to satisfy the preset normal condition.

[0058] Specifically, by trying to adjust the lowest-level influencing variables that meet the second preset abnormal condition, a first maintenance plan that makes the abnormal node meet the preset normal condition can be obtained. The preset normal condition defines the conditions that need to be met when the top-level characteristic variable is in a normal state. For example, when the conditional probability that the end-to-end delay is very low is greater than or equal to 0.9, the end-to-end delay, the top-level characteristic variable, is judged to be in a normal state. 0.9 can also be set to 0.8 or 0.7. The preset normal condition can be set according to actual needs. For example, it can be set to the discrete value range of the top-level characteristic variable under normal conditions, such as temperature. Assuming that the temperature is 50-80 degrees as the normal temperature, when the discrete value of the top-level characteristic variable is within this range after adjustment, the adjustment plan is considered feasible.

[0059] The method of the embodiment of the present invention, based on the representation, learning, and reasoning capabilities of the Bayesian network, can fully and comprehensively utilize expert prior information and field observation data, lower the threshold for collecting high-quality data, and reduce human subjectivity. It also mines and analyzes the collected and monitored data, and makes inferences and judgments on the performance degradation of the system components of the wireless sensor network, detects the bottom-level influencing variables in a degraded state, locates faults, and provides specific strategies for improving system maintenance, thereby improving maintenance efficiency, reducing maintenance costs, and reducing the probability of system failure within the rated life cycle, providing a research focus for the improvement of the next generation of wireless sensor network systems.

[0060] In a possible embodiment, the present invention provides a specific Bayesian network, referring to Figure 2 , where latency refers to the time required to successfully complete a single task. When the latency is infinite, it indicates a failure; when the latency is higher than the average, it indicates poor performance; and when the latency is lower than the average, it indicates normal / redundancy.

[0061] In the embodiment of the present invention, the top layer of the Bayesian network uses the "end-to-end delay" as the characterization quantity of the performance of the wireless sensor network system. Of course, there may be multiple other characterization quantities. End-to-end delay refers to the time to successfully complete a data upload to the concentrator. If the task fails, the delay is considered to be infinite. The "end-to-end delay" is affected by the three characteristic quantities of "total number of routing hops", "point-to-point transmission delay", and "node processing delay" (the characteristic quantities are selected as convenient for measurement, collection and quantification as much as possible). The "total number of routing hops" depends on the three characteristic quantities of the "radio coverage radius", "node distance from the concentrator", and "routing optimization factor" of the radio frequency module. The "point-to-point transmission delay" is affected by the "number of retransmissions / packet loss rate", "data packet length", and "transmission rate". The sensor node is mainly composed of sensors and radio frequency modules, so the "node processing delay" is affected by the "sensor data acquisition delay" and the "radio frequency module processing delay". When any of them fails or an abnormality occurs, the "node processing delay" will become longer or even infinite. The 'sensor data collection delay' is affected by 'sensor failure' and 'operating temperature'; the 'RF module processing delay' is affected by 'operating temperature' and 'RF module failure'.

[0062] In this application method, the establishment of the cause-effect relationship diagram can be constructed according to actual needs and current situation. Figure 2 The examples are provided for reference only; the setting of the fault base is also relatively flexible. Generally speaking, the observable characteristic quantities and the quantities that can improve maintenance can be in different forms, and changes in their forms are all within the scope of the method of this application.

[0063] In one possible embodiment, before determining the initial condition probability table, it is necessary to first determine the quantization method of each node, or to determine the quantization method and discretization method of each node, that is, to digitize the data of each node variable, and the data of all variables are ultimately discrete values. Specifically, all variables must be quantized, but only continuous variables need to be discretized; for example, the presence or absence of obstacles is a discrete value after quantization. All variables are ultimately discrete values.

[0064] For example, a feature, such as performance, needs to be described in a digital way. For example, the transmission delay can be used to describe the performance; the transmission success rate can also be used to describe the performance; or these two indicators can be used to calculate a value through a certain algorithm to become a "performance value"; performance needs a value to describe, which is quantification. Since the Bayesian network can only accept discrete input values, the continuous variable needs to be discretized.

