Grain Drying Tower Operation Status Data Processing Method and System

Abstract

This invention relates to the field of intelligent monitoring and industrial data processing of grain drying equipment, specifically a method and system for processing operational status data of grain drying towers. The method includes: acquiring operational status data from various zones and multiple areas, material flow data, equipment operation data, and environmental data to generate real-time status data; reconstructing a theoretical baseline matrix based on the mass conservation rule of grain dry matter and the energy conservation rule based on enthalpy-humidity balance; injecting damaged states based on fault mechanism parameterization rules corresponding to multiple fault categories to obtain multiple theoretical damaged matrices; constructing a real residual matrix and a theoretical residual matrix, calculating topological similarity by combining temporal morphological distance and spatial structural distance, determining the status assessment result, and updating the fault mechanism parameterization rules and dynamic judgment threshold groups. This invention effectively distinguishes between environmental fluctuations and hidden faults, improving the stability and interpretability of status assessment.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and industrial data processing of grain drying equipment, specifically to a method and system for processing operational status data of grain drying towers. Background Technology

[0002] Existing grain drying towers typically involve multiple zones such as drying and cooling sections during operation, and are equipped with devices for temperature, humidity, air pressure, grain inlet and outlet detection, heat source supply, and environmental monitoring to monitor the status and control the operation of the drying process.

[0003] In related technologies, in order to determine whether there is an operational abnormality in a grain drying tower, the real-time detection values ​​of sensors in each zone can be compared with thresholds, or historical experience curves, individual equipment parameters and manual inspection results can be combined to analyze temperature fluctuations, humidity changes, fan conditions and heat source conditions, thereby determining whether there is a fault or abnormal operating condition.

[0004] However, under conditions of fluctuating grain moisture content, sudden changes in ambient temperature and humidity, and strong coupling of heat and mass exchange in multiple zones, the former method is prone to misjudging normal operating disturbances as equipment failures, while the latter method does not make sufficient use of multi-source data correlation and has low real-time performance and accuracy in fault identification. Therefore, the grain drying tower operation status processing methods in related technologies are difficult to balance the ability to distinguish normal fluctuations with the ability to identify hidden faults. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for processing operational status data of grain drying towers, avoiding misjudging normal operating disturbances as equipment failures under conditions of fluctuating grain moisture content, sudden changes in ambient temperature and humidity, and strong coupling of heat and mass exchange in multiple zones. It balances the ability to distinguish normal fluctuations with the ability to identify hidden faults, thereby simultaneously improving the accuracy of fault identification and environmental adaptability. Specifically, the technical solution of this invention is as follows:

[0006] Methods for processing operational status data of grain drying towers include:

[0007] Acquire multi-zone operation status data, material flow data, equipment operation data and environmental data of each zone of the grain drying tower, and generate real-time status data, which includes temperature value of each zone, humidity value of each zone, moisture value of ingested grain, moisture value of outgested grain, fan speed, fuel consumption, ambient temperature and ambient humidity;

[0008] Based on the conservation of dry matter mass and the conservation of enthalpy-humidity balance energy, the real-time state data is reconstructed into a theoretical normal state to obtain a theoretical baseline matrix. The theoretical baseline matrix includes the theoretical temperature and theoretical humidity values ​​corresponding to each time point and each region.

[0009] Based on the preset fault mechanism parameterization rules, the theoretical benchmark matrix is ​​subjected to damaged state injection processing to obtain the theoretical damaged matrix. The real residual matrix is ​​generated based on the real-time state data and the theoretical benchmark matrix, and the theoretical residual matrix is ​​generated based on the theoretical damaged matrix and the theoretical benchmark matrix.

[0010] Based on the real residual matrix and the theoretical residual matrix, temporal morphological features and spatial structural features are extracted respectively, and the temporal morphological distance and spatial structural distance are calculated and fused to obtain the topological similarity, and the state assessment result is determined accordingly.

[0011] Based on the condition assessment results, update the fault mechanism parameterization rules and dynamic judgment threshold group. The dynamic judgment threshold group includes at least fault judgment threshold and safety judgment threshold.

[0012] Optionally, multi-zone operating status data, material flow data, equipment operation data, and environmental data of each section of the grain drying tower are acquired to generate real-time status data of the grain drying tower, including:

[0013] Based on temperature sensors, humidity sensors, and wind pressure sensors deployed in each zone of the grain drying tower's drying and cooling sections, multi-zone operating status data is collected.

[0014] Based on the grain inlet and outlet detection devices, material flow data is collected;

[0015] The material flow data includes grain inflow, grain outflow, grain inflow moisture content, and grain outflow moisture content.

[0016] Collect equipment operation data based on the fan drive unit and heat source supply unit;

[0017] The equipment operation data includes fan speed and heat source fuel consumption;

[0018] Collect environmental data using environmental monitoring devices;

[0019] The environmental data includes ambient temperature and ambient humidity.

[0020] Real-time status data is generated based on multi-zone operation status data, material flow data, equipment operation data, and environmental data.

[0021] Optionally, based on the mass conservation rule and the energy conservation rule, a theoretical normal state reconstruction is performed on the real-time state data to obtain a theoretical baseline matrix, including:

[0022] The amount of moisture migration in the grain drying tower is determined based on the infeed moisture value, the outfeed moisture value in the real-time status data, and the infeed amount and outfeed amount in the material flow data.

[0023] The heat input of the grain drying tower is determined based on the heat source fuel consumption in the real-time status data.

[0024] Based on the ambient temperature, ambient humidity, and multi-zone operation status data in the real-time status data, determine the heat exchange conditions for each zone. The heat exchange conditions include one or more of the following: inlet air temperature, inlet air humidity, air pressure, and heat transfer coefficient for each zone.

[0025] Based on the amount of moisture migration, the amount of heat input, and the heat exchange conditions, the theoretical temperature and humidity values ​​for each time period and each zone are calculated.

[0026] A theoretical baseline matrix is ​​generated based on the theoretical temperature and humidity values ​​for each time period and each zone.

[0027] Optionally, based on the fault mechanism parameterization rules corresponding to multiple fault categories, the theoretical baseline matrix is ​​subjected to damaged state injection processing to obtain multiple theoretical damaged matrices, including:

[0028] According to the preset sensor drift rules, time drift features are injected into the data nodes in the theoretical reference matrix corresponding to the target partition temperature parameters according to the drift function that increases with time.

[0029] According to the preset airflow impedance rules, spatial impedance injection is performed on adjacent partition nodes in the theoretical reference matrix according to the impedance increment function between adjacent partitions.

[0030] According to the preset thermal efficiency decay rule, the global heat distribution in the theoretical baseline matrix is ​​injected nonlinearly according to the thermal efficiency decay function.

[0031] Based on the injected partition node data and global heat distribution data, multiple theoretical damage matrices corresponding to different fault categories are generated;

[0032] Among them, the fault mechanism parameterization rules include at least one or more of the following: sensor drift parameters, airflow impedance parameters, and thermal efficiency attenuation parameters.

[0033] Optionally, a real residual matrix is ​​generated based on real-time state data and a theoretical baseline matrix, and a corresponding theoretical residual matrix is ​​generated based on each theoretical damage matrix and the theoretical baseline matrix, including:

[0034] By mapping real-time state data and theoretical baseline matrix to a unified sampling time axis and a unified partition numbering system, the actual alignment matrix and the baseline alignment matrix are obtained respectively.

[0035] The actual residual matrix is ​​generated based on the element-wise difference or normalized element-wise difference between the actual alignment matrix and the benchmark alignment matrix at the corresponding time position and the corresponding partition position.

[0036] By mapping each theoretical damage matrix and theoretical baseline matrix to a unified sampling time axis and a unified partition numbering system, the damage alignment matrix and the baseline alignment matrix are obtained respectively.

[0037] Based on the element-wise difference or normalized element-wise difference between the damaged alignment matrix and the baseline alignment matrix at the corresponding time position and the corresponding partition position, a theoretical residual matrix corresponding to the fault category is generated.

[0038] Optionally, based on the temporal morphological distance and spatial structural distance between the actual residual matrix and each theoretical residual matrix, the corresponding topological similarity is calculated, and the state evaluation result is determined based on the topological similarity, including:

[0039] Based on the preset dynamic time warping rules, the temporal morphological distance between the actual residual matrix and each theoretical residual matrix is ​​calculated;

[0040] Based on the physical adjacency structure of each zone in the grain drying tower, a zone adjacency network is constructed. Based on the spatial manifold alignment rules used to maintain the adjacency relationship between zones, the spatial structural distance between the actual residual matrix and each theoretical residual matrix is ​​calculated.

[0041] Based on temporal morphological distance and spatial structural distance, the topological similarity corresponding to each theoretical residual matrix is ​​calculated.

[0042] The state assessment results are determined based on the topological similarity of each topology.

[0043] Optionally, the state evaluation results are determined based on each topological similarity, including:

[0044] The maximum value among the topological similarities corresponding to each theoretical residual matrix is ​​determined as the target topological similarity.

[0045] When the target topology similarity is greater than or equal to the fault determination threshold, the fault category corresponding to the target topology similarity is output as the fault status result.

[0046] When the target topology similarity is less than or equal to the safety determination threshold, output the environmental noise status result.

[0047] When the safety judgment threshold is less than the target topology similarity and the target topology similarity is less than the fault judgment threshold, the status result to be reviewed is output.

[0048] Among them, the fault determination threshold is greater than the safety determination threshold.

[0049] Optionally, based on the state assessment results, the fault mechanism parameterization rules and dynamic judgment threshold groups are updated, including:

[0050] Record the real-time state data, actual residual matrix, theoretical residual matrix, and state assessment results corresponding to the state assessment results;

[0051] When the condition assessment result is a fault condition result, the drift amplitude, impedance increment and / or thermal efficiency attenuation coefficient corresponding to the fault category are corrected based on real-time condition data and actual residual matrix.

[0052] When the status assessment result is the environmental noise status result, the safety judgment threshold and / or fault judgment threshold shall be adjusted according to the real-time status data.

