Online diagnosis and retrieval system for PET (positron emission tomography) processing exception logs

Abstract

The invention relates to the technical field of information retrieval, in particular to an online diagnosis retrieval system for PET processing abnormal logs, which comprises a difference preprocessing module for calculating first-order difference absolute values of adjacent time points, extracting second-order difference symbol change times and generating a key change characteristic parameter set based on a temperature parameter sequence and a pressure parameter sequence of PET processing equipment. According to the method, the difference absolute value of the adjacent time points of the temperature and pressure parameters is calculated in real time, the number of second-order difference symbol changes is extracted, the associated code is generated in combination with the timestamp and the equipment number, the time-space relevance of abnormal fluctuation is enhanced, and the weak process deviation is accurately captured in the early stage. According to the method, interval hash indexes are divided on the basis of temperature difference absolute values, meanwhile, B + tree range indexes are constructed for pressure standard deviations, multi-dimensional joint retrieval is achieved, the two abnormal modes of high-frequency oscillation and steady-state offset can be screened independently or in a combined mode, and the false detection rate under the complex working condition is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of information retrieval, and particularly to an online diagnostic retrieval system for PET processing exception logs. Background Art

[0002] An online diagnostic retrieval system for PET processing exception logs is used to perform structured indexing, exception feature extraction, and dynamic retrieval on real-time log data generated by polyester (PET) film production equipment, quickly locate process parameter anomalies (such as temperature fluctuations, pressure deviations), the root causes of equipment failures, and associated event chains, assist engineers in real-time diagnosing production line problems and optimizing process parameters, thereby reducing production downtime and improving product quality consistency.

[0003] The prior art relies on direct indexing and keyword matching of raw log data, lacking quantitative extraction of the dynamic change characteristics of time-series parameters, resulting in slow-changing anomalies of temperature and pressure being easily masked by noise. For example, when the temperature parameter shows a small but continuous positive offset within a continuous time window, it is difficult to identify the potential trend of deteriorating thermal stability because the difference value and the sign change frequency are not calculated, and it is necessary to manually review the waveform diagram to discover. Most existing index structures are of a single type and are difficult to support both equal-value queries and range retrievals at the same time. When it is necessary to jointly screen the temperature fluctuation range and the pressure standard deviation threshold, it is necessary to traverse the full amount of data multiple times, increasing the response delay and affecting the timeliness of online diagnosis. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the deficiencies existing in the prior art and propose an online diagnostic retrieval system for PET processing exception logs.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions: The online diagnostic retrieval system for PET processing exception logs includes: A differential preprocessing module, based on the temperature parameter sequence and pressure parameter sequence of the PET processing equipment, calculates the absolute value of the first-order difference between adjacent time points, extracts the number of sign changes of the second-order difference, generates a key change characteristic parameter set, and associates and encodes the key change characteristic parameter set with the original log timestamp and equipment number to generate a differential associated log sequence; An index dynamic construction module, based on the key change characteristic parameter set, establishes an interval hash index for the absolute value of the temperature first-order difference and a B+ tree range index for the pressure standard deviation, maps the index and the differential associated log sequence to generate a dynamic index address mapping table; The query feature mapping module, based on the abnormal oscillation frequency threshold and the state transition coding sequence input by the user, calls the dynamic index address mapping table to retrieve the log segments whose main oscillation frequency is greater than the abnormal oscillation frequency threshold, matches the log entries whose state transition paths conform to the preset coding, generates a preliminary abnormal candidate set, sorts according to the temperature change rate and pressure standard deviation of the entries in the preliminary abnormal candidate set, and generates a dynamic query parameter vector; The feedback-driven adjustment module, based on the click duration and marking behavior of the user on the entries in the preliminary abnormal candidate set, statistically calculates the average value of the temperature change rate and the extreme value of the pressure standard deviation of the marked entries, calculates the feature difference degree of the unclicked entries, generates a feedback weight coefficient set, superimposes the feedback weight coefficient set and the dynamic query parameter vector, reorders, and generates an optimized result sequence.

[0006] Preferably, the steps for obtaining the key change feature parameter set are as follows: Based on the temperature parameter sequence and pressure parameter sequence collected in real time by the PET processing equipment, respectively call the temperature values at adjacent time points in the temperature parameter sequence and to calculate the absolute difference between adjacent time points of temperature and , generate an absolute value sequence of the first-order difference of temperature; at the same time, call the pressure values at adjacent time points in the pressure parameter sequence and and to calculate the absolute difference between adjacent time points of pressure and , generate an absolute value sequence of the first-order difference of pressure; Based on the absolute value sequence of the first-order difference of temperature and the absolute value sequence of the first-order difference of pressure, traverse each absolute value of the first-order difference of temperature, calculate the second-order difference sign direction with the previous absolute value of the first-order difference of temperature. If the current second-order difference direction is inconsistent with the previous second-order difference direction , record a sign change once, and cumulatively count the frequency of all consecutive three inconsistent sign directions in the temperature parameter sequence to generate the number of second-order difference sign changes of temperature and the number of second-order difference sign changes of pressure; Combine the maximum value in the absolute value sequence of the first-order difference of temperature, the standard deviation of the absolute value sequence of the first-order difference of pressure, the number of second-order difference sign changes of temperature and the number of second-order difference sign changes of pressure to generate a key change feature parameter set.

[0007] Preferably, the steps for obtaining the differential correlation log sequence are as follows: Parse the timestamp and device number fields in the original log entry, call the unique identifier strings of the timestamp and device number, and generate the original log timestamp set and device number set; Based on the maximum value of the absolute value of the first-order difference of temperature, the standard deviation of the absolute value of the first-order difference of pressure, the number of sign changes of the second-order difference of temperature, and the number of sign changes of the second-order difference of pressure in the key change feature parameter set, splice them into a feature string in a fixed order, and perform string concatenation with the corresponding timestamp and device number identifier to generate a differential correlation log sequence.

[0008] Preferably, the steps for obtaining the dynamic index address mapping table are as follows: Traverse the absolute value sequence of the first-order difference of temperature in the key change feature parameter set, set a preset interval division rule, divide the absolute value of the first-order difference of temperature into equal-width intervals according to the numerical range, and call the hash function to perform hash calculation on the start value and end value of each interval to generate a hash index for the absolute value interval of the first-order difference of temperature; Based on the standard deviation of the absolute value of the first-order difference of pressure in the key change feature parameter set, arrange the standard deviations in ascending order, and use the standard deviation as the key value and the corresponding log entry address as the leaf node pointer to generate a B+ tree range index for pressure standard deviation; Combine and map the hash key of the absolute value interval hash index of the first-order difference of temperature with the key value range of the B+ tree range index of pressure standard deviation, establish a two-way pointer association relationship between the hash key and the B+ tree key value and the storage address of the differential correlation log sequence, and generate a dynamic index address mapping table.