[0065] Further, refer to Figure 3 , Figure 3 is a schematic diagram of quantization and discretization of a variable provided by an embodiment of the present invention; for the convenience of explanation, Figure 3It was intercepted Figure 2 As a part of Figure 3 As shown in the figure, for various delays and temperatures, they are quantized into {VL, L, M, H, VH}, which means {very low, low, medium, high, very high}. Since delay is a continuous variable, it needs to be discretized, that is, to determine the four boundaries of {VL, L, M, H, VH}. It can be discretized based on the expectation and variance of statistical delay data. For example, the delay less than the "expectation-variance" delay is discretized into VL, which means very low. Whether "sensor failure" and "RF module failure" occur or not is quantified into two values: {normal, abnormal}. The more discretized values ​​there are, the more accurate the inference result is, but the amount of calculation will increase geometrically; the number of discretized values ​​for each node can be different.

[0066] In the method of the present application, the quantization and discretization methods of different variables can be flexibly set according to needs, and the different setting methods are all within the protection scope of the method of the present application.

[0067] In one possible embodiment, step 102 includes:

[0068] The initial condition probability table for each node is determined based on expert prior knowledge and / or historical statistical data.

[0069] Specifically, after determining the quantization and discretization methods of the variables, it is necessary to determine the initial condition probability table corresponding to each node in the Bayesian network (causal relationship diagram). It is preferred to use historical statistical data to determine the initial condition probability table. When there are no historical statistical data, the initial condition probability table of each node is set according to expert experience. When there is no expert prior knowledge, the initial condition probability table of each node can be set randomly, but the reasoning effect of this method is not good; other methods for determining the initial condition probability table can also be used, without special limitation. It is also possible to combine historical statistical data with expert prior knowledge to set the initial condition probability table of each node. For example, in the Bayesian network, some nodes do not have historical statistical data. At this time, the initial condition probability table of the node can be set using expert prior knowledge.

[0070] More specifically, in the directed acyclic graph (causal relationship graph) established in step 101, each node needs to establish an initial conditional probability table (CPT). The so-called conditional probability table is explained as follows: Figure 4 , Figure 4 yes Figure 3The initial condition probability indicates the intention; if there is a variable A, which is affected by variables B and C, and variables B and C are independent of each other, then in the causal relationship diagram, node B has an arrow pointing to node A, and node C has an arrow pointing to node A. Node A is affected by nodes B and C, so its corresponding CPT is in the form of P(A|B,C), that is, the probability of A taking different values ​​when B and C take different values. For example, P(A=VL|B=VL,C=VL) indicates the probability of A: node processing delay being very low under the conditions of B: sensor data collection delay being very low, C: RF module processing delay being very low. At this time, its initial value can be set to 0.8 (here is an assumed value) based on expert experience, or it can be set based on historical statistical data. It should be noted that there is a constraint: P(A=VL|B=VL,C=VL)+P(A=L|B=VL,C=VL)+P(A=M|B=VL,C=VL)+P(A=HL|B=VL,C=VL)+P(A=VH|B=VL,C=VL) = 1. These initial CPT values ​​will be optimized by big data in the following steps to reduce the subjectivity of expert experience and the errors caused by the inconsistency between historical data and field data.

[0071] Assume that the discretized value of node A is {L, H}, the discretized value of node B is {L, H}, and the discretized value of node C is {L, H}. At this time, the initial CPT of node A includes the following conditional probabilities:

[0072] P(A=L|B=L,C=L),

[0073] P(A=H|B=L,C=L),

[0074] P(A=L|B=H,C=L),

[0075] P(A=H|B=H,C=L),

[0076] P(A=L|B=L,C=H),

[0077] P(A=H|B=L,C=H),

[0078] P(A=L|B=H,C=H),

[0079] P(A=H|B=H,C=H);

[0080] Among them, P(A=L|B=L,C=L) represents the probability that the value of node A is L when the value of node B is L and the value of node C is L, and the other probabilities are similar. The initial CPT of each node can be obtained based on expert experience, historical statistical data, or other methods.