[0053] When the status assessment result is a status result to be reviewed, the fault mechanism parameterization rules and dynamic judgment threshold group are modified simultaneously.

[0054] The grain drying tower operation status data processing system includes a data acquisition interface, a processor, and a memory. The memory stores a program, which is executed by the processor to implement the above method.

[0055] The data acquisition interface is used to communicate with temperature sensors, humidity sensors, wind pressure sensors, grain in / out detection devices, fan drive devices, heat source supply devices, and environmental monitoring devices deployed in each zone of the drying and cooling sections of the grain drying tower.

[0056] The processor includes:

[0057] The data sensing module is used to acquire multi-zone operating status data, material flow data, equipment operation data and environmental data of each section of the grain drying tower, and generate real-time status data of the grain drying tower.

[0058] The ideal benchmark reconstruction module is used to reconstruct the theoretical normal state from real-time state data based on the mass conservation rule and the energy conservation rule, so as to obtain the theoretical benchmark matrix.

[0059] The parameterized injection module is used to perform damaged state injection processing on the theoretical baseline matrix based on the fault mechanism parameterization rules corresponding to multiple preset fault categories, so as to obtain multiple theoretical damaged matrices.

[0060] The dual-track differential module is used to generate the actual residual matrix based on real-time state data and the theoretical reference matrix, and to generate the corresponding theoretical residual matrix based on each theoretical damage matrix and the theoretical reference matrix.

[0061] The coupling determination module is used to calculate the corresponding topological similarity based on the temporal morphological distance and spatial structural distance between the actual residual matrix and each theoretical residual matrix, and to determine the state evaluation result based on the topological similarity.

[0062] The feedback update module is used to update the fault mechanism parameterization rules and dynamic judgment threshold group based on the status assessment results. The dynamic judgment threshold group includes at least the fault judgment threshold and the safety judgment threshold.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. This invention acquires multi-zone operating status data, material flow data, equipment operation data, and environmental data from each zone to generate real-time status data including temperature, humidity, feed moisture content, discharge moisture content, fan speed, heat source fuel consumption, and ambient temperature and humidity for each zone. This enables a comprehensive characterization of the drying tower's operation process from the data source, avoiding insufficient information caused by relying solely on a single temperature and humidity threshold, thereby improving the ability to distinguish complex operating conditions such as feed moisture fluctuations and sudden changes in ambient temperature and humidity.

[0065] 2. Based on the laws of mass conservation and energy conservation, this invention reconstructs the theoretical normal state from real-time state data, and obtains a theoretical benchmark matrix that dynamically changes with the moisture content of the grain, the amount of heat input and the heat exchange conditions. This matrix can form a health reference surface that matches the current operating conditions, and no longer simply uses fixed thresholds or historical averages for judgment. This effectively reduces the probability that normal disturbances such as the entry of high-moisture grains or the arrival of cold waves are misjudged as equipment failures.

[0066] 3. This invention uses parameterization rules based on multiple preset fault mechanism categories to perform damaged state injection processing on the theoretical benchmark matrix to obtain multiple theoretical damaged matrices. Furthermore, it constructs the actual residual matrix and the corresponding theoretical residual matrix, which can transform empirical fault mechanisms such as sensor drift, local airflow blockage, and thermal efficiency decay into calculable and comparable fault templates. It can not only determine whether a deviation has occurred, but also determine which type of fault the deviation is closer to, thereby improving the pertinence and interpretability of hidden fault identification.

[0067] 4. This invention calculates the corresponding topological similarity based on the temporal and spatial structural distances between the actual residual matrix and each theoretical residual matrix, and determines the state evaluation result based on the topological similarity. It can simultaneously preserve the evolution trend of the residual over time and the spatial correlation between each partition, avoiding misjudgments caused by comparing only single-moment values ​​or overall averages. This enables the system to effectively distinguish between normal fluctuations caused by sudden weather changes and hidden faults in equipment or sensors, improving the real-time performance and accuracy of fault identification.

[0068] 5. This invention updates the fault mechanism parameterization rules and dynamic judgment threshold groups based on the state assessment results, and combines the fault state results, environmental noise state results, and state results to be verified for hierarchical output. It can continuously correct the drift amplitude, impedance increment, thermal efficiency attenuation coefficient, safety judgment threshold, and fault judgment threshold during continuous operation, so that the model and judgment boundary continuously approach the actual working conditions on site, thereby taking into account both fault identification capability and false alarm suppression capability, and improving the long-term stability and environmental adaptability of the system. Attached Figure Description

[0069] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0070] Figure 1 This is a flowchart illustrating the grain drying tower operation status data processing method provided in the embodiments of this application;

[0071] Figure 2 This is a module architecture diagram of the grain drying tower operation status data processing system in the embodiments of this application. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0073] A method for processing operational status data of a grain drying tower includes: acquiring multi-zone operational status data, material flow data, equipment operation data, and environmental data for each section of the grain drying tower; generating real-time status data, which includes temperature and humidity values ​​for each zone, inlet and outlet moisture content, fan speed, fuel consumption, ambient temperature, and ambient humidity; reconstructing a theoretical normal state from the real-time status data based on the conservation of grain dry matter mass and the enthalpy-humidity balance energy conservation rules, obtaining a theoretical baseline matrix, which includes theoretical temperature and humidity values ​​for each time point and each zone; and reconstructing a theoretical normal state based on preset fault conditions. The mechanism parameterization rules are used to inject damaged states into the theoretical baseline matrix to obtain the theoretical damaged matrix. A real residual matrix is ​​generated based on the real-time state data and the theoretical baseline matrix, and a theoretical residual matrix is ​​generated based on the theoretical damaged matrix and the theoretical baseline matrix. Based on the real residual matrix and the theoretical residual matrix, temporal morphological features and spatial structural features are extracted respectively, and the temporal morphological distance and spatial structural distance are calculated and fused to obtain the topological similarity, which is used to determine the state assessment result. Based on the state assessment result, the fault mechanism parameterization rules and the dynamic judgment threshold group are updated. The dynamic judgment threshold group includes at least the fault judgment threshold and the safety judgment threshold.

[0074] This embodiment provides a mechanism for processing operational status data of a grain drying tower, such as... Figure 1 As shown; specifically, the main scenario is a continuous grain drying tower, which operates continuously during the peak autumn grain storage period. The tower is divided into three sections from top to bottom: drying section one, drying section two, and cooling section three. The system collects operating data every 5 seconds and performs status assessment on the data window of the most recent 10 minutes to distinguish between normal fluctuations caused by sudden weather changes and hidden faults in equipment or sensors.

[0075] Specifically, it involves acquiring multi-source data and forming real-time status data; this real-time status data is not a single-point reading, but rather a composite result of various zones, material flows, and equipment operating conditions at the same sampling time.

[0076] Data at a certain moment can be organized into a row vector. For example, at time t1, the temperatures of zones 1, 2, and 3 are 78℃, 70℃, and 32℃, respectively, with corresponding humidity levels of 18%, 22%, and 40%. The moisture content of the ingested grain is 26%, the moisture content of the outgested grain is 14%, the fan speed is 1450 rpm, the fuel consumption is 18 liters per hour, the ambient temperature is 5℃, and the ambient humidity is 82%. Thus, a real-time status record can be formed for this moment.

[0077] Multiple time points can be superimposed to form a real-time state matrix, where rows represent time and columns represent different physical quantities. After obtaining the real-time state data, the theoretical normal state is reconstructed using the conservation of mass and energy. The reconstruction goal here is not to replay historical averages, but to calculate the temperature and humidity that each zone should theoretically exhibit, based on the premise that the equipment is in a healthy state, combined with real-time feed moisture content, feed moisture content, feed flow rate, fuel consumption, and environmental conditions.

[0078] For ease of explanation, let's assume that time points t1 and t2 correspond to three zones. We can construct a 2×3 theoretical temperature submatrix: the first row represents the theoretical temperature at time t1 as [80, 69, 31], and the second row represents the theoretical temperature at time t2 as [79, 68, 30]. Similarly, we can construct a theoretical humidity submatrix, for example: the first row is [17, 23, 41], and the second row is [18, 24, 42]. Combining the two forms the theoretical baseline matrix. This matrix essentially represents the distribution of physical quantities that each zone should exhibit under the current feed and environmental conditions, assuming no equipment malfunctions.

[0079] Furthermore, damaged states are injected into the theoretical baseline matrix; the system pre-stores several fault categories and their mechanism parameterization rules, such as temperature sensor drift, local airflow blockage, and reduced thermal efficiency of the burner; during injection, the original collected data is not directly changed, but the corresponding fault modes are superimposed on the theoretical baseline matrix to obtain multiple theoretical damaged matrices.

[0080] For example, if the temperature sensor in zone 1 is simulated to drift in the positive direction, the theoretical temperature value of zone 1 can be increased by 0.5℃ and 1.0℃ over time. If the airflow blockage between zone 2 and zone 3 is simulated, the temperature in zone 2 can be made higher and the temperature in zone 3 can be made lower, forming a reverse shift between adjacent zones. If the global thermal efficiency is simulated to decrease, the theoretical temperature of the three zones can be reduced as a whole, but the rate of decrease is non-linear.

[0081] Therefore, each type of fault corresponds to a theoretical damage matrix that characterizes a typical fault pattern; calculate the residual; the actual residual is obtained by subtracting the theoretical benchmark matrix from the real-time status data, reflecting the degree to which the actual operation deviates from the healthy state; the theoretical residual is obtained by subtracting the theoretical benchmark matrix from the theoretical damage matrix, reflecting the ideal deviation pattern of a certain type of fault itself;

[0082] For example, at times t1 and t2, the real-time temperature of zone 1 is [81, 82], and the theoretical temperature is [80, 79], then the actual temperature residual of zone 1 is [1, 3]; if the theoretical damage value after sensor drift injection is [80.5, 81], then its theoretical residual is [0.5, 2]; combine the temperature and humidity residuals according to zone and time to form the actual residual matrix and the theoretical residual matrix of each category;

[0083] In the similarity determination stage, the system does not directly compare the absolute difference of the values, but simultaneously measures the temporal form and spatial structure. The temporal form distance is used to observe whether the trend of the residual changes over time is consistent, and the spatial structure distance is used to observe whether the distribution of deviations between different partitions conforms to a certain fault mechanism.