[0009] Preferably, the steps for obtaining the preliminary anomaly candidate set are as follows: Parse the anomaly oscillation frequency threshold and state transition coding sequence input by the user, traverse the hash key values of the absolute value interval hash index of the first-order difference of temperature in the dynamic index address mapping table, and extract the set of log segment addresses where the maximum value of the absolute value of the first-order difference of temperature exceeds the preset temperature fluctuation intensity threshold; Based on the set of log segment addresses, calculate the spectral main frequency of the temperature parameter sequence in each log segment, and filter out the log segments with a spectral main frequency greater than the anomaly oscillation frequency threshold; According to the log segments with a spectral main frequency greater than the anomaly oscillation frequency threshold, filter out the entries with a state transition path consistent with the user input coding sequence to generate a preliminary anomaly candidate set.

[0010] Preferably, the steps for obtaining the dynamic query parameter vector are as follows: Traverse each log entry in the preliminary anomaly candidate set, call the temperature difference between adjacent time points of the temperature parameter sequence in the log entry, and calculate the absolute value of the temperature change amount per unit time to generate a temperature change rate sequence; Sort the standard deviations of pressure for each log entry in descending order based on the standard deviation of pressure in the preliminary anomaly candidate set, calculate the sorting serial numbers, and generate a sorting sequence of the standard deviations of pressure; Merge the temperature change rate sequence and the sorting sequence of the standard deviations of pressure to generate a dynamic query parameter vector.

[0011] Preferably, the steps for obtaining the set of feedback weight coefficients are as follows: Traverse the operation logs of the user on the preliminary anomaly candidate set, extract the log entries that are clicked and marked as anomalies, generate a set of marked log entries, and record the click duration of each entry at the same time; Based on the set of marked log entries, calculate the mean value of the temperature change rate and the extreme value of the standard deviation of pressure , and the calculation formulas are: ; and ; wherein, is the temperature change rate of the th marked entry, is the standard deviation of pressure, is the click duration, is the maximum value of the click duration within the candidate set, is the total number of marked entries; According to the mean value of the temperature change rate and the extreme value of the standard deviation of pressure, traverse the unclicked log entries, and calculate the distance between the temperature change rate of the unclicked log entry and the mean value of the temperature change rate, and the ratio of the standard deviation of pressure to the extreme value of the standard deviation of pressure , and generate a feature difference degree

[0012] as the set of feedback weight coefficients. Preferably, the steps for obtaining the optimized result sequence are as follows: Based on the feature difference degree of each entry in the set of feedback weight coefficients, call the temperature change rate and the standard deviation of pressure of the corresponding entry in the dynamic query parameter vector, and linearly superimpose the feature difference degree, the temperature change rate, and the standard deviation of pressure according to a preset weight ratio to generate a set of superimposed parameters; Arrange the superimposed parameters in descending order according to the magnitude of each superimposed parameter in the set of superimposed parameters to generate an intermediate sorting sequence;

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, by calculating the absolute value of the difference between adjacent time points of temperature and pressure parameters in real time and extracting the number of sign changes of the second-order difference, an association code is generated by combining the time stamp and the device number, enhancing the spatio-temporal correlation of abnormal fluctuations, so that weak process deviations can be accurately captured at an early stage. Based on the absolute value of the temperature difference, an interval hash index is divided, and at the same time, a B+tree range index is constructed for the pressure standard deviation to realize multi-dimensional joint retrieval, so that two abnormal modes of high-frequency oscillation and steady-state offset can be screened independently or in combination, reducing the false detection rate under complex working conditions. After screening the candidate set by using the oscillation frequency threshold and the state coding sequence input by the user, a dynamic query vector is generated by combining the temperature change rate sorting and the extreme value of the pressure standard deviation, improving the differential response ability to sudden abnormalities and progressive faults. By introducing the user click duration and marked behavior data, the characteristic mean and extreme value of the verified abnormal entries are statistically calculated, the difference degree of the unclicked entries is calculated and the feedback weight is generated, and the sorting priority is reconstructed by superimposing the original query parameters, forming a closed-loop optimization mechanism, so that the retrieval result gradually fits the abnormal distribution law in the actual production scenario, reducing the cost of manual secondary screening. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0015] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0016] Please refer to Figure 1 , the present invention provides a technical solution: an online diagnostic retrieval system for PET processing abnormal logs includes: A differential preprocessing module, based on the temperature parameter sequence and pressure parameter sequence of the PET processing equipment, calculates the absolute value of the first-order difference between adjacent time points, extracts the number of sign changes of the second-order difference, generates a key change characteristic parameter set, and associates and codes the key change characteristic parameter set with the original log time stamp and device number to generate a differential association log sequence; An index dynamic construction module, based on the key change characteristic parameter set, establishes an interval hash index for the absolute value of the temperature first-order difference and a B+tree range index for the pressure standard deviation, maps the index and the differential association log sequence to generate a dynamic index address mapping table; The query feature mapping module, based on the abnormal oscillation frequency threshold and the state transition coding sequence input by the user, calls the dynamic index address mapping table to retrieve the log segments where the main oscillation frequency is greater than the abnormal oscillation frequency threshold, matches the log entries whose state transition paths conform to the preset coding, generates a preliminary abnormal candidate set, sorts according to the temperature change rate and pressure standard deviation of the entries in the preliminary abnormal candidate set, and generates a dynamic query parameter vector; The feedback-driven adjustment module, based on the click duration and marking behavior of the user on the entries in the preliminary abnormal candidate set, statistically calculates the average value of the temperature change rate and the extreme value of the pressure standard deviation of the marked entries, calculates the feature difference degree of the unclicked entries, generates a feedback weight coefficient set, superimposes the feedback weight coefficient set and the dynamic query parameter vector, re-sorts, and generates an optimized result sequence.

[0017] The steps for obtaining the key change feature parameter set are as follows: Based on the temperature parameter sequence and pressure parameter sequence collected in real time by the PET processing equipment, respectively call the temperature values at adjacent time points in the temperature parameter sequence and to calculate the absolute difference between adjacent time points of temperature and , and generate an absolute value sequence of the first-order difference of temperature; at the same time, call the pressure values at adjacent time points in the pressure parameter sequence and and to calculate the absolute difference between adjacent time points of pressure and , and generate an absolute value sequence of the first-order difference of pressure; Based on the absolute value sequence of the first-order difference of temperature and the absolute value sequence of the first-order difference of pressure, traverse each absolute value of the first-order difference of temperature, calculate the second-order difference symbol direction with the previous absolute value of the first-order difference of temperature. If the current second-order difference direction is inconsistent with the previous second-order difference direction , record a symbol change once, and cumulatively count the frequency of all consecutive three inconsistent symbol directions in the temperature parameter sequence to generate the number of second-order difference symbol changes of temperature and the number of second-order difference symbol changes of pressure; Merge the maximum value in the absolute value sequence of the first-order difference of temperature, the standard deviation of the absolute value sequence of the first-order difference of pressure, the number of second-order difference symbol changes of temperature, and the number of second-order difference symbol changes of pressure to generate a key change feature parameter set.