[0081] In one possible embodiment, after determining the method for obtaining a discretized value for each node, discrete feature data can be obtained by collecting the quantized values ​​of observable variables in the wireless sensor network at different times and then discretizing the collected data sets (data preprocessing).

[0082] In a possible embodiment, step 103 includes:

[0083] According to the discrete characteristic data and the expected maximum algorithm, the initial conditional probability table is optimized to obtain the optimized conditional probability table.

[0084] Specifically, the expectation-maximization algorithm is suitable for optimization scenarios where model parameters are missing. The EM (Expectation-Maximization) algorithm is used to reduce the problem of excessive subjectivity in the maintenance of wireless sensor networks.

[0085] like Figure 4 In the example, a large amount of data is collected for nodes A and C, while nodes B, D, E, and F are all unknown data. Because the expert prior information is too subjective, and the historical data may be inaccurate or inconsistent with the on-site information, the initial CPT cannot be used directly to infer P(D|A,C), P(E|A,C), and P(F|A,C). The initial CPT needs to be optimized using on-site observation data. Only after the EM algorithm is used to optimize the CPT value of the node, can the subjectivity and data errors be removed as much as possible.

[0086] Assume that A, B, C are observed nodes, D, E, F are unobservable hidden nodes, and the directed edge set is {(B, A), (C, A), (D, B), (E, B), (E, C)(F, C)}, and n samples of {A, B, C} are collected. The main idea of ​​using the EM algorithm on the Bayesian network model is as follows:

[0087] ① “Expectation” step

[0088] From the known nodes of the Bayesian network, the precise inference algorithm such as elimination method or clique tree algorithm is used to infer the status of the unknown nodes of the Bayesian network according to the initial CPT set by the expert. For all the n {A, B, C} samples collected, the status of their corresponding unknown nodes {D, E, F} is inferred, and the samples become n {A, B, C, D, E, F} without incomplete unknown nodes.

[0089] ② "Maximum" step (maximum likelihood estimation)

[0090] The joint probability of all nodes in the Bayesian network can be converted into the product of CPT, that is,

[0091] P(Si)=P(Ai,Bi,Ci,Di,Ei,Fi)=P(Ai|Bi,Ci)P(Bi|Di,Ei)P(Ci|Fi)P(Di)P(Ci)P(Fi)(i=1,…,n)(Formula 4-1)

[0092] In order to maximize the joint probability P(S1, S2, ..., Sn) of the n intact samples in the Bayesian network model, it is equivalent to finding the maximum product of the individual probabilities of the n samples, that is, P(S1, S2, ..., Sn) = P(S1)P(S2)...P(Sn). The individual probability of each sample (that is, the joint probability of all nodes) can be converted into the product of CPT (as shown in Formula 4-1), so the joint probability of the n samples can be converted into the product of CPT. As long as the best CPT is found to maximize this probability, the new CPT can replace the original CPT.

[0093] ③ Repeat steps ① and ②, CPT will eventually converge. Set a threshold and stop when the CPT update amplitude is less than the threshold.

[0094] The above process is to optimize the CPT obtained based on expert prior information or historical statistical data through big data mining technology.

[0095] In one possible embodiment, step 104 includes:

[0096] According to the abnormal nodes and the Bayesian network reasoning algorithm, the lowest-level influencing variables that meet the second preset abnormal condition are inferred downward.

[0097] Specifically, based on the discrete feature data, the nodes that are abnormal in the top-level feature variables, that is, the abnormal nodes that meet the first preset abnormal condition, can be determined; then, using the Bayesian network inference algorithm, the state of the corresponding fault base can be inferred from the upper abnormal node downward, that is, the lowest-level influencing variable that meets the second abnormal condition can be found. Furthermore, the Bayesian network inference algorithm includes an elimination inference algorithm or a clique tree inference algorithm.