[0084] For ease of understanding, we can assume that the temporal distance between the actual residual and the theoretical residual of sensor drift is 0.2, and the spatial distance is 0.1; the temporal distance with the theoretical residual of airflow blockage is 0.15, and the spatial distance is 0.45; and the temporal distance with the theoretical residual of thermal efficiency reduction is 0.5, and the spatial distance is 0.2. The system fuses the two types of distances into a topological similarity; the smaller the distance, the higher the similarity. For example, if the similarities of the three are calculated to be 0.88, 0.62, and 0.51 respectively, then the current window is closer to the sensor drift mode.

[0085] The fault mechanism parameterization rules and dynamic judgment threshold groups are updated based on the state assessment results. If a certain type of fault is judged to be true multiple times in a row, it means that the matching degree between the original parameterization rules and the actual fault in the field meets the preset approximation conditions. The drift slope, impedance increment or thermal attenuation coefficient can be further fine-tuned to make the subsequent simulation closer to the actual situation. If the environment is in a state of noise for a long time, it means that the original threshold does not meet the tolerance requirements. The safety judgment threshold or fault judgment threshold should be dynamically adjusted according to the error distribution to avoid frequent false alarms.

[0086] Regarding the data anomaly handling mechanism, if individual sensor data is missing at a certain moment, it can be supplemented by interpolation of adjacent moments or estimation of neighboring partitions before entering the reconstruction process; if the missing proportion exceeds the preset limit, for example, exceeding 20% ​​of the total number of measurement points in the window, the window is directly marked as unreliable data, no fault category is output, and only the mark to be supplemented is retained.

[0087] If the theoretical residual matrix corresponding to a certain fault category is empty or has no effective difference from the baseline matrix after injection, the category will be removed in this round of comparison to prevent invalid comparisons. If the similarity of multiple categories is very close, for example, the difference between the maximum and the second largest value is less than 0.03, they will first enter the pending review state to avoid premature output of the judgment result.

[0088] During the nighttime operation of the aforementioned three-zone drying tower, the external ambient temperature suddenly dropped from 8°C to 1°C, causing significant fluctuations in the temperature of the cooling sections in the three zones. If a fixed threshold is used for monitoring directly, this fluctuation could easily be misjudged as an abnormality in the cooling system. However, in this embodiment, the theoretical benchmark is first calculated based on the ambient temperature and humidity and fuel input at that time. Then, through residual comparison, it is found that the actual residual does not meet the spatial structure of local blockage causing the adjacent zones to shift in opposite directions. Therefore, no mechanical fault is output, but it is classified as environmental noise.

[0089] Conversely, when the temperature measurement point in zone 1 shows a monotonically upward shift in four consecutive windows and other physical quantities do not change synchronously, the shape of the actual residual gradually matches the theoretical residual of the drift type, and the system can identify it as early sensor drift.

[0090] The purpose of this step is to break down the complex, noisy, and strongly coupled operation process of the drying tower into a closed-loop process of health baseline—damage injection—residual comparison—feedback update, thereby enabling effective differentiation of hidden faults and environmental fluctuations, and improving the stability and interpretability of condition assessment.

[0091] Furthermore, multi-zone operational status data, material flow data, equipment operation data, and environmental data of each section of the grain drying tower are acquired to generate real-time status data of the grain drying tower, including:

[0092] Based on temperature sensors, humidity sensors, and wind pressure sensors deployed in each zone of the grain drying tower's drying and cooling sections, multi-zone operating status data is collected.

[0093] Based on the grain inlet and outlet detection devices, material flow data is collected;

[0094] The material flow data includes grain inflow, grain outflow, grain inflow moisture content, and grain outflow moisture content.

[0095] Collect equipment operation data based on the fan drive unit and heat source supply unit;

[0096] The equipment operation data includes fan speed and heat source fuel consumption; environmental data is collected based on environmental monitoring devices.

[0097] The environmental data includes ambient temperature and ambient humidity.

[0098] Real-time status data is generated based on multi-zone operation status data, material flow data, equipment operation data, and environmental data.

[0099] This embodiment provides a multi-source sensing convergence step; specifically, in the aforementioned continuous grain drying tower scenario, temperature sensors, humidity sensors and wind pressure sensors are respectively installed in the three zones of the tower body, flow and moisture detection devices are respectively installed at the grain inlet and grain outlet, the fan driver provides real-time speed, the heat source supply device provides fuel flow, and the environmental monitoring device provides ambient temperature and ambient humidity.

[0100] If only the temperature and humidity inside the tower are collected, while ignoring the moisture content of the feed grain, the fan speed and the external environment, although monitoring curves can be generated, these curves will deviate in a way similar to a fault when cold air invades or the grain layer is uneven, resulting in an incomplete input basis for subsequent reconstruction. Therefore, this embodiment expands the collection objects to material flow, equipment operating conditions and environmental operating conditions to reduce ambiguity from the data source.

[0101] Specifically, within each sampling cycle, the system first collects multi-zone operating status data from the three zones; for example, at time t1, the temperatures in the three zones are 78℃, 70℃, and 32℃, the humidity is 18%, 22%, and 40%, and the wind pressure is 420Pa, 380Pa, and 110Pa, respectively; it obtains material flow data from the grain inlet and outlet detection device, such as grain inlet of 6 tons / hour, grain outlet of 5.8 tons / hour, grain inlet moisture content of 26%, and grain outlet moisture content of 14%; it then obtains fan speed of 1450 rpm and fuel consumption of 18 liters / hour from the equipment side; and it obtains ambient temperature of 5℃ and ambient humidity of 82% from the environmental monitoring device; the system assembles these data into a complete record with a unified timestamp t1.

[0102] For ease of explanation, a complete record at any given time can be defined as a real-time status record vector R, which can be simplified as: Real-time status record vector R(t1) = [78, 70, 32, 18, 22, 40, 420, 380, 110, 6, 5.8, 26, 14, 1450, 18, 5, 82]; where the first 9 items correspond to the temperature, humidity, and wind pressure of the three zones, the middle 4 items correspond to grain flow and moisture, and the last 4 items correspond to equipment and environment; multiple time-based records are stacked in chronological order to form the real-time status data matrix required for subsequent processing;

[0103] In practical engineering, the reporting frequencies of different data sources are usually inconsistent; for example, temperature and humidity are reported every 5 seconds, grain moisture is reported every 30 seconds, and fuel consumption is reported every 60 seconds. To solve this problem, the system can use a unified sampling time axis for alignment, and supplement low-frequency data by keeping the most recent valid value, linear interpolation, or window averaging. For example, if there is no new moisture value at time t2, the most recent valid value after t1 is used. If both moisture measurement points are available, short-time interpolation is performed to align them with high-frequency temperature data.

[0104] Regarding the data anomaly handling mechanism, if the wind pressure sensor in a certain zone is disconnected, but the temperature and humidity signals are normal, the operating status record of that zone can be retained first, and the wind pressure field can be marked as missing. When the continuous missing time is less than a preset threshold, such as less than 2 minutes, it can be estimated using the wind pressure difference between adjacent zones and the fan speed. If it exceeds the threshold, the wind pressure dimension of that zone will be marked as low confidence in the real-time status data, so that its weight can be reduced in the subsequent reconstruction stage.

[0105] If there is an abnormal difference between the amount of grain entering and leaving the grain that exceeds the normal grain storage range, such as an instantaneous difference that is far beyond the historical allowable range, the status of the metering device should be checked first, and the time period should be marked as the flow abnormality window to avoid introducing abnormal data to interfere with subsequent analysis.

[0106] When a batch of high-moisture corn entered the tower at night, the moisture content of the feed increased from 24% to 28%. At the same time, the outdoor weather conditions changed abruptly, and the ambient temperature dropped from 7°C to 2°C. If only the temperature drop in one zone is observed, there is a risk of false alarm that it is a heat source failure. However, since this embodiment collects the feed moisture content and ambient temperature simultaneously, the subsequent modules can identify that this is a combination of increased load and cooling of the environment, rather than a simple equipment failure.

[0107] The purpose of this step is to provide a complete, homogeneous, and alignable data foundation for subsequent theoretical reconstruction and fault comparison, thereby achieving a comprehensive characterization of the drying tower's operating status.

[0108] Furthermore, based on the laws of mass conservation and energy conservation, the theoretical normal state is reconstructed from the real-time state data to obtain a theoretical benchmark matrix. This includes: determining the moisture migration of the grain drying tower based on the inlet and outlet moisture values ​​in the real-time state data, as well as the inlet and outlet quantities in the material flow data; determining the heat input of the grain drying tower based on the heat source fuel consumption in the real-time state data; determining the heat exchange conditions for each zone based on the ambient temperature, ambient humidity, and multi-zone operating status data in the real-time state data, including one or more of the inlet air temperature, inlet air humidity, air pressure, and heat transfer coefficient for each zone; calculating the theoretical temperature and theoretical humidity values ​​for each time point and each zone based on the moisture migration, heat input, and heat exchange conditions; and generating the theoretical benchmark matrix based on the theoretical temperature and theoretical humidity values ​​for each time point and each zone.

[0109] This embodiment provides a step for reconstructing a theoretically normal state. Specifically, in the aforementioned scenario, real-time state data alone is insufficient to directly identify faults, because the normal range of temperature and humidity inside the tower varies significantly under different batches of grain and different weather conditions. If the historical average is simply used as a reference, the system will misidentify normal fluctuations as abnormalities when high-moisture grains are introduced or cold air is suddenly introduced. Therefore, this embodiment introduces a dynamic reconstruction mechanism based on mass conservation and energy conservation.

