A Remote Monitoring Method and System Based on the Internet of Things

Abstract

The present invention relates to the field of monitoring and early warning, and particularly to a remote monitoring method and system based on the Internet of Things. The method includes the steps of: obtaining a plurality of operation index data points of a machine; using the operation index data points to complete the clustering iterative training of the SOM algorithm to achieve operation monitoring; in any clustering training stage, obtaining the activated neurons of the current clustering training stage, obtaining the reference clustering training stage, and obtaining the data sets of each neuron in each clustering training stage; obtaining fuzzy neurons according to the intersection relationship between the data set of the current clustering training stage of the activated neurons and the data sets of the reference training stages of other neurons; calculating the learning rates of the fuzzy neurons; and updating and training the fuzzy neurons and the activated neurons according to the learning rates. By accurately updating the neurons, the accuracy of clustering is improved, and further the monitoring accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of monitoring and early warning, and particularly to a remote monitoring method and system based on the Internet of Things. Background Art

[0002] With the development of technology, the degree of machine intelligence is getting higher and higher, and enterprises increasingly advocate the unmanned management of machines. However, it is inevitable that abnormalities will occur during the operation of machines. If the abnormalities cannot be detected in time, a series of subsequent problems will occur; for example, if an abnormal reaction kettle causes the produced compound to not meet the quality requirements, the compound can only be sold at a low price as a defective product, resulting in a waste of resources.

[0003] In order to timely understand the operating conditions of machines, it is necessary to monitor the operating conditions of machines. Since there are differences between the operating index data generated when the machine operates abnormally and the operating index data generated during normal operation, it is possible to judge whether the machine is in an abnormal condition by screening for outliers. Since the machine has multiple control states and the operating index data corresponding to each control state is different, when using the method of screening outliers to judge the operating state of the machine, misjudgment of outliers is often caused by differences in control states, thus misjudging the normal operation of the machine as abnormal operation. To solve this problem, it is necessary to perform clustering processing on the operating index data. The Self-organizing map (SOM) neural network algorithm is commonly used for data clustering processing. This algorithm selects activated neurons based on the similarity or distance between the data and the neurons, and then uses the data to update and train the activated neurons until the neurons converge and the update training ends, and then it can be used for clustering processing. However, during the training stage, the neurons are not stable, and the nearest neurons or the most similar neurons of the data are different at different stages, resulting in different activated neurons being selected based on the similarity or distance between the data and the neurons at different stages. If the data only updates the activated neurons, the neurons of the true category will not be updated, resulting in an unsatisfactory training effect of the SOM algorithm. Therefore, how to screen out the neurons that need to be updated and control the update of the neurons that need to be updated has become the research focus of the present invention.

[0004] The patent application document with the publication number CN101846995A discloses an industrial field remote monitoring method. The method in this patent application document mainly shows the monitoring process, and the abnormal detection involved in the monitoring process is only the application of existing abnormal detection algorithms, and does not involve the content of SOM neuron update control. Therefore, the method in this patent application document cannot solve the technical problems in this solution. Summary of the Invention

[0005] To solve the problem of how to screen out neurons that need to be updated and perform update control on the neurons that need to be updated, the present invention provides a remote monitoring method and system based on the Internet of Things.

[0006] In a first aspect, the present invention provides a remote monitoring method based on the Internet of Things, adopting the following technical solution:

[0007] A remote monitoring method based on the Internet of Things includes the steps of:

[0008] Obtain a number of operating index data points of the reactor;

[0009] Use the operating index data points to complete the clustering iterative training of the SOM algorithm to achieve operating monitoring;

[0010] In any clustering training stage, obtain the activated neurons in the current clustering training stage, obtain the previous preset number of clustering training stages, denoted as reference clustering training stages, and divide the operating index data points into the data sets of the corresponding neurons according to the distances between the operating index data points and the neurons in each clustering training stage; obtain fuzzy neurons according to the intersection relationship between the data set of the current clustering training stage of the activated neurons and the data sets of other neurons in the reference training stages;

[0011] Calculate the learning rate of each fuzzy neuron , , represents the number of data points in the intersection of the data set of the i-th reference clustering training stage of the k-th fuzzy neuron and the data set of the current clustering training stage of the activated neuron, N represents the preset number, represents the number of data points in the data set of the current clustering training stage of the activated neuron, , respectively represent the distances between the operating index data points used for the clustering training in the current clustering training stage and the activated neuron and the k-th fuzzy neuron, respectively represent the variances of the position changes of the k-th fuzzy neuron and the activated neuron in all reference clustering training stages, represents the preset learning rate of the activated neuron;

[0012] Update and train the fuzzy neurons and the activated neurons according to the learning rate.

