A method and system for analyzing the correlation of alarms of intelligent lamp posts based on deep learning

Abstract

The present application relates to a method and system for analyzing the correlation of warnings of intelligent lamp posts based on deep learning. The method includes at least one deep learning process, and the deep learning process includes the steps of: sampling warnings for intelligent lamp posts with physical relevance, and taking a batch of sampled warnings as a temporary warning sample set; performing forward calculation on the temporary warning sample set based on the current neural network to obtain a warning correlation analysis result; marking the warning correlation analysis result according to the situation where the warning is cleared to obtain a marking result; using the temporary warning sample set and the marking result as training samples, and starting the training of the deep neural network when the number of training samples reaches a set value to optimize the model parameters of the current neural network. The present invention can effectively improve the accuracy of the correlation analysis result while avoiding a large amount of one-time manual marking and input work.

Description

Technical Field

[0001] The present invention relates to the technical field of road lamp post resource management, and particularly to a method and system for analyzing the correlation of intelligent lamp post alarms based on deep learning. Background Art

[0002] Currently, the types of devices connected to intelligent lamp posts are increasing, and the correlation between devices is also getting higher. A fault in one device may cause a chain reaction of faults in a series of devices or devices on other lamp posts. For example, a short circuit or open circuit in a device on a certain lamp post may cause the devices on other lamp posts in the same circuit to go offline or have communication failures, and these devices will all send fault alarms to the intelligent lamp post management system; in fact, there is a correlation between these alarms, that is, the root alarm of these alarms is a short circuit or short circuit in a device on a certain lamp post, and the alarms on other devices or lamp posts are related alarms derived from this root alarm. How to accurately analyze the correlation between alarms is an important function of the intelligent lamp post management system.

[0003] In related technologies, some adopt manual setting of static correlation rules, that is, artificially define in advance which alarm combinations are relevant when they appear, and when such an alarm combination appears, it is prompted that these alarms are relevant. This method requires manual continuous summarization and analysis of historical alarm data, extraction of correlation rules, and then entry into the system to continuously adapt to new correlation changes. It has high requirements for personnel analysis ability, and the manual workload will also become huge when the amount of historical alarm data is large; some rely on artificial intelligence algorithms to train and learn from a large amount of historical alarm data, enabling the system to obtain the ability to analyze alarm correlation through self-learning. Although this method can automatically obtain the analysis ability from historical alarm data, it still requires manual cleaning and marking of a large amount of historical alarm data in advance to prepare sample data for training before AI training and learning can be carried out. Moreover, this kind of data for specific business functions can only be processed by a specific few professional personnel, unlike image, text, etc. data that can be assisted by a large number of non-professional personnel, which also makes the workload of preparing this sample data still huge and will also hinder the application and implementation of this method; others rely on data mining algorithms to mine frequently occurring alarm combinations from a large amount of historical alarm data, and then manually determine whether they are relevant and generate static correlation determination rules based on this; this method can save a large amount of manual work, but still relies on the professional ability of the determination personnel. Because this is a post hoc determination, the actual situation at the time when the alarm occurred may no longer be verified, and it cannot be guaranteed whether the frequently occurring alarm combinations mined are truly relevant. Therefore, the accuracy of this method cannot be guaranteed, and there is no method to gradually improve the accuracy. Summary of the Invention

[0004] An embodiment of the present invention provides a method for analyzing the correlation of warnings of intelligent lamp posts based on deep learning, which can effectively improve the accuracy of the correlation analysis results while avoiding a large amount of one-time manual marking and input work.

[0005] In a first aspect, a method for analyzing the correlation of warnings of intelligent lamp posts based on deep learning is provided. The method includes at least one deep learning process, and the deep learning process includes the following steps:

[0006] Sample warnings for intelligent lamp posts with physical relevance, and use a batch of sampled warnings as a temporary warning sample set.

[0007] Perform forward calculation on the temporary warning sample set based on the current neural network to obtain the warning correlation analysis result.

[0008] Mark the warning correlation analysis result according to the situation where the warning is cleared to obtain a marking result.

[0009] Use the temporary warning sample set and the marking result as training samples, and start training the deep neural network when the number of training samples reaches a set value to optimize the model parameters of the current neural network.

[0010] In some embodiments, sampling warnings for intelligent lamp posts with physical relevance and using a batch of sampled warnings as a temporary warning sample set includes the following steps:

[0011] When a warning occurs on an intelligent lamp post, the intelligent lamp post where the warning occurs samples the warning at a preset sampling duration and sends a warning collection instruction to other intelligent lamp posts with physical relevance.