[0110] Specifically, the amount of moisture migration is calculated; the amount of water evaporated per unit time can be estimated by the difference between the amount of grain entering and leaving the grain and the moisture content; the principle of dry matter conservation is followed during the drying process. For example, if the amount of grain entering the grain in a certain minute is 100 kg and the moisture content of the grain is 26%, it means that it contains 74 kg of dry matter and 26 kg of moisture.

[0111] Under steady-state flow, if the moisture content of the discharged grain is 14%, in order to maintain 74 kg of dry matter without loss, the corresponding discharged grain amount for this batch should be calculated as 86 kg, i.e., 74 / (1-0.14); therefore, the calculated moisture content of the discharged grain is 12 kg; considering a small amount of retention, the effective moisture migration amount per minute can be calculated as 14 kg, i.e., 26 kg minus 12 kg; this amount reflects the evaporation load in the tower; calculate the heat input; if the heat source device consumes 0.3 liters of fuel per minute, combined with the preset lower heating value of the fuel and the heat transfer efficiency, the corresponding heat input can be calculated;

[0112] The underlying logic here is: the higher the fuel consumption, the greater the total heat that can be provided to the air and grain layer inside the tower; if the fuel consumption remains unchanged but the temperature inside the tower decreases, it indicates that the heat may be carried away by the environment or that there is an efficiency loss in the equipment; determine the heat exchange conditions of each zone; the heat exchange conditions are not a single parameter, but are composed of ambient temperature, ambient humidity, wind pressure of each zone and the current state of each zone.

[0113] To illustrate this more clearly, we can assume that the theoretical inlet temperature of Zone 1 reaches the first preset temperature threshold, the inlet temperature of Zone 2 is between the first and second preset temperature thresholds, and the temperature of Zone 3 in the cooling section is below the second preset temperature threshold. The higher the air pressure, the stronger the airflow exchange. The higher the ambient humidity, the lower the air's moisture-carrying capacity. Based on these inputs, the system establishes a current heat exchange state for each zone. For example, when a cold wave arrives, the heat exchange between Zone 3 and the outside environment increases, so its theoretical temperature will naturally decrease and should not be considered a fault.

[0114] After obtaining the amount of moisture migration, the amount of heat input, and the heat exchange conditions, the theoretical temperature and humidity values ​​for each time period and each zone are calculated. The underlying structured deduction logic is based on thermodynamic enthalpy-humidity balance: the amount of heat input is regarded as the total available energy. After deducting the structural heat loss to the environment through the tower body, which is determined by the ambient temperature and the heat exchange coefficient, part of the remaining net heat is converted into the latent heat required for grain moisture evaporation, the size of which is determined by the amount of moisture migration calculated above. The other part is converted into the sensible heat for heating the air and grain layer in the tower.

[0115] In the specific calculation, the initial hot air enthalpy of each zone is first determined using the fan speed and inlet air temperature. The heat dissipation per unit structure at the current moment is calculated based on the ambient temperature and heat transfer coefficient and deducted from the total enthalpy. The heat transfer coefficient is an empirical constant pre-calibrated based on the thickness and thermal conductivity of the drying tower's outer wall material, or a dynamic coefficient obtained by the system through real-time mapping and table lookup based on ambient wind speed and wind pressure data. The latent heat consumption calculated by the product of moisture migration and the latent heat constant of water vaporization is further deducted from the remaining total enthalpy. The final remaining enthalpy is combined with the air mass inside the tower and the specific heat capacity of the grain layer to deduce the theoretical final air temperature, i.e., the theoretical temperature value, at each moment and in each zone.

[0116] Simultaneously, the amount of evaporated moisture is accumulated and added to the initial absolute humidity of the incoming air. Then, it is converted into relative humidity by combining the saturated vapor pressure corresponding to the theoretical temperature value, thus obtaining the theoretical humidity value. When the hot air flows through the grain layer in each zone, it loses sensible heat, causing the theoretical air temperature to drop. At the same time, it carries away the evaporated moisture, causing the theoretical air humidity to rise. The rate of this heat and mass exchange is limited by the wind pressure and heat transfer coefficient of each zone. Through the above dynamic distribution process of heat and moisture, a mechanism reconstruction that does not rely on the fitting of implicit parameters can be achieved.

[0117] To facilitate understanding, a simplified calculation model with two time points and three zones can be used to illustrate this: When the moisture content of the feed is high, the fuel input is normal, and the ambient temperature is 5℃ at time t1, the system calculates the theoretical temperature of the three zones as [80, 69, 31] and the theoretical humidity as [17, 23, 41]; when the ambient temperature drops sharply to 2℃ at time t2 and the moisture content of the feed continues to rise, the system calculates the theoretical temperature of the three zones as [79, 68, 28] and the theoretical humidity as [18, 24, 44];

[0118] As can be seen, even if the equipment is not malfunctioning, the theoretical values ​​will change in real time with the operating conditions; this is the key difference between dynamic benchmarks and fixed thresholds. The theoretical temperature and humidity values ​​at all times are organized into a theoretical benchmark matrix according to time and zone. The matrix can be constructed by splicing temperature first and then humidity, or by constructing a multi-channel matrix. For example, a 2×6 matrix can be represented as: t1: [80, 69, 31, 17, 23, 41]; t2: [79, 68, 28, 18, 24, 44]; where the first three columns are the theoretical temperatures of the three zones, and the last three columns are the theoretical humidity of the three zones. This matrix will serve as the basis for comparing with real data.

[0119] Regarding the data anomaly handling mechanism, if the difference between the grain inflow and outflow within a preset time window exceeds a preset difference threshold, it may be caused by grain layer stagnation rather than measurement anomalies. In this case, smoothing can be performed within multiple time windows to avoid moisture migration caused by calculation deviations in a single window. If fuel consumption data is missing, heat input can be estimated in a short time based on the opening of the heat source control valve or historical adjacent values.

[0120] If the environmental monitoring device malfunctions, the measured air parameters at the air inlet of the drying tower can be used as a temporary substitute; if the calculated theoretical temperature or theoretical humidity exceeds the physical allowable range of the equipment, for example, the theoretical temperature of the cooling section is higher than that of the drying section, or the humidity is negative, the reconstruction result for the current period is considered abnormal, subsequent similarity determination is suspended and the model self-check is triggered.

[0121] In the seventh hour of continuous operation, a batch of corn with a moisture content higher than the preset benchmark value entered the tower, and the outdoor humidity increased simultaneously. If the average temperature of the previous batch of low-moisture grains is used as the benchmark, it will be considered that the temperature in Zone 2 is too low and the humidity in Zone 3 is too high. However, the theoretical value obtained by this embodiment after reconstruction based on the new moisture migration and heat exchange conditions already shows a slight decrease in temperature in Zone 2 and a slight increase in humidity in Zone 3. Therefore, the residual between the actual data and the theoretical value is small and there will be no false alarm.

[0122] The purpose of this step is to construct a health reference surface that evolves dynamically with the operating conditions, thereby enabling the prior deduction of normal production disturbances and providing a calculable and interpretable benchmark for subsequent fault mode extraction.

[0123] Furthermore, based on the fault mechanism parameterization rules corresponding to multiple fault categories, the theoretical benchmark matrix is ​​subjected to damaged state injection processing to obtain multiple theoretical damaged matrices, including: injecting time drift features into the data nodes corresponding to the target partition temperature parameters in the theoretical benchmark matrix according to a time-increasing drift function based on a preset sensor drift rule; injecting spatial impedance into adjacent partition nodes in the theoretical benchmark matrix according to an adjacent partition impedance increment function based on a preset airflow impedance rule; injecting nonlinear decay into the global heat distribution in the theoretical benchmark matrix according to a thermal efficiency decay function based on a preset thermal efficiency decay rule; and generating multiple theoretical damaged matrices corresponding to different fault categories based on the injected partition node data and global heat distribution data; wherein, the fault mechanism parameterization rules include at least one or more of sensor drift parameters, airflow impedance parameters, and thermal efficiency decay parameters.

[0124] This embodiment provides a damaged state injection step; specifically, based on the aforementioned established dynamic benchmark, if only the difference between the actual data and the benchmark is calculated, it can only be known that a deviation has occurred, but it cannot be known which type of fault the deviation resembles; to solve this problem, this embodiment further transforms maintenance experience and process knowledge into computable parameterized rules, and actively generates multiple fault samples on the theoretical benchmark matrix;

[0125] First, examine the sensor drift rules. Early sensor aging often manifests as a gradual shift in readings over time, rather than a sudden, abrupt change. Therefore, a time-increasing drift injection can be performed on the target zone temperature nodes. For example, at times t1, t2, and t3, the theoretical temperature of zone one was originally [80, 79, 78]. If the drift function increases by [0.3, 0.6, 0.9] at each time point, it becomes [80.3, 79.6, 78.9] after injection. The general expression for the time-increasing drift function is:

[0126]

[0127] in, The preset drift slope parameter, The injection start time for the drift fault; for The drift injection amount corresponding to the time. The current sampling time; drift slope parameter The initial value is calibrated according to the annual drift rate indicated in the sensor manual, and iteratively corrected in the subsequent feedback update stage based on the average slope of the actual residual matrix;

[0128] The resulting damage matrix corresponds to the positive drift category of the temperature sensor in Zone 1. To simulate negative drift, simply change the drift increment to a negative value. Then look at the airflow impedance rules. Baffle jamming, dust accumulation in the air duct, or local blockage often do not change all zones of the tower, but rather change the hot air distribution between adjacent zones.

[0129] To address this, spatial impedance injection can be performed on adjacent partition nodes. For example, if there is an increase in impedance between Zone 2 and Zone 3, the theoretical temperature at t1, which was originally [80, 69, 31], can be adjusted to [80, 72, 27]; the corresponding humidity, which was originally [17, 23, 41], can be adjusted to [17, 20, 45]. This reflects that the blockage causes more heat and drying capacity to remain in the upstream partition, resulting in insufficient heat gain and insufficient dehumidification in the downstream partition, thus forming a fault pattern with spatial asymmetry.