[0018] ​Specifically, based on the temperature parameter sequence and pressure parameter sequence collected in real time by the PET processing equipment in a specific time period, for example, from 10:00:00 to 10:10:00 on April 14, 2025, for example, the collection frequency is 1Hz, and 600 data points are obtained, of which some temperature data are shown in Table 1 and some pressure data are shown in Table 2. First, each time point in the temperature parameter sequence is called Temperature value and the previous time point Temperature value , perform a subtraction operation on these two values and take their absolute value, that is, calculate , for example, for the time point (10:00:01), the temperature is °C, previous time point The temperature at (10:00:00) is °C, then the absolute value of the first temperature first-order difference is calculated , for the time point (10:00:02), the temperature is °C, previous time point The temperature at (10:00:01) is °C, calculate the absolute value of the first-order difference of the second temperature , perform this operation on the entire temperature sequence [280.0, 280.5, 280.2, 280.8, 281.0, 280.7, ...] (600 points in total) to generate a temperature first-order difference absolute value sequence containing 599 values. For example, the sequence is initially [0.5, 0.3, 0.6, 0.2, 0.3, ...]. Then, process the pressure parameter sequence in the same way and call each time point in the pressure parameter sequence Pressure value and the previous time point Pressure value , perform subtraction and take the absolute value operation, calculate , for example, for the time point (10:00:01), the pressure is MPa, previous time point The pressure at (10:00:00) is MPa, calculate the absolute value of the first-order difference of the first pressure , for the time point (10:00:02), the pressure is MPa, previous time point The pressure at (10:00:01) is MPa, calculate the absolute value of the second pressure first-order difference This operation is then performed successively on the entire pressure sequence [5.0, 5.1, 5.05, 5.15, 5.2, 5.18,...] (a total of 600 points), generating a sequence of absolute values of the first-order differences of pressure containing 599 values. For example, the initial sequence is [0.1, 0.05, 0.1, 0.05, 0.02,...], generating a sequence of absolute values of the first-order differences of temperature and a sequence of absolute values of the first-order differences of pressure.

[0019] Based on the sequence of absolute values of the first-order differences of temperature generated in the previous paragraph, denoted as and the sequence of absolute values of the first-order differences of pressure, denoted as traverse each value in the sequence of absolute values of the first-order differences of temperature starting from since the first two values are required for comparison, calculate the difference between the current value and the previous value and determine the sign direction of this difference Meanwhile, obtain the sign direction of the second-order difference obtained in the previous calculation and compare whether these two sign directions are inconsistent, that is, determine whether holds. If it holds, record a sign change. For example, for the sequence calculate the second-order difference: the sign is positive (+1), the sign is negative (-1), the sign is positive (+1), and the sequence of signs of the second-order difference starts from as [+1, -1, +1,...]. Compare adjacent signs: at the current sign (-1) is inconsistent with the previous sign (+1), record a change, and at the current sign (+1) is inconsistent with the previous sign (-1), record a change. Perform this operation on the entire sequence of length 599 (from to , a total of 597 comparisons are made), accumulate the total number of recorded changes to obtain the total number of sign changes of the second-order difference of temperature. For example, after complete calculation, the total number is 85 times. Then, perform the exact same calculation process on the sequence of absolute values of the first-order differences of pressure calculate the second-order difference of pressure obtain its sign and compare with Whether it is inconsistent, the number of times that all second-order difference signs in the cumulative pressure sequence are inconsistent. For example, after complete calculation, the total number of times is 112 times, generating the number of times of change of the second-order difference sign of temperature (85 times) and the number of times of change of the second-order difference sign of pressure (112 times).

[0020] Find the maximum value in the generated absolute value sequence of the first-order difference of temperature By traversing all 599 elements in the sequence and comparing their sizes, the maximum value is finally determined. For example, if the maximum value in the sequence is 0.8, record this value , at the same time, for the generated absolute value sequence of the first-order difference of pressure Calculate its standard deviation. First, calculate the average value of this sequence , for example , for example, the calculated result is , then calculate the square of the difference between each element and the mean value , and then divide this sum by (where is the sequence length), and finally take the square root to get the standard deviation , for example, the calculated result is , then, call the calculated number of times of change of the second-order difference sign of temperature, the value of which is 85 times, and the number of times of change of the second-order difference sign of pressure, the value of which is 112 times. Combine these four numerical values: the maximum value (0.8) in the absolute value sequence of the first-order difference of temperature, the standard deviation (0.021) of the absolute value sequence of the first-order difference of pressure, the number of times of change of the second-order difference sign of temperature (85), and the number of times of change of the second-order difference sign of pressure (112) together to form a set or vector, generating the key change feature parameter set {0.8, 0.021, 85, 112}.

[0021] The steps for obtaining the differential correlation log sequence are as follows: Parse the timestamp and device number fields in the original log entries, call the unique identifier strings of the timestamp and device number, and generate the original log timestamp set and device number set; Based on the maximum value of the absolute value of the first-order difference of temperature, the standard deviation of the absolute value of the first-order difference of pressure, the number of times of change of the second-order difference sign of temperature, and the number of times of change of the second-order difference sign of pressure in the key change feature parameter set, splice them in a fixed order into a feature string, and perform string concatenation with the corresponding timestamp and device number identifier to generate the differential correlation log sequence.

[0022] Specifically, parse each original log entry generated by the PET processing equipment within a specific time period (e.g., from 10:00:00 to 10:10:00 on April 14, 2025). For example, a log is "2025-04-14 10:00:05, DevicePET001, T=281.0, P=5.2, State=Extruding". Extract the timestamp field "2025-04-14 10:00:05" and the device number field "DevicePET001" from it. Collect these extracted timestamp strings as unique identifiers to form a set, such as {"2025-04-14 10:00:00", "2025-04-14 10:00:01",..., "2025-04-14 10:09:59"}. Similarly, collect the extracted device number strings to form a set. If there is only one device, it is {"DevicePET001"}. If there are multiple devices, it includes all involved device numbers, generating an original log timestamp set and a device number set.

[0023] Based on the obtained set of key change feature parameters {0.8, 0.021, 85, 112}, which corresponds to a specific time period (e.g., from 10:00:00 to 10:10:00) or a specific log segment, extract the four numerical values: the maximum absolute value of the first-order difference of temperature , the standard deviation of the absolute value of the first-order difference of pressure , the number of sign changes of the second-order difference of temperature , the number of sign changes of the second-order difference of pressure , in a pre-set fixed order, such as , , , 's order. Convert these numerical values to strings and concatenate them with a specific delimiter (such as underscore "_") to form a feature string "0.8_0.021_85_112". Then, find the representative timestamp corresponding to the calculation period of this feature set, such as the start timestamp "2025-04-14 10:00:00" of this period, and the corresponding device number identifier "DevicePET001". Use another delimiter (such as hash "#") to concatenate the feature string, the timestamp string, and the device number string to generate a differential correlation log entry "0.8_0.021_85_112#2025-04-14 10:00:00#DevicePET001". Repeat this process for all sets of key change feature parameters calculated for the device in different time periods to generate a series of such correlation log entries, constituting a differential correlation log sequence.