[0012] Other intelligent lamp posts that receive the warning collection instruction sample their own warnings within the preset sampling duration according to the warning collection instruction.

[0013] After the preset sampling duration, determine a batch of warnings from all the sampled warnings as the temporary warning sample set.

[0014] In some embodiments, performing forward calculation on the temporary warning sample set based on the current neural network to obtain the warning correlation analysis result includes the following steps:

[0015] Use the attribute values of the warnings in the temporary warning sample set as the input layer of the current neural network.

[0016] Use the correlation between the warnings in the temporary warning sample set as the output layer of the current neural network.

[0017] After performing forward calculation on the input layer based on the current neural network, the result of the output layer is used as the alarm correlation analysis result.

[0018] In some embodiments, before performing forward calculation on the temporary alarm sample set based on the current neural network, the method further includes the steps of:

[0019] Sort the alarms in the temporary alarm sample set according to the chronological order.

[0020] In some embodiments, marking the alarm correlation analysis result according to the situation of alarm clearance to obtain a marking result includes the steps of:

[0021] Automatically mark the association information between alarms according to the change situation of other alarms after an alarm is cleared;

[0022] Manually correct the alarm correlation analysis result based on the actual situation of alarm handling on the basis of the automatically marked result to form the marking result.

[0023] In some embodiments, automatically marking the association information between alarms according to the change situation of other alarms after an alarm is cleared includes the steps of:

[0024] If there are other alarms that are automatically cleared within a preset time threshold after an alarm is cleared, mark the other alarms that are automatically cleared within the preset time threshold as related alarms and use this as the result of automatic marking.

[0025] In some embodiments, the deep neural network is trained using the backpropagation algorithm.

[0026] In a second aspect, a smart street light pole alarm correlation analysis system based on deep learning is provided, which is characterized in that it includes:

[0027] An alarm sampler, which is used to sample alarms for smart street light poles with physical relevance and send the alarm sampling result;

[0028] An alarm correlation analysis device, which communicates with the alarm sampler and is used for:

[0029] Receive the alarm sampling result, and determine a batch of alarms from it as a temporary alarm sample set;

[0030] Perform forward calculation on the temporary alarm sample set based on the current neural network to obtain an alarm correlation analysis result;

[0031] A manual processing terminal, which communicates with the alarm sampler and the alarm correlation analysis device respectively, and is used to mark the alarm correlation analysis result according to the situation of alarm clearance to obtain a marking result;

[0032] The alarm correlation analysis device is further configured to use the temporary alarm sample set and the marking result as training samples, and start deep neural network training when the number of training samples reaches a set value, so as to optimize the model parameters of the current neural network.

[0033] In some embodiments, the alarm sampler is disposed on the intelligent street lamp pole, and is configured to:

[0034] When an alarm occurs in the set intelligent street lamp pole, sample the occurred alarm at a preset sampling duration, and send an alarm collection instruction to other intelligent street lamp poles with physical relevance;

[0035] Sample the alarm occurring in the set intelligent street lamp pole within the preset sampling duration according to the alarm collection instruction;

[0036] After the preset sampling duration, send the alarm sampling result.

[0037] In some embodiments, the alarm collector is further configured to automatically mark the association information between alarms according to the change situation of other alarms after the alarm is cleared;

[0038] The manual processing terminal is further configured to correct the alarm correlation analysis result based on the actual situation of alarm processing on the basis of the automatically marked result, and form the marking result accordingly.

[0039] Through the embodiments of the present invention, the process of marking, whether it is carried out manually or by an automatic algorithm, is carried out after each collection and forward calculation, and is a process of gradually participating and accumulating during the daily operation of the system, rather than a one-time marking process that requires manual operation of a large number of historical alarms at the initial stage of method implementation or system operation. Therefore, the difficulty of sample marking work in the related art is greatly reduced. At the same time, the objects of sample training include the temporary alarm sample set and the marking result, and the authenticity and accuracy of the training samples can be improved through the continuously accumulated marking results, thereby further improving the accuracy of the neural network training result. The result of the neural network training is used to optimize the neural network model parameters, so as to gradually improve the accuracy of the alarm correlation analysis result calculated by each forward calculation. This embodiment can gradually improve the system correlation analysis ability during the daily operation process, without the need for a large amount of manual marking and input work in the early stage, and at the same time reduces the dependence on the ability of technicians to analyze historical alarm data. Only routine operations in daily life are required to gradually improve the alarm correlation analysis ability. Description of the Drawings