[0130] Next, let's look at the rules of thermal efficiency decay. Carbon buildup in the burner, scaling in the heat exchanger, or a decrease in heating efficiency often manifest as insufficient overall heating, but the degree of impact on different zones is not exactly the same. Therefore, a non-linear decay injection can be performed on the global heat distribution. For example, if the original theoretical temperature is [80, 69, 31], after injection in a manner where the decay is small at the beginning and large at the end, it can become [78.5, 66, 27]. This type of pattern is different from local blockage because it is not a deviation in the opposite direction of a certain adjacent zone, but rather a decline in the overall heat level.

[0131] Multiple theoretical damage matrices can be generated using the different rules mentioned above. For easy comparison, they can be uniformly stored as a ternary relationship of fault category, parameter group, and matrix. For example, a type of drift fault can be configured as zone one, positive drift, and slope of 0.3℃ / window; a type of impedance fault can be configured as zone two to three with impedance increment of 15%; and a type of thermal decay fault can be configured as an 8% decrease in overall tower thermal efficiency. This not only distinguishes the categories but also reflects different degrees such as mild, moderate, and severe.

[0132] Regarding the data anomaly handling mechanism, if the difference between the damaged matrix generated after a certain injection and the baseline matrix is ​​less than the preset feature extraction lower limit, for example, if the absolute value of all offset values ​​is lower than the sensor resolution, then the parameter combination is not identifiable and can be discarded in this round of simulation; if a physically unreasonable result is generated after a certain injection, for example, if the theoretical humidity of a certain zone exceeds 100% or the temperature gradient reversal exceeds the feasible boundary of the equipment, then the system will automatically reduce the amplitude of the parameter or mark the group as an illegal rule;

[0133] If multiple rules overlap and mask each other, such as the simultaneous injection of global thermal decay and local drift causing ambiguity in category features, a single-fault prior mode can be used to prioritize the output of independent categories, and the mixed categories can be placed in an extended library for separate training.

[0134] When the drying tower was running continuously until the early morning, maintenance personnel suspected that the temperature probe in Zone 1 was aging, there was a small amount of bran powder accumulation in the air duct in Zone 2, and the thermal efficiency of the burner had recently decreased slightly. In this embodiment, three sets of theoretical damage matrices can be constructed respectively: one set shows that the temperature residual in Zone 1 gradually increases over time; one set shows that Zone 2 is too hot and Zone 3 is too cold, indicating spatial splitting; another set shows that Zones 3 cool down simultaneously but the gradient is relatively smooth. Subsequent actual data only need to be compared with these pure fault templates to determine which one is closer to the actual situation on site.

[0135] The purpose of this step is to transform empirical fault mechanisms into structured, iterable mathematical objects, thereby enabling the proactive generation and interpretable identification of fault modes.

[0136] Further, a real residual matrix is ​​generated based on the real-time status data and the theoretical reference matrix, and a corresponding theoretical residual matrix is ​​generated based on each theoretical damage matrix and the theoretical reference matrix. This includes: mapping the real-time status data and the theoretical reference matrix to a unified sampling time axis and a unified partition numbering system to obtain a real alignment matrix and a reference alignment matrix, respectively; generating a real residual matrix based on the element-wise difference or normalized element-wise difference between the real alignment matrix and the reference alignment matrix at the corresponding time position and the corresponding partition position; mapping each theoretical damage matrix and the theoretical reference matrix to a unified sampling time axis and a unified partition numbering system to obtain a damage alignment matrix and a reference alignment matrix, respectively; and generating a theoretical residual matrix corresponding to the fault category based on the element-wise difference or normalized element-wise difference between the damage alignment matrix and the reference alignment matrix at the corresponding time position and the corresponding partition position.

[0137] This embodiment provides a step for dual-track differential extraction. Specifically, given an existing theoretical benchmark matrix and multiple theoretical damage matrices, directly comparing field data with fault templates is easily affected by inconsistent sampling frequencies, inconsistent partition numbers, and different dimensions. Therefore, this embodiment first performs unified alignment, then calculates the actual residual and theoretical residual to make the two differential tracks comparable. Time axis and partition number alignment are performed. Taking a three-zone drying tower as an example, the field sampling times may be t1, t2, and t4, lacking t3; while the theoretical benchmark generates four timestamps: t1, t2, t3, and t4 according to rules.

[0138] The system can establish a unified sampling time axis [t1, t2, t3, t4]. For the missing t3 in the field, it can be filled by nearest neighbor interpolation, nearest value filling, or missing measurement marking. In terms of partition numbering, the field equipment may use the first drying zone identifier, the second drying zone identifier, and the first cooling zone identifier to represent drying zone 1, drying zone 2, and cooling zone 1, while the theoretical model uses 1, 2, and 3. The system completes the mapping before entering the difference to ensure that the same column always corresponds to the same physical partition. After alignment, the actual residual matrix is ​​calculated.

[0139] For ease of demonstration, assume the actual temperature alignment matrix for the three zones within a window is: [81, 70, 30; 82, 69, 29]; and the baseline alignment matrix is: [80, 69, 31; 79, 68, 28]. After subtracting element by element, the actual temperature residual is: [1, 1, -1; 3, 1, 1]. If dimensional normalization is considered, it can be further divided by the allowable fluctuation range of each variable or the sensor range. For example, if the allowable temperature fluctuation is normalized to 5℃, then the actual normalized residual in the first row is [0.2, 0.2, -0.2].

[0140] Humidity, wind pressure, and other dimensions can be processed in the same way, and finally combined into a real residual matrix under a unified scale; similarly, a theoretical residual matrix is ​​calculated for each theoretical damage matrix; for example, the drift damage temperature matrix of Zone 1 is: [80.5, 69, 31; 81, 68, 28], and after subtracting it from the benchmark alignment matrix, the theoretical temperature residual is obtained: [0.5, 0, 0; 2, 0, 0];

[0141] For example, the temperature matrix of impedance increase damage from zone 2 to zone 3 is: [80, 72, 27; 79, 71, 25], and the corresponding theoretical temperature residual is: [0, 3, -4; 0, 3, -3]. From these results, it can be seen that the residual structure of different fault categories is significantly different: the former gradually increases in the time dimension and concentrates in zone 1, while the latter shows a reverse shift between adjacent zones in the spatial dimension.

[0142] If it is necessary to integrate multi-dimensional variables such as temperature, humidity, and wind pressure, the matrix can be organized by dividing it into blocks according to variables; for example, the first 3 columns are temperature residuals, the middle 3 columns are humidity residuals, and the 3 columns are wind pressure residuals; in this way, subsequent similarity calculations can not only see temperature deviations, but also the linked deviations of humidity and wind pressure.

[0143] Regarding the data anomaly handling mechanism, if the variance of a variable is lower than the preset lower limit within the current window, direct normalization may lead to an excessively small denominator, amplifying minor noise. In this case, a minimum normalization baseline can be set, such as not lower than the sensor resolution or an empirically stable value. If the field data and the theoretical matrix are inconsistent in length at a certain partition, for example, if temporary measurement points are added in the field but the theoretical model has not yet been expanded, only the common dimensions will participate in the difference calculation, and the newly added dimensions will retain the marker for expansion. If the difference cannot be reliably calculated for some elements due to missing measurements, a mask matrix can be used to record its validity and reduce or eliminate the contribution of that position in subsequent similarity calculations.

[0144] Under the influence of the cold wave, the actual temperature in Zone 3 was 1°C lower than the theoretical value, but the temperature in Zone 1 was 1°C and 3°C higher than the theoretical value. After expanding the actual residuals, it can be seen that Zone 1 showed a continuous upward trend, while Zone 3 only fluctuated for a short time. At this time, the time pattern is more consistent with the theoretical residual of the drift in Zone 1. However, the spatial structure is obviously inconsistent with the theoretical residual of the global thermal efficiency decline. Therefore, the difference results themselves lay the foundation for subsequent judgment.

[0145] The purpose of this step is to transform complex raw monitoring data into comparable and quantifiable deviation representations, thereby achieving a unified expression between the actual operating status and the fault template.

[0146] Furthermore, based on the temporal morphological distance and spatial structural distance between the actual residual matrix and each theoretical residual matrix, the corresponding topological similarity is calculated, and the state assessment result is determined according to the topological similarity. This includes: calculating the temporal morphological distance between the actual residual matrix and each theoretical residual matrix according to a preset dynamic time warping rule; constructing a partition adjacency network based on the physical adjacency structure of each partition of the grain drying tower; calculating the spatial structural distance between the actual residual matrix and each theoretical residual matrix according to the spatial manifold alignment rule used to maintain the partition adjacency relationship; calculating the topological similarity corresponding to each theoretical residual matrix based on the temporal morphological distance and spatial structural distance; and determining the state assessment result based on the topological similarity.

[0147] This embodiment provides a coupling determination step; specifically, after having a real residual matrix and multiple theoretical residual matrices, if only the numerical difference at the same time is compared, it is easy to miss the case where the fault occurrence time is slightly off but the overall shape is similar; if only the overall mean is compared, the structural relationship between adjacent partitions will be ignored; therefore, this embodiment introduces both time morphological distance and spatial structural distance for coupling calculation; let's look at the time morphological distance first;

[0148] Real-world faults often do not occur precisely at the same sampling point in the theoretical template. For example, theoretically, drift may start from t1, but in reality, it may gradually appear starting from t2. To address this, dynamic time warping can be performed on the real residual sequence and the theoretical residual sequence to ensure that local time misalignment does not disrupt the overall similarity assessment.

[0149] Specifically, the execution logic of the dynamic time warping rule is as follows: A two-dimensional distance matrix is ​​constructed, with the actual residual sequence as the horizontal axis and the theoretical residual sequence as the vertical axis. Each element in the matrix represents the absolute value of the numerical difference between the two sequences at a specific sampling point. Based on the dynamic programming algorithm, an alignment path is searched from the upper left corner of this distance matrix to the lower right corner, minimizing the cumulative value of the elements along the path. The state transition equation of the dynamic programming algorithm is as follows:

[0150]

[0151] in, For the real sequence number The sampling point and the theoretical sequence The absolute difference of the residuals at each sampling point. The minimum cumulative distance to reach the current coordinate point. These represent the minimum cumulative distances to the adjacent preceding coordinate point. Through this minimum alignment path, the system can correlate fault features that have undergone minor misalignment, stretching, or compression on the time axis, and ultimately use the average cumulative distance of this path as the output temporal morphology distance. ;

[0152] In a simple example, the real residual in zone 1 is [1, 3, 4], and the theoretical drift residual is [0.5, 2, 3.5]. Although they are not completely consistent point by point, both show a continuous increase, and the minimum cumulative distance after dynamic alignment is small. Compared with the blocking residual [0, 0, 0], the minimum cumulative path distance is significantly larger. Now let's look at the spatial structure distance.