[0024] The steps for obtaining the dynamic index address mapping table are as follows: Traverse the absolute value sequence of the first-order temperature difference in the key change feature parameter set, set a preset interval division rule, divide the absolute value of the first-order temperature difference into equal-width intervals according to the numerical range, and call a hash function to perform hash calculations on the start value and end value of each interval to generate a hash index for the absolute value interval of the first-order temperature difference; Based on the standard deviation of the absolute value of the first-order pressure difference in the key change feature parameter set, sort the standard deviations in ascending order, use the standard deviation as the key value and the corresponding log entry address as the leaf node pointer to generate a range index for the pressure standard deviation B+ tree; Combine and map the hash key of the hash index for the absolute value interval of the first-order temperature difference with the key value range of the range index of the pressure standard deviation B+ tree, establish a two-way pointer association relationship between the hash key and the B+ tree key value and the storage address of the differential correlation log sequence, and generate a dynamic index address mapping table.

[0025] Specifically, traverse the entire obtained absolute value sequence of the first-order temperature difference , set a preset interval division rule, which is determined based on the analysis of the normal temperature fluctuation range during the PET processing process. For example, set equal-width intervals with a width of 0.1 °C / s (for example, the difference value represents the change rate per unit time), then the divided intervals are [0, 0.1), [0.1, 0.2), [0.2, 0.3), [0.3, 0.4), [0.4, 0.5), [0.5, 0.6), [0.6, 0.7), [0.7, 0.8), [0.8, 0.9),..., for each divided interval, such as the interval [0.5, 0.6), call a hash function, which takes the start value 0.5 and end value 0.6 of the interval as inputs and performs hash calculations. For example, use a simple hash function , then for the interval [0.5, 0.6), the hash value is , perform this hash calculation for all preset intervals to generate a series of hash keys. At the same time, map each value in the original sequence to the interval it belongs to and its corresponding hash key. For example , the values 0.5, 0.55 (for example, existing) in the sequence belong to the interval [0.5, 0.6), and its hash key is 506. Organize the storage addresses (or indexes) of the differential correlation log entries with the same hash key together, such as {... 506: [address1, address5,...],...}, to generate a hash index for the absolute value interval of the first-order temperature difference Based on the calculated standard deviation of the absolute value of the first-order pressure difference included in the key change feature parameter set numerical values, for example, there is a series of Values: [0.021, 0.035, 0.018, 0.025, 0.040, 0.021,...]. Arrange these standard deviation values in ascending order to obtain a sorted sequence, such as [0.018, 0.021, 0.021, 0.025, 0.035, 0.040,...]. Construct a B+ tree index structure. Use these sorted pressure standard deviation values as the keys (key) of the B+ tree. Each key corresponds to a pointer in the leaf node, and this pointer points to the address in the storage system of the differential correlation log entry that calculated this value (or its index in the differential correlation log sequence). For example, the leaf node pointer corresponding to the key value 0.018 points to the address Addr3, the pointer corresponding to the key value 0.021 points to the addresses Addr1 and Addr6 (if there are duplicate values), the pointer corresponding to the key value 0.025 points to the address Addr4, and so on. The internal nodes of the B+ tree store the range information of the key values for efficient range queries. For example, to find all log entry addresses within the range [0.020, 0.030], generate a B+ tree range index for the pressure standard deviation.

[0026] Combine the generated hash index for the absolute value interval of the first-order difference of temperature (for example, the hash key 506 corresponds to the address list [address1, address5]) with the generated B+ tree range index for the pressure standard deviation (for example, the address list [Addr1, Addr4, Addr6] corresponding to the key value range [0.020, 0.030]) to establish an association mechanism, so that the storage address of the differential correlation log sequence that simultaneously meets these two conditions can be quickly located through a hash key of a temperature interval and a range of pressure standard deviations. Specifically in implementation, a higher-level data structure can be designed. This structure allows input of a hash key (representing a temperature change range) and a key value range (representing a range of pressure standard deviations), and then obtains the set of addresses that meet the temperature condition by querying the hash index , and then obtains the set of addresses that meet the pressure condition by querying the B+ tree index , and finally calculates the intersection of these two address sets , to obtain the final result. At the same time, a reverse association is also established, that is, given an address in the differential correlation log sequence, its corresponding temperature hash key and pressure standard deviation key value can be quickly found. This two-way pointer association relationship makes it efficient to query the log from the features or to reverse query the features from the log, and generates a dynamic index address mapping table.

[0027] The steps to obtain the preliminary abnormal candidate set are as follows: Parse the abnormal oscillation frequency threshold and state transition encoding sequence of the user input, traverse the temperature first-order difference absolute value interval hash index key values in the dynamic index address mapping table, and extract the set of log segment addresses where the maximum value of the temperature first-order difference absolute value exceeds the preset temperature fluctuation intensity threshold; Based on the set of log segment addresses, calculate the spectral main frequency of the temperature parameter sequence in each log segment, and filter out the log segments where the spectral main frequency is greater than the abnormal oscillation frequency threshold. The calculation formula is: ; where, is the th temperature parameter value in the log segment, is the log segment length, is the frequency component, is the spectral main frequency; According to the log segments where the spectral main frequency is greater than the abnormal oscillation frequency threshold, filter out the entries whose state transition paths are consistent with the user input encoding sequence, and generate a preliminary abnormal candidate set.

[0028] Specifically, parse the abnormal screening conditions input by the user, including the abnormal oscillation frequency threshold and the state transition encoding sequence. The setting of the abnormal oscillation frequency threshold refers to the temperature spectral characteristics during the normal operation of the PET processing equipment and the characteristic frequencies under typical fault modes. For example, through historical data analysis, the main temperature frequency during normal operation is usually lower than 0.05 Hz, while a certain specific extrusion fault may cause the temperature to oscillate at a frequency of about 0.1 Hz. Therefore, the abnormal oscillation frequency threshold is set to Hz, and the state transition encoding sequence is defined by the process engineer according to experience, representing an abnormal working condition conversion mode, such as "Stabilizing->Extruding_HighPressure->Cooling_Fast". At the same time, the user also needs to input a temperature fluctuation intensity threshold, which is set based on the allowed maximum temperature change rate. For example, it is set to °C / s. First, traverse the temperature first-order difference absolute value interval hash index part in the established dynamic index address mapping table, query the temperature intervals corresponding to all hash keys, and filter out the hash keys corresponding to the intervals whose upper limit value or representative value (such as the midpoint or maximum value of the interval) is greater than the preset temperature fluctuation intensity threshold °C / s. For example, if the hash key for the interval [1.2, 1.3) is 1213 and the hash key for the interval [1.3, 1.4) is 1314, then extract the addresses of all differential-associated log segments associated with these hash keys (and higher interval hash keys) to form a preliminary address set , such as .