[0013] In view of the instability of neurons, the activated neurons selected in the current training stage are not necessarily the cluster centers to which the data belongs. Any neuron that has an intersection relationship with the dataset of the activated neurons may become the cluster center to which the data point belongs. Therefore, by analyzing the intersection relationship between each neuron and the dataset of the activated neurons in different training stages, the fuzzy neurons that may become the cluster centers of the data points are accurately screened out, providing a basis for accurately updating the neurons. Further, by analyzing the possibility that each fuzzy neuron is the cluster center of the data point, the learning rate for updating each fuzzy neuron by the data point is accurately calculated. Further, when analyzing the possibility that each fuzzy neuron is the cluster center of the data point, by introducing the number of intersection data points between the fuzzy neuron and the dataset of the activated neurons, the possibility of the data point flowing to the category to which the fuzzy neuron belongs is accurately reflected, and thus the possibility that each fuzzy neuron is the cluster center of the data point is accurately reflected. Further, when analyzing the possibility that each fuzzy neuron is the cluster center of the data point, by introducing the distance from the data point to the fuzzy neuron, the possibility that the data point becomes the category to which the fuzzy neuron belongs is accurately reflected, and thus the possibility that each fuzzy neuron is the cluster center of the data point is accurately reflected.

[0014] Preferably, the step of dividing the operation index data points into the datasets corresponding to the neurons according to the distances between the operation index data points and the neurons in each clustering training stage includes:

[0015] In each clustering training stage, obtain the distances between each operation index data point and each neuron, and divide each operation index data point into the dataset of the neuron with the closest distance to it, so as to obtain the datasets of each neuron in each clustering training stage.

[0016] The present invention divides the data points into the datasets corresponding to the neurons through the distance relationship, and this implementation method is relatively simple and has higher implementation efficiency.

[0017] Preferably, the step of obtaining the fuzzy neurons according to the intersection relationship between the dataset of the activated neuron in the current clustering training stage and the datasets of other neurons in the reference training stages includes:

[0018] If there is an intersection between the dataset of the activated neuron in the current clustering training stage and the datasets of other neurons in the reference clustering training stages, then mark the other neurons as suspected fuzzy neurons;

[0019] Mark the data points that exist in both the dataset of the activated neuron in the current clustering training stage and the dataset of the reference training stage of the fuzzy neuron as jump data points;

[0020] Calculate the abnormality possibility of the jump data points; mark the jump data points with the abnormality possibility less than the preset possibility threshold as target jump data points;

[0021] Mark the suspected fuzzy neurons with target jump data points in the dataset during the reference clustering training phase as fuzzy neurons.

[0022] The present invention accurately screens out fuzzy neurons that may become the clustering centers of data points by analyzing the intersection relationship between the datasets of each neuron and the activated neurons in different training phases; further, considering that abnormal data points may also jump between the corresponding categories of different neurons, and this kind of jump is not caused by insufficient neuron update, it is necessary to exclude the interference of abnormal data points in the intersection relationship and accurately extract the fuzzy neurons.

[0023] Preferably, the obtaining of the abnormal possibility of the jump data points includes:

[0024] Obtain the single-sided neighborhood data points of the jump data points; record the average distance between the jump data points and the single-sided neighborhood data points as the metric distance, and determine the normalized value of the ratio of the metric distance of the jump data points to the average metric distance of all operation index data points as the abnormal possibility of the jump data points.

[0025] The present invention excludes the influence of neighborhood data of other categories on distance calculation by introducing single-sided neighborhood data points, and relatively accurately analyzes the abnormal conditions of data points.