[0153] The drying tower zones are not isolated from each other; zones one, two, and three are actually adjacent. If a fault occurs between zones two and three, it often manifests as zones two and three shifting simultaneously in opposite directions. If the fault occurs in a zone one sensor, only zone one will show a continuous shift.

[0154] To preserve this structural relationship, the system constructs spatial manifold alignment rules based on the partition adjacency relationship. At the specific computational level, the spatial manifold alignment rules abstract each partition as a node in the graph and the hot air flow path between adjacent partitions as directed edges connecting nodes, thereby constructing an adjacency graph that reflects the spatial structure of the drying tower.

[0155] To eliminate computational biases caused by different physical dimensions, all residual values ​​used in subsequent topological similarity calculations are normalized element-wise differences. When calculating spatial structure distance, the system not only calculates the absolute difference between the actual and theoretical residuals for the same node, but also synchronously calculates the gradient difference between adjacent nodes using the adjacency graph. The mean of the absolute differences between nodes and the mean of the gradient differences between nodes—that is, the difference between the upstream and downstream residuals—are compared in reality and in theory, and then added together according to preset spatial weights to obtain the spatial structure distance. :

[0156]

[0157] in, and These are the preset weights for the absolute difference between nodes and the gradient difference between nodes, respectively. This represents the total number of nodes in the adjacency graph. Let be the total number of directed edges. and These represent the normalized residual values ​​of the same node in the actual residual matrix and the theoretical residual matrix, respectively. and These are the normalized residual gradient values ​​between adjacent nodes in the actual residual matrix and the theoretical residual matrix, respectively;

[0158] This mechanism decouples the absolute fluctuation of a single measuring point from the relative correlation structure between measuring points, making the spatial distribution of residuals on the adjacency chain more dominant than the absolute value. For example, a three-zone tower can be abstracted as an adjacency chain 1-2-3, and any residual distribution needs to be compared on this chain topology. If the actual residual distribution is [3, 0, 0], the node difference on the manifold is mainly concentrated at the entrance, and the spatial structure is closer to the local anomaly of zone one. If it is [0, 3, -3], a huge adjacent gradient difference is formed between nodes 2 and 3, which is closer to the blockage anomaly from zone two to zone three.

[0159] By fusing the two types of distances, the topological similarity can be obtained. The fusion method can adopt a nonlinear mapping approach. To clarify the logical calculation mechanism of the algorithm, the system uses a quantized fusion formula:

[0160]

[0161] in, and These are preset weight coefficients for the time and space dimensions, for example, set to 1.0 and 2.0 respectively to enhance the sensitivity of spatial adjacency features; this mechanism maps unbounded dissimilarity to topological similarity in the range of 0 to 1 through the inverse of distance weighting.

[0162] In a simplified calculation example, comparing the actual residuals with the three types of theoretical residuals yields the following results: a temporal distance of 0.15 and a spatial distance of 0.10 with the drift class, corresponding to a topological similarity of 0.90; a temporal distance of 0.18 and a spatial distance of 0.42 with the blockage class, corresponding to a topological similarity of 0.67; and a temporal distance of 0.40 and a spatial distance of 0.25 with the thermal decay class, corresponding to a topological similarity of 0.58. Therefore, the actual data can be considered to be closest to the drift class fault. If multiple minor faults occur simultaneously, a single topological similarity may not be sufficient to fully describe the situation. In this case, the system can first output the most similar category as the primary anomaly, and then retain the second most similar category as a candidate for accompanying anomalies for subsequent manual review or model training.

[0163] Regarding boundary condition handling, if a category has insufficient comparable elements due to mask missing measurements, for example, if the effective elements are less than 60% of the total elements, then that category will not participate in the similarity comparison in this round. If the temporal distance is lower than the first distance threshold and the spatial distance is higher than the second distance threshold, it indicates that the on-site change trend is similar to the theoretical template but the location is incorrect. This type of situation can be judged as a fault-like trend but with spatial mismatch. Conversely, if the spatial distance is small but the temporal distance is large, it can be judged as a location match but a different evolution rhythm. Both situations can enter the pending review state. If the similarity of all categories is lower than the low confidence threshold, it indicates that the current deviation is more like environmental noise, unknown operating conditions, or unmodeled faults, and the system will not force it to be classified into an existing template.

[0164] During the early morning shift of the drying tower, the temperature in Zone 1 was successively higher in four consecutive windows, while Zones 2 and 3 remained relatively stable. After dynamic time normalization, it was found that although the actual residual and the drift template in Zone 1 were misaligned in the initial time by one window, their upward trajectories were highly consistent.

[0165] After spatial alignment, it was confirmed that the deviation was concentrated in one zone rather than adjacent zones; therefore, the category with the highest final topological similarity was sensor drift. Conversely, when the cold air suddenly intensifies and causes the temperature in the three zones to drop, its temporal variation may be obvious, but the spatial structure does not conform to any modeled fault, so it will not be misjudged as blockage or thermal efficiency decay.

[0166] The purpose of this step is to preserve the evolutionary rhythm and spatial layout of the residuals, thereby enabling robust identification of real failure modes.

[0167] Furthermore, based on each topological similarity, the state assessment result is determined, including: determining the maximum value among the topological similarities corresponding to each theoretical residual matrix as the target topological similarity; when the target topological similarity is greater than or equal to the fault determination threshold, outputting the fault category corresponding to the target topological similarity as the fault state result; when the target topological similarity is less than or equal to the safety determination threshold, outputting the environmental noise state result; when the safety determination threshold is less than the target topological similarity and the target topological similarity is less than the fault determination threshold, outputting the state result to be reviewed; wherein, the fault determination threshold is greater than the safety determination threshold.

[0168] This embodiment provides a step for status output. Specifically, obtaining the similarity of each category alone is not enough to be directly used for operation and maintenance decisions, because the site needs to know clearly whether to determine the fault, determine safety, or need manual verification. Therefore, this embodiment sets up a dual threshold judgment mechanism on top of the similarity results.

[0169] Specifically, the system first finds the maximum value among all categories of topological similarity and uses it as the target topological similarity. For example, if the comparison results are: drift class 0.90, blockage class 0.67, thermal decay class 0.58, then the target topological similarity is 0.90, and the corresponding category is drift class.

[0170] The system then compares it with the fault judgment threshold and safety judgment threshold in the dynamic judgment threshold group. For example, if the fault judgment threshold is 0.85 and the safety judgment threshold is 0.55, then 0.90 is greater than 0.85, and the drift-type fault status result is directly output. If the target topology similarity is less than the preset judgment lower limit, for example, 0.41, 0.46 and 0.38 for the three categories, then the maximum value of 0.46 is still lower than the safety judgment threshold of 0.55. At this time, the environmental noise status result is output. This result does not mean that there is no fluctuation at all, but rather that the current deviation cannot form a sufficient isomorphic relationship with any fault template. It is more likely caused by weather, uneven grain layer or random measurement noise.

[0171] If the target topological similarity is between two thresholds, for example, the maximum value is 0.73, which is greater than 0.55 but less than 0.85, then the result of pending review is output. This state can trigger the extension of the observation window, request manual review of the trend chart, or call a higher frequency sampling strategy. Its value lies in avoiding mistaking early weak faults for safety, and also avoiding prematurely escalating anomalies that have not been fully developed into confirmed faults.

[0172] To enhance the stability of decision-making, the output can be further combined with the consistency of continuous windows; for example, when a single window reaches the fault threshold, it is first marked as a level one warning, and if it still meets the threshold for three consecutive windows, it is upgraded to a confirmed fault; when a single window falls into the area to be reviewed, it can continue to be tracked in the next two windows. If the trend strengthens, it is upgraded to a fault, otherwise it falls back to environmental noise.

[0173] Regarding the data anomaly handling mechanism, if the difference between the maximum similarity and the second largest similarity is too small, for example, less than 0.02, even if the target topological similarity exceeds the fault threshold, it can first output a request for review with two candidate categories to prevent category confusion; if the safety judgment threshold and the fault judgment threshold overlap or are too close due to dynamic updates, for example, the difference is less than the preset minimum interval, the system will automatically perform threshold repair to ensure that the fault threshold is always higher than the safety threshold; if the maximum similarity is high but the corresponding category switches frequently in a short period of time, a category stability constraint can be added, and a formal fault will only be output when the same category is continuously dominant.

[0174] During the first hour of the cold wave hitting the drying tower, the maximum similarity hovered between 0.60 and 0.72. The system continuously output a status pending verification and prompted the on-duty personnel to pay attention to the temperature probe in Zone 1. In the second hour, the residual in Zone 1 continued to rise, and the maximum similarity stabilized above 0.88 with no change in category. Only then did the system officially output a temperature sensor drift fault in Zone 1. During another nighttime period, Zone 3 fluctuated due to the influence of external cold wind, but the similarity of all categories did not exceed 0.50. The system then output the environmental noise status result.

[0175] The purpose of this step is to transform continuous similarity metrics into clear and actionable operational conclusions, thereby achieving a balance between fault identification and false alarm suppression.

[0176] Furthermore, based on the state assessment results, the fault mechanism parameterization rules and dynamic judgment threshold groups are updated, including:

[0177] Record the real-time state data, actual residual matrix, theoretical residual matrix, and state assessment results corresponding to the state assessment results;

[0178] When the condition assessment result is a fault condition result, the drift amplitude, impedance increment and / or thermal efficiency attenuation coefficient corresponding to the fault category are corrected based on real-time condition data and actual residual matrix.