[0029] For the formula : This formula is used to calculate the main frequency of the temperature time series within the log segment. .

[0030] Parameter description: : Represents the temperature value at the th sampling time point within the log segment, in °C. The subscript represents the sample index in the time series.

[0031] : Represents the length of the log segment, i.e., the total number of sample points it contains. This length is set according to the analysis requirements, for example, covering a complete processing cycle or a concerned time window, such as sample points.

[0032] : Represents the frequency component, in Hz. The formula needs to calculate the spectral amplitude corresponding to different frequencies .

[0033] : Represents the imaginary unit, i.e., .

[0034] : Represents the summation of the time series index from 1 to .

[0035] : Is the core part of the discrete Fourier transform (DFT), a complex exponential function, used to project the time-domain signal onto the frequency .

[0036] : Calculates the magnitude of the result (a complex number) of the DFT at the frequency , representing the amplitude or energy of that frequency component.

[0037] : Represents finding the frequency that makes the following expression (i.e., the spectral amplitude) reach the maximum value. This is the main frequency of the temperature series of the log segment .

[0038] Operation logic: This formula essentially performs a discrete Fourier transform to calculate the amplitude of the temperature signal at different frequencies . It projects the signal Dot - multiply and sum with a series of complex - exponential basis functions of different frequencies to decompose the frequency components of the signal. The modulus operation is used to obtain the intensity of each frequency component. The operation then finds the frequency with the maximum intensity, that is, the frequency component with the most concentrated energy in the signal, also known as the dominant frequency.

[0039] Parameter acquisition and calculation example: For example, select a log - segment address Addr25 from the address set . The corresponding original temperature sequence within seconds (i.e., sample points) has the data part: [285.0, 285.5, 286.0, 285.6, 285.1,..., 285.8]. Here, is obtained by consulting the original log or the stored time - series database.

[0040] is the pre - set segment length according to the analysis requirements.

[0041] It is necessary to calculate the amplitudes at different frequencies . The frequency resolution is usually . If the sampling frequency is 1 Hz and the sampling period is 1 second, the frequency range is from 0 to 0.5 Hz (Nyquist frequency), and a series of frequency points can be tested, such as Hz.

[0042] Taking the calculation of the amplitude at Hz as an example (simplifying the calculation and only showing the first few items schematically): Amplitude ≈ Amplitude ≈ For example, through complete calculation (usually using the fast Fourier transform FFT algorithm), the amplitudes at different frequencies are obtained, such as: f = 0.05 Hz, amplitude = 15.2 f = 0.10 Hz, amplitude = 35.8 f = 0.15 Hz, amplitude = 12.1 ... Comparing all the calculated amplitudes, it is found that Hz has the maximum amplitude (35.8), so the dominant frequency of this log segment is

[0043] Then, based on the address set obtained in the previous step , for For each log segment corresponding to an address, calculate the spectral dominant frequency of its temperature parameter sequence , using the formula and calculation method described above, for example, calculate the dominant frequency of Addr10 Hz, the dominant frequency of Addr25 Hz, the dominant frequency of Addr50 Hz. Compare each calculated dominant frequency with the preset abnormal oscillation frequency threshold Hz, and filter out log segments. In this example, Addr25( ) and Addr50( ) are retained, while Addr10( ) is excluded to form a new set of addresses . Then, according to the state transition coding sequence input by the user, such as "Stabilizing->Extruding_HighPressure->Cooling_Fast", check the device state sequence recorded in each log segment corresponding to the addresses in the set . For example, the state sequence in the log segment corresponding to Addr25 is [..., "Stabilizing", "Extruding_HighPressure", "Cooling_Fast",...]. If this sequence is consistent with the coding sequence input by the user, then retain Addr25. For example, the state sequence corresponding to Addr50 is [..., "Heating", "Stabilizing", "Extruding_Normal",...], which is inconsistent with the user input, so exclude Addr50. Collect all the log segment addresses that pass the state sequence matching check to generate a preliminary abnormal candidate set. For example, the final candidate set is {Addr25,...}.

[0044] The benefit of the formula is that by calculating the spectral dominant frequency of the temperature signal , the periodic characteristics of temperature changes can be quantified, especially identifying those working conditions with abnormal high-frequency oscillations. Such oscillations may be related to equipment instability, control disorders, or specific fault modes, and it may be difficult to capture this periodic behavior only by first-order or second-order differences.

[0045] This result Hz indicates that in the log segment corresponding to Addr25, there is a significant periodic fluctuation of temperature with a frequency of 0.10 Hz. This frequency value Hz is greater than the set threshold Hz, meaning that the temperature fluctuation frequency of this log segment is considered abnormal, so this log segment is retained for subsequent state sequence checks.

[0046] The steps for obtaining the dynamic query parameter vector are as follows: Traverse each log entry in the preliminary anomaly candidate set, call the temperature difference between adjacent time points of the temperature parameter sequence within the log entry, calculate the absolute value of the temperature change per unit time, and generate a temperature change rate sequence; Based on the pressure standard deviation of the preliminary anomaly candidate set, sort the pressure standard deviations of each log entry in descending order, calculate the sorting serial number, and generate a pressure standard deviation sorting sequence. The calculation formula is: ; Wherein, is the pressure standard deviation of the th entry, is the average value of the pressure standard deviations within the candidate set, is the standard deviation change step, is the sorting serial number; Merge the temperature change rate sequence and the pressure standard deviation sorting sequence to generate a dynamic query parameter vector.

[0047] Specifically, traverse the generated preliminary anomaly candidate set, for example, it contains log entry addresses {Addr25, Addr68,...}. For each log entry address in the candidate set, such as Addr25, call the temperature parameter sequence recorded in its corresponding original log entry or associated data , for example, the sequence of Addr25 is [285.0, 285.5, 286.0, 285.6, 285.1,..., 285.8]. Calculate the absolute value of the temperature difference between adjacent time points, that is , to obtain the temperature change rate sequence of this entry, for example, the sequence is [|285.5 - 285.0| = 0.5, |286.0 - 285.5| = 0.5, |285.6 - 286.0| = 0.4, |285.1 - 285.6| = 0.5,..., |285.8 - Tn-1|]). Perform this operation on all entries in the preliminary anomaly candidate set to generate a temperature change rate sequence for each entry.

[0048] For the formula : This formula is used to calculate the pressure standard deviation sorting serial number of the th preliminary anomaly candidate log entry .

[0049] Parameter description: : is the calculated pressure standard deviation sorting serial number (or rank, grading) of the th candidate entry.