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0041] Figure 1 Schematic flowchart of a method for analyzing the correlation of warnings of intelligent lamp posts based on deep learning provided by an embodiment of the present invention;

[0042] Figure 2 Schematic flowchart of warning sampling for intelligent lamp posts with physical relevance provided by an embodiment of the present invention;

[0043] Figure 3 Schematic structural diagram of the current neural network provided by an embodiment of the present invention;

[0044] Figure 4 Schematic flowchart of marking the analysis results of warning correlation provided by an embodiment of the present invention.

[0045] Figure 5 Schematic flowchart of the training of the deep neural network provided by an embodiment of the present invention;

[0046] Figure 6 Schematic structural diagram of a system for analyzing the correlation of warnings of intelligent lamp posts based on deep learning provided by an embodiment of the present invention;

[0047] Figure 7 Schematic structural diagram of a system for analyzing the correlation of warnings of intelligent lamp posts based on deep learning provided by an embodiment of the present invention. Detailed implementation manners

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention 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 of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0049] As Figure 1 shown, an embodiment of the present invention provides a method for analyzing the correlation of warnings of intelligent lamp posts based on deep learning. This method includes at least one deep learning process, and one deep learning process includes the steps:

[0050] S100: Perform warning sampling on intelligent lamp posts with physical relevance, and use a batch of warnings obtained from the sampling as a temporary warning sample set;

[0051] S200: Perform forward calculation on the temporary alarm sample set based on the current neural network to obtain the alarm correlation analysis result;

[0052] S300: Mark the alarm correlation analysis result according to the situation of the alarm being cleared to obtain the marking result;

[0053] S400: Use the temporary alarm sample set and the marking result as training samples, and start training the deep neural network when the number of training samples reaches the set value to optimize the model parameters of the current neural network.

[0054] It should be noted that in step S100, the physical relevance can be understood by referring to the relevant concepts of the Internet of Things. The smart lamp posts with physical relevance refer to those that can collect any alarm, positioning and other information that needs to be monitored in real time through various information sensors, radio frequency identification technologies, etc. among smart lamp posts, and realize intelligent perception, recognition and management among smart lamp posts through various possible network accesses. The alarm can be the alarm issued by the devices connected to the smart lamp post, or the alarm issued by the sampling device or sensor used to sample the smart lamp post alarm. In step S200, forward calculation refers to calculating the influence of the input layer on the hidden layer in the neural network using the forward algorithm.

[0055] It can be understood that since the neural network will adjust (optimize) the model parameters based on the results of sample training during each deep learning process, the current neural network mentioned here refers to the neural network at the current moment, which may be the initial neural network without model parameter optimization, or the neural network that has undergone model parameter optimization, depending on the number of deep learning process times experienced by this method.

[0056] In this embodiment, whether the marking process is carried out manually or by an automatic algorithm, it is carried out after each acquisition and forward calculation, and it is a process of gradually participating and accumulating during the daily operation of the system, rather than a one-time marking process that requires manual marking of a large number of historical alarms at the initial stage of method implementation or system operation. Therefore, the difficulty of sample marking work in related technologies is greatly reduced. At the same time, the objects of sample training include the temporary alarm sample set and the marking results. The authenticity and accuracy of the training samples can be improved through the continuously accumulated marking results, thereby further improving the accuracy of the neural network training results. The results of neural network training are used to optimize the neural network model parameters, thereby gradually improving the accuracy of the alarm correlation analysis results calculated by each forward calculation. This embodiment can gradually improve the system correlation analysis ability during the daily operation process, without the need for a large amount of manual marking and input work in the early stage, and at the same time reduces the dependence on the ability of technicians to analyze historical alarm data. Only routine operations in daily life are required to gradually improve the alarm correlation analysis ability.

[0057] In some embodiments, step S100 includes:

[0058] S110: When an alarm occurs in a smart light pole, the smart light pole where the alarm occurs samples the occurred alarm at a preset sampling duration and sends an alarm collection instruction to other smart light poles with physical relevance;

[0059] S120: Other smart light poles that receive the alarm collection instruction sample their own alarms within the preset sampling duration according to the alarm collection instruction;

[0060] S130: After the preset sampling duration, a batch of alarms are determined from all the sampled alarms as the temporary alarm sample set.