[0026] Preferably, the obtaining of the single-sided neighborhood data points of the jump data points includes:

[0027] Obtain the area within the preset radius of the jump data points as the preset neighborhood, and record the operation index data points within the preset neighborhood as neighborhood data points; evenly divide the preset neighborhood into a preset number of regions with the jump data points as the center;

[0028] Calculate the average value of the distances between the neighborhood data points and the jump data points in each region as the distance of the region;

[0029] Take the neighborhood data points in the region where the distance is less than the average value of the distances of all regions as the single-sided neighborhood data points.

[0030] The present invention accurately screens out data points of other categories by analyzing the data points in the neighborhood regionally, providing a basis for accurate abnormal analysis.

[0031] Preferably, the obtaining method of the position change variance includes:

[0032] Obtain the positions of each neuron in each clustering reference phase, calculate the distances between the positions of each neuron in two reference clustering training phases, and record the variance of the position distances of each neuron in all two reference clustering training phases as the position change variance of the neuron in all reference clustering training phases.

[0033] Preferably, the implementation of operation monitoring includes:

[0034] Inputting the newly collected operation index data points of the reactor into the SOM algorithm completed by clustering training to obtain the category to which the newly collected operation index data points belong; obtaining the lower quartile of the distance between each operation index data point and the clustering center of its category;

[0035] Obtaining the distance between the newly collected operation index data point and the clustering center of its category. If the distance between the newly collected operation index data point and the clustering center of its category is greater than the lower quartile, it is determined that the newly collected operation index data point is an abnormal data point;

[0036] If the number of continuously collected abnormal data points is greater than the preset quantity threshold, it is determined that there is an abnormality in the operation of the reactor.

[0037] In a second aspect, the present invention provides a remote monitoring system based on the Internet of Things, adopting the following technical solution:

[0038] A remote monitoring system based on the Internet of Things includes: a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned remote monitoring method based on the Internet of Things is implemented.

[0039] By adopting the above technical solution, the above-mentioned remote monitoring method based on the Internet of Things is generated into a computer program and stored in the memory to be loaded and executed by the processor, so as to manufacture a terminal device according to the memory and the processor, which is convenient to use.

[0040] The present invention has the following technical effects:

[0041] Considering that when neurons are unstable, the activated neurons selected in the current training stage are not necessarily the clustering centers to which the data belongs, and any neuron having an intersection relationship with the data set of the activated neurons may become the clustering center to which the data point belongs. Therefore, by analyzing the intersection relationship between each neuron and the data set of the activated neurons in different training stages, the fuzzy neurons that may become the clustering centers of the data points are accurately screened out, providing a basis for accurately updating the neurons;

[0042] Furthermore, by analyzing the possibility of each fuzzy neuron being the clustering center of the data point, the learning rate for accurately updating each fuzzy neuron by the data point is calculated;

[0043] Further, when analyzing the possibility of each fuzzy neuron being the clustering center of data points, the number of intersection data points of the data sets of the fuzzy neurons and the activated neurons is introduced to accurately reflect the possibility of the data points flowing to the categories to which the fuzzy neurons belong, and further accurately reflect the possibility of each fuzzy neuron being the clustering center of the data points;

[0044] Further, when analyzing the possibility of each fuzzy neuron being the clustering center of data points, the distance from the data points to the fuzzy neurons is introduced to accurately reflect the possibility of the data points becoming the categories to which the fuzzy neurons belong, and further accurately reflect the possibility of each fuzzy neuron being the clustering center of the data points. Brief Description of the Drawings

[0045] By referring to the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts.

[0046] Figure 1 is a flowchart of the method in a remote monitoring method based on the Internet of Things according to an embodiment of the present invention. Detailed Embodiments

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0048] It should be understood that when the claims, specifications and drawings of the present invention use the terms "first", "second", etc., they are only used to distinguish different objects and not to describe a specific order. The terms "comprising" and "including" used in the specifications and claims of the present invention indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their combinations.

[0049] An embodiment of the present invention discloses a remote monitoring method based on the Internet of Things. Refer to Figure 1 , including steps S1 - step S2:

[0050] S1: Obtain a number of operating index data points of the machine.