[0050] : It represents the standard deviation of the absolute value sequence of the first-order difference of the pressure of the th candidate entry, with the unit of MPa.

[0051] : It represents the mean value of the pressure standard deviations of all entries in the preliminary abnormal candidate set in MPa.

[0052] : It represents the step size of the standard deviation change, which is a preset constant used to determine the granularity of sorting or grading, with the unit of MPa. For example, if it is considered that a change of 0.005 MPa in the pressure standard deviation represents a significant hierarchical difference, then MPa can be set.

[0053] : It represents the absolute value operation.

[0054] : It represents the floor function (downward rounding operation).

[0055] : It represents the sign function, which outputs +1 when the input value is greater than 0, -1 when it is less than 0, and 0 when it is equal to 0.

[0056] Operation logic: First, calculate the difference between the of the th entry and the average value of the candidate set , and determine its sign (positive or negative) to obtain . Then calculate the absolute value of this difference and divide it by the step size , which represents how many "step sizes" the of this entry deviates from the average value. Round down this ratio to obtain the integer number of steps of deviation. Finally, multiply this integer number of steps by the sign obtained previously. The result is an integer representing the th entry's deviation level from the average value, where a positive number indicates higher than the average value, a negative number indicates lower than the average value, and 0 indicates close to the average value.

[0057] Parameter acquisition and calculation example: For example, the preliminary abnormal candidate set contains 3 entries, and their corresponding pressure standard deviations are respectively: MPa (corresponding to Addr25), MPa (corresponding to Addr68), MPa (corresponding to Addr90). These values are extracted from the key feature parameter set or recalculated here for the candidate segments.

[0058] First, calculate the mean value within the candidate set : MPa.

[0059] Set the step size of the standard deviation change . Referring to historical data or expert experience, it is considered that the pressure standard deviation fluctuation exceeding 0.005 MPa is considered a significant level, so set MPa.

[0060] Now calculate the for each entry: For entry 1 ( , Addr25, ): ; ; ; ; For entry 2 ( , Addr68, ): ; ; ; ; For entry 3 ( , Addr90, ): ; ; ; ; In this way, the pressure standard deviation sorting sequence is obtained.

[0061] Then, based on the pressure standard deviation of the preliminary abnormal candidate set (for example, [0.035, 0.048, 0.025]), its descending order is [0.048, 0.035, 0.025], but the formula is to calculate the sorting serial number , so use the As the pressure standard deviation sorting sequence (here, "sorting" refers to the binning number based on the distance from the mean), generate the pressure standard deviation sorting sequence [0, 2, -2]. Finally, merge the temperature change rate sequence corresponding to each candidate entry with the calculated pressure standard deviation sorting number , for example, for Addr25, if the maximum value of its temperature change rate sequence is 0.5 °C / s, its , then a part of its vector is [0.5, 0]. For Addr68, if its maximum rate is 0.7 °C / s, its , then a part of its vector is [0.7, 2]. For Addr90, if its maximum rate is 0.6 °C / s, its , then a part of its vector is [0.6, -2], and generate a set of dynamic query parameter vectors containing this combined information.

[0062] The benefit of the formula is that 's calculation provides a method to discretize and standardize the continuous pressure standard deviation . It not only considers the 's absolute magnitude, but also pays more attention to its deviation degree and direction relative to the overall fluctuation level of the candidate set (represented by ), and quantifies it in units of step size , so that different magnitudes of values can be converted into comparable integer ranks , which is convenient for subsequent model processing or rule formulation.

[0063] This result indicates that in the candidate set, the pressure fluctuation of Addr25 is close to the average level (rank 0), the pressure fluctuation of Addr68 is significantly higher than the average level (rank +2), and the pressure fluctuation of Addr90 is significantly lower than the average level (rank -2). Each rank represents a multiple of approximately . These rank information constitute a part of the dynamic query parameter vector for subsequent weighted sorting.

[0064] The steps to obtain the feedback weight coefficient set are as follows: Traverse the operation logs of the user on the preliminary abnormal candidate set, extract the logged entries that are clicked and marked as abnormal, generate a set of logged entries that are marked, and record the click duration of each entry; Based on the set of logged entries that are marked, calculate the mean value of the temperature change rate and the extreme value of the pressure standard deviation respectively, and the calculation formula is: ; and ; wherein, is the temperature change rate of the th marked entry, is the standard deviation of pressure, is the click duration, is the maximum value of the click duration within the candidate set, is the total number of marked entries; According to the mean value of the temperature change rate and the extreme value of the standard deviation of pressure, traverse the unclicked log entries, and calculate the distance between the temperature change rate of the unclicked log entries and the mean value of the temperature change rate , and the ratio of the standard deviation of pressure to the extreme value of the standard deviation of pressure , to generate the feature difference degree , as the feedback weight coefficient set.

[0065] Specifically, traverse the operation logs generated by the user when reviewing the preliminary anomaly candidate set (such as including Addr25, Addr68, Addr90 and other unclicked entries Addr100, Addr105), identify those log entries that are clicked by the user to view details and are finally marked as "anomaly", for example, if the user clicks and marks Addr25 and Addr68 as anomalies, then generate the set of marked log entries {Addr25, Addr68}, and at the same time, record the click duration of the user viewing each marked entry. For example, the click duration of Addr25 seconds, and the click duration of Addr68 seconds.

[0066] For the formulas and : These two formulas are used to calculate the feedback reference benchmark based on the anomaly entries marked by the user.

[0067] The first formula calculates the average temperature change rate of the marked anomaly entries .

[0068] Parameter description: : The calculated average temperature change rate of the marked anomaly entries, unit °C / s.

[0069] : The total number of log entries marked as anomalies.

[0070] : The representative temperature change rate of the th marked anomaly entry, unit °C / s. This value is usually a certain statistic of the temperature change rate sequence of this entry, such as the average value or the maximum value. Here, for example, the average value is used. Subscript Identify the marked entries.

[0071] : Indicates the sum of all marked abnormal entries.

[0072] Operation logic: Simply calculate the rate of change of temperature for all marked abnormal entries and obtain the arithmetic mean to get a typical rate of temperature fluctuation representing the abnormal situation confirmed by the user .

[0073] The second formula calculates the extreme value of the weighted standard deviation of pressure for the marked abnormal entries .

[0074] Parameter description: : The extreme value of the weighted standard deviation of pressure obtained by calculation, unit MPa.

[0075] : The standard deviation of pressure for the

[0076] th marked abnormal entry, unit MPa. : The duration in seconds when the user clicks to view the

[0077] th marked abnormal entry.

[0078] : The maximum click duration in seconds among all marked abnormal entries (or sometimes among all candidate entries). : Calculate the normalized ratio of the click duration of the

[0079] th entry to the maximum duration, with a value range between [0, 1]. : Calculate the standard deviation of pressure for the

[0080] th entry and weight it according to the user's attention duration. The longer the attention time, the higher the weight.