[0061] It should be noted that the sampling instruction may include information such as the preset sampling duration, sampling start time, and sampling ID. It can be understood that alarms with the same sampling ID represent that the collection of this alarm belongs to the correlation alarm collection behavior for the same alarm. Thus, by limiting the sampling start time, sampling duration, and sampling ID, corresponding identification and processing can be carried out after receiving and processing the alarm collection instruction, only alarms with the same sampling ID are collected, and unnecessary alarms are discarded, realizing the correlation alarm collection process for one alarm.

[0062] This embodiment realizes the sampling of alarms for smart lamp posts with physical relevance. Compared with the related art, in this embodiment, the alarm sampling process is carried out simultaneously when an alarm occurs, and the sampling work can be carried out by a device (such as an alarm sampler) responsible for sampling and communication on the smart lamp post to convey instructions and sample between lamp posts with physical relevance, avoiding the system pressure brought by the upper management system sampling alarms on all lamp posts.

[0063] As Figure 2 shown, in a specific embodiment, the alarm sampler is installed on the smart lamp post, which samples and collects the alarms of the devices connected to the smart lamp post and the alarms occurring on the alarm sampler itself, and performs alarm sampling on the smart lamp posts with physical relevance through the following process.

[0064] S1: The alarm sampler waits for the smart lamp post to generate an alarm;

[0065] S2: Determine whether an alarm has occurred. If an alarm has occurred, first determine whether alarm sampling is currently in progress. If alarm sampling is not in progress, then perform S4; otherwise, perform S6. If no alarm has occurred, determine whether an alarm sampling instruction has been received (from other alarm samplers). If so, perform S3; otherwise, continue to wait;

[0066] S3: Perform instruction conflict detection to ensure that only the earliest issued sampling instruction among the alarm sampling instructions received from physically related lamp posts is valid, and other later issued instructions are discarded. After the instruction conflict detection passes, execute S4;

[0067] S4: The alarm sampler on the smart lamp post where the alarm occurs sends an alarm sampling instruction to other smart lamp posts with physical relevance through local networking. Among them, the instruction includes: sampling duration T, sampling start time T s and sampling ID;

[0068] S5: Mark that the current system is performing alarm sampling, and at the same time start a timer with a timing time of T seconds (this time T can be configured), where T is the sampling duration and the sampling start time is T s ;

[0069] S6: Collect alarm information that occurs within T seconds starting from time T within the sampling duration T, record the sampling ID into these occurring alarm information and perform alarm caching; s S7: If an alarm whose occurrence time is earlier than T

[0070] is collected, discard this alarm (alarm during non-sampling period); s S8: If the sampling duration T arrives, report the alarms collected in the sampling alarm cache.

[0071] S8: If the sampling duration T reaches, report the alarms collected in the sampling alarm cache.

[0072] S9: Cancel the current system sampling flag and continue with S1 to wait for new alarms to occur.

[0073] In some embodiments, step S200 includes:

[0074] S210: Use the attribute values of the alarms in the temporary alarm sample set as the input layer of the current neural network;

[0075] S220: Use the correlation between alarms in the temporary alarm sample set as the output layer of the current neural network;

[0076] S230: After performing forward calculation on the input layer based on the current neural network, use the result of the output layer as the alarm correlation analysis result.

[0077] It should be noted that the current neural network can be, for example, Figure 3 the deep neural network 300 shown in the figure, which is composed of an input layer 302, a hidden layer 305, and an output layer 303; among them, the input layer 302 is defined as an input matrix I(301) composed of the alarms in the temporary alarm sample set A N and the attribute values in the alarm information. Among them, x NM is the value of the Mth alarm attribute of the Nth alarm.

[0078] Preferably, alarms 1 to N can be sorted in the order of the alarm occurrence time; and if the number of alarms in a single alarm sampling data is less than N, the remaining part of the matrix is filled with 0s;

[0079] The vector composed of all row vectors in the input matrix I(301) concatenated together, which is a vector of M×N elements, is the input vector of the neural network input layer 302, and is specifically defined as:

[0080] i = [x 11 x 12 ... x 21 x 22 ... x NM , where the elements in the vector i will be used as the neurons of the neural network input layer 302, and the number of neurons in the input layer 302 is M×N;

[0081] The output of the deep neural network 300 is defined as a matrix Y(304) of the correlation between alarms in the temporary alarm sample set A N and is specifically defined as:

[0082]

[0083] Among them, the row vectors in the output result matrix Y(304) represent the correlation between the alarms corresponding to each row and other alarms. For example, the row vector value of the i-th row is as follows:

[0084] y i =[01 12 0.73...1 i ... 0.8 N , which means that alarm i is correlated with alarms 2, 3, and N. The magnitude of the element values in the vector represents the magnitude of the correlation probability.