[0051] Specifically, obtain each type of operation index data of the machine at each moment. In this embodiment, the machine is a reactor as the monitoring object for description. In other embodiments, other monitoring objects can be used, and this embodiment does not make specific restrictions. When the reactor is used as the monitoring object, the types of operation index data mainly collected include but are not limited to the following aspects: temperature, pressure, liquid level, stirring speed, and flow rate.

[0052] S2: Use the operation index data points to complete the clustering iterative training of the SOM algorithm to achieve operation monitoring.

[0053] S20: Use the operation index data points to complete the clustering iterative training of the SOM algorithm.

[0054] S200: In any clustering training stage, obtain the activated neuron in the current clustering training stage, obtain the previous preset number of clustering training stages, denoted as the reference clustering training stages, and divide the operation index data points into the data sets corresponding to the neurons according to the distances between the operation index data points and the neurons in each clustering training stage; obtain the fuzzy neurons according to the intersection relationship between the data set of the current clustering training stage of the activated neuron and the data sets of the reference training stages of other neurons.

[0055] It should be noted that when the neurons are unstable, the affiliated data between different neurons will switch back and forth in different training stages. In other words, some data has not yet determined which neuron is its clustering center at this time, resulting in some data belonging to one neuron in some training stages and belonging to another neuron in the next training stage. Therefore, in each training stage, it is not accurate enough to only update the data for the activated neurons. Therefore, it is necessary to determine which neurons are its possible clustering centers and then update all its possible clustering centers.

[0056] First, in any clustering training stage, obtain the activated neuron in the current clustering training stage, obtain the previous preset number of clustering training stages, denoted as the reference clustering training stages, and divide the operation index data points into the data sets corresponding to the neurons according to the distances between the operation index data points and the neurons in each clustering training stage.

[0057] Preferably, as an example, in any clustering training stage, obtain the activated neuron in the current clustering training stage, obtain the previous preset number of clustering training stages, denoted as the reference clustering training stages, and divide the operation index data points into the data sets corresponding to the neurons according to the distances between the operation index data points and the neurons in each clustering training stage, including:

[0058] In any clustering training stage, obtain the activated neurons in the current clustering training stage, and obtain the previous preset number of clustering training stages, denoted as reference clustering training stages. In this embodiment, the preset number is taken as 5 for description, and other values can be taken in other embodiments, which are not specifically limited in this embodiment.

[0059] Obtain the distances between each operation index data point and each neuron in each clustering training stage, and divide each operation index data point into the data set of the neuron with the closest distance to it, so as to obtain the data sets of each neuron in each clustering training stage.

[0060] It should be noted that the data points in the data sets of each neuron in each clustering training stage are the affiliated data points of each neuron in each clustering training stage.

[0061] It should be further noted that the method for obtaining the activated neurons in the current clustering training stage is a prior art and will not be elaborated here.

[0062] Then, in any clustering training stage, obtain fuzzy neurons according to the intersection relationship between the data set of the activated neuron in the current clustering training stage and the data sets of other neurons in the reference training stage.

[0063] It should be noted that in different training stages, there will be a switch of affiliated data points between some neurons, indicating that the affiliated data will flow back and forth among these neurons, thus indicating that these neurons may all be the clustering centers of the data. Therefore, when updating the neurons, these neurons need to be updated. Therefore, the interfering neurons of the neurons can be determined according to the switching relationship of the affiliated data points of different neurons.

[0064] It should be further noted that since some data are abnormal data, some data cannot be accurately divided into the data sets of which neurons, resulting in the switching of this data between different neurons. This is caused by the data rather than the instability of the neurons. Therefore, the influence of this data on the acquisition of interfering neurons needs to be excluded.

[0065] Preferably, as an example, in any clustering training stage, obtaining fuzzy neurons according to the intersection relationship between the data set of the activated neuron in the current clustering training stage and the data sets of other neurons in the reference training stage includes:

[0066] If there is an intersection between the data set of the activated neuron in the current clustering training stage and the data sets of other neurons in the reference clustering training stage, then mark the other neurons as suspected fuzzy neurons;

[0067] Data points in the dataset of the current clustering training stage for activating neurons and in the dataset of the reference training stage for fuzzy neurons that exist simultaneously are denoted as jump data points;

[0068] Calculate the anomaly likelihood of the jump data points; Denote the jump data points with an anomaly likelihood less than a preset likelihood threshold as target jump data points; In this embodiment, taking the preset likelihood threshold as 0.85 as an example for description, other embodiments can take other values, and this embodiment does not make specific limitations.