[0179] When the status assessment result is the environmental noise status result, the safety judgment threshold and / or fault judgment threshold shall be adjusted according to the real-time status data.

[0180] When the status assessment result is a status result to be reviewed, the fault mechanism parameterization rules and dynamic judgment threshold group are modified simultaneously.

[0181] This embodiment provides a feedback update step; specifically, if the system uses the initial fault parameters and fixed thresholds for a long time, although it can complete the initial identification, the field features will gradually deviate from the initial template with seasonal changes, changes in grain varieties and equipment aging, resulting in a decrease in identification accuracy; therefore, this embodiment introduces a closed-loop update mechanism that adaptively adjusts according to the evaluation results and archives the results.

[0182] After each status output, the system records the real-time status data of that window, the actual residual matrix, the theoretical residual matrix to be compared, and the final status result. This not only facilitates traceability but also provides training samples for subsequent parameter correction. Each record can be understood as a complete set of samples of on-site performance, theoretical matching, and final conclusion.

[0183] When the status result is a confirmed fault, it indicates that a certain type of theoretical template is close to the field. At this time, the parameters of this type can be further aligned with the field. For example, if the drift fault in Zone 1 is continuously confirmed, and the actual residual growth rate is slightly faster than the existing template, the drift amplitude can be corrected from 0.3℃ per window to 0.4℃.

[0184] If a blockage fault is confirmed and the actual spatial offset is stronger than the template, the impedance increment can be appropriately increased; if a thermal attenuation fault is confirmed and the temperature drop in the later stage is more significant than that in the earlier stage, the nonlinear intensity of the thermal efficiency attenuation coefficient can be increased, so that the theoretical damage matrix generated later can be closer to the actual equipment degradation process.

[0185] When the status result is environmental noise, it indicates that the current deviation should not be attributed to a fault, but is more likely to be a fluctuation in normal operating conditions. If such results occur frequently under certain weather conditions, such as a similarity of 0.50 to 0.58 often occurring on cold nights, it indicates that the safety judgment threshold may be set below the current environmental noise distribution range, or the fault judgment threshold setting range does not cover the actual fault characteristics.

[0186] At this point, the sensitive area near the safety judgment threshold can be appropriately lowered, or the fault threshold can be increased to reduce false alarms caused by environmental disturbances; conversely, if the environmental noise window is very stable for a long time, the safety judgment threshold can also be appropriately increased so that the system can eliminate invalid alarms earlier.

[0187] When the status result is "pending review", it indicates that the current template and threshold may be in a boundary state. At this time, the system corrects both the fault mechanism parameters and the threshold group. For example, if a certain window repeatedly falls around 0.78 and the actual residual is similar to the drift template trend but the amplitude is weaker, the amplitude slope of the drift template can be reduced by a preset step size to make it closer to the early fault. On the other hand, the dynamic range of the fault judgment threshold can be appropriately adjusted to make the boundary judgment more stable.

[0188] To facilitate understanding, a simplified set of data can be used: the initial drift slope is 0.3℃ / window, the fault determination threshold is 0.85, and the safety determination threshold is 0.55; after one week of operation, five confirmed fault samples show a value closer to 0.4℃ / window, so the system updates the drift slope to 0.38℃ / window; at the same time, the maximum similarity of 20 cold wave environmental noise samples is concentrated between 0.52 and 0.60, so the system fine-tunes the safety determination threshold from 0.55 to 0.50, or raises the fault determination threshold from 0.85 to 0.87; through this gradual update, the model will become increasingly adapted to the field situation.

[0189] Regarding the data anomaly handling mechanism, parameter updates should not be too large in one step, otherwise it may cause template drift. To this end, an upper limit can be set for a single update, such as each drift slope correction not exceeding 10%, and the threshold change not exceeding 0.03. If the number of a certain type of fault samples is too small, such as less than 3, then only record it without updating to avoid random samples dominating the model. If the false alarm rate is found to increase during replay verification after the update, then roll back to the previous version of the parameters. If the results pending review are backlogged for a long time, a manual annotation channel can be set up to write the manual conclusions back into the parameter library to accelerate rule convergence.

[0190] After the drying tower entered winter, environmental fluctuations increased significantly. The system found that the original safety judgment threshold caused frequent nighttime checks, increasing the burden on the on-duty personnel. After analyzing archived samples for several days, the system automatically identified that these windows were mainly due to cold wave noise rather than equipment failure, so the noise judgment boundary was appropriately tightened. At the same time, after multiple confirmations of faults, the drift template slope of the temperature sensor in Zone 1 was also corrected to be closer to the actual situation. The number of false alarms from the system was significantly reduced, while the real faults could still be reliably identified.

[0191] The purpose of this step is to ensure that the model and thresholds are not set only once, but are continuously calibrated based on the results of field operation, thereby achieving a simultaneous improvement in recognition accuracy and environmental adaptability.

[0192] The grain drying tower operation status data processing system includes a data acquisition interface, a processor, and a memory. The memory stores a program, which is executed by the processor to implement the above method.

[0193] The data acquisition interface is used to communicate with temperature sensors, humidity sensors, wind pressure sensors, grain in / out detection devices, fan drive devices, heat source supply devices, and environmental monitoring devices deployed in each zone of the drying and cooling sections of the grain drying tower.

[0194] The processor includes: a data sensing module for acquiring multi-zone operating status data, material flow data, equipment operation data, and environmental data of each section of the grain drying tower, generating real-time status data of the grain drying tower; an ideal benchmark reconstruction module for reconstructing the theoretical normal state of the real-time status data based on mass conservation and energy conservation rules, obtaining a theoretical benchmark matrix; a parameterization injection module for injecting damaged states into the theoretical benchmark matrix based on preset fault mechanism parameterization rules corresponding to multiple fault categories, obtaining multiple theoretical damaged matrices; a dual-track difference module for generating a real residual matrix based on the real-time status data and the theoretical benchmark matrix, and generating corresponding theoretical residual matrices based on each theoretical damaged matrix and the theoretical benchmark matrix; a coupling judgment module for calculating the corresponding topological similarity based on the temporal morphological distance and spatial structural distance between the real residual matrix and each theoretical residual matrix, and determining the status assessment result based on the topological similarity; and a feedback update module for updating the fault mechanism parameterization rules and the dynamic judgment threshold group based on the status assessment result, wherein the dynamic judgment threshold group includes at least a fault judgment threshold and a safety judgment threshold.

[0195] This embodiment provides a grain drying tower operation status data processing system, such as... Figure 2 As shown; specifically, the system can be deployed in the industrial control cabinet at the drying tower site. The system executes logic control instructions through a programmable logic controller to achieve real-time scheduling of data from various sensors. Alternatively, it can adopt a distributed architecture of edge industrial control computer + host computer server. The system includes a data acquisition interface, a processor, and a memory. After the processor executes the program in the memory, it completes the aforementioned steps.

[0196] From the perspective of module collaboration, the data acquisition interface is responsible for establishing communication connections with the tower's sensors and devices; the connection method can be a wired industrial bus, serial port, Ethernet, or industrial wireless link; the data received by the interface is first timestamped and parsed, and then transmitted to the data sensing module;

[0197] The output of the data sensing module is not a collection of scattered points, but a real-time status data packet with a unified structure; for example, a data packet is generated every 5 seconds, and each data packet contains temperature and humidity in three zones, wind pressure, grain inlet and outlet quantity, grain inlet and outlet moisture, fan speed, fuel consumption and environmental parameters; after receiving the data packet, the ideal benchmark reconstruction module generates a theoretical benchmark matrix according to the laws of mass conservation and energy conservation; this module can have built-in thermal calculation subroutines, parameter tables and outlier filters;

[0198] The parameterized injection module calls the fault mechanism library to generate multiple theoretical damage matrices on the theoretical benchmark matrix. The mechanism library can pre-store multiple sets of mild, moderate and severe fault templates to support comparisons of different severity levels. The dual-track differential module is responsible for subtracting the field data from the theoretical benchmark and the theoretical damage from the theoretical benchmark to form the actual residual matrix and the theoretical residual matrix.

[0199] The coupling determination module further calculates the temporal morphological distance and spatial structural distance, and outputs the topological similarity and state results. After receiving the state results, the feedback update module corrects the parameter library and threshold library, and writes the updated version back to the memory.

[0200] For ease of explanation, the single processing chain of this system can be understood as follows: the acquisition interface inputs the raw data packet A, the data perception module outputs the real-time state matrix B, the ideal benchmark reconstruction module outputs the benchmark matrix C, the parameterization injection module outputs the damaged matrix set D1, D2, D3, the dual-track difference module outputs the actual residual E and the theoretical residual set F1, F2, F3, the coupling judgment module outputs the similarity set G and the state result H, and the feedback update module outputs the new parameter library I, thus forming a complete closed loop.

[0201] In terms of engineering implementation, the processor can be an industrial controller, an embedded processor, a general-purpose server central processing unit, or an accelerator with parallel computing capabilities; the memory can include a program storage area, a parameter library storage area, a sample archiving area, and a log area; if only rapid early warning is required on site, the first four modules can be executed on the edge device, and the residual matrix can be uploaded to the host computer to complete the coupling judgment and feedback update; if the on-site network is unstable, all modules can run independently locally, and the archived data can be synchronized in batches after the network is restored.

[0202] Regarding the data anomaly handling mechanism, if a certain module fails to run, the system will not necessarily be interrupted in its entirety; for example, if the parameterized injection module is temporarily unavailable, the system can still retain the functions of data acquisition, theoretical reconstruction and actual residual calculation, and output the basic deviation monitoring status.

[0203] If the feedback update module is disabled, the system can still continue to work using the parameters and thresholds of the previous stable version; if the data acquisition interface is interrupted, the system can enter a degraded mode, maintaining short-term monitoring only based on the most recent valid data that has been cached, and stopping status output after the cache time limit is exceeded; if the storage capacity is close to the limit, it can archive on a rolling basis over time, compressing or transferring historical raw data to an external server.

