A fault diagnosis method for environmental monitoring equipment based on artificial intelligence

Abstract

A method for diagnosing environmental monitoring equipment faults based on artificial intelligence belongs to the technical field of atmospheric environmental monitoring equipment. The method comprises the following steps: fault fingerprint extraction, decision tree construction, decision tree self-learning and decision tree model calculation, and solution giving step. The advantage of the present invention is that it can be applied to environmental quality online monitoring products: it can be used for dust online monitors, atmospheric micro-monitoring stations, noise online monitors, greenhouse gas online monitors, etc., and applied to environmental quality online equipment. It has a wide range of applications, a high degree of artificial intelligence, and a high fault diagnosis accuracy. It can automatically determine the cause of on-site faults, automatically provide solutions, and guide operation and maintenance personnel to quickly troubleshoot. The solutions provided are highly targeted, and the workload and cost of on-site operation and maintenance are greatly reduced. The method can perform remote fault diagnosis and intelligent positioning, and guide on-site personnel to go to the site with solutions, directly and quickly solve faults, and quickly restore the normal operation of the equipment.

Description

Technical Field

[0001] The present invention relates to an artificial intelligence-based fault diagnosis method for environmental monitoring equipment, and belongs to the technical field of environmental quality online monitoring equipment. Background Art

[0002] Explanation of terms:

[0003] Environmental quality online monitoring instrument: A general term for instruments used to monitor various parameters of indoor and outdoor environmental quality. By measuring the representative values ​​of factors affecting environmental quality, the environmental quality (or pollution level) and its changing trends and the specific values ​​of the monitoring factors are determined.

[0004] The decision tree algorithm is an artificial intelligence technique that approximates the value of discrete functions. It is a typical classification method that first processes data, uses an inductive algorithm to generate readable rules and decision trees, and then uses these decisions to analyze new data. Essentially, a decision tree classifies data using a series of rules.

[0005] Currently, online environmental quality monitoring instruments are deployed in large numbers but lack effective maintenance, leading to data distortion, data anomalies, low device uptime, and potential equipment failures. Remote fault diagnosis is impossible, requiring on-site personnel or equipment return to the factory for location analysis. Inaccurate on-site location analysis and repeated on-site repairs waste significant manpower and resources, yet the results remain far from ideal. AI-based remote fault diagnosis and intelligent location analysis digitize the years of experience of R&D engineers through AI methods, allowing operations engineers to bring solutions to the site, quickly resolve faults, and restore equipment to normal operation. Summary of the Invention

[0006] In order to overcome the deficiencies of the prior art, the present invention provides an artificial intelligence-based method for diagnosing faults in online environmental quality monitoring equipment.

[0007] An artificial intelligence-based fault diagnosis method for online environmental quality monitoring equipment combines the fault phenomena of on-site equipment with artificial intelligence to automatically diagnose and provide fault solutions remotely. It uses a fault fingerprint extraction algorithm and a decision tree artificial intelligence method to formulate fault solutions, model and self-learn the faults of online environmental quality monitoring instruments, and form an intelligent decision-making model.

[0008] An artificial intelligence-based fault diagnosis method for environmental monitoring equipment comprises the following steps: a decision tree construction step, a decision tree learning step, and a decision tree model establishment step.

[0009] The steps of fault fingerprint extraction include:

[0010] The first step is to obtain fault information, which is obtained through on-site equipment;

[0011] The second step is to extract fault fingerprints and compare them with the fingerprint library for classification processing;

[0012] In the third step, the fault fingerprint library is updated and the best matching fault fingerprint is extracted and input into the decision tree.

[0013] The decision tree construction steps include:

[0014] The first step is the generation of decision tree: the process of generating a decision tree from a training sample set.

[0015] The second step is pruning the decision tree: Decision tree pruning is the process of testing, correcting and modifying the decision tree generated in the previous stage. The preliminary rules generated during the decision tree generation process are verified with data from a new sample data set (called a test data set), and branches that affect the accuracy of the prediction are pruned.

[0016] The decision tree learning steps include:

[0017] Build a decision tree model based on the given training data set, correctly classify the instances, summarize a set of classification rules from the training data set, and select a decision tree that has the least contradiction with the training data.

[0018] Decision tree learning usually includes three steps: feature selection step, decision tree generation step and decision tree pruning step.

[0019] Decision tree feature selection step: If the number of features is large, the features are selected at the beginning of decision tree learning, leaving only the features that have sufficient classification ability for the training data.

[0020] The steps of generating a decision tree: corresponding to the local selection of the model, the pruning of the decision tree corresponds to the global selection of the model. The generation of the decision tree only considers the local optimum, while the pruning of the decision tree considers the global optimum.

[0021] The pruning steps of the decision tree are: by removing the leaf nodes that are too subdivided, making them fall back to the parent node or even a higher node, and then changing the parent node or higher node to a new leaf node.

[0022] Steps to build a decision tree model:

[0023] Among thousands of sets of field equipment, common fault characteristics were selected: number of equipment offline times, data exceedance rate, data anomaly rate, data quality control exceedance rate, data mean square deviation exceedance rate, horizontal exceedance ratio, vertical exceedance ratio, and analysis accuracy rate.

[0024] The fault information is refined and the decision tree algorithm C4.5 is used to extract the features of the fault fingerprint. A fault fingerprint library is established and input into the decision tree model.

[0025] Fault information X is a random variable. Based on the decision tree algorithm C4.5, the information entropy of the random variable X is calculated:

[0026]

[0027] Where n represents the value of X, pi represents the probability of taking the value i,

[0028] Conditional entropy is calculated using the following formula, which represents the uncertainty of random variable X under the condition of random variable Y.

[0029]

[0030] G(X, Y) = H(X, Y) - H(X|Y), information gain,

[0031] Assume that the sample set on node t is D = (x, y), where x = (x1, x2, ..., xN) represents the feature variable.

[0032] y=(y1,y2,………,yN) represents the response variable,

[0033] For any split point m = (a, fm), a and fm correspond to the characteristic variable and the split point critical value respectively, m divides the sample set into the left and right sides, namely Dleft(m) and Dright(m).

[0034] Dleft(m)=(x,y)|xa≤fm;

[0035] Dright(m)=D\Dleft(m).

[0036] Considering that the sample sets of child nodes are impure in practice, the weight of impurity is added in the classification.

[0037] In the case of probability p, the impurity I expression is:

[0038]

[0039] Where K represents the category, pk represents the probability of the lower k categories,

[0040] I(p) represents the impurity of probability p.

[0041] Assume that the split point impurity function I is as follows:

[0042]

[0043] Where |D| represents the number of samples at node t, and the optimal split point that minimizes the impurity function is calculated

[0044] m * =argmin m I(D, m)

[0045] Use the same method to iteratively split Dleft(m * ) and Dright(m * ), guiding the decision tree depth to reach the upper limit or the number of child node samples is less than the number of samples specified in advance.

[0046] The advantage of the present invention is that it can be applied to online environmental quality monitoring products: it can be used for online dust monitors, atmospheric micro-monitoring stations, online noise monitors, online greenhouse gas monitors, etc., and is applied to online environmental quality equipment. It has a wide range of applications, a high degree of artificial intelligence, and a high fault diagnosis accuracy. It can automatically determine the cause of on-site faults, automatically provide solutions, and guide operation and maintenance personnel to quickly troubleshoot; the solutions provided are highly targeted, and the workload and cost of on-site operation and maintenance are greatly reduced; the method can remotely diagnose faults and intelligently locate them, and guide on-site personnel to go to the site with solutions, directly and quickly solve faults, and quickly restore normal operation of the equipment.

[0047] By adopting this method, the fault diagnosis accuracy rate has increased from the original 72.65% to 97.85%. On-site fault diagnosis is accurate, on-site operation and maintenance are more targeted, and the timeliness of fault resolution is significantly improved. As a result, the average online rate of equipment has increased from the original 84.7% to 94.55%, the data missing situation has been significantly improved, the data integrity has been significantly improved, and the post-accident analysis has been significantly improved.

[0048] By adopting this method, the workload of on-site operation and maintenance is reduced by 70%, which greatly reduces the operation and maintenance costs and has great economic value. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] When considered in conjunction with the accompanying drawings, the present invention can be more completely and better understood and its many attendant advantages can be easily known by referring to the following detailed description. However, the drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention, as shown in the figure:

[0050] Figure 1 It is a flow chart of the program of the present invention.

[0051] Figure 2 This is a comparison curve of the present invention.

[0052] Figure 3 This is the second comparison curve of the present invention.

[0053] Figure 4 These are three figures showing the comparison curves of the present invention. DETAILED DESCRIPTION

[0054] The present invention will be further described below with reference to the accompanying drawings and examples.

[0055] Obviously, many modifications and variations made by those skilled in the art based on the purpose of the present invention fall within the protection scope of the present invention.

[0056] The terms "first" and "second" are used for descriptive purposes only and should not be construed to indicate or imply relative importance or implicitly specify the quantity of the technical features being referred to. Thus, a feature specified as "first" or "second" may explicitly or implicitly include one or more of such features. In the description, "plurality" means two or more, unless otherwise specifically defined.

[0057] Unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," and "fixed" should be interpreted broadly. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediary; and internal communication between two components. Those skilled in the art will understand the specific meanings of these terms in the present invention based on specific circumstances.

[0058] Those skilled in the art will understand that unless otherwise defined, all terms used herein, including technical and scientific terms, have the same meaning as commonly understood by those skilled in the art.

[0059] To facilitate understanding of the embodiments, further explanations will be given below, and each embodiment does not constitute a limitation of the embodiments.

[0060] Example 1: The decision tree algorithm is an artificial intelligence technique that approximates the value of discrete functions. This is a typical classification method that first processes data, uses an inductive algorithm to generate readable rules and decision trees, and then uses the decision trees to analyze new data. Essentially, a decision tree is a process of classifying data using a series of rules.

[0061] Decision tree algorithms first emerged in the 1960s and continued to be used until the late 1970s. Decision tree algorithms construct decision trees to discover classification rules implicit in data. The core of decision tree algorithms is how to construct highly accurate and small-scale decision trees. Typical decision tree algorithms include ID3, C4.5, and CART.

[0062] like Figure 1 、 Figure 2、 Figure 3 and Figure 4 As shown, an artificial intelligence-based environmental monitoring equipment fault diagnosis method includes a decision tree construction step, a decision tree learning step, and a decision tree model establishment step.

[0063] 1. The decision tree construction steps include:

[0064] The first step is the generation of a decision tree: the process of generating a decision tree from a training sample set. Generally speaking, the training sample data set is a data set that has a history and a certain degree of comprehensiveness based on actual needs and is used for data analysis and processing.

[0065] The second step is pruning the decision tree: Decision tree pruning is the process of testing, correcting and modifying the decision tree generated in the previous stage. It mainly uses the data in the new sample data set (called the test data set) to verify the preliminary rules generated during the decision tree generation process, and prunes those branches that affect the accuracy of the prediction.

[0066] Decision tree learning usually includes three steps: feature selection step, decision tree generation step and decision tree pruning step.

[0067] 2. Decision tree learning steps:

[0068] Objective: Based on a given training dataset, construct a decision tree model that correctly classifies instances. Decision tree learning essentially involves inducing a set of classification rules from the training dataset. There may or may not be multiple decision trees that correctly classify the training data. When selecting a decision tree, one should minimize conflicts with the training data and exhibit good generalization capabilities. Furthermore, the selected conditional probability model should not only provide a good fit for the training data but also provide good predictions for unseen data.

[0069] The decision tree learning algorithm typically recursively selects the optimal features and segments the training data based on these features, resulting in the best classification for each subset of the dataset. This includes feature selection, decision tree generation, and decision tree pruning.

[0070] If the number of features is large, features are selected at the beginning of decision tree learning, leaving only features that have sufficient classification ability for the training data.

[0071] The steps of building a decision tree correspond to local selection of the model, while pruning the decision tree corresponds to global selection of the model. Building a decision tree only considers the local optimum, while pruning the decision tree considers the global optimum.

[0072] Decision tree pruning steps: Decision tree pruning is to make the tree simpler so that it has better generalization ability. It does this by removing overly detailed leaf nodes, making them fall back to the parent node or even a higher node, and then changing the parent node or higher node to a new leaf node.

[0073] 3. Steps to build a decision tree model:

[0074] Among thousands of sets of field equipment, common fault characteristics were selected: number of equipment offline times, data exceedance rate, data anomaly rate, data quality control exceedance rate, data mean square deviation exceedance rate, horizontal exceedance ratio, vertical exceedance ratio, and analysis accuracy rate.

[0075] The fault information is refined and the decision tree algorithm C4.5 is used to extract the characteristic fingerprint of the fault. A fault fingerprint library is established and input into the decision tree model.

[0076] Fault information X is a random variable. Based on the decision tree algorithm C4.5, the information entropy of the random variable X is calculated:

[0077]

[0078] Where n represents the value of X, and pi represents the probability of taking the value i.

[0079] Conditional entropy is calculated using the following formula, which represents the uncertainty of random variable X under the condition of random variable Y.

[0080]

[0081] G(X, Y) = H(X, Y) - H(X|Y), information gain

[0082] Assume that the sample set on node t is D = (x, y), where x = (x1, x2, ..., xN) represents the feature variable,

[0083] y=(y1,y2,………,yN) represents the response variable.

[0084] For any splitting point m = (a, fm), a and fm correspond to the characteristic variable and the splitting point critical value respectively, and m divides the sample set into the left and right sides, namely Dleft(m) and Dright(m).

[0085] Dleft(m)=(x,y)|xa≤fm;

[0086] Dright(m)=D\Dleft(m).

[0087] Considering that the sample sets of child nodes are impure in practice, the weight of impurity is added in the classification.

[0088] In the case of probability p, the impurity I expression is:

[0089]

[0090] Where K represents the category and pk represents the probability of the lower k categories.

[0091] I(p) represents the impurity of probability p;

[0092] Assume that the split point impurity function I is as follows:

[0093]

[0094] Where |D| represents the number of samples at node t, and the optimal split point that minimizes the impurity function is calculated

[0095] m * =argmin m I(D, m)

[0096] Use the same method to iteratively split Dleft(m * ) and Dright(m * ), guiding the decision tree depth to reach the upper limit or the number of child node samples is less than the number of samples specified in advance.

[0097] Example 2: Figure 1 、 Figure 2 、 Figure 3 and Figure 4 As shown, an artificial intelligence-based fault diagnosis method for environmental monitoring equipment using Example 1 includes the following steps: combining the fault phenomena of on-site equipment, using artificial intelligence to automatically diagnose and provide fault solutions remotely, and using a decision tree artificial intelligence method to model and self-learn the faults of environmental monitoring instruments to form an intelligent decision-making model.

[0098] Fault classification and solutions include:

[0099] 1.1 Model parameter class

[0100] Zero drift: Adjust the sensor zero point.

[0101] Temperature Drift: Correct drift model parameters.

[0102] Time Drift: Correct drift model parameters.

[0103] Sensitivity Drift: Adjust the sensitivity coefficient.

[0104] 1.2 Sensor failure

[0105] Electrolyte exhausted: Replace sensor.

[0106] The UV lamp is contaminated: return the UV lamp to the factory for cleaning.

[0107] The UV lamp has reached the end of its life: Replace the UV lamp.

[0108] Grid failure: Replace the grid.

[0109] 1.3 Communication failure

[0110] Tariff card is in arrears: recharge.

[0111] Tariff card failure: The tariff card is magnetized, please replace the card.

[0112] DTU hardware failure: Replace the DTU.

[0113] DTU offline bug: Replace the DTU.

[0114] 1.4 Device hardware failure

[0115] Power supply failure: Replace the AC-DC power converter.

[0116] Solar panel failure: Replace the charge and discharge manager.

[0117] Lithium battery failure: Replace the aged lithium battery.

[0118] Core control board failure: Replace the main board.

[0119] Sensor base plate failure: Replace the sensor base plate.

[0120] LED display card failure: Replace the display card.

[0121] 1.5 Device software failure

[0122] DTU firmware failure: Upgrade the DTU firmware.

[0123] The main core control board firmware is faulty: Upgrade the firmware.

[0124] Display unit firmware failure: Upgrade the firmware.

[0125] Actual operation effect:

[0126] like Figure 2 The column comparison curve shown shows that the operation and maintenance workload is reduced: the operation and maintenance workload is compared for 8 consecutive months, and the average monthly operation and maintenance workload is: 5.875 days. After adopting the present invention, it is 1.75 days, which is a 70% reduction in operation and maintenance workload.

[0127] like Figure 3 The column comparison curves shown demonstrate improved fault diagnosis accuracy:

[0128] average value 72.65 97.85

[0129] The average fault accuracy is 72.65%. After adopting the present invention, it is 97.85%, which is an increase of 25.2% in accuracy. The data online rate is also improved. The same device is continuously running for one month for online statistics.

[0130] like Figure 4 The column comparison curve shown shows that the online rate of the equipment without the intelligent operation and maintenance technology is 84.7%. After adopting the intelligent operation and maintenance technology of the present invention, the online rate of the same equipment is 94.55, which is an increase of nearly 10%, and the effect is obvious.

[0131] Example 3: Figure 1 、 Figure 2 、 Figure 3 and Figure 4 As shown, the artificial intelligence-based environmental monitoring equipment fault diagnosis method described in the above embodiment further includes the following steps:

[0132] Step 1: Enter fault information

[0133] Fault information input is divided into two methods: manual input and automatic collection input. Manual input of fault information can be simulated to verify the accuracy of the process. Automatic collection input means that the information collector receives the operating status information of the equipment, automatically judges the equipment fault, and inputs the fault information code into the diagnostic model information port.

[0134] Step 2: Fault fingerprint extraction

[0135] By acquiring a large amount of fault information, we can obtain a massive amount of fault information data; then we use the decision tree model to refine and focus the fault information and eliminate invalid information; until the fault information reaches the upper limit of the decision tree depth.

[0136] Step 3: Fingerprint database collection and random test fingerprints

[0137] The fault fingerprint information is extracted and added to the fault fingerprint library; if it belongs to an existing fault fingerprint, the fault information is assigned a corresponding fault fingerprint code; if no matching fingerprint information is found in the fingerprint library, a new fault fingerprint and corresponding code are created.

[0138] The phenomena of the acquired fault information are aggregated and processed to minimize the number of fingerprint codes.

[0139] Randomly input a fault phenomenon for testing to verify whether the fingerprint extraction and matching are correct. If the test is correct, the fingerprint library will not be modified; otherwise, the fingerprint library will be modified again until the correct recognition rate of the random test fault fingerprint is greater than 99%.

[0140] Fingerprint library collection processing includes: fault fingerprint feature comparison, new fingerprint normalization,

[0141] Fault fingerprints are similar and overlapping, and fingerprints with overlapping information need to be further aggregated and processed; the characteristics of the fault fingerprints are compared one by one, and the similarity of all fingerprint features is compared. If the number of features with a similarity of no more than 1% exceeds 99%, it can be determined to be the same fault fingerprint; otherwise, it should be identified as two fingerprints.

[0142] Step 4: Compare fault fingerprint characteristics and normalize new fingerprints

[0143] Fault fingerprints are similar and overlapping, and fingerprints with overlapping information need to be further aggregated and processed; the characteristics of the fault fingerprints are compared one by one, and the similarity of all fingerprint features is compared. If the number of features with a similarity of no more than 1% exceeds 99%, it can be determined to be the same fault fingerprint; otherwise, it should be identified as two fingerprints.

[0144] Step 5: Determine whether the termination condition is met

[0145] Compare the characteristics of the fault fingerprints one by one, and obtain the similarity of all fingerprint features. If the number of features with a similarity of no more than 1% exceeds 99%, it can be determined to be the same fault fingerprint; otherwise, it should be determined to be two fingerprints and execute step 6.

[0146] Step 6: Generate new fingerprint

[0147] After a new fingerprint is generated, it is assigned a fingerprint code and the identity information of the fault fingerprint, such as the cause of the fault, the nature of the fault, and the fault solution.

[0148] Step 7: Update the fault fingerprint library

[0149] For the newly generated fault fingerprint, use the database update instruction to update the current fingerprint database;

[0150] Step 8: Best matching fault fingerprint: For the latest fault pattern stored in the database, the matching degree is calculated with the existing fingerprints in the database. If the average matching degree of all features is not less than 90%, it is listed as the best matching fault fingerprint, and the faults of the same type are grouped and given the same fingerprint information.

[0151] Step 9: Determine whether the fingerprint database (features) already has information

[0152] Determine whether there is a matching solution for the fault fingerprint. If there is a solution, execute step 10 to input the fault fingerprint, calculate through the decision tree model, find the best solution, and output the fault diagnosis plan;

[0153] If the existing information in the database does not match, decision tree feature selection is performed, decision tree pruning is performed, the latest decision tree model is updated, and then the decision tree model calculation is performed.

[0154] Step 10: Troubleshooting Fingerprint Input

[0155] Input the fault fingerprint information into the decision tree and use the decision tree model to calculate and obtain the best solution.

[0156] Step 11: Decision Tree Feature Selection

[0157] Decision tree feature selection uses a recursive selection method to find the optimal features and then splits the training data based on these features to achieve the best classification for each sub-dataset. This process includes feature selection, decision tree generation, and decision tree pruning.

[0158] If the number of features is large, features are selected at the beginning of decision tree learning, leaving only features that have sufficient classification ability for the training data.

[0159] Step 12: Determine decision tree pruning

[0160] Decision tree pruning steps: Decision tree pruning is to make the tree simpler so that it has better generalization ability. It does this by removing overly detailed leaf nodes, making them fall back to the parent node or even a higher node, and then changing the parent node or higher node to a new leaf node.

[0161] If the condition is met, proceed to step 13; if not, proceed to step 14.

[0162] Step 13: Update the decision tree model

[0163] Add new leaf nodes to the decision tree model, improve and update the decision tree model, and generate the latest leaf node array.

[0164] Step 14: Decision tree model calculation (decision tree feature selection, pruning)

[0165] The fault diagnosis model determines the specific cause of the fault based on the input fault problem, extracts the characteristics of the fault information, uses the decision tree algorithm C4.5 to extract the characteristic fingerprint of the fault, establishes a fault fingerprint library, and inputs it into the decision tree model.

[0166] Fault information X is a random variable. Based on the decision tree algorithm C4.5, the information entropy of the random variable X is calculated:

[0167]

[0168] Where n represents the value of X, and pi represents the probability of taking the value i.

[0169] Conditional entropy is calculated using the following formula, which represents the uncertainty of random variable X under the condition of random variable Y.

[0170]

[0171] G(X, Y) = H(X, Y) - H(X|Y), information gain,

[0172] Assume that the sample set on node t is D = (x, y), where x = (x1, x2, ..., xN) represents the feature variable,

[0173] y=(y1,y2,………,yN) represents the response variable.

[0174] For any splitting point m = (a, fm), a and fm correspond to the characteristic variable and the splitting point critical value respectively, and m divides the sample set into the left and right sides, namely Dleft(m) and Dright(m).

[0175] Dleft(m)=(x,y)|xa≤fm;

[0176] Dright(m)=D\Dleft(m),

[0177] Considering that the sample sets of child nodes are impure in practice, the weight of impurity is added in the classification.

[0178] The expression of impurity I under probability p is:

[0179]

[0180] Where K represents the category and pk represents the probability of the kth category.

[0181] I(p) represents the impurity with probability p.

[0182] Assume that the split point impurity function I is as follows:

[0183]

[0184] Where |D| represents the number of samples at node t, and the optimal split point that minimizes the impurity function is calculated

[0185] m * =argmin m I(D, m)

[0186] Use the same method to iteratively split Dleft(m * ) and Dright(m * ), guiding the decision tree depth to reach the upper limit or the number of child node samples is less than the number of samples specified in advance.

[0187] Step 15: Troubleshooting Solutions

[0188] Output the solution to the problem:

[0189] 1.1 Model parameter class

[0190] Zero drift: Adjust the sensor zero point.

[0191] Temperature Drift: Correct drift model parameters.

[0192] Time Drift: Correct drift model parameters.

[0193] Sensitivity Drift: Adjust the sensitivity coefficient.

[0194] 1.2 Sensor failure

[0195] Electrolyte exhausted: Replace sensor.

[0196] The UV lamp is contaminated: return the UV lamp to the factory for cleaning.

[0197] The UV lamp has reached the end of its life: Replace the UV lamp.

[0198] Grid failure: Replace the grid.

[0199] 1.3 Communication failure

[0200] Tariff card is in arrears: recharge.

[0201] Tariff card failure: The tariff card is magnetized, please replace the card.

[0202] DTU hardware failure: Replace the DTU.

[0203] DTU offline bug: Replace the DTU.

[0204] 1.4 Device hardware failure

[0205] Power supply failure: Replace the AC-DC power converter.

[0206] Solar panel failure: Replace the charge and discharge manager.

[0207] Lithium battery failure: Replace the aged lithium battery.

[0208] Core control board failure: Replace the main board.

[0209] Sensor base plate failure: Replace the sensor base plate.

[0210] LED display card failure: Replace the display card.

[0211] 1.5 Device software failure

[0212] DTU firmware failure: Upgrade the DTU firmware.

[0213] The main core control board firmware is faulty: Upgrade the firmware.

[0214] Display unit firmware failure: Upgrade the firmware.

[0215] As described above, the embodiments of the present invention have been described in detail. However, it is obvious to those skilled in the art that many variations are possible without departing from the spirit and effects of the present invention. Therefore, all such variations are included within the scope of protection of the present invention.

Claims

1. A method for fault diagnosis of environmental monitoring equipment based on artificial intelligence, characterized in that ,Contains the following steps: including fault fingerprint extraction, decision tree construction, decision tree self-learning and decision tree model calculation, solution given; It also contains the following steps: Step 1: Enter fault information; Step 2: Obtain massive amounts of fault information data through a large amount of fault information acquisition; Then, the decision tree model is used to refine and focus the fault information and eliminate invalid information until the fault information reaches the upper limit of the decision tree depth. Step 3: The focused and refined fault information phenomena are aggregated and processed to minimize the number of fingerprint codes. The termination condition is determined by comparing the characteristics of the fault fingerprints one by one. The similarity of all fingerprint features is determined. If the number of features with a similarity of no more than 1% exceeds 99%, they are considered to be the same fault fingerprint. Otherwise, they are considered to be two fingerprints. A random fault phenomenon is input for testing to verify the correctness of fingerprint extraction and matching. If the test is correct, the fingerprint library is not modified. Otherwise, the fingerprint library is modified again until the correct recognition rate of the random fault fingerprint test exceeds 99%. Step 4: Extract the fingerprint information of the collected fault and add it to the fault fingerprint library. If it is an existing fault fingerprint, assign the corresponding fault fingerprint code to the fault information. If no matching fingerprint information is found in the fingerprint library, create a new fault fingerprint and corresponding code. Step 5: After the new fingerprint is generated, it is assigned a fingerprint code, the identity information of the fault fingerprint, the cause of the fault, the nature of the fault, and the fault solution; Step 6: For the newly generated fault fingerprint, use the database update instruction to update the current fingerprint database; Step 7: Calculate the matching degree of the latest fault pattern with the existing fingerprints in the database. If the average matching degree of all features is not less than 90%, it will be listed as the best matching fault fingerprint, grouped according to the same type of fault, and assigned the same fingerprint information; Step 8: Determine whether the fingerprint database has existing information and whether there is a matching solution for the fault fingerprint. If there is a solution, proceed to step 9 to input the fault fingerprint. If it does not meet the existing information in the database, proceed to step 10 to select decision tree features and the subsequent decision tree pruning steps, update the decision tree model step, and calculate the decision tree model. Step 9: Fault fingerprint input: Input the best matching fault fingerprint information into the decision tree, and use the decision tree model in step 13 to calculate and obtain the best solution; Step 10: Decision tree feature selection: Decision tree learning includes feature selection, decision tree pruning and decision tree update processes; Feature selection uses a recursive selection method to find the optimal feature and split the training data according to the feature, so that each sub-data set has the best classification process. If there are many features, the features are selected at the beginning of decision tree learning, leaving only the features that have sufficient classification ability for the training data; Step 11: Pruning the decision tree: By removing the leaf nodes that are too detailed, make them fall back to the parent node or even a higher node, and then change the parent node or higher node to a new leaf node. If it meets the requirements, execute step 12, otherwise execute step 10. Step 12: Update the decision tree model: add the new leaf nodes to the decision tree model, improve and update the decision tree model, and generate the latest leaf node array; Step 13: Decision tree model calculation: The fault diagnosis model determines the specific cause of the fault based on the input fault problem, extracts the characteristics of the fault information, and uses the decision tree algorithm C4.5 to extract the characteristic fingerprint of the fault. The fingerprint is input into the decision tree to obtain the best solution. Step 14: Given a fault diagnosis solution, output a solution to the fault.

2. The method for diagnosing environmental monitoring equipment faults based on artificial intelligence according to claim 1 is characterized in that , The steps of fault fingerprint extraction include: The first step is to obtain fault information, which is obtained through on-site equipment; The second step is to extract fault fingerprints and compare them with the fingerprint library for classification processing; In the third step, the fault fingerprint library is updated and the best matching fault fingerprint is extracted and input into the decision tree; The decision tree construction steps include: The first step is the generation of decision tree: the process of generating a decision tree from a training sample set; The second step is pruning the decision tree: Decision tree pruning is the process of testing, correcting, and modifying the decision tree generated in the previous stage. A new sample data set, called the test data set, is used to verify the preliminary rules generated during the decision tree generation process, and branches that affect the accuracy of the prediction are pruned. The decision tree learning steps include: building a decision tree model based on a given training data set, correctly classifying instances, inducing a set of classification rules from the training data set, and selecting a decision tree that has the least conflict with the training data; Decision tree learning usually includes three steps: feature selection step, decision tree generation step and decision tree pruning step; Decision tree feature selection step: If the number of features is large, select the features at the beginning of decision tree learning, leaving only the features that have sufficient classification ability for the training data; The steps of generating a decision tree: corresponding to the local selection of the model, the pruning of the decision tree corresponds to the global selection of the model. The generation of the decision tree only considers the local optimum, while the pruning of the decision tree considers the global optimum; The pruning steps of the decision tree are: by removing the leaf nodes that are too detailed, making them fall back to the parent node or even a higher node, and then changing the parent node or higher node to a new leaf node; Steps for establishing a decision tree model: Select common fault characteristics from thousands of sets of field equipment: number of equipment offline times, data exceeding standard rate, data abnormality rate, data quality control exceeding standard rate, data mean square error exceeding standard rate, horizontal exceeding standard ratio, vertical exceeding standard ratio, and analysis accuracy rate. Massive fault information is focused and refined, with the following characteristics: Fault information X is a random variable. Based on the decision tree algorithm C4.5, the information entropy of the random variable X is calculated: （1） Where n represents the value of X, pi represents the probability of taking the value i, Conditional entropy is calculated using the following formula to express the uncertainty of random variable X under the condition of random variable Y: （2） G(X, Y) = H(X, Y) - H(X|Y), information gain, Assume that the sample set on node t is D=(x,y), where x=(x1,x2,……,xN) represents the feature variable, y=(y1,y2,………,yN) represents the response variable, For any split point m=(a,fm), a and fm correspond to the characteristic variable and the split point critical value respectively, m divides the sample set into the left and right sides, namely Dleft(m) and Dright(m), Dleft(m)=(x,y)|xa≤fm; Dright(m)=D\Dleft(m), Considering that the sample set of the child node is impure in practice, the weight of impurity is added in the classification. The expression of impurity I under probability p is: Where K represents the category, pk represents the probability of the kth category, I(p) represents the impurity with probability p, Assume that the split point impurity function I is as follows: ， Where |D| represents the number of samples at node t, and the optimal split point that minimizes the impurity function is calculated: m * =argmin m I(D,m), Use the same method to iteratively split Dleft(m * ) and Dright (m * ), guide the decision tree depth to reach the upper limit or the number of child node samples is less than the pre-specified number of samples; Add new leaf nodes to the decision tree model, improve and update the decision tree model, and generate the latest leaf node array.

3. The method for diagnosing environmental monitoring equipment faults based on artificial intelligence according to claim 2 is characterized in that ,Input fault information: Fault information input is divided into two ways: manual input and automatic collection input; manual input of fault information is used to perform simulation and verify the accuracy of the process; automatic collection input, the information collector receives the operating status information of the equipment, automatically judges the equipment fault, and inputs the fault information code into the diagnostic model information port.

Images