[0069] Denote the suspected fuzzy neurons in the dataset of the reference clustering training stage that have target jump data points as fuzzy neurons.

[0070] It should be noted that a fuzzy neuron is a neuron with an affiliated data switch with an activating neuron. The data belongs to the activating neuron in the current training stage and may belong to the fuzzy neuron in subsequent training stages. Therefore, it cannot be determined whether the fuzzy neuron or the activating neuron is the clustering center of the data in the current training stage. Thus, the data needs to be updated for both the fuzzy neuron and the activating neuron.

[0071] The above embodiments involve the anomaly likelihood of jump data points. Next, the calculation method of the anomaly likelihood of jump data points needs to be described.

[0072] Optionally, as an example, calculating the anomaly likelihood of jump data points includes:

[0073] Obtain the operation index data points within the preset radius of the jump data points and denote them as neighborhood data points; In this embodiment, taking the preset radius as 20 as an example for description, other embodiments can take other values, and this embodiment does not make specific limitations.

[0074] Calculate the mean of the distances between the neighborhood data points and the jump data points, and denote it as the neighborhood distance of the jump data points;

[0075] Calculate the neighborhood distances of other operation index data points, and use the normalized data of the ratio of the neighborhood distance of the jump data points to the mean of the neighborhood distances of all other operation index data points as the anomaly likelihood of the jump data points.

[0076] It should be noted that since some jump data points are at the classification boundary, there may be data points of other categories within the neighborhood of the jump data points, and it is normal for the distances between the jump data points and the data points of other categories to be large. However, a large distance between the jump data points and the data points of other categories will result in a large calculated anomaly likelihood of the jump data points. Thus, it is easy to misjudge the jump data points at the classification boundary as abnormal data points in this way, and then it is impossible to accurately eliminate the influence of abnormal data points on the acquisition of interfering neurons.

[0077] Preferably, as an example, calculating the anomaly possibility of the jump data points includes:

[0078] Obtain the area within the preset radius of the jump data point and denote it as the preset neighborhood, and denote the operation index data points within the preset neighborhood as neighborhood data points; evenly divide the preset neighborhood into a preset number of regions with the jump data point as the center; in this embodiment, the preset number is taken as 30 for description, and other values can be taken in other embodiments, and this embodiment does not make specific limitations.

[0079] Calculate the mean value of the distances between the neighborhood data points and the jump data point in each region and denote it as the distance of the region;

[0080] Take the neighborhood data points within the region where the distance is less than the mean value of the distances of all regions as the single-sided neighborhood data points.

[0081] Denote the mean value of the distances between the jump data point and the single-sided neighborhood data points as the metric distance, and judge the normalized value of the ratio of the metric distance of the jump data point to the mean value of the metric distances of all operation index data points as the anomaly possibility of the jump data point.

[0082] It should be noted that by dividing the region to exclude data of other categories in the neighborhood, the distance relationship between the jump data point and other surrounding data points can be accurately measured, so as to relatively accurately reflect the anomaly situation of the jump data, and further relatively accurately exclude the influence of abnormal data points on the interference of neurons.

[0083] S201: Calculate the learning rate of each fuzzy neuron.

[0084] It should be noted that the neurons that need to be updated have been screened out in the above process, and the following is to perform update control on each neuron that needs to be updated. The learning rate is used to determine the update degree of each neuron, so the update control of each neuron that needs to be updated is performed by adjusting the learning rate.

[0085] Preferably, as an example, calculating the learning rate of each fuzzy neuron includes:

[0086]

[0087] Wherein, represents the number of data points in the intersection of the data sets of the i-th reference clustering training stage of the k-th fuzzy neuron and the current clustering training stage of the activated neuron, N represents the preset number, represents the number of data points in the data set of the current clustering training stage of the activated neuron, represents the distance between the operation index data points used for the clustering training of the current clustering training stage and the activated neuron, Represents the distance between the running metric data point for the clustering training in the current clustering training phase and the k-th fuzzy neuron, Represents the variance of the position change of the k-th fuzzy neuron in all reference clustering training phases, Represents the variance of the position change of the activated neuron in all reference clustering training phases, Represents the preset learning rate of the activated neuron, Represents the learning rate of the k-th fuzzy neuron.