[0204] At the winter operation site of the aforementioned continuous grain drying tower, the data acquisition interface receives status information from three zones every 5 seconds, and the processor completes theoretical reconstruction, fault template injection, and residual comparison on the local industrial control computer.

[0205] When the temperature probe in Zone 1 shows continuous drift, the coupling judgment module provides high similarity results in three consecutive windows, and then prompts the temperature measurement point in Zone 1 for verification through the on-site human-machine interface; if the fluctuation in the cooling section of Zone 3 is only caused by the cold wave, the system displays the ambient noise on the human-machine interface, and there is no need to stop the system; after the maintenance personnel handle it, the feedback update module writes the parameters corresponding to this event back to the parameter library for use in the next round of operation.

[0206] The purpose of this step is to solidify the aforementioned methods into a deployable, scalable, and degradeable system architecture, thereby enabling the on-site implementation of drying tower operation status management.

[0207] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for processing operational status data of a grain drying tower, characterized in that, include: The system acquires multi-zone operating status data, material flow data, equipment operation data, and environmental data for each zone of the grain drying tower, and generates real-time status data, which includes temperature values, humidity values, inlet moisture values, outlet moisture values, fan speed, fuel consumption, ambient temperature, and ambient humidity for each zone. Based on the conservation of dry matter mass and the conservation of enthalpy-humidity balance energy, the real-time state data is reconstructed into a theoretical normal state to obtain a theoretical reference matrix. The theoretical reference matrix includes the theoretical temperature and theoretical humidity values ​​corresponding to each time and each region. Based on the preset fault mechanism parameterization rules, the theoretical reference matrix is ​​subjected to damaged state injection processing to obtain the theoretical damaged matrix. The real residual matrix is ​​generated according to the real-time state data and the theoretical reference matrix, and the theoretical residual matrix is ​​generated according to the theoretical damaged matrix and the theoretical reference matrix. Based on the actual residual matrix and the theoretical residual matrix, temporal morphological features and spatial structural features are extracted respectively, and the temporal morphological distance and spatial structural distance are calculated and fused to obtain the topological similarity, and the state evaluation result is determined accordingly. Based on the state assessment results, the fault mechanism parameterization rules and dynamic judgment threshold group are updated. The dynamic judgment threshold group includes at least a fault judgment threshold and a safety judgment threshold.

2. The method for processing grain drying tower operation status data according to claim 1, characterized in that, Acquire multi-zone operating status data, material flow data, equipment operation data, and environmental data for each section of the grain drying tower, and generate real-time status data for the grain drying tower, including: Based on temperature sensors, humidity sensors, and wind pressure sensors deployed in each zone of the grain drying tower's drying and cooling sections, multi-zone operating status data is collected. Based on the grain inlet and outlet detection devices, material flow data is collected; The material flow data includes grain inflow, grain outflow, grain inflow moisture content, and grain outflow moisture content. Collect equipment operation data based on the fan drive unit and heat source supply unit; The equipment operating data includes fan speed and heat source fuel consumption; Collect environmental data using environmental monitoring devices; The environmental data includes ambient temperature and ambient humidity; The real-time status data is generated based on the multi-zone operation status data, the material flow data, the equipment operation data, and the environmental data.

3. The method for processing grain drying tower operation status data according to claim 1, characterized in that, Based on the mass conservation rule and the energy conservation rule, the real-time state data is reconstructed into a theoretical normal state to obtain a theoretical baseline matrix, including: The amount of moisture migration in the grain drying tower is determined based on the ingested and outgested moisture values ​​in the real-time status data and the ingested and outgested amounts in the material flow data. The heat input of the grain drying tower is determined based on the heat source fuel consumption in the real-time status data. Based on the ambient temperature, ambient humidity and multi-zone operation status data in the real-time status data, the heat exchange conditions of each zone are determined. The heat exchange conditions include one or more of the following: air inlet temperature, air inlet humidity, air pressure and heat exchange coefficient of each zone. Based on the moisture migration, the heat input, and the heat exchange conditions, calculate the theoretical temperature and humidity values ​​for each time period and each zone. The theoretical baseline matrix is ​​generated based on the theoretical temperature and humidity values ​​for each time period and each zone.

4. The method for processing grain drying tower operation status data according to claim 1, characterized in that, Based on the fault mechanism parameterization rules corresponding to multiple fault categories, the theoretical baseline matrix is ​​subjected to damaged state injection processing to obtain multiple theoretical damaged matrices, including: According to the preset sensor drift rules, time drift features are injected into the data nodes in the theoretical reference matrix corresponding to the target partition temperature parameters according to the drift function that increases with time. According to the preset airflow impedance rules, spatial impedance injection is performed on adjacent partition nodes in the theoretical reference matrix according to the impedance increment function between adjacent partitions. According to the preset thermal efficiency decay rule, the global heat distribution in the theoretical benchmark matrix is ​​injected with nonlinear decay according to the thermal efficiency decay function. Based on the injected partition node data and global heat distribution data, multiple theoretical damage matrices corresponding to different fault categories are generated; The fault mechanism parameterization rules include at least one or more of the following: sensor drift parameters, airflow impedance parameters, and thermal efficiency attenuation parameters.

5. The method for processing grain drying tower operation status data according to claim 1, characterized in that, A real residual matrix is ​​generated based on the real-time state data and the theoretical baseline matrix, and a corresponding theoretical residual matrix is ​​generated based on each of the theoretical damage matrices and the theoretical baseline matrix, including: The real-time status data and the theoretical benchmark matrix are mapped to a unified sampling time axis and a unified partition numbering system to obtain the actual alignment matrix and the benchmark alignment matrix, respectively. The real residual matrix is ​​generated based on the element-wise difference or normalized element-wise difference between the real alignment matrix and the reference alignment matrix at the corresponding time position and the corresponding partition position. The theoretical damage matrix and the theoretical reference matrix are mapped to the unified sampling time axis and the unified partition numbering system to obtain the damage alignment matrix and the reference alignment matrix, respectively. Based on the element-wise difference or normalized element-wise difference between the damaged alignment matrix and the reference alignment matrix at the corresponding time position and the corresponding partition position, a theoretical residual matrix corresponding to the fault category is generated.

6. The method for processing grain drying tower operation status data according to claim 1, characterized in that, Based on the temporal morphological distance and spatial structural distance between the actual residual matrix and each of the theoretical residual matrices, the corresponding topological similarity is calculated, and the state evaluation result is determined based on the topological similarity, including: Based on the preset dynamic time warping rules, the temporal morphological distance between the actual residual matrix and each theoretical residual matrix is ​​calculated; Based on the physical adjacency structure of each zone of the grain drying tower, a zone adjacency network is constructed. Based on the spatial manifold alignment rules used to maintain the adjacency relationship between zones, the spatial structural distance between the actual residual matrix and each theoretical residual matrix is ​​calculated. Based on the temporal morphological distance and the spatial structural distance, the topological similarity corresponding to each theoretical residual matrix is ​​calculated; The state evaluation result is determined based on the topological similarity of each of the above.

7. The method for processing grain drying tower operation status data according to claim 6, characterized in that, The state evaluation result is determined based on the topological similarity of each of the aforementioned factors, including: The maximum value among the topological similarities corresponding to each theoretical residual matrix is ​​determined as the target topological similarity. When the target topology similarity is greater than or equal to the fault determination threshold, the fault category corresponding to the target topology similarity is output as the fault status result. When the target topology similarity is less than or equal to the security determination threshold, output the environmental noise status result; When the security determination threshold is less than the target topology similarity and the target topology similarity is less than the fault determination threshold, the status result to be reviewed is output. The fault determination threshold is greater than the safety determination threshold.

8. The method for processing grain drying tower operation status data according to claim 1, characterized in that, Based on the state assessment results, the fault mechanism parameterization rules and dynamic judgment threshold groups are updated, including: Record the real-time state data, actual residual matrix, theoretical residual matrix, and state assessment result corresponding to the state assessment result; When the state assessment result is a fault state result, the drift amplitude, impedance increment and / or thermal efficiency attenuation coefficient corresponding to the fault category are corrected according to the real-time state data and the actual residual matrix. When the state assessment result is an environmental noise state result, the safety judgment threshold and / or fault judgment threshold are corrected based on the real-time state data. When the state assessment result is a state result to be reviewed, the fault mechanism parameterization rule and the dynamic judgment threshold group are simultaneously modified.

9. A grain drying tower operation status data processing system, characterized in that, It includes a data acquisition interface, a processor, and a memory, wherein the memory stores a program that, when executed by the processor, implements the method according to any one of claims 1 to 8; The data acquisition interface is used to communicate with temperature sensors, humidity sensors, wind pressure sensors, grain in / out detection devices, fan drive devices, heat source supply devices, and environmental monitoring devices deployed in each section of the drying and cooling sections of the grain drying tower. The processor includes: The data sensing module is used to acquire multi-zone operating status data, material flow data, equipment operation data and environmental data of each section of the grain drying tower, and generate real-time status data of the grain drying tower. The ideal benchmark reconstruction module is used to reconstruct the theoretical normal state of the real-time state data based on the mass conservation rule and the energy conservation rule, so as to obtain the theoretical benchmark matrix; The parameterized injection module is used to perform damaged state injection processing on the theoretical benchmark matrix based on the fault mechanism parameterization rules corresponding to multiple preset fault categories, so as to obtain multiple theoretical damaged matrices. The dual-track differential module is used to generate a real residual matrix based on the real-time state data and the theoretical reference matrix, and to generate a corresponding theoretical residual matrix based on each of the theoretical damage matrices and the theoretical reference matrix. The coupling determination module is used to calculate the corresponding topological similarity based on the temporal morphological distance and spatial structural distance between the actual residual matrix and each of the theoretical residual matrices, and to determine the state evaluation result based on the topological similarity. The feedback update module is used to update the fault mechanism parameterization rules and the dynamic judgment threshold group according to the state assessment results. The dynamic judgment threshold group includes at least a fault judgment threshold and a safety judgment threshold.

Images