A facility state anomaly diagnosis method and system based on robot inspection data

Abstract

The present application relates to the technical field of industrial process control and automation monitoring, and in particular to a facility state anomaly diagnosis method and system based on robot inspection data, the method comprising: constructing a historical normal data set, constructing a mapping model mapping local density to neighborhood parameters based on the local density extreme range of the data set; for a new inspection data point, determining an adaptive neighborhood parameter using the mapping model according to the real-time local density of the new inspection data point, determining a neighborhood set based on the parameter and calculating a local anomaly factor; and determining the facility state in response to the comparison result of the local anomaly factor and an anomaly threshold. The method solves the false alarm and missed alarm problems caused by density differences in multiple working conditions through an adaptive adjustment mechanism of neighborhood parameters, and improves the accuracy of facility state anomaly diagnosis.

Description

Technical Field

[0001] This invention relates to the field of industrial process control and automation monitoring technology. Specifically, it relates to a method and system for diagnosing facility status anomalies based on robot inspection data. Background Technology

[0002] In the grain and oil processing process, in automated production workshops such as high-pressure boilers and flour milling systems, inspection robots have become a key tool to ensure the continuous and stable operation of facilities. These robots periodically collect multi-dimensional operational data from key facility locations (such as boiler drums, deaerators, and feed airlocks) according to predetermined routes. These data include furnace temperature, drum pressure, fan frequency, and motor current. Based on this inspection data, abnormal conditions of these facilities are diagnosed.

[0003] In existing technologies, the simplest method for diagnosing anomalies in inspection data is to use a fixed threshold for judgment. However, taking a high-pressure boiler as an example, its operating conditions, i.e., production load, will dynamically switch between multiple levels according to production needs, such as 17T, 14T, 10T, 4T, etc. The range of normal operating parameters (such as furnace temperature and pressure) under different operating conditions varies greatly. If a fixed threshold is used, it is very easy to misjudge normal operating data as abnormal when switching operating conditions or running at low load, resulting in the central control room receiving frequent false alarms and seriously interfering with production scheduling.

[0004] To address the false alarm problem caused by using fixed thresholds for anomaly detection, the industry has begun to adopt density-based anomaly detection algorithms, especially... (Local outlier) algorithm, The core idea of ​​the algorithm is that normal data points should have a local density similar to that of their neighbors; while outliers should have a local density significantly lower than that of their neighbors. Theoretically, The algorithm can treat different operating conditions (such as 17T high load and 4T low load) as two or more high-density normal areas in different locations in the data space, while the actual equipment abnormality point will deviate from all these normal areas, thereby achieving condition-independent abnormality diagnosis.

[0005] However, traditional When the algorithm is applied to industrial scenarios with multiple operating conditions, its effectiveness heavily depends on a globally fixed neighborhood parameter K, i.e. the number of neighborhood data points. During boiler inspection, the data density varies greatly under different operating conditions. Under high load conditions, the operating parameters are stable, and the collected data points are highly clustered, forming a high-density area. Under low load conditions, the parameters fluctuate relatively much and the amount of data may be less, forming a low-density area.

[0006] This multi-density coexistence characteristic leads to false alarms and false negatives in the fixed neighborhood parameter K, resulting in inaccurate diagnosis of facility status anomalies. Specifically, if K is set too small to suit high-density operating conditions, when the equipment is operating normally in low-density conditions, a normal inspection data point will have very sparse surroundings, making it difficult to find K neighboring data points within a small radius. This will cause its local reachability density to be incorrectly calculated as very low, resulting in a large LOF value, which will be misjudged as an anomaly, generating a false alarm. If K is set too large to suit low-density operating conditions, when the equipment is operating in high-density conditions, a real, small anomaly point, although deviating from the dense core, will still have a large number of normal neighboring data points within its neighborhood due to the large K. These normal neighboring data points will overwhelm the small anomaly data point, causing the calculated LOF value to be close to 1, which will be misjudged as a normal data point, resulting in a false negative.

