Hardware and software intelligent alarm method for warehouse ex-warehouse and warehouse-in equipment

Abstract

The invention provides a hardware and software intelligent alarm method for warehouse ex-warehouse and warehouse-in equipment, which comprises the following steps of: firstly, collecting hardware data such as equipment temperature and running speed and software data such as system response time through a sensor and a system interface, sampling once every 10 milliseconds, and storing by utilizing an SQL (Structured Query Language) program; secondly, cleaning the data according to a 3 sigma principle and performing Z-score standardization processing by applying an SQL query statement and a related function; and then an anomaly detection model is constructed based on a hidden Markov model (HMM) and support vector data description (SVDD), the HMM is trained by normal data to determine a state transition and observation probability matrix, a minimum hypersphere is found by the SVDD, and parameters are optimized through cross validation. In the real-time monitoring link, the SQL continuously obtains a new data input model, the HMM calculates probability distribution, the SVDD measures the distance, the sliding window technology is combined for comprehensive judgment, when abnormal conditions are met, voice alarm is triggered, abnormal information is displayed at the same time, abnormal events can be recorded and analyzed to improve the system, and high efficiency and stability of warehousing operation are effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of hardware and software intelligent alarm, and in particular to a hardware and software intelligent alarm method for warehouse inbound and outbound equipment. Background Art

[0002] During the warehouse outbound and inbound operations, the normal operation of the equipment's hardware and software is crucial to the efficiency and accuracy of the entire warehousing operation. In terms of hardware, abnormalities in parameters such as equipment temperature, operating speed, and load may cause equipment failure, performance degradation, or even damage, which in turn affects the efficiency of outbound and inbound operations and the safe storage of goods. For example, excessively high equipment temperatures may cause the equipment to overheat and shut down, causing delays in the transportation or storage of goods. In terms of software, problems with system response time, data transmission rate, process status, etc. will cause data processing errors, information delays, or interruptions to the operation process. For example, too slow a data transmission rate may result in untimely updates of inventory information, affecting the allocation and management decisions of goods.

[0003] Traditional monitoring methods often rely on manual inspections or simple threshold alarms. Manual inspections are difficult to achieve real-time and comprehensiveness, and are prone to missing abnormal situations. Simple threshold alarms have poor adaptability to complex dynamic data changes and high false alarm and missed alarm rates. With the expansion of the scale of warehousing business and the improvement of automation, there is an urgent need for an intelligent method that can accurately and timely detect hardware and software anomalies and automatically alarm to ensure the smooth operation of warehouse outbound and inbound operations, improve the reliability and efficiency of warehouse management, and reduce economic losses and operational risks caused by equipment failures and software problems. Summary of the invention

[0004] The main purpose of the present invention is to provide a hardware and software intelligent alarm method for warehouse inbound and outbound equipment to solve the problem that traditional monitoring methods often rely on manual inspections or simple threshold alarms, and manual inspections are difficult to achieve real-time and comprehensiveness and are prone to missing abnormal situations.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a hardware and software intelligent alarm method for warehouse outbound and inbound equipment, the method comprising: S1. Use sensors and system interfaces to collect equipment operating status and system performance data at fixed time intervals, create SQL programs, create special data tables to store the collected data, and insert the real-time collected data into the corresponding data tables through SQL statements; S2. Use SQL query statements to obtain data within a certain period of time from the data table for cleaning operations, and perform standardization on the cleaned data; S3. An anomaly detection model is constructed based on HMM and SVDD descriptions. During the training process, the model parameters of HMM and SVDD are adjusted through cross-validation and other techniques to improve the accuracy and generalization ability of the model. HMM is a hidden Markov model and SVDD is support vector data. S4, continuously obtain the latest data from the SQL database as the input of the model, use HMM to calculate the probability distribution of the data, and use SVDD to measure the distance from the data point to the center of the hypersphere to determine whether there is an anomaly; In order to avoid false positives, SQL window functions are used to analyze multiple consecutive data points for comprehensive judgment; S5. According to the detection of S4 above, when an abnormality is detected, the voice alarm module is activated to play a corresponding voice prompt; Display detailed exception details on the control terminal to help staff understand the problem and take action; Use SQL programs to record all abnormal events in log tables to facilitate subsequent analysis and system improvements.

[0006] In the preferred embodiment, in step S1: Hardware data includes device temperature, operating speed, load, and software data; The software data includes system response time, data transmission rate, and process status. It is sampled at intervals of 10 milliseconds, and a data table containing data collection timestamps, hardware parameter names and corresponding values, software parameter names and corresponding value fields is created through the developed SQL program. The collected data is inserted into the data table in real time for storage.