[0085] When the alarms are sorted in chronological order of occurrence, and the root alarm among the correlated alarms occurs before the correlated alarms, the position corresponding to the element that is the first and has an element value greater than the set probability value in the i-th row vector y is the root alarm. For example, in the above example, the element value at the first position 2 is 1, so it is considered that alarm 2 is the root alarm of alarms 3, i, and N; the vector y = [y 11 y 12 ... y 21 y 22 ... y NN composed of all row vectors from top to bottom in the output result matrix Y(304) is the output vector of the neural network output layer 303, and the elements in this vector are the neurons of the output layer 303.

[0086] In a specific embodiment, after each smart light pole has completed sampling, it uploads its own sampled alarm data. The platform saves the sampled data with the same ID together (considered as the sampled data of the correlated alarms of one alarm) according to the sampling ID in the received sampled data. The alarm correlation analysis device can be set to receive the alarm sampling result. When the sampled data with alarm ID information is received for the first time, an overtime timer is started, and the sampled data with the same alarm ID received within this overtime period is recorded as a complete sampled data. After confirming a complete alarm sampled data, the alarm data in it is sorted in chronological order of alarm occurrence; if the number of alarms in the sampled data is greater than or equal to the set sampling alarm number N, the first N alarm data are taken out from the alarm sampled data to generate a temporary alarm sample set. Define the temporary alarm sample set as a set composed of N alarms sampled, denoted here as:

[0087] A N =[a1 a2... a i ... a N , where each element a N in A i represents an alarm. After storing the temporary alarm sample set in the temporary alarm sample library, the temporary alarm sample set A NConvert it into the input matrix I(301) and input it into the deep neural network 300 (such as Figure 3 the neural network shown) for forward calculation to obtain the output result matrix Y(304); finally, obtain A based on the result matrix Y(304). N The correlation between alarms in A is used as the result of the alarm correlation analysis and pushed to the marking processing device (which can be the mobile terminal device of the alarm processing personnel or other fixed terminal devices).

[0088] In some embodiments, step S300 includes:

[0089] S310: Automatically mark the association information between alarms according to the changes of other alarms after an alarm is cleared;

[0090] S320: Manually correct the alarm correlation analysis result based on the automatically marked result in combination with the actual situation of alarm processing to form the marked result.

[0091] It should be noted that the clearing of an alarm is generally carried out by the alarm processing personnel according to the alarm processing work order to the smart light pole where the alarm occurs for on-site inspection or fault handling, clearing the device alarm or fault.

[0092] In this embodiment, the correlation between alarms can be automatically marked through an automatic processing program according to the changes of other alarms (whether they are also automatically cleared) after an alarm is cleared, so as to further reduce the workload of subsequent manual correction.

[0093] In some embodiments, the process of automatic marking is as follows: after an alarm is cleared, if there are other alarms that are automatically cleared within a preset time threshold, mark the other alarms that are automatically cleared within the preset time threshold as related alarms and use this as the result of automatic marking.

[0094] Such as Figure 4 shown, in a specific embodiment, the specific process of marking the alarm correlation analysis result is as follows:

[0095] S10: Process and clear a certain alarm a, specifically: the alarm processing personnel go to the smart light pole site according to the received alarm processing work order for alarm inspection or fault handling, clearing the device alarm or fault (in some preferred embodiments, the alarm processing order in the work order has been sorted according to the time of alarm occurrence).