[0088] It can be understood that, Reflects the possibility that the fuzzy neuron is the clustering center of the data for the clustering training in the current clustering training phase. The larger this value, the more necessary it is to use this data to update the fuzzy neuron, so as to more accurately locate each clustering center; Reflects the relative distance relationship between the data for clustering training and the fuzzy neuron and the activated neuron in the current clustering training phase. The larger this value, the closer the data for clustering training is to this fuzzy neuron, and the more it should be used to update this fuzzy neuron; Reflects the stability of this fuzzy neuron. The larger this value, the more unstable this fuzzy neuron is, and the more it needs to be updated. Therefore, its learning rate should be larger, Reflects the necessity of updating this fuzzy neuron compared with the activated neuron. By multiplying by the preset learning rate of the activated neuron, the learning rate of this fuzzy neuron can be obtained.

[0089] It should be added that the method for obtaining the variance of position change includes:

[0090] Obtain the positions of each neuron in each clustering reference phase, calculate the distance between the positions of each neuron in two reference clustering training phases, and record the variance of the position distances of each neuron in all two-reference clustering training phases as the variance of the position change of the neuron in all reference clustering training phases.

[0091] S202: Update and train the fuzzy neuron and the activated neuron according to the learning rate.

[0092] It should be noted that updating and training the fuzzy neuron and the activated neuron according to the learning rate is an existing technology and will not be elaborated here.

[0093] S21: Implement running monitoring.

[0094] Preferably, as an example, implementing running monitoring includes:

[0095] Input the operation index data points of the newly collected reactor into the SOM algorithm that has completed clustering training to obtain the category to which the newly collected operation index data points belong; obtain the lower quartile of the distances between each operation index data point and the cluster center of its category.

[0096] Obtain the distance between the newly collected operation index data point and the cluster center of its category. If the distance between the newly collected operation index data point and the cluster center of its category is greater than the lower quartile, then determine that the newly collected operation index data point is an abnormal data point.

[0097] If the number of continuously collected abnormal data points is greater than the preset quantity threshold, then determine that there is an abnormality in the operation of the reactor, and notify the relevant personnel for adjustment. In this embodiment, the preset quantity threshold is taken as 5 for description. Other embodiments can take other values, and this embodiment does not make specific limitations.

[0098] An embodiment of the present invention also discloses an Internet of Things-based remote monitoring system, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, an Internet of Things-based remote monitoring method according to the present invention is implemented.

[0099] The above system also includes other components well known to those skilled in the art such as a communication bus and a communication interface. Their settings and functions are known in the art, so they will not be elaborated here.

[0100] In the present invention, the foregoing memory can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory, a dynamic random access memory, a static random access memory, an enhanced dynamic random access memory, a high-bandwidth memory, a hybrid storage cube, etc., or any other medium that can be used to store the required information and can be accessed by an application program, a module, or both. Any such computer storage medium can be part of the device or accessible or connectable to the device.

[0101] Although this specification has shown and described multiple embodiments of the present invention, it is obvious to those skilled in the art that such embodiments are provided only by way of example. Those skilled in the art will think of many changes, alterations, and alternative ways without departing from the spirit and idea of the present invention. It should be understood that various alternative solutions to the embodiments of the present invention described herein can be adopted in the process of practicing the present invention.

[0102] The above are all preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention shall be covered within the protection scope of the present invention.