[0098] Due to the fluctuation of system parameters and the degradation of overall system performance, poorly performing data samples will appear. For these samples, in step 103, on the Bayesian network of CPT optimized, an accurate inference algorithm such as elimination inference algorithm or clique tree inference algorithm is used to use known abnormal nodes as inference evidence to infer the state of the corresponding bottom-level node (fault base). For example, refer to Figure 4, given nodes A, B, and C, find the posterior probabilities of unknown nodes D, E, and F, that is, find P(D|A,B,C), P(E|A,B,C), and P(F|A,B,C), respectively. P(D|A,B,C)=P(A,B,C,D) / P(A,B,C), and P(A,B,C,D) can be obtained by eliminating E and F from the joint probability P(A,B,C,D, E,F) using the elimination inference algorithm. The joint probability can be converted into a CPT product operation, so the inference operation is finally converted into an operation on the CPT. With the known abnormal nodes as evidence and the CPT of each known abnormal node, the posterior probability of the unknown node can be inferred. In this way, the state of the underlying fault base can be roughly determined, and the parameters that are degraded can be located (this is the process of inferring the "cause" from the "effect").

[0099] Among them, select the data with poor performance, such as Figure 4 The data with A=VL in the table is the data with very low node processing latency. The evidence of each data is the value of A and C. Using the optimized CPT, the probabilities of P(D|A,C), P(E|A,C), and P(F|A,C) are inferred, that is, the state of the fault base. For a large amount of data with poor performance, the degradation state of the fault base can be obtained, that is, the lowest-level impact vector (fault base) that causes the poor performance of node A can be found.

[0100] In a possible embodiment, step 105 includes:

[0101] Step 1051, try to adjust the lowest-level influencing variables that meet the second preset abnormal condition, and determine at least one second maintenance plan that makes the abnormal node meet the preset normal condition;

[0102] Specifically, since the Bayesian network can probabilistically infer the status of other nodes from any given one or more nodes, based on the status of the fault base inferred in step 104, each fault base is adjusted continuously, and the network performance of the corresponding top-level feature variable is inferred upward from the adjusted fault base (this is the process of inferring the "effect" from the "cause").

[0103] That is, by trying to adjust the fault base, the upper abnormal node meets the preset normal condition, that is, it is in the preset normal state, so that a feasible second maintenance plan can be obtained; through multiple trial adjustments, at least one second maintenance plan can be obtained. In the method of the present application, the way to adjust the fault base can be a single adjustment or a combined adjustment. Different ways of adjusting the fault base are all within the protection scope of the method of the present application.

[0104] For example Figure 4In the case of a large amount of bad evidence of A and C, it is inferred that P(D=abnormal|A,C)=0.9 and P(D=normal|A,C)=0.1. At this time, according to the preset probability threshold (i.e., the second preset abnormal condition, assumed to be 0.8), it can be judged that D: sensor failure has occurred. In other words, the sensor failure needs to be checked. For example, it is inferred that P(E=VH|A,C)=0.9, P(E=VL|A,C)=0.02, P(E=L|A,C)=0.02, P(E=M|A,C)=0.03, and P(E=H|A,C)=0.03, which means that the temperature is too high. However, it is unknown whether the high temperature affects the node processing delay. At this time, the fault basis E can be continuously adjusted to make E=H, M, L, and VL respectively, and when E=H, P(E=VH|A,C)=0.9, P(E=VL|A,C)=0.02, P(E=L|A,C)=0.02, P(E=M|A,C)=0.03, and P(E=H|A,C)=0.03. =H) is set to 1, and the rest of the probabilities are set to 0; when E = M, L, VL, the same applies, and other fault bases such as D and F remain unchanged, and then the adjusted values ​​of D, E, and F are used as evidence to infer the probability of P(A|D,E,F). If the probability of P(A|D,E,F) does not change, it indicates that the temperature reduction will hardly affect the node processing delay. Otherwise, it means that the ambient temperature of the system operation needs to be reduced. This is equivalent to providing a maintenance strategy for the wireless sensor network.