[0081] Operation logic: This formula attempts to find an upper limit or representative value of the pressure instability reflected by the abnormal entry that the user is most concerned about (the longest click time) and has the largest pressure fluctuation ( high). It weights by duration and then takes the maximum value to capture the intensity of the pressure anomaly features that the user focuses on.

[0082] Parameter acquisition and calculation example: The set of marked log entries is {Addr25, Addr68}, so .

[0083] It is necessary to obtain the and of these entries: Addr25: MPa. For example, its average temperature change rate °C / s. Click duration s.

[0084] Addr68: MPa. For example, its average temperature change rate °C / s. Click duration s.

[0085] Calculate : °C / s.

[0086] Calculate : First, determine . Among the marked entries, s.

[0087] Calculate the weighted of each marked entry: For Addr25: MPa.

[0088] For Addr68: MPa.

[0089] Take the maximum value among them: MPa.

[0090] Then, based on the calculated average temperature change rate °C / s and the extreme value of pressure standard deviation MPa, traverse the log entries in the preliminary abnormal candidate set that have not been clicked by the user, such as Addr90, Addr100, Addr105. For each unclicked entry , obtain its temperature change rate (for example °C / s) and pressure standard deviation (for example MPa), and calculate the distance between the temperature change rate of the unclicked entry and the average rate of the marked abnormal rate For Addr90, , then calculate the ratio of its pressure standard deviation to the extreme value of the standard deviation of the marked abnormal pressure , for Addr90 , add these two values to generate the feature difference degree , for Addr90 , repeat this calculation for all unclicked entries (Addr100, Addr105, etc.) to obtain their feature difference degrees , collect the feature difference degrees of all unclicked entries to generate a feedback weight coefficient set {Addr90: 0.571, Addr100: , Addr105: ,...}.

[0091] The benefits of the formula are as follows: and learn the feature center (average temperature rate) and feature boundary (weighted maximum pressure fluctuation) of the abnormal pattern from the actually marked abnormalities by the user, while the subsequent calculated quantifies the similarity / difference degree between other unmarked (possibly ignored or unnoticed) candidate entries and this confirmed abnormal pattern The smaller it is, the closer the entry is to the abnormal pattern confirmed by the user in terms of temperature rate and pressure fluctuation, and it may also be a potential abnormality

[0092] This result °C / s and MPa become the benchmarks for evaluating other unclicked entries. The calculated represents the degree of feature difference between Addr90 and the abnormal pattern confirmed by the user, and this value will be used as the feedback weight for subsequent optimization sorting

[0093] The steps to obtain the optimized result sequence are as follows: Based on the feature difference degree of each entry in the feedback weight coefficient set, call the temperature change rate and pressure standard deviation of the corresponding entry in the dynamic query parameter vector, and linearly superimpose the feature difference degree with the temperature change rate and pressure standard deviation according to the preset weight ratio to generate a set of superimposed parameters According to the magnitude of each superimposed parameter in the set of superimposed parameters, sort the superimposed parameters in descending order to generate an intermediate sorting sequence Traverse each entry index in the intermediate sorting sequence, call the hash address mapping table of the differential correlation log sequence, and map the sorted entry index back to the original log storage location to generate the optimized result sequence

[0094] Specifically, based on the feedback weight coefficient set, which contains each unclicked entry Feature difference degree (for example , and for example, calculate ), and combine with the information of the corresponding entries in the generated dynamic query parameter vector, specifically, the temperature change rate of each entry (for example (for example °C / s) and the pressure standard deviation (for example MPa) (Note: What is used here is instead of the previously calculated rank . It is necessary to confirm the consistency of the data source or perform conversion. Here, use the original text ), set the preset weight ratio. These weights reflect the importance of the feature difference degree (user feedback), the temperature change rate (inherent feature), and the pressure standard deviation (inherent feature) in the final ranking. The setting of the weights is based on expert experience or experimental optimization. For example, if more emphasis is placed on user feedback, it can be set , and , for each entry (including the previously marked entries, its can be set or other processing methods can be adopted. Here, only the unclicked entries are calculated) perform a linear superposition calculation to generate the superposition parameter Score , for example, calculate the superposition parameters of Addr90, Addr100, Addr105: Score ; Score ; Score ; Collect all the calculated superposition parameter values to form a superposition parameter set {Addr90: 0.4576, Addr100: 0.576, Addr105: 0.249,...}.

[0095] According to the superposition parameter set {Addr90: 0.4576, Addr100: 0.576, Addr105: 0.249,...} generated in the previous step and the scores calculated for the marked entries (Addr25, Addr68) according to the rules (for example, set , Score , Score ), sort the superimposed parameters of all these entries in descending order to obtain the sorting result: Addr100(0.576), Addr90(0.4576), Addr105(0.249), Addr68(0.1296), Addr25(0.087). This sorting reflects the anomaly possibility ranking after combining user feedback and original features, generating an intermediate sorting sequence (by address / index): [Addr100, Addr90, Addr105, Addr68, Addr25,...]. Then, traverse each entry index in this intermediate sorting sequence (such as Addr100, Addr90,...), using the established dynamic index address mapping table or a lookup mechanism that can map the entry index back to its specific location (address) in the original log storage (or differential correlation log sequence storage). For example, Addr100 is mapped to storage address 0xAddress100, Addr90 is mapped to 0xAddress90, Addr105 is mapped to 0xAddress105, Addr68 is mapped to 0xAddress68, Addr25 is mapped to 0xAddress25. Collect these storage addresses arranged in the order of the intermediate sorting sequence, [0xAddress100, 0xAddress90, 0xAddress105, 0xAddress68, 0xAddress25,...], to generate the final optimized result sequence.

Claims

1. An online diagnostic retrieval system for PET processing exception logs, characterized in that, The system includes: A differential preprocessing module, which calculates the absolute value of the first-order difference between adjacent time points based on the temperature parameter sequence and the pressure parameter sequence of the PET processing equipment, extracts the number of sign changes in the second-order difference, generates a key change feature parameter set, and associates and encodes the key change feature parameter set with the original log timestamp and the equipment number to generate a differential association log sequence; An index dynamic construction module, which based on the key change feature parameter set, establishes an interval hash index for the absolute value of the temperature first-order difference and a B+ tree range index for the pressure standard deviation, maps the index and the differential association log sequence to generate a dynamic index address mapping table; A query feature mapping module, which based on the abnormal oscillation frequency threshold and the state transition coding sequence input by the user, calls the dynamic index address mapping table to retrieve the log segments that meet the condition that the main oscillation frequency is greater than the abnormal oscillation frequency threshold, matches the log entries whose state transition paths conform to the preset coding, generates a preliminary abnormal candidate set, and sorts according to the temperature change rate and the pressure standard deviation of the entries in the preliminary abnormal candidate set to generate a dynamic query parameter vector; A feedback-driven adjustment module, which based on the click duration and marking behavior of the user on the entries in the preliminary abnormal candidate set, statistically calculates the mean value of the temperature change rate and the extreme value of the pressure standard deviation of the marked entries, calculates the feature difference degree of the unclicked entries, generates a feedback weight coefficient set, superimposes the feedback weight coefficient set and the dynamic query parameter vector, and re-sorts to generate an optimized result sequence.