Claims

1. A remote monitoring method based on the Internet of Things, characterized in that: Includes steps: Obtaining several operating indicator data points of the reactor; Use the operation indicator data points to complete the clustering iterative training of the SOM algorithm to achieve operation monitoring; In any cluster training stage, the activated neurons of the current cluster training stage are obtained, and the previous preset number of cluster training stages are obtained and recorded as reference cluster training stages. The operation index data points are divided into the data sets of the corresponding neurons according to the distance between each operation index data point and the neuron in each cluster training stage; the fuzzy neurons are obtained according to the intersection relationship between the data set of the current cluster training stage of the activated neurons and the data sets of the reference training stage of other neurons; Calculate the learning rate of each fuzzy neuron , , represents the number of data points in the intersection of the data set of the i-th reference cluster training stage of the k-th fuzzy neuron and the current cluster training stage of the activated neuron, N represents the preset number, represents the number of data points in the dataset for the current cluster training phase that activated the neuron, , They represent the distances between the running indicator data points of cluster training and the activated neurons and the kth fuzzy neuron used in the current cluster training phase, respectively. Respectively represent the position change variance of the kth fuzzy neuron and the activated neuron in all reference cluster training stages, Represents the preset learning rate for activating neurons; The fuzzy neurons and activation neurons are updated and trained according to the learning rate.

2. The remote monitoring method based on the Internet of Things according to claim 1, characterized in that: The step of dividing the operation index data points into data sets corresponding to neurons according to the distances between the operation index data points and neurons in each cluster training stage includes: In each cluster training stage, the distance between each operation index data point and each neuron is obtained, and each operation index data point is divided into the data set of the neuron closest to it, so as to obtain the data set of each neuron in each cluster training stage.

3. The remote monitoring method based on the Internet of Things according to claim 1, characterized in that: The method of obtaining a fuzzy neuron according to the intersection relationship between the data set of the current cluster training phase of the activated neuron and the data set of the reference training phase of other neurons includes: If the data set of the current cluster training phase of the activated neuron intersects with the data set of the reference cluster training phase of other neurons, the other neurons are recorded as suspected fuzzy neurons; The data points in the data set of the current cluster training phase where activated neurons exist and in the data set of the reference training phase where fuzzy neurons exist are recorded as jump data points; Calculate the abnormal probability of the jumping data point; record the jumping data point whose abnormal probability is less than the preset probability threshold as the target jumping data point; The suspected fuzzy neurons with target jump data points in the data set of the reference cluster training phase are recorded as fuzzy neurons.

4. The remote monitoring method based on the Internet of Things according to claim 3 is characterized in that: The calculating of the abnormal possibility of the jumping data point includes: Obtain the unilateral neighborhood data points of the jumping data point; record the mean distance between the jumping data point and the unilateral neighborhood data points as the metric distance, and determine the normalized value of the ratio of the metric distance of the jumping data point to the mean metric distance of all running indicator data points as the abnormal possibility of the jumping data point.

5. The remote monitoring method based on the Internet of Things according to claim 4 is characterized in that: The step of obtaining a single biased neighborhood data point of a jumping data point includes: The area within the preset radius of the jump data point is recorded as a preset neighborhood, and the operation indicator data points within the preset neighborhood are recorded as neighborhood data points; the preset neighborhood is evenly divided into a preset number of areas with the jump data point as the center; Calculate the mean of the distance between the neighborhood data points and the jump data points in each region and record it as the distance of the region; The neighborhood data points in the area whose distance is less than the mean of the distances of all areas are regarded as unilateral neighborhood data points.

6. The remote monitoring method based on the Internet of Things according to claim 1, characterized in that: The method for obtaining the position change variance includes: The position of each neuron in each cluster reference stage is obtained, the distance between the positions of each neuron in the two reference cluster training stages is calculated, and the variance of the position distance of each neuron in all two reference cluster training stages is recorded as the variance of the position change of the neuron in all reference cluster training stages.

7. The remote monitoring method based on the Internet of Things according to claim 1, characterized in that: The implementation of operation monitoring includes: Input the newly collected operation index data points of the reactor into the SOM algorithm after cluster training to obtain the category to which the newly collected operation index data points belong; obtain the lower quartile of the distance between each operation index data point and the cluster center of the category to which it belongs; Obtaining the distance between the newly collected operation index data point and the cluster center of the category to which it belongs. If the distance between the newly collected operation index data point and the cluster center of the category to which it belongs is greater than the lower quartile, determining that the newly collected operation index data point is an abnormal data point; If the number of continuously collected abnormal data points is greater than a preset number threshold, it is determined that the reactor operation is abnormal.

8. A remote monitoring system based on the Internet of Things, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a remote monitoring method based on the Internet of Things according to any one of claims 1 to 7 is implemented.

Images