[0096] S20: Monitor that after alarm a is cleared and there are other alarms that are automatically cleared within time t, record these cleared alarms as related alarms of alarm a, and after exceeding time t, feedback the marked related alarm result to the mobile terminal of the alarm processing personnel;

[0097] S30: Compare with other alarms in the temporary alarm sample set where the cleared alarm a is located, mark other alarms associated with this alarm, and set the corresponding association degree (value range 0 to 1) accordingly. Specifically: According to the on-site situation of alarm handling, the alarm handler compares with the list of alarms in the temporary alarm sample set A listed in the mobile terminal N in the alarm list, and according to the actual cause of the currently cleared alarm, checks whether there are alarms in the list that are associated with the currently cleared alarm. If so, mark that the alarm in the list is relevant to the currently cleared alarm, and set a correlation degree (value range 0 to 1) at the same time; At the same time, the alarm handler checks whether the relevant alarms in the automatically marked results are accurate, and corrects them if there are inaccurate situations

[0098] S40: Upload the final marked result to the platform, and the platform stores the marked result corresponding to the temporary alarm sample set in the temporary alarm sample library. Specifically: After the marked result is uploaded to the platform, the platform converts the marked result into a vector d i , which is defined as the marked vector here, where i represents the i-th alarm in the temporary alarm sample set A N in this vector d i is defined in the same way as the row vector in the output result matrix Y (see the input and output matrices of the neural network in the appendix Figure 3 ); After the marked vector d i is generated, store it in the record corresponding to the temporary alarm sample set A in the temporary alarm sample library N .

[0099] For example Figure 5 as shown, in a specific embodiment, the backpropagation algorithm is used to train the deep neural network. Step S400 includes:

[0100] S410: The platform monitors that all the alarms in a certain temporary alarm sample set have been cleared and marked

[0101] S420: Generate the input matrix I and the expected result matrix Y according to the temporary alarm sample set and the marked data. Specifically: The input matrix I is composed of the alarms in the temporary alarm sample set A N and the attribute values in the alarm information; The expected result matrix Y is composed of the marked vectors d corresponding to each alarm in the taken-out temporary alarm sample set i , and the specific definition is:

[0102]

[0103] S430: Store the input matrix I and the expected result matrix Y together as a new tagged alarm correlation sample data into the global alarm sample library, and increment the number of new sample storage events k by one; if k ≥ K (K is a set threshold) and the current system is not being trained, execute S440; if k ≥ K and the system is being trained, execute S480; if k < K, continue to wait for a new set of temporary alarm samples to be cleared and tagged;

[0104] S440: Take out all samples from the global alarm sample library and use the backpropagation algorithm to train the deep neural network;

[0105] S450: Mark that the current system is in the training process;

[0106] S460: After training is completed, adjust the parameters in the deep neural network algorithm in reverse according to the training results;

[0107] S470: Clear the mark that the current system is in training;

[0108] S480: Queue the training plan. Specifically: when the training trigger condition is reached, if it is found that the previous training of the system is still in progress, then the new training plan this time is put into the queue for queuing, and wait for the current training to be completed before starting the new training.

[0109] As Figure 6 shown, an embodiment of the present invention also provides an intelligent street lamp alarm correlation analysis system based on deep learning, which includes:

[0110] An alarm sampler 100, which is used to sample alarms for intelligent street lamps 400 with physical relevance and send the alarm sampling results;

[0111] An alarm correlation analysis device 200, which communicates with the alarm sampler 100 and is used for:

[0112] Receive the alarm sampling results and determine a batch of alarms as a temporary alarm sample set from them;

[0113] Perform forward calculation on the temporary alarm sample set based on the current neural network to obtain the alarm correlation analysis result;

[0114] An artificial processing terminal 300, which communicates with the alarm sampler 100 and the alarm correlation analysis device 200 respectively, and is used to mark the alarm correlation analysis result according to the situation of alarm clearance to obtain the marked result;

[0115] The alarm correlation analysis device 200 is further configured to use the temporary alarm sample set and the marking result as training samples, and start deep neural network training when the number of training samples reaches a set value, so as to optimize the model parameters of the current neural network.

[0116] Through the system provided by this embodiment, the system correlation analysis ability can be gradually improved during the daily operation process. It neither requires a large amount of manual marking and input work in the early stage, nor relies on the ability of technicians to analyze historical alarm data. Only routine operations in daily life are needed to gradually improve the alarm correlation analysis ability.

[0117] In some embodiments, the alarm sampler 100 is disposed on the smart light pole 400, and is further configured to:

[0118] When an alarm occurs in the set smart light pole 400, sample the occurred alarm at a preset sampling duration, and send an alarm collection instruction to other smart light poles 400 with physical relevance;

[0119] Sample the alarm occurring in the set smart light pole 400 within the preset sampling duration according to the alarm collection instruction;

[0120] After the preset sampling duration, send the alarm sampling result.