[0105] For another example, the method of adjusting the fault base can also be to move the probability distribution of the fault base as a whole. Assuming that the probability of the temperature being abnormal is 0.2 and the probability of the temperature being normal is 0.8 based on the abnormal node, after the overall probability distribution of the temperature is moved to the right by 20%, the probability of the temperature being abnormal is 0.4 and the probability of the temperature being normal is 0.6. Based on the adjusted probability distribution of the temperature, the discrete value corresponding to the top abnormal node is inferred upward, and the discrete value and the corresponding probability are multiplied and averaged. For example, assuming that the discrete value inferred upward when the temperature is abnormal is A and the discrete value inferred upward when the temperature is normal is B, then the sum of 0.4A+0.6B is the discrete value of the top abnormal node. By comparing the discrete value with the preset discrete value range under the normal state, it can be determined whether the top abnormal node is in a normal state. When the abnormal node is in a normal state, it indicates that the adjustment plan is feasible, and the adjustment plan is "moving the overall probability distribution of the temperature to the right by 20%".

[0106] Step 1052, determining a cost function value corresponding to the second maintenance solution;

[0107] Specifically, there is a cost function for the improvement of each cause. According to the cost functions of all fault bases in the second maintenance plan, the overall cost of the second maintenance plan is calculated. This is used for comparison. For example, if the fault base is temperature, assuming that the temperature was originally 80 degrees, and the second maintenance plan recommends reducing the temperature to 50 degrees, the adjustment step is 30 degrees, and 30 degrees is the cost value corresponding to the temperature. By analogy, the overall cost corresponding to the second maintenance plan can be obtained.

[0108] Step 1053: Use the second maintenance solution with the minimum cost function value as the first maintenance solution.

[0109] Specifically, in the embodiment of the present invention, the adjustment is made according to the size of the fault adjustment base cost, and the solution with the minimum cost is selected to perform wireless sensor network maintenance, which can reduce the maintenance cost.

[0110] In step 105 of the embodiment of the present invention, the degradation state of the fault base is inferred downwards by using the on-site abnormal data, and the fault base is adjusted and inferred upwards until the result of the corresponding top-level characteristic variable is improved to be non-abnormal. The scheme for adjusting the fault base can be used as a maintenance strategy, and finally the maintenance strategy with the lowest cost is selected as the maintenance improvement strategy of the wireless sensor network. By using the ability of the Bayesian network to infer from any known node to other unknown nodes, the adjustment effect is inferred by adjusting the fault base, so that the maintenance strategy finally obtained has a theoretical basis.

[0111] The method of the embodiment of the present invention solves the problems of high maintenance cost and subjective maintenance strategy in large-scale wireless sensor network maintenance; based on Bayesian network and data-driven technology, it reduces human subjectivity; when using the maximum expectation algorithm, it fully mines the information contained in historical data and reduces the cost of traditional high-quality data collection; based on the representation, learning and reasoning capabilities of Bayesian network, it realizes a high-efficiency and low-cost maintenance method; and provides a research focus for the maintenance improvement of wireless sensor networks. In addition, through the simulation experiment of wireless sensor network, the reasoning accuracy of a single Bayesian network node can reach an average of more than 80%.

[0112] Based on the description of the above wireless sensor network maintenance method embodiment, the present invention also discloses a wireless sensor network maintenance device. Figure 5 , Figure 5 1 is a schematic diagram of the structure of a wireless sensor network maintenance device provided by an embodiment of the present invention, the wireless sensor network maintenance device comprising:

[0113] Establishing module 501, used to establish a Bayesian network between observable top-level characteristic variables of the wireless sensor network and influencing variables affecting the top-level characteristic variables according to expert prior knowledge, wherein the top-level characteristic variables are used to characterize the performance of the wireless sensor network, and the influencing variables include unobservable variables;

[0114] A determination module 502 is used to determine an initial condition probability table for each node in the Bayesian network;

[0115] The optimization module 503 is used to optimize the initial conditional probability table according to the discrete characteristic data of the wireless sensor network to obtain the optimized conditional probability table of each node, wherein the discrete characteristic data is the observed data of the observable variables in the Bayesian network;

[0116] The inference module 504 is used to determine the abnormal node that meets the first preset abnormal condition in the top-level feature variables, and infer the lowest-level influencing variables that meet the second preset abnormal condition from the abnormal node downward according to the optimized condition probability table;

[0117] The adjustment module 505 is used to adjust the lowest-level influencing variables that meet the second preset abnormal condition, and determine a first maintenance plan that enables the abnormal node to meet the preset normal condition.