[0007] Therefore, it is urgent to design a method that enables the neighborhood parameter K of the LOF algorithm to adapt to the local density of the working conditions in which the data points are located, so as to achieve high-precision and low-false-alarm-rate anomaly diagnosis under multiple working conditions. Summary of the Invention

[0008] To address the problem of false alarms and missed alarms in facility status diagnosis caused by the inability of existing technologies to adapt to data density differences under multiple operating conditions due to the use of fixed neighborhood parameters, this invention proposes a facility status anomaly diagnosis method and system based on robot inspection data.

[0009] In a first aspect, the present invention provides a method for diagnosing facility status anomalies based on robot inspection data, comprising:

[0010] Acquire historical inspection data points consisting of multi-dimensional operational data collected by the robot during facility inspections, and construct a historical normal dataset based on the historical inspection data points;

[0011] Traverse the historical normal dataset, count the number of neighbors of each historical inspection data point based on the preset exploration radius to determine the local density, obtain the extreme value range of the local density in the historical normal dataset, and build a mapping model to map the local density to neighborhood parameters based on the extreme value range.

[0012] For newly collected inspection data points by the robot, the real-time local density of the new inspection data points is calculated based on the historical normal dataset. The adaptive neighborhood parameters of the new inspection data points are calculated using a mapping model. Based on the adaptive neighborhood parameters, the neighborhood set of the new inspection data points is determined in the historical normal dataset. The local reachability density of the new inspection data points is calculated through the neighborhood set. The local anomaly factor of the new inspection data points is determined based on the relative difference between the local reachability density of the new inspection data points and the local reachability density of the historical inspection data points in its neighborhood set.

[0013] Based on the comparison results of local anomaly factors of new inspection data points and preset anomaly thresholds, it is determined whether the facility status has become abnormal, so as to complete the facility status anomaly diagnosis.

[0014] This technical solution first constructs a historical normal dataset as a benchmark and statistically analyzes its density distribution characteristics. Then, a crucial mapping model is built, which transforms the physical density attributes of the data into neighborhood parameters for the algorithm. In the real-time diagnostic phase, instead of using rigid, fixed parameters, the optimal adaptive neighborhood parameters are dynamically matched based on the local density environment of new data points through the mapping model. This mechanism enables the algorithm to automatically shrink the neighborhood to capture subtle deviations when processing high-density operating condition data, and automatically expand the neighborhood to find sufficient statistical support when processing low-density operating condition data. This eliminates statistical biases caused by differences in operating condition density when calculating local reachability density and local anomaly factors, ultimately achieving accurate diagnosis of facility anomalies in complex multi-condition environments.

[0015] Preferably, the determination of local density based on the number of neighbors of each historical inspection data point according to the preset detection radius is carried out in the following manner:

[0016] For any historical inspection data point in the historical normal dataset, calculate the Euclidean distance between the historical inspection data point and other historical inspection data points in the historical normal dataset (excluding itself); count the number of historical inspection data points whose Euclidean distance is less than or equal to the preset detection radius, and determine this number as the local density of the historical inspection data point.

[0017] This technical solution uses a sample counting method within a fixed detection radius to quantify local density. It has a clear physical meaning and can intuitively reflect the degree of aggregation of data points in a multi-dimensional feature space. Through the statistics of Euclidean distance, it can provide accurate quantitative input for the subsequent construction of the mapping relationship between density and neighborhood parameters, thus ensuring the accuracy of density assessment.

[0018] Preferably, the mapping model satisfies the following relationship:

[0019] ;

[0020] in, For new inspection data points Adaptive neighborhood parameters, The preset minimum neighborhood parameter, The preset maximum neighborhood parameter, For new inspection data points Real-time local density, and These represent the maximum and minimum values ​​within the extreme range of local density in the historical normal dataset, respectively. The rounding up symbol.