[0007] In the preferred embodiment, in step S2: Its data preprocessing module obtains data within a certain period of time from the data table through SQL query statements, uses SQL aggregation functions and conditional screening statements, cleans the data according to the 3σ principle, removes noise and outliers, and standardizes the cleaned data by writing custom functions or using mathematical functions supported by the database to implement the Z-score standardization method, and stores the processed data in a new data table or temporary table.

[0008] In the preferred embodiment, in step S3: Its anomaly detection model builds and trains HMM and SVDD. It uses the data obtained from the SQL database under normal operating conditions to train the HMM, determine the state transition probability matrix and observation probability matrix of the model, and use SVDD to model the normal data to find the minimum hypersphere. It also adjusts the model parameters such as the number of hidden states of the HMM and the kernel function parameters of the SVDD through cross-validation technology to improve accuracy and generalization ability.

[0009] In the preferred embodiment, the HMM construction steps are: S31. Extract the time series data under normal operation from the SQL database. The data series is ,in Indicates at time The observed value of is the length of the data sequence; S32. Determine the model structure: Assume that the number of hidden states of the hidden Markov model is ; First, you need to initialize the probability distribution of the hidden state ,in , represents the initial hidden state, ; Initialize the state transition probability matrix at the same time ,in , indicating that from the state Transfer to state The probability of And the observation probability matrix ,in , indicating that in the hidden state Observed below The probability of Using the initialization method based on data frequency, for , the initial state in the calculation data is ratio; , calculated from the state Transfer to state The number of times from the state The ratio of the total number of transfers; , in the state Observed below The number of times the state The proportion of the total number of occurrences; S33. Training model: Use the Baum-Welch algorithm to iteratively update model parameters , and , to maximize the probability of observing the data ,in is the parameter set of HMM; In each iteration, the forward probability is calculated and the backward probability : Forward Probability The calculation formula is: ; Its function is to calculate the time by recursion In hidden state And the observed sequence The probability of Backward Probability The calculation formula is: ; It is used to calculate the time In hidden state And the subsequent observation sequence is known The probability of Then calculate and : ; The above formula means that at time In hidden state The probability of ; The above formula means that at time In hidden state And at the moment Move to hidden state The probability of Based on these probabilities, update the model parameters: ; ; ; in is an indicator function if Then , otherwise ; The above process is repeated until the model converges, that is, the change in parameters is less than the set threshold.

[0010] In the preferred solution, the SVDD construction steps are: S301. Also obtain data from the SQL database under normal operation ,in is the number of data points; S302. Select kernel function and determine hypersphere: Select appropriate kernel function , Gaussian kernel function ,in is the kernel parameter; introduce slack variables and penalty parameters , construct the optimization problem: ; ; in is the radius of the hypersphere, is the representation of the center of the hypersphere in the feature space, Is the data A function that maps to a high-dimensional feature space; by solving this optimization problem, we can find the smallest hypersphere that contains as much normal data as possible; Using the Lagrange dual method to solve the above optimization problem, we get the dual problem: ; ; in is the Lagrange multiplier; After solving the dual problem, we can get the center of the hypersphere ,radius ,in is satisfied Any data point of .

[0011] In the preferred embodiment, the cross-validation technique steps are: The normal operation data obtained from the SQL database is randomly divided into a training set, a validation set, and a test set, for example, in a ratio of 60%, 20%, and 20%; For the number of hidden states of HMM , take values ​​within a certain range, use the training set to train the HMM model, and evaluate the performance of the model on the validation set; For the kernel parameters of SVDD and penalty parameters , using the grid search method; within a certain range of values, different parameter combinations are used to train the SVDD model, and the performance of the model is evaluated on the validation set; Repeat the above process several times to ensure the stability and reliability of parameter adjustment.

[0012] In the preferred embodiment, in step S4: Its real-time monitoring and abnormal judgment mechanism uses SQL programs to continuously query newly collected real-time data and input it into the model trained in step S3. It uses HMM to calculate the probability distribution of the data under the current model and judge whether it conforms to the normal dynamic change pattern. At the same time, it calculates the distance from the data point to the center of the SVDD hypersphere; When the probability of data in HMM is lower than 0.01 or the distance from the center of the hypersphere is greater than the set radius, it is judged as abnormal data. The sliding window technology is used to analyze continuous data points by writing SQL complex query statements combined with window functions. When 3 or more of 5 consecutive data points are judged to be abnormal, it is finally determined that an abnormal situation has occurred.

[0013] In the preferred embodiment, in step S4: Including data acquisition and HMM probability calculation method: Use SQL program to continuously obtain the latest production and manufacturing outbound and inbound operation data sequence from the SQL database , and input it into the trained HMM model, and calculate the probability distribution of the data under the current model through the following formula : First calculate the forward probability , ,in is the parameter set of HMM, is the initial probability distribution of the hidden state, is the state transition probability matrix, is the observation probability matrix, is the number of hidden states; obtained by recursive calculation , and then calculate ,when When the data is initially judged to be abnormal; SVDD distance calculation method for new data points , using the kernel function determined during the model training phase Map it to the feature space of SVDD and calculate the data points using the following formula To the center of the SVDD hypersphere Distance : ,in is the Lagrange multiplier in the SVDD model, is a training data point, when When , it is determined that the data point may deviate from the normal data distribution and there is an anomaly. is the radius of the SVDD hypersphere, which is determined during the training phase; Its sliding window and comprehensive judgment mechanism uses SQL window function to create a sliding window of size w = 5, and uses the statement `OVER (ORDER BY timestamp ROWS BETWEEN 4 PRECEDING AND CURRENT ROW)` to store the abnormal judgment results of the last 5 data points. Each new data point is marked as abnormal 1 or normal 0 according to the judgment results of HMM and SVDD and stored in the window. Then, the SQL query statement `SUM(abnormal_flag) OVER(ORDER BY timestamp ROWS BETWEEN 4 PRECEDING AND CURRENT ROW)` is used to count the number of abnormal data points c in the window. When the abnormal situation occurs, it is finally determined.

[0014] In the preferred embodiment, in step S5: The voice alarm and information prompt function triggers the voice alarm module immediately after determining that an abnormality has occurred. According to the correspondence between the predefined abnormality type and the voice prompt information, the appropriate voice content is selected for playback. At the same time, the control terminal obtains the specific value, occurrence time, and possible cause analysis information of the abnormal data from the database through SQL query for display, so that the staff can take intervention measures; It also includes abnormal event recording and analysis functions, using SQL programs to record alarm time, abnormal type, and processing result abnormal event information in a special log table, so as to facilitate subsequent analysis of the frequency and cause of abnormal occurrence, thereby adjusting and improving the model and improving system reliability and accuracy.

[0015] The present invention provides a hardware and software intelligent alarm method for warehouse outbound and inbound equipment. The intelligent alarm method has many beneficial effects. First, it can monitor the hardware and software operating status of warehouse outbound and inbound equipment in real time and accurately, effectively avoiding the untimeliness and limitations of manual inspections, as well as the high false alarm and missed alarm problems of traditional simple threshold alarms. By adopting an algorithm combining advanced hidden Markov model (HMM) and support vector data description (SVDD), and using SQL programs for data processing and analysis, the accuracy and reliability of anomaly detection are greatly improved.

[0016] In terms of hardware, timely detection of abnormalities in equipment parameters such as temperature, operating speed, and load can enable measures to be taken in advance to prevent equipment failure, ensure stable operation of the equipment, and reduce equipment downtime, thereby improving the efficiency of warehousing and outbound operations and reducing equipment maintenance costs and the risk of cargo loss due to equipment failure.

[0017] In terms of software, precise monitoring of system response time, data transmission rate, process status, etc. ensures the normal operation of the software system, avoids data processing errors and information delays, ensures timely updating of inventory information and continuity of operating processes, helps optimize warehouse management decisions, improves the reliability and efficiency of overall warehouse operations, and enhances the competitiveness of enterprises in the warehousing and logistics links. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is a flow chart of the hardware and software intelligent algorithm of the storage and warehousing equipment of the present invention; Figure 2 It is a flow chart of the data collection and storage algorithm of the present invention; Figure 3 It is a flow chart of the data preprocessing algorithm of the present invention; Figure 4 It is a flow chart of the algorithm for constructing and training a model in the present invention; Figure 5 It is a flow chart of the real-time monitoring and abnormality judgment algorithm of the present invention; Figure 6 It is a flow chart of the voice alarm and information prompt algorithm of the present invention; Figure 7 It is a flow chart of abnormal event recording and analysis of the present invention. DETAILED DESCRIPTION

[0019] Example 1 like Figure 1-7 As shown, a method for intelligent alarm of hardware and software of warehouse inbound and outbound equipment, the method comprising: S1. Use sensors and system interfaces to collect equipment operating status and system performance data at fixed time intervals, create SQL programs, create special data tables to store the collected data, and insert the real-time collected data into the corresponding data tables through SQL statements; S2. Use SQL query statements to obtain data within a certain period of time from the data table for cleaning operations, and perform standardization on the cleaned data; S3. An anomaly detection model is constructed based on HMM and SVDD descriptions. During the training process, the model parameters of HMM and SVDD are adjusted through cross-validation and other techniques to improve the accuracy and generalization ability of the model. HMM is a hidden Markov model and SVDD is support vector data. S4, continuously obtain the latest data from the SQL database as the input of the model, use HMM to calculate the probability distribution of the data, and use SVDD to measure the distance from the data point to the center of the hypersphere to determine whether there is an anomaly; In order to avoid false positives, SQL window functions are used to analyze multiple consecutive data points for comprehensive judgment; S5. According to the detection of S4 above, when an abnormality is detected, the voice alarm module is activated to play a corresponding voice prompt; Display detailed exception details on the control terminal to help staff understand the problem and take action; Use SQL programs to record all abnormal events in log tables to facilitate subsequent analysis and system improvements.

[0020] Through the above steps, the SQL program is fully utilized for data storage, query and processing in the intelligent voice abnormality alarm method. The HMM and SVDD algorithms are combined to realize the effective detection and alarm of hardware and software abnormalities in the production and manufacturing outbound and inbound operations, and provide timely intervention prompts for the staff.

[0021] In the preferred embodiment, in step S1: Hardware data includes device temperature, operating speed, load, and software data; The software data includes system response time, data transmission rate, and process status. It is sampled at intervals of 10 milliseconds, and a data table containing data collection timestamps, hardware parameter names and corresponding values, software parameter names and corresponding value fields is created through the developed SQL program. The collected data is inserted into the data table in real time for storage.

[0022] It can collect equipment operation data comprehensively and timely to avoid data omissions, which helps to accurately grasp the operating status of the equipment. It is the cornerstone of the entire intelligent alarm system, providing a stable data source for subsequent links and ensuring the effective operation of the system.

[0023] In the preferred embodiment, in step S2: Its data preprocessing module obtains data within a certain period of time from the data table through SQL query statements, uses SQL aggregation functions and conditional screening statements, cleans the data according to the 3σ principle, removes noise and outliers, and standardizes the cleaned data by writing custom functions or using mathematical functions supported by the database to implement the Z-score standardization method, and stores the processed data in a new data table or temporary table.

[0024] It improves data quality, reduces the interference of abnormal data on subsequent model training and judgment, enhances the stability and accuracy of the system, and makes subsequent analysis and model building based on these data more reliable.

[0025] Example 2 Further illustrate with reference to Example 1, Figure 1-7As shown, in step S3: its anomaly detection model constructs and trains HMM and SVDD, uses the data obtained from the SQL database under normal operating conditions to train the HMM, determines the state transition probability matrix and observation probability matrix of the model, and uses SVDD to model the normal data to find the minimum hypersphere, and adjusts the model parameters such as the number of hidden states of the HMM and the kernel function parameters of the SVDD through cross-validation technology to improve accuracy and generalization ability.

[0026] An anomaly detection model is constructed based on HMM and SVDD. The normal operation data in the SQL database is used to train HMM to obtain the state transition and observation probability matrix. At the same time, SVDD is used to find the minimum hypersphere of normal data, and the model parameters are optimized through cross-validation.

[0027] HMM can effectively capture the dynamic changes of data, and SVDD has a good description of the boundaries of normal data distribution. The combination of the two improves the model's ability to identify anomalies. Through parameter adjustment, the accuracy and generalization ability of the model are further enhanced, and it can better adapt to different operating scenarios.

[0028] In the preferred embodiment, the HMM construction steps are: S31. Extract the time series data under normal operation from the SQL database. The data series is ,in Indicates at time The observed value of is the length of the data sequence; S32. Determine the model structure: Assume that the number of hidden states of the hidden Markov model is ; First, you need to initialize the probability distribution of the hidden state ,in , represents the initial hidden state, ; Initialize the state transition probability matrix at the same time ,in , indicating that from the state Transfer to state The probability of And the observation probability matrix ,in , indicating that in the hidden state Observed below The probability of Using the initialization method based on data frequency, for , the initial state in the calculation data is ratio; , calculated from the state Transfer to state The number of times from the state The ratio of the total number of transfers; , in the state Observed below The number of times the state The proportion of the total number of occurrences; S33. Training model: Use the Baum-Welch algorithm to iteratively update model parameters , and , to maximize the probability of observing the data ,in is the parameter set of HMM; In each iteration, the forward probability is calculated and the backward probability : Forward Probability The calculation formula is: ; Its function is to calculate the time by recursion In hidden state And the observed sequence The probability of Backward Probability The calculation formula is: ; It is used to calculate the time In hidden state And the subsequent observation sequence is known The probability of Then calculate and : ; The above formula means that at time In hidden state The probability of ; The above formula means that at time In hidden state And at the moment Move to hidden state The probability of Based on these probabilities, update the model parameters: ; ; ; in is an indicator function if Then , otherwise ; The above process is repeated until the model converges, that is, the change in parameters is less than the set threshold.

[0029] By extracting time series data from the SQL database and initializing model parameters based on data frequency, the foundation for subsequent training is laid. The Baum-Welch algorithm is used to iteratively update parameters based on forward probability, backward probability, and The calculation of etc. realizes the gradual optimization of model parameters, and finally obtains the state transition probability matrix that can reflect the dynamic change law of data and the observation probability matrix .

[0030] In the real-time monitoring stage, the probability distribution of new input data under the trained model can be calculated , by comparing with the set threshold (such as $0.01$), we can preliminarily judge whether the data is abnormal, providing an important basis for subsequent comprehensive judgment.

[0031] It effectively captures the dynamic change patterns of data, improves the sensitivity to changes in equipment operating status, can promptly discover possible abnormal trends, reduces the risk of underreporting due to dynamic changes in data, and ensures the stability of equipment hardware and software operation in warehouse operations.

[0032] The training process is based on actual production and manufacturing warehousing and outbound operation data, which makes the model more targeted and adaptable, and can better serve the anomaly detection needs in the warehouse environment.

[0033] In the entire intelligent alarm system, HMM is mainly responsible for modeling and analyzing the dynamic changes of data. By learning historical data and calculating the probability of new data, it provides key quantitative indicators for judging whether the data conforms to the normal operating mode. It is an important part of anomaly detection.

[0034] In the preferred solution, the SVDD construction steps are: S301. Also obtain data from the SQL database under normal operation ,in is the number of data points; S302. Select kernel function and determine hypersphere: Select appropriate kernel function , Gaussian kernel function ,in is the kernel parameter; introduce slack variables and penalty parameters , construct the optimization problem: ; ; in is the radius of the hypersphere, is the representation of the center of the hypersphere in the feature space, Is the data A function that maps to a high-dimensional feature space; by solving this optimization problem, we can find the smallest hypersphere that contains as much normal data as possible; Using the Lagrange dual method to solve the above optimization problem, we get the dual problem: ; ; in is the Lagrange multiplier; After solving the dual problem, we can get the center of the hypersphere ,radius ,in is satisfied Any data point of .

[0035] After obtaining normal data from the SQL database, the optimization problem is constructed by selecting a suitable kernel function, introducing slack variables and penalty parameters, and solving it using the Lagrangian dual method to determine the center a and radius R of the minimum hypersphere containing the normal data.

[0036] During real-time monitoring, new data points can be mapped to the feature space, and their distance $d_t$ to the center of the hypersphere can be calculated. This can be compared with the radius $R$ to determine whether the data deviates from the normal distribution, providing another important dimension for abnormality judgment.

[0037] The distribution boundary of normal data is accurately described, which can effectively identify data points with large differences from normal data distribution, supplement the shortcomings of HMM in anomaly detection, and improve the overall anomaly detection accuracy of the system.

[0038] Its data-driven modeling approach enables the model to adapt to the data distribution of different types of equipment, enhancing the generalization ability of the system.

[0039] SVDD is mainly used in the system to define the distribution range of normal data. By judging the relationship between new data points and this range, it assists HMM in completing the anomaly detection task, thereby improving the reliability and comprehensiveness of anomaly detection.

[0040] In the preferred embodiment, the cross-validation technique steps are: The normal operation data obtained from the SQL database is randomly divided into a training set, a validation set, and a test set, for example, in a ratio of 60%, 20%, and 20%; For the number of hidden states of HMM , take values ​​within a certain range, use the training set to train the HMM model, and evaluate the performance of the model on the validation set; For the kernel parameters of SVDD and penalty parameters , using the grid search method; within a certain range of values, different parameter combinations are used to train the SVDD model, and the performance of the model is evaluated on the validation set; Repeat the above process several times to ensure the stability and reliability of parameter adjustment.

[0041] The normal operation data is reasonably divided into training set, validation set and test set, and the number of hidden states of HMM is The kernel parameters of SVDD are tested using the range of values. and penalty parameters A grid search method was used to train the model under different parameter combinations and evaluate the performance on the validation set.

[0042] By repeating the above process many times, the stability and reliability of parameter adjustment are ensured, and finally the optimized model parameters are obtained.

[0043] It avoids the problem of model overfitting or underfitting, improves the accuracy and generalization ability of the model in different data scenarios, and enables the model to better adapt to the complexity and diversity of warehouse equipment hardware and software data.

[0044] It provides a better-performing anomaly detection model for the entire intelligent alarm system, enhancing the reliability and practicality of the system.

[0045] Cross-validation technology plays a key optimization and verification role in the model training process. By systematically adjusting model parameters, the performance of HMM and SVDD models is improved, which is an important link in ensuring the effectiveness of the entire intelligent alarm method.

[0046] Example 3 Further illustrate with reference to Example 1, Figure 1-7 As shown, in step S4: its real-time monitoring and abnormal judgment mechanism uses SQL program to continuously query the newly collected real-time data and input it into the model trained in step S3, uses HMM to calculate the probability distribution of the data under the current model and judge whether it conforms to the normal dynamic change mode, and at the same time calculates the distance from the data point to the center of the SVDD hypersphere; When the probability of data in HMM is lower than 0.01 or the distance from the center of the hypersphere is greater than the set radius, it is judged as abnormal data. The sliding window technology is used to analyze continuous data points by writing SQL complex query statements combined with window functions. When 3 or more of 5 consecutive data points are judged to be abnormal, it is finally determined that an abnormal situation has occurred.

[0047] New data is continuously input into the trained model through SQL program, combined with HMM probability calculation and SVDD distance measurement, and the sliding window technology and SQL complex query statements and window functions are used to conduct comprehensive analysis and judgment on continuous data points.

[0048] It realizes real-time monitoring of equipment operation data and accurate abnormality judgment, effectively reduces the false alarm rate, and can promptly discover potential hardware and software anomalies, providing guarantees for timely intervention measures and ensuring the continuity and stability of warehouse operations.

[0049] In the preferred embodiment, in step S4: Including data acquisition and HMM probability calculation method: Use SQL program to continuously obtain the latest production and manufacturing outbound and inbound operation data sequence from the SQL database , and input it into the trained HMM model, and calculate the probability distribution of the data under the current model through the following formula : First calculate the forward probability , ,in is the parameter set of HMM, is the initial probability distribution of the hidden state, is the state transition probability matrix, is the observation probability matrix, is the number of hidden states; obtained by recursive calculation , and then calculate ,when When the data is initially judged to be abnormal; SVDD distance calculation method for new data points , using the kernel function determined during the model training phase Map it to the feature space of SVDD and calculate the data points using the following formula To the center of the SVDD hypersphere Distance : ,in is the Lagrange multiplier in the SVDD model, is a training data point, when When , it is determined that the data point may deviate from the normal data distribution and there is an anomaly. is the radius of the SVDD hypersphere, which is determined during the training phase; Its sliding window and comprehensive judgment mechanism uses SQL window function to create a sliding window of size w = 5, and uses the statement `OVER (ORDER BY timestamp ROWS BETWEEN 4 PRECEDING AND CURRENT ROW)` to store the abnormal judgment results of the last 5 data points. Each new data point is marked as abnormal 1 or normal 0 according to the judgment results of HMM and SVDD and stored in the window. Then, the SQL query statement `SUM(abnormal_flag) OVER(ORDER BY timestamp ROWS BETWEEN 4 PRECEDING AND CURRENT ROW)` is used to count the number of abnormal data points c in the window. , it is finally determined that an abnormal situation has occurred.

[0050] Use SQL window functions to create a sliding window of size $5$, which can store the abnormal judgment results of the most recent $5$ data points, and use specific SQL query statements to count the number of abnormal data points in the window. Statistics.

[0051] According to the set rules (when The system finally determines whether an abnormal situation occurs (when the abnormal situation occurs), comprehensively considers the situation of continuous data points, and avoids false alarms that may be caused by single data point judgment.

[0052] It significantly reduces the false alarm rate, improves the accuracy and reliability of abnormal judgment, ensures that the alarm is triggered only when multiple consecutive data points show abnormal trends, reduces unnecessary interference, and improves staff's trust in the alarm and response efficiency.

[0053] It ensures the continuity of warehouse operations and avoids frequent interruptions to the operating process due to false alarms. It can also promptly detect potential hardware and software anomalies and buy time for intervention measures.

[0054] In the entire intelligent alarm process, the sliding window and comprehensive judgment mechanism serve as the last line of defense, integrating and optimizing the preliminary judgment results of HMM and SVDD, making the final abnormal judgment results more scientific and reasonable, which is an important guarantee for the stable operation of the system and effective alarm.

[0055] In the preferred embodiment, in step S5: The voice alarm and information prompt function triggers the voice alarm module immediately after determining that an abnormality has occurred. According to the correspondence between the predefined abnormality type and the voice prompt information, the appropriate voice content is selected for playback. At the same time, the control terminal obtains the specific value, occurrence time, and possible cause analysis information of the abnormal data from the database through SQL query for display, so that the staff can take intervention measures; It also includes abnormal event recording and analysis functions, using SQL programs to record alarm time, abnormal type, and processing result abnormal event information in a special log table, so as to facilitate subsequent analysis of the frequency and cause of abnormal occurrence, thereby adjusting and improving the model and improving system reliability and accuracy.

[0056] After detecting an abnormality, the voice alarm module is triggered quickly, and the corresponding voice is played according to the predefined relationship. At the same time, SQL query is used to obtain detailed abnormal information from the database and display it on the control terminal. This enables staff to know the abnormal situation and related information in the first place, without the need for manual attention to the data at all times, which improves the response speed and facilitates staff to quickly take targeted intervention measures to reduce the impact of abnormalities on operations.

[0057] Use SQL programs to record key information of abnormal events into special log tables to provide data support for subsequent analysis.

[0058] It is convenient to conduct statistical analysis on the frequency and causes of abnormal occurrences, which helps to discover the weak links in the system, so as to adjust and improve the model and the entire system in a targeted manner, continuously improve the reliability and accuracy of the system, and reduce the probability and impact of abnormal occurrences.

[0059] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for intelligent alarm of hardware and software of warehouse inbound and outbound equipment, characterized by: The method includes: S1. Use sensors and system interfaces to collect equipment operating status and system performance data at fixed time intervals, create SQL programs, create special data tables to store the collected data, and insert the real-time collected data into the corresponding data tables through SQL statements; S2. Use SQL query statements to obtain data within a certain period of time from the data table for cleaning operations, and perform standardization on the cleaned data; S3. An anomaly detection model is constructed based on HMM and SVDD descriptions. During the training process, the model parameters of HMM and SVDD are adjusted through cross-validation and other techniques to improve the accuracy and generalization ability of the model. HMM is a hidden Markov model and SVDD is support vector data. S4, continuously obtain the latest data from the SQL database as the input of the model, use HMM to calculate the probability distribution of the data, and use SVDD to measure the distance from the data point to the center of the hypersphere to determine whether there is an anomaly; In order to avoid false positives, SQL window functions are used to analyze multiple consecutive data points for comprehensive judgment; S5. According to the detection of S4 above, when an abnormality is detected, the voice alarm module is activated to play a corresponding voice prompt; Display detailed exception details on the control terminal to help staff understand the problem and take action; Use SQL programs to record all abnormal events in log tables to facilitate subsequent analysis and system improvements.

2. According to claim 1, a method for intelligent alarm of hardware and software of warehouse storage and outbound storage equipment, characterized in that: in step S1: Hardware data includes device temperature, operating speed, load, and software data; The software data includes system response time, data transmission rate, and process status. It is sampled at intervals of 10 milliseconds, and a data table containing data collection timestamps, hardware parameter names and corresponding values, software parameter names and corresponding value fields is created through the developed SQL program. The collected data is inserted into the data table in real time for storage.

3. According to the method for intelligent alarm of hardware and software of warehouse storage and outbound storage equipment in claim 1, it is characterized in that: in step S2: Its data preprocessing module obtains data within a certain period of time from the data table through SQL query statements, uses SQL aggregation functions and conditional screening statements, cleans the data according to the 3σ principle, removes noise and outliers, and standardizes the cleaned data by writing custom functions or using mathematical functions supported by the database to implement the Z-score standardization method, and stores the processed data in a new data table or temporary table.

4. According to claim 1, a method for intelligent alarm of hardware and software of warehouse inbound and outbound equipment, characterized in that: In step S3: Its anomaly detection model constructs and trains HMM and SVDD. It uses the data obtained from the SQL database under normal operating conditions to train the HMM, determine the state transition probability matrix and observation probability matrix of the model, and use SVDD to model the normal data to find the minimum hypersphere. It also adjusts the model parameters such as the number of hidden states of the HMM and the kernel function parameters of the SVDD through cross-validation technology to improve accuracy and generalization ability.

5. The method for intelligent alarm of hardware and software of warehouse inbound and outbound equipment according to claim 4 is characterized by: The steps of HMM construction are: S31. Extract the time series data under normal operation from the SQL database. The data series is ,in Indicates at time The observed value of is the length of the data sequence; S32. Determine the model structure: Assume that the number of hidden states of the hidden Markov model is ; First, you need to initialize the probability distribution of the hidden state ,in , represents the initial hidden state, ; Initialize the state transition probability matrix at the same time ,in , indicating that from the state Transfer to state The probability of And the observation probability matrix ,in , indicating that in the hidden state Observed below The probability of Using the initialization method based on data frequency, for , the initial state in the calculation data is ratio; , calculated from the state Transfer to state The number of times from the state The ratio of the total number of transfers; , in the state Observed below The number of times the state The proportion of the total number of occurrences; S33. Training model: Use the Baum-Welch algorithm to iteratively update model parameters , and , to maximize the probability of observing the data ,in is the parameter set of HMM; In each iteration, the forward probability is calculated and the backward probability : Forward Probability The calculation formula is: ； Its function is to calculate the time by recursion In hidden state And the observed sequence The probability of Backward Probability The calculation formula is: ； It is used to calculate the time In hidden state And the subsequent observation sequence is known The probability of Then calculate and : ； The above formula means that at time In hidden state The probability of ； The above formula means that at time In hidden state And at the moment Move to hidden state The probability of Based on these probabilities, update the model parameters: ； ； ； in is an indicator function if Then , otherwise ; The above process is repeated until the model converges, that is, the change in parameters is less than the set threshold.

6. According to claim 5, a method for intelligent alarm of hardware and software of warehouse inbound and outbound equipment is characterized by: The steps to construct SVDD are: S301. Also obtain data from the SQL database under normal operation ,in is the number of data points; S302. Select kernel function and determine hypersphere: Select appropriate kernel function , Gaussian kernel function ,in is the kernel parameter; introduce slack variables and penalty parameters , construct the optimization problem: ； ； in is the radius of the hypersphere, is the representation of the center of the hypersphere in the feature space, Is the data A function that maps to a high-dimensional feature space; by solving this optimization problem, we can find the smallest hypersphere that contains as much normal data as possible; Using the Lagrange dual method to solve the above optimization problem, we get the dual problem: ； ； in is the Lagrange multiplier; After solving the dual problem, we can get the center of the hypersphere ,radius ,in is satisfied Any data point of .

7. According to claim 6, a method for intelligent alarm of hardware and software of warehouse inbound and outbound equipment, characterized in that: The steps of cross validation technique are: The normal operation data obtained from the SQL database is randomly divided into a training set, a validation set, and a test set, for example, in a ratio of 60%, 20%, and 20%; For the number of hidden states of HMM , take values ​​within a certain range, use the training set to train the HMM model, and evaluate the performance of the model on the validation set; For the kernel parameters of SVDD and penalty parameters , using the grid search method; within a certain range of values, different parameter combinations are used to train the SVDD model, and the performance of the model is evaluated on the validation set; Repeat the above process several times to ensure the stability and reliability of parameter adjustment.

8. According to claim 1, a method for intelligent alarm of hardware and software of warehouse storage and outbound storage equipment, characterized in that: in step S4: Its real-time monitoring and abnormal judgment mechanism uses SQL programs to continuously query newly collected real-time data and input it into the model trained in step S3. It uses HMM to calculate the probability distribution of the data under the current model and judge whether it conforms to the normal dynamic change pattern. At the same time, it calculates the distance from the data point to the center of the SVDD hypersphere; When the probability of data in HMM is lower than 0.01 or the distance from the center of the hypersphere is greater than the set radius, it is judged as abnormal data. The sliding window technology is used to analyze continuous data points by writing SQL complex query statements combined with window functions. When 3 or more of 5 consecutive data points are judged to be abnormal, it is finally determined that an abnormal situation has occurred.

9. According to claim 8, a method for intelligent alarm of hardware and software of warehouse inbound and outbound equipment, Its characteristics are: in step S4: Including data acquisition and HMM probability calculation method: Use SQL program to continuously obtain the latest production and manufacturing outbound and inbound operation data sequence from the SQL database , and input it into the trained HMM model, and calculate the probability distribution of the data under the current model through the following formula : First calculate the forward probability , ,in is the parameter set of HMM, is the initial probability distribution of the hidden state, is the state transition probability matrix, is the observation probability matrix, is the number of hidden states; obtained by recursive calculation , and then calculate ,when When the data is initially judged to be abnormal; SVDD distance calculation method for new data points , using the kernel function determined during the model training phase Map it to the feature space of SVDD and calculate the data points using the following formula To the center of the SVDD hypersphere Distance : ,in is the Lagrange multiplier in the SVDD model, is a training data point, when When , it is determined that the data point may deviate from the normal data distribution and there is an anomaly. is the radius of the SVDD hypersphere, which is determined during the training phase; Its sliding window and comprehensive judgment mechanism uses SQL window function to create a sliding window of size w = 5, and uses the statement `OVER (ORDER BY timestamp ROWS BETWEEN 4 PRECEDING AND CURRENT ROW)` to store the abnormal judgment results of the last 5 data points. Each new data point is marked as abnormal 1 or normal 0 according to the judgment results of HMM and SVDD and stored in the window. Then, the SQL query statement `SUM(abnormal_flag) OVER (ORDERBY timestamp ROWS BETWEEN 4 PRECEDING AND CURRENT ROW)` is used to count the number of abnormal data points c in the window. When the abnormal situation occurs, it is finally determined.

10. The method for intelligent alarm of hardware and software of warehouse inbound and outbound equipment according to claim 8, characterized in that: in step S5: The voice alarm and information prompt function triggers the voice alarm module immediately after determining that an abnormality has occurred. According to the correspondence between the predefined abnormality type and the voice prompt information, the appropriate voice content is selected for playback. At the same time, the control terminal obtains the specific value, occurrence time, and possible cause analysis information of the abnormal data from the database through SQL query for display, so that the staff can take intervention measures; It also includes abnormal event recording and analysis functions, using SQL programs to record alarm time, abnormal type, and processing result abnormal event information in a special log table, so as to facilitate subsequent analysis of the frequency and cause of abnormal occurrence, thereby adjusting and improving the model and improving system reliability and accuracy.

Images