2. The online diagnostic retrieval system for PET processing exception logs according to claim 1, wherein The steps for obtaining the key change feature parameter set are as follows: Based on the temperature parameter sequence and pressure parameter sequence collected in real time by the PET processing equipment, the adjacent time points in the temperature parameter sequence are called respectively. and Temperature value and , calculate the absolute difference of temperature at adjacent time points , generate the absolute value sequence of the first-order difference of temperature; at the same time, call the adjacent time points in the pressure parameter sequence and Pressure value and , calculate the absolute difference of pressure at adjacent time points , generate the absolute value sequence of the first-order difference of pressure; Based on the absolute value sequence of the first-order difference of temperature and the absolute value sequence of the first-order difference of pressure, traverse each absolute value of the first-order difference of temperature, calculate the second-order difference sign direction with the previous absolute value of the first-order difference of temperature. If the current second-order difference direction is inconsistent with the previous second-order difference direction record a sign change once, accumulate and count the frequency of all consecutive three inconsistent sign directions in the temperature parameter sequence, and generate the number of second-order difference sign changes of temperature and the number of second-order difference sign changes of pressure; Combine the maximum value in the absolute value sequence of the temperature first-order difference, the standard deviation of the absolute value sequence of the pressure first-order difference, the number of sign changes in the temperature second-order difference, and the number of sign changes in the pressure second-order difference to generate a key change feature parameter set.

3. The online diagnostic retrieval system for PET processing exception logs according to claim 1, characterized in that The steps for obtaining the differential association log sequence are as follows: Parse the timestamp and equipment number fields in the original log entry, call the unique identifier strings of the timestamp and the equipment number to generate an original log timestamp set and an equipment number set; Based on the maximum value of the absolute value of the temperature first-order difference, the standard deviation of the absolute value of the pressure first-order difference, the number of sign changes in the temperature second-order difference, and the number of sign changes in the pressure second-order difference in the key change feature parameter set, splice them in a fixed order into a feature string, and concatenate the feature string with the corresponding timestamp and equipment number identifier to generate a differential association log sequence.

4. The online diagnostic retrieval system for PET processing exception logs according to claim 1, characterized in that, The steps for obtaining the dynamic index address mapping table are as follows: Traverse the absolute value sequence of the temperature first-order difference in the key change feature parameter set, set a preset interval division rule, divide the absolute value of the temperature first-order difference into equal-width intervals according to the numerical range, and call the hash function to calculate the hash values of the start value and the end value of each interval to generate an interval hash index for the absolute value of the temperature first-order difference; Based on the standard deviation of the absolute value of the pressure first-order difference in the key change feature parameter set, sort the standard deviations in ascending order, and generate a B+ tree range index for the pressure standard deviation with the standard deviation as the key value and the corresponding log entry address as the leaf node pointer; Combine the hash key of the absolute value interval hash index of the first-order temperature difference with the key value range of the pressure standard deviation B+ tree range index to establish a two-way pointer association relationship between the hash key and the B+ tree key value and the storage address of the differential correlation log sequence, and generate a dynamic index address mapping table.

5. The online diagnostic retrieval system for PET processing exception logs according to claim 1, wherein The steps for obtaining the preliminary anomaly candidate set are as follows: Parse the anomaly oscillation frequency threshold and the state transition coding sequence input by the user, traverse the hash index key values of the absolute value interval of the first-order temperature difference in the dynamic index address mapping table, and extract the set of log segment addresses where the maximum value of the absolute value of the first-order temperature difference exceeds the preset temperature fluctuation intensity threshold; Based on the set of log segment addresses, calculate the dominant frequency of the spectrum of the temperature parameter sequence in each log segment, and filter out the log segments whose dominant frequency of the spectrum is greater than the anomaly oscillation frequency threshold; According to the log segments whose dominant frequency of the spectrum is greater than the anomaly oscillation frequency threshold, filter out the entries whose state transition paths are consistent with the user-input coding sequence to generate a preliminary anomaly candidate set.

6. The online diagnostic retrieval system for PET processing exception logs according to claim 1, wherein The steps for obtaining the dynamic query parameter vector are as follows: Traverse each log entry in the preliminary anomaly candidate set, call the temperature difference between adjacent time points of the temperature parameter sequence in the log entry, and calculate the absolute value of the temperature change per unit time to generate a temperature change rate sequence; Based on the pressure standard deviation of the preliminary anomaly candidate set, sort the pressure standard deviations of each log entry in descending order, calculate the sorting sequence number, and generate a pressure standard deviation sorting sequence; Merge the temperature change rate sequence and the pressure standard deviation sorting sequence to generate a dynamic query parameter vector.

7. The online diagnostic retrieval system for PET processing abnormal logs according to claim 1, characterized in that, The steps for obtaining the feedback weight coefficient set are as follows: Traverse the operation log of the user on the preliminary anomaly candidate set, extract the log entries that are clicked and marked as anomalies to generate a set of marked log entries, and record the click duration of each entry; Calculate the mean value of the temperature change rate and the extreme value of the pressure standard deviation respectively based on the set of the marked log entries and the extreme value of the pressure standard deviation , and the calculation formula is as follows: ； and ； Among them, is the temperature change rate of the th marked entry, is the pressure standard deviation, is the click duration, is the maximum value of the click duration in the candidate set, is the total number of marked entries; Traverse the unclicked log entries according to the mean value of the temperature change rate and the extreme value of the pressure standard deviation, and calculate the distance between the temperature change rate of the unclicked log entry and the mean value of the temperature change rate , and the ratio of the pressure standard deviation to the extreme value of the pressure standard deviation , and generate a feature difference degree , which is used as a feedback weight coefficient set.

8. The online diagnostic retrieval system for PET processing abnormal logs according to claim 1, characterized in that The steps for obtaining the optimized result sequence are as follows: Based on the feature difference degree of each entry in the feedback weight coefficient set, call the temperature change rate and pressure standard deviation of the corresponding entry in the dynamic query parameter vector, and linearly superimpose the feature difference degree with the temperature change rate and pressure standard deviation according to the preset weight ratio to generate a set of superimposed parameters; Arrange the superimposed parameters in descending order according to the size of each superimposed parameter in the set of superimposed parameters to generate an intermediate sorting sequence; Traverse each entry index in the intermediate sorting sequence, call the hash address mapping table of the differential correlation log sequence, and map the sorted entry index back to the original log storage location to generate an optimized result sequence.

Images