[0121] In some embodiments, the alarm collector 100 is further configured to automatically mark the association information between alarms according to the change situation of other alarms after the alarm is cleared;

[0122] The manual processing terminal 300 is further configured to correct the alarm correlation analysis result based on the automatically marked result in combination with the actual situation of alarm processing, and form the marking result accordingly.

[0123] As Figure 7 shown, in a specific embodiment, the alarm sampler 100 includes: an alarm sampling module 102, an alarm diagnosis module 101, a system communication module 103, and a local communication module 104; the alarm correlation analysis device 200 (which can be a hardware system independently performing the alarm correlation analysis function or a sub-module of the smart light pole management platform), includes: a temporary alarm sample processing module 106, an alarm correlation analysis module 107, and a global alarm sample training module 108; the manual processing terminal 300 includes a mobile terminal device 310.

[0124] Among them, the alarm sampling module 102 samples and collects the alarms of the devices connected to the smart street lamp and the alarms of the alarm sampler itself; the alarm diagnosis module 101 collects the change situations of other alarms when a certain alarm is cleared, finds out the possible related alarms of the cleared alarm, and generates alarm correlation marking data (automatically marks according to the situation of the cleared alarm and generates an automatic marking result); the system communication module 103 sends the sampling result to the alarm correlation analysis device 200; the local communication module 104 communicates with other physically related smart street lamps through local networking, and sends or receives sampling instructions. Here, the local networking method can be power line carrier communication networking, wired local area networking, zigbee wireless networking, Lora wireless networking, or WiFi wireless networking.

[0125] Among them, the temporary alarm sample processing module 106 receives the alarm sampling data (sampling result) reported by the alarm sampler 100 on the smart street lamp 400 and generates temporary alarm sample set data. At the same time, it receives the alarm marking data reported by the mobile terminal device 310 of the alarm handler 109 and stores the marking result in the corresponding temporary alarm sample set; the alarm correlation analysis module 107 receives the output of the temporary alarm sample processing module 106, performs forward calculation on it using a deep neural network model, and pushes the result to the mobile terminal device 310 of the alarm handler 109; at the same time, the alarm correlation analysis module 107 can also receive the training result of the global alarm sample training module 108 to adjust the model parameters of the deep neural network; when the number of newly added alarm samples reaches the set value K, the global alarm sample training module 108 performs deep learning training on the samples in the global alarm sample library, and imports the training result into the deep neural network in the alarm correlation analysis module 107.

[0126] Such as Figure 7 As shown, the functions of the mobile terminal device 310 of the alarm handler 109 specifically include: 1) Displaying alarm correlation, specifically: receiving the alarm correlation analysis result sent by the alarm correlation analysis device 200, and displaying the association relationship between the alarms to be processed by the current alarm handler 109 and other alarms in a tree structure; 2) Alarm clearance marking, specifically: when the alarm handler 109 clears a certain alarm, mark the related alarms in the temporary alarm sample set where the cleared alarm is located and set the degree of association, and at the same time correct the alarm data automatically marked by the alarm diagnosis module. Finally, upload the marking result to the alarm correlation analysis device 200. Here, the function implementation method in the mobile terminal device 310 can be an independent mobile application or a sub-function module embedded in the existing smart street lamp management application.

[0127] It can be understood that the intelligent street lamp alarm correlation analysis system based on deep learning provided by the embodiments of the present invention can be used to implement the foregoing intelligent street lamp alarm correlation analysis method based on deep learning.