[0021] This technical solution constructs a negatively correlated linear mapping function to achieve adaptive inverse adjustment of the neighborhood parameters with local density. When the local density is large, the adaptive neighborhood parameters tend to the minimum value, avoiding abnormal flooding caused by an excessively large neighborhood; when the local density is small, the adaptive neighborhood parameters tend to the maximum value, ensuring that enough neighbors can be found for statistical comparison in sparse regions. This design cleverly balances the sensitivity and stability of the algorithm in different density regions.

[0022] Preferably, the locally reachable density satisfies the following relationship:

[0023] ;in, For new inspection data points Locally achievable density, and They are respectively and Adaptive neighborhood parameters, For based on Determined in historical normal datasets The neighborhood set, This represents the total number of historical inspection data points contained in this neighborhood set. For any historical inspection data point contained in this neighborhood set, for and The Euclidean distance between them Centralized historical normal data and The most recent Historical inspection data points and The Euclidean distance between them.

[0024] This technical solution not only utilizes adaptive neighborhood parameters to determine the neighborhood range when calculating local reachability density, but also introduces the adaptive neighborhood parameters of the neighboring points themselves in the calculation of reachability distance. This dual adaptive mechanism ensures that the density calculation can fully consider the local environmental characteristics of the data points and their neighbors, and smooths the distance fluctuations through the maximum value function, thereby obtaining a density index that better reflects the true distribution characteristics.

[0025] Preferably, the local anomaly factors satisfy the following relationship:

[0026] ;in, For new inspection data points Local anomalous factors, For based on Determined in historical normal datasets The neighborhood set, This represents the total number of historical inspection data points contained in this neighborhood set. For any historical inspection data point contained in this neighborhood set, for Locally achievable density, for The locally achievable density.

[0027] This technical solution constructs the final anomaly criterion by comparing the average ratio of the local reachability density of a new inspection data point to that of its neighboring points: if the new inspection data point is normal, its density should be similar to that of its neighbors, with a ratio close to 1; if it is abnormal, its density will be significantly lower than that of its neighbors, resulting in a ratio significantly greater than 1. This can effectively eliminate the interference of operating condition density differences and directly reflect the outlier degree of the new inspection data point.

[0028] Preferably, based on adaptive neighborhood parameters, the neighborhood set of new inspection data points is determined in the historical normal dataset, including: calculating the neighborhood set of each historical inspection data point in the historical normal dataset and... Determine the Euclidean distance between them. The Recent historical inspection data points and The distance between them is used as the target distance. All historical inspection data points whose Euclidean distance is less than or equal to the target distance are considered as the target distance. The neighborhood set.

[0029] Preferably, the real-time local density of new inspection data points is calculated based on the following method:

[0030] Calculate the Euclidean distance between the new inspection data point and each historical inspection data point in the historical normal dataset, count the number of historical inspection data points whose Euclidean distance is less than or equal to the preset detection radius, and use this number as the real-time local density of the new inspection data point.

[0031] Preferably, in response to the comparison result of the local anomaly factor of the new inspection data point and the preset anomaly threshold, it is determined whether the facility status has become abnormal, including:

[0032] The local anomaly factor is compared with the preset anomaly threshold. If the local anomaly factor is greater than the preset anomaly threshold, the facility status is determined to be abnormal. If the local anomaly factor is not greater than the preset anomaly threshold, the facility status is determined to be normal.

[0033] Preferably, the multi-dimensional operational data collected by the robot during facility inspection includes:

[0034] According to the preset inspection route, the robot periodically collects the operating parameters of key points of the facility. The operating parameters include at least one or more of temperature, pressure, frequency and current.

[0035] Secondly, the present invention also provides a facility status anomaly diagnosis system based on robot inspection data. The intelligent detection system includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of any of the intelligent detection methods described above.