[0118] In a possible embodiment, the influencing variables corresponding to the top-level characteristic variables are an influencing variable set including at least one level, and one level includes at least one influencing variable.

[0119] In a possible embodiment, the determination module is specifically used for:

[0120] The initial condition probability table for each node is determined based on expert prior knowledge and / or historical statistical data.

[0121] In a possible embodiment, the optimization module is specifically used for:

[0122] According to the discrete characteristic data and the expected maximum algorithm, the initial conditional probability table is optimized to obtain the optimized conditional probability table.

[0123] In a possible embodiment, the reasoning module is specifically used for:

[0124] According to the abnormal nodes and the Bayesian network reasoning algorithm, the lowest-level influencing variables that meet the second preset abnormal condition are inferred downward.

[0125] In a possible embodiment, the Bayesian network inference algorithm includes an elimination inference algorithm or a clique tree inference algorithm.

[0126] In a possible embodiment, the adjustment module includes:

[0127] The first submodule is used to try to adjust the lowest-level influencing variables that meet the second preset abnormal condition, and determine at least one second maintenance plan that makes the abnormal node meet the preset normal condition;

[0128] The second submodule is used to determine the cost function value corresponding to the second maintenance plan;

[0129] The third submodule is used to use the second maintenance plan with the smallest cost function value as the first maintenance plan.

[0130] It is worth pointing out that, among them, the specific functional implementation mode of the wireless sensor network maintenance device can refer to the description of the above-mentioned wireless sensor network maintenance method, which will not be repeated here. Each unit or module in the wireless sensor network maintenance device can be separately or completely merged into one or several other units or modules to constitute, or one (some) of the units or modules can also be split into multiple smaller units or modules to constitute, which can achieve the same operation without affecting the realization of the technical effect of the embodiment of the present invention. The above-mentioned units or modules are divided based on logical functions. In practical applications, the function of a unit (or module) can also be realized by multiple units (or modules), or the function of multiple units (or modules) is realized by one unit (or module).

[0131] Based on the description of the above method embodiment and device embodiment, an embodiment of the present invention further provides a wireless sensor network maintenance device.

[0132] See also Figure 6 , is a schematic diagram of the structure of a wireless sensor network maintenance device provided by an embodiment of the present invention. Figure 6 As shown, the above-mentioned wireless sensor network maintenance device can be applied to the wireless sensor network maintenance device 600, and the wireless sensor network maintenance device 600 may include: a processor 601, a network interface 604 and a memory 605. In addition, the wireless sensor network maintenance device 600 may also include: a user interface 603, and at least one communication bus 602. Among them, the communication bus 602 is used to realize the connection and communication between these components. Among them, the user interface 603 may include a display screen (Display), a keyboard (Keyboard), and the user interface 603 may optionally include a standard wired interface and a wireless interface. The network interface 604 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 605 may be a high-speed RAM memory, or it may be a non-volatile memory (non-volatile memory), such as at least one disk storage. The memory 605 may optionally also be at least one storage device located away from the aforementioned processor 601. As Figure 6 As shown, the memory 605 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a device control application program.

[0133] exist Figure 6In the wireless sensor network maintenance device 600 shown, the network interface 604 can provide network communication functions; the user interface 603 is mainly used to provide an input interface for the user; and the processor 601 can be used to call the device control application stored in the memory 605 to implement the steps in the above-mentioned wireless sensor network maintenance method.

[0134] It should be understood that the wireless sensor network maintenance device 600 described in the embodiment of the present invention can execute the wireless sensor network maintenance method described above, and can also execute the description of the wireless sensor network maintenance device described above, which will not be repeated here. In addition, the description of the beneficial effects of the same method will not be repeated.

[0135] In addition, it should be pointed out here that: an embodiment of the present invention also provides a computer storage medium, and the computer storage medium stores a computer program executed by the wireless sensor network maintenance device mentioned above, and the computer program includes program instructions. When the processor executes the program instructions, it can execute the description of the wireless sensor network maintenance method mentioned above, so it will not be repeated here. In addition, the description of the beneficial effects of using the same method will not be repeated. For technical details not disclosed in the computer storage medium embodiment involved in the present invention, please refer to the description of the method embodiment of the present invention.

[0136] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0137] The above disclosure is only the preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A wireless sensor network maintenance method, characterized in that: include: According to the expert prior knowledge, a Bayesian network is established between the observable top-level characteristic variables of the wireless sensor network and the influencing variables affecting the top-level characteristic variables, wherein the top-level characteristic variables are used to characterize the performance of the wireless sensor network, and the influencing variables include unobservable variables; Determining an initial condition probability table for each node in the Bayesian network; Optimizing the initial conditional probability table according to the discrete feature data of the wireless sensor network to obtain an optimized conditional probability table for each node, wherein the discrete feature data is observation data of observable variables in the Bayesian network; Determine an abnormal node in the top-level feature variable that satisfies a first preset abnormal condition, and infer the lowest-level influencing variable that satisfies a second preset abnormal condition from the abnormal node according to the optimized condition probability table; The bottom-level influencing variables are adjusted to determine a first maintenance plan that enables the abnormal node to meet a preset normal condition.

2. The method according to claim 1, characterized in that The step of determining an initial condition probability table for each node in the Bayesian network comprises: The initial condition probability table of each node is determined according to the expert prior knowledge and / or historical statistical data.

3. The method according to claim 1, characterized in that The step of optimizing the initial condition probability table according to the discrete characteristic data of the wireless sensor network to obtain an optimized condition probability table for each node includes: The initial conditional probability table is optimized according to the discrete feature data and the expected maximum algorithm to obtain the optimized conditional probability table.

4. The method according to any one of claims 1 to 3, characterized in that: The step of inferring downward from the abnormal node to obtain the lowest-level influencing variable that meets the second preset abnormal condition includes: According to the abnormal nodes and the Bayesian network reasoning algorithm, the bottom-level influencing variables are inferred downwards.

5. The method according to claim 4, characterized in that The Bayesian network reasoning algorithm includes an elimination reasoning algorithm or a clique tree reasoning algorithm.

6. The method according to any one of claims 1 to 3, characterized in that: Adjusting the bottom-level influencing variables to determine a first maintenance plan that enables the abnormal node to meet a preset normal condition includes: Try to adjust the bottom-level influencing variables to determine at least one second maintenance plan that enables the abnormal node to meet the preset normal conditions; Determining a cost function value corresponding to the second maintenance solution; The second maintenance plan with the smallest cost function value is used as the first maintenance plan.

7. The method according to any one of claims 1 to 3, characterized in that: The influencing variables corresponding to the top-level characteristic variables are an influencing variable set including at least one level, and the level includes at least one influencing variable.

8. A wireless sensor network maintenance device, characterized in that: include: An establishment module is used to establish a Bayesian network between observable top-level characteristic variables of the wireless sensor network and influencing variables affecting the top-level characteristic variables according to expert prior knowledge, wherein the top-level characteristic variables are used to characterize the performance of the wireless sensor network, and the influencing variables include unobservable variables; A determination module, used to determine an initial condition probability table for each node in the Bayesian network; An optimization module, configured to optimize the initial condition probability table according to discrete feature data of the wireless sensor network to obtain an optimized condition probability table for each node, wherein the discrete feature data is observation data of observable variables in the Bayesian network; An inference module, used to determine an abnormal node in the top-level feature variable that satisfies a first preset abnormal condition, and infer the lowest-level influencing variable that satisfies a second preset abnormal condition from the abnormal node downward according to the optimized condition probability table; The adjustment module is used to adjust the bottom-level influencing variables to determine a first maintenance plan that enables the abnormal node to meet a preset normal condition.

9. A wireless sensor network maintenance device, characterized in that: include: Processor and memory; The processor is connected to a memory, wherein the memory is used to store program codes, and the processor is used to call the program codes to execute the wireless sensor network maintenance method according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that: The computer storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, the wireless sensor network maintenance method according to any one of claims 1 to 7 is executed.

Images