[0128] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by terms such as "upper" and "lower" is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation to the present invention. Unless otherwise clearly specified and defined, the terms "installation", "connection" and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0129] It should be noted that in the present invention, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0130] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for analyzing the correlation of warnings of intelligent lamp poles based on deep learning, characterized in that The method includes at least one deep learning process, and the deep learning process includes the steps of: Performing alarm sampling on intelligent lamp posts with physical relevance, and taking a batch of alarms obtained from the sampling as a temporary alarm sample set; Performing forward calculation on the temporary alarm sample set based on the current neural network to obtain an alarm correlation analysis result; Marking the alarm correlation analysis result according to the situation where the alarm is cleared to obtain a marking result; Using the temporary alarm sample set and the marking result as training samples, and starting the training of the deep neural network when the number of training samples reaches a set value to optimize the model parameters of the current neural network; Marking the alarm correlation analysis result according to the situation where the alarm is cleared to obtain a marking result, including the steps of: Automatically marking the association information between alarms according to the change situation of other alarms after the alarm is cleared; Manually correcting the alarm correlation analysis result based on the automatically marked result in combination with the actual situation of alarm handling to form the marking result; Automatically marking the association information between alarms according to the change situation of other alarms after the alarm is cleared, including the steps of: If there are other alarms that are automatically cleared within a preset time threshold after one alarm is cleared, marking the other alarms that are automatically cleared within the preset time threshold as relevant alarms and taking this as the automatically marked result; Performing alarm sampling on intelligent lamp posts with physical relevance, and taking a batch of alarms obtained from the sampling as a temporary alarm sample set, including the steps of: When an alarm occurs on an intelligent lamp post, the intelligent lamp post where the alarm occurs samples the occurring alarm with a preset sampling duration and sends an alarm collection instruction to other intelligent lamp posts with physical relevance; Other intelligent lamp posts that receive the alarm collection instruction sample their own alarms within the preset sampling duration according to the alarm collection instruction; After the preset sampling duration, determining a batch of alarms from all the sampled alarms as a temporary alarm sample set; Wherein, the alarm collection instruction includes a preset sampling duration, a sampling start time, and sampling ID information, and only alarms with the same sampling ID are collected; Wherein, the alarm sampling process is carried out simultaneously when the alarm occurs, and the sampling work is carried out by the alarm sampler on the intelligent lamp post for instruction transmission and sampling between physically related lamp posts.

2. The method for analyzing the correlation of intelligent lamp post alarms based on deep learning according to claim 1, characterized in that Performing forward calculation on the temporary alarm sample set based on the current neural network to obtain an alarm correlation analysis result, including the steps of: Taking the attribute values of the alarms in the temporary alarm sample set as the input layer of the current neural network; Taking the correlation between the alarms in the temporary alarm sample set as the output layer of the current neural network; Taking the result of the output layer as the alarm correlation analysis result after performing forward calculation on the input layer based on the current neural network.

3. The method for analyzing the warning correlation of an intelligent street lamp based on deep learning according to claim 1, characterized in that, Before performing forward calculation on the temporary alarm sample set based on the current neural network, it further includes the step of: sorting the alarms in the temporary alarm sample set according to the chronological order.

4. The method for analyzing the correlation of warnings of an intelligent light pole based on deep learning according to claim 1, characterized in that the backpropagation algorithm is used for training the deep neural network.

5. A wisdom lamp post alarm correlation analysis system based on deep learning, characterized in that, It includes: A warning sampler, which is used to sample warnings of intelligent light poles with physical relevance and send the warning sampling results; A warning correlation analysis device, which communicates with the warning sampler and is used for: Receiving the warning sampling results and determining a batch of warnings therefrom as a temporary warning sample set; Performing forward calculation on the temporary warning sample set based on the current neural network to obtain a warning correlation analysis result; An artificial processing terminal, which communicates with the warning sampler and the warning correlation analysis device respectively, and is used for marking the warning correlation analysis result according to the situation where the warning is cleared to obtain a marking result; The warning correlation analysis device is also used to use the temporary warning sample set and the marking result as training samples, and start training the deep neural network when the number of training samples reaches a set value to optimize the model parameters of the current neural network; The warning sampler is also used to automatically mark the association information between warnings according to the change situation of other warnings after the warning is cleared; The artificial processing terminal is also used to correct the warning correlation analysis result based on the actual situation of warning processing on the basis of the automatically marked result and form the marking result therefrom; The warning sampler is also used to automatically mark the association information between warnings according to the change situation of other warnings after the warning is cleared, specifically: If there are other warnings that are automatically cleared within a preset time threshold after one warning is cleared, mark the other warnings that are automatically cleared within the preset time threshold as relevant warnings and use this as the automatically marked result; The warning sampler is arranged on the intelligent light pole and is used for: When a warning occurs on the set intelligent light pole, sampling the occurring warning with a preset sampling duration and sending a warning collection instruction to other intelligent light poles with physical relevance; Other intelligent light poles that receive the warning collection instruction sample the warning occurring on the set intelligent light pole within the preset sampling duration according to the warning collection instruction; After the preset sampling duration, send the warning sampling result; Among them, the warning collection instruction includes a preset sampling duration, a sampling start time, and sampling ID information, and only warnings with the same sampling ID are collected; Among them, the warning sampling process is carried out simultaneously when the warning occurs, and the sampling work is carried out by the warning sampler on the intelligent light pole for instruction transmission and sampling between physically related light poles.

Images