[0036] The present invention has the following effects:

[0037] This invention improves upon the limitations of fixed parameters in the traditional LOF algorithm by introducing a mapping model from local density to neighborhood parameters, enabling adaptive adjustment of neighborhood parameters according to operating condition density. Under low-density operating conditions, it automatically expands the neighborhood range to suppress false alarms, and under high-density operating conditions, it automatically reduces the neighborhood range to prevent missed alarms. This significantly improves the accuracy and robustness of facility condition anomaly diagnosis in complex industrial scenarios with multiple operating conditions. Attached Figure Description

[0038] Figure 1 This is a flowchart of the present invention;

[0039] Figure 2 This is a schematic diagram illustrating the anomaly diagnosis effect when the fixed neighborhood parameter K is small in the existing technology;

[0040] Figure 3 This is a schematic diagram illustrating the anomaly diagnosis effect when the fixed neighborhood parameter K is large in the existing technology;

[0041] Figure 4 This is a schematic diagram illustrating the anomaly diagnosis effect of the present invention based on adaptive neighborhood parameters. Detailed Implementation

[0042] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0043] Reference Figure 1 A method for diagnosing facility status anomalies based on robot inspection data, specifically including the following steps:

[0044] S1: Obtain historical inspection data points consisting of multi-dimensional operational data collected by the robot during facility inspections, and construct a historical normal dataset based on the historical inspection data points.

[0045] In large-scale grain and oil processing, sensor data from boiler systems often exhibit multi-dimensional and multi-dimensional characteristics. For example, furnace temperature is measured in degrees Celsius (often in the hundreds), while steam drum pressure is measured in megapascals (often in single digits), representing a significant difference in magnitude. Directly performing Euclidean distance calculations would allow large numerical features to completely dominate the results, masking changes in smaller numerical features and causing the algorithm to fail. Furthermore, to enable the algorithm to learn the normal behavior patterns of the equipment, it is essential to accumulate historical data covering various normal operating conditions (such as loads of different tonnages). Therefore, this step aims to perform data acquisition and preprocessing. By constructing a historical database containing rich operating conditions and performing Z-score standardization on the data to eliminate the influence of dimensions, data with different physical meanings are mapped to a unified dimensionless statistical space, thus establishing an accurate mathematical benchmark for subsequent distance- and density-based calculations.

[0046] Specifically, the inspection robot is first activated and controlled to periodically collect operating parameters from key points of the facilities, following a preset route: boiler control room, high-pressure boiler main steam drum, high-pressure deaerator, rice husk feeding airlock, etc. Each key point will collect a historical inspection data point consisting of multiple dimensions of operating parameters. The collected historical inspection data points, including all normal operating conditions (high, medium, and low load conditions) and manually verified, gathered over a month by the robot, constitute the historical normal dataset. The multiple dimensions of operating parameters include at least one or more of temperature, pressure, frequency, and current.

[0047] Since the LOF algorithm is distance-based, and the physical dimensions of values ​​differ significantly across dimensions, it is necessary to perform Z-Score standardization on the historical normal dataset according to each dimension. The purpose of Z-Score standardization is to eliminate the influence of dimensions, ensuring that the processed data has a mean of 0 and a variance of 1. For example, if the mean furnace temperature in historical data is 800℃ and the standard deviation is 50℃, and the furnace temperature at a certain inspection data point is 825℃, its standardized value would be (825-800) / 50 = 0.5. Thus, through data collection and standardization, a foundation is laid for subsequent density assessment and... Value computation provides a unified and dimensionless data foundation.

[0048] S2: Traverse the historical normal dataset, count the number of neighbors of each historical inspection data point based on the preset exploration radius to determine the local density, and obtain the extreme range of local density in the historical normal dataset.

[0049] After standardizing the data, to address the issue of fixed-parameter algorithms failing to adapt to varying density across different operating conditions, it's crucial to avoid simply assuming uniform distribution characteristics across all data regions. Therefore, this step aims to establish a macroscopic operating condition density sensing mechanism. By introducing a fixed detection radius of 0.5, the surrounding environment of each historical inspection data point in the historical normal dataset is explored, and the number of its neighbors is counted to quantify the density of the area where that historical inspection data point is located, i.e., local density. This step serves as a bridge connecting the original data distribution with algorithm parameter adjustments. Its purpose is to define the dynamic range of equipment operating conditions by statistically analyzing the maximum and minimum values ​​of local density, providing accurate boundary conditions for subsequently building an adaptive parameter mapping model.

[0050] Specifically, for any historical inspection data point in the historical normal dataset, calculate the Euclidean distance between the historical inspection data point and other historical inspection data points in the historical normal dataset (excluding itself); count the number of historical inspection data points whose Euclidean distance is less than or equal to 0.5, and determine this number as the local density of the historical inspection data point.

[0051] After traversal, the maximum local density of all historical inspection data points in the historical normal dataset is calculated. and minimum value To form the extreme range of local density [ , ],in, This typically corresponds to a high-load, stable operating condition for equipment, where historical inspection data points are highly clustered. Historical data points are typically sparse, usually corresponding to low load or operating condition switching processes.

[0052] S3: Construct a mapping model based on the extreme value range to map local density to neighborhood parameters in order to determine the adaptive neighborhood parameters of new inspection data points.

[0053] After determining the extreme range of local density of all historical inspection data points, given that under high-density conditions, historical inspection data points are compact and minor anomalies are easily overwhelmed by a large number of normal neighbors, while under low-density conditions, historical inspection data points are sparse and normal data points are easily misjudged because they cannot find neighbors.

[0054] Therefore, this step aims to construct an inverse adaptive regulator by establishing a mathematical mapping function that maps the local density of the data to the neighborhood parameters of the algorithm. The values ​​are associated, making The value can automatically scale according to the environmental density of the inspection data points. This process gives the algorithm adaptive adjustment capability, ensuring that the algorithm can use the most reasonable observation scale for detection under any working condition.

[0055] Specifically, the mapping model satisfies the following relationship:

[0056]

[0057] in, For new inspection data points Adaptive neighborhood parameters, The preset minimum neighborhood parameter is 5. The preset maximum neighborhood parameter is 100. For new inspection data points Real-time local density, and These represent the maximum and minimum values ​​within the extreme range of local density in the historical normal dataset, respectively. The rounding up symbol.

[0058] This relation is a linear inverse interpolation model, where the fractional term is essentially a normalized sparsity coefficient, and the numerator term... Indicates new inspection data points The difference between the real-time local density and the maximum local density, the denominator term This represents the span of the local density of all historical inspection data points. When a new inspection data point... When in a high-density area, The numerator term decreases, causing the overall ratio to approach 0, and the final calculated... Approaching This small Value-based strategies can narrow the observation range, finely characterize local features, and prevent minor anomalies in dense areas from being masked, thereby solving the problem of missed reports. When new inspection data points... When in a low-density area, The numerator term increases, causing the overall ratio to approach 1, and the final calculated value is... Approaching This kind of big Value-based strategies can expand the search scope, ensuring that sparse points can find enough neighbors to maintain statistical stability and prevent them from being misclassified as isolated points, thereby solving the false alarm problem.

[0059] After constructing the adaptive mapping model, whenever the robot collects a new inspection data point in real time, it needs to perform real-time adaptive neighborhood parameter operation. By calculating the local density of the new inspection data point relative to the historical normal dataset, the working environment (dense or sparse) is located. Then, using the mapping model, the optimal neighborhood parameters of the new inspection data point are dynamically calculated. This is the entry point for the algorithm to run online. Its purpose is to ensure that each newly arrived inspection data point can obtain a customized detection parameter, laying the foundation for subsequent accurate calculations.

[0060] For example, , , , Substitute this into the mapping model:

[0061]

[0062] When the local density of new inspection data points is very high, such as ,get ,near Suitable for use in high-density areas The expected value. When the local density of new inspection data points is very low, such as... ,get ,equal Suitable for use in low-density areas Expected value.

[0063] Thus, by constructing density estimation and K-value mapping function, the core settings for the algorithm's adaptive capability were completed, laying the foundation for subsequent diagnosis.

[0064] S4: Based on the adaptive neighborhood parameters of the new inspection data points, determine the neighborhood set of the new inspection data points in the historical normal dataset, and calculate the local reachability density of the new inspection data points through the neighborhood set.

[0065] After determining the adaptive neighborhood parameters of the new inspection data point, it is necessary to analyze the local geometric features of the new inspection data point based on these adaptive neighborhood parameters. This step introduces the concept of local reachability density, which describes the density of data points more accurately by considering the distribution of neighboring points.

[0066] In particular, this step introduces an reachability distance assessment mechanism in the calculation and emphasizes the use of the adaptive neighborhood parameters of the neighboring points themselves. This fully reflects the differences in the local characteristics of the data and avoids the bias of measuring different regions with a uniform standard.

[0067] Specifically, firstly, the relationship between each historical inspection data point in the historical normal dataset and... Determine the Euclidean distance between them. The Recent historical inspection data points and The distance between them is used as the target distance. All historical inspection data points whose Euclidean distance is less than or equal to the target distance are considered as the target distance. neighborhood set .

[0068] For example, The adaptive neighborhood parameters are At that time, in the historical normal dataset, calculate the sum of each historical inspection data point. The Euclidean distances between them are sorted in ascending order of Euclidean distance as distances. The data is sorted from nearest to farthest, and then the 8th historical inspection data point is selected for calculation. The Euclidean distance d8 between the 8th historical inspection data point and the 8th historical normal data point is then used to centralize and... All historical inspection data points whose Euclidean distance between them is less than d8 constitute neighborhood set .

[0069] Then, the reachability distance between the new inspection data point and each historical inspection data point in its neighborhood set is calculated, and the reciprocal of the average reachability distance between the new inspection data point and all historical inspection data points in its neighborhood set is taken as the local reachability density of the new inspection data point.

[0070] Specifically, the locally reachable density is calculated using the following relationship:

[0071]

[0072] in, For new inspection data points Locally achievable density, and They are respectively and Adaptive neighborhood parameters, For based on Determined in historical normal datasets The neighborhood set, This represents the total number of historical inspection data points contained in this neighborhood set. For any historical inspection data point contained in this neighborhood set, for and The Euclidean distance between them Centralized historical normal data and The most recent Historical inspection data points and The Euclidean distance between them is calculated by dividing all historical inspection data points in the historical normal dataset by distance. Sort by nearest and farthest, then select the first... Calculate the number of historical inspection data points. Historical inspection data points and The Euclidean distance between them.

[0073] The physical meaning of this relation is the number of neighbors within a unit reachable distance, and the denominator is the number of all neighbors relative to... The sum of reachable distances, yes The reachability distance between the algorithm and any neighbor o in its neighborhood set is calculated using a smoothing process in the denominator to prevent the density from approaching infinity due to an excessively small denominator, thus enhancing the algorithm's robustness to noise.

[0074] It is important to note that this involves all historical data points within the forced neighborhood set. When calculating its own reachability, the new inspection data points are used uniformly. Adaptive neighborhood parameters This ensures that the density comparison of new inspection data points and their neighbors is performed under the same physical scale. Without this unification, the micro-density under high-density conditions (based on small K values) and the macro-density under low-density conditions (based on large K values) cannot be directly compared due to their different defined spatial ranges. This is achieved through locking... This eliminated density differences across multiple operating conditions, resulting in the final product... Able to truly reflect Has it deviated from its current local environment? The larger the value, the better. The more crowded the surroundings; The smaller the value, the better. The surrounding area is more open.

[0075] S5: Determine the local anomaly factor of the new inspection data point based on the relative difference between the local reachability density of the new inspection data point and the local reachability density of the historical inspection data points in its neighborhood set, so as to diagnose the facility status anomaly.

[0076] This is the final decision-making step in the diagnostic process. Since local reachability density is significantly affected by operating conditions, the density in high-load areas is naturally much higher than in low-load areas, making direct comparison of local reachability density meaningless. This step aims to isolate the influence of the operating condition background by calculating the relative density ratio, constructing a dimensionless index that purely reflects the degree of anomaly. Regardless of whether the operating conditions are sparse or dense, if a new inspection data point exhibits a significant sense of alienation relative to its surrounding environment, it is judged as an anomaly.

[0077] Specifically, for all historical inspection data points in the neighborhood set of the new inspection data point, their respective neighborhood sets are divided according to the adaptive neighborhood parameters of the new inspection data point. The local reachability density of each historical inspection data point in the neighborhood set of the new inspection data point is calculated according to the process of step S4. The mean of the local reachability density of all historical inspection data points in the neighborhood set of the new inspection data point is calculated. The ratio of this mean to the local reachability density of the new inspection data point itself is used as the local anomaly factor of the new inspection data point.

[0078] It is important to note that this involves all historical data points within the forced neighborhood set. When calculating its own local reachability density, the new inspection data points are used uniformly. Adaptive neighborhood parameters Each node is divided into its own neighborhood set, and then its local reachability density is calculated. This is to establish a unified benchmark for comparing local reachability densities. If the local reachability density is directly calculated using the adaptive neighborhood parameters of historical inspection data points within the neighborhood set, the resulting density values ​​are not physically comparable due to the different volume definitions of sparse and dense regions. This is addressed by enforcing a unified benchmark. The algorithm removes the influence of differences in working conditions, so that the final local anomaly factor can purely reflect the degree of outlier of the new inspection data point relative to its current local environment, thereby solving the problem of false alarms and missed alarms under multiple working conditions.

[0079] The local anomaly factor is calculated using the following relationship:

[0080]

[0081] in, For new inspection data points Local anomalous factors, For based on Determined in historical normal datasets The neighborhood set, This represents the total number of historical inspection data points contained in this neighborhood set. For any historical inspection data point contained in this neighborhood set, for Locally achievable density, for The locally achievable density.

[0082] The relationship is a ratio structure. The numerator is the mean of the local reachability of all historical inspection data points in the neighborhood set of the new inspection data point, and the denominator is the local reachability of the new inspection data point itself. The whole is used to reflect the relative difference between the local reachability of the new inspection data point and the average level of the local reachability of its neighborhood set. This design enables the algorithm to measure the degree of anomaly under different working conditions and achieves true working condition adaptation.

[0083] For example, when This is a normal point in a high-density area, with local density reaching [a certain level]. Since their neighborhood sets are also normal points in high-density regions, their respective local reachability densities are also around 50. For example, the average local reachability density of the neighborhood set is... ,but The local anomaly factor is ;when It is a small outlier in a high-density area, and its neighborhood set is still composed of normal points in those high-density areas. Therefore, the average local reachability density of the neighborhood set is still around 50. However, Its deviation from the dense core increases the reachability distance between it and historical inspection data points within its neighborhood set. Its local reachability density will be significantly reduced, for example, ,but The local anomaly factor is .

[0084] Finally, a global LOF anomaly threshold is preset. ,Will Local anomalous factors and If a comparison is made, Greater than The system determines that the facility is in an abnormal state and immediately sends an alarm signal to the central control room system. Not greater than The facility was determined to be in normal condition.

[0085] Reference Figures 2 to 4 With furnace temperature as the X-axis and steam drum pressure as the Y-axis, the background points include a dense point cloud in the lower left corner (normal operating condition 1, high load) and a sparse point cloud in the upper right corner (normal operating condition 2, low load). The test point is P1, a normal point in a dense region; P2, a normal point in a sparse region; P3, a global anomaly; and P4, a tiny anomaly in a dense region. The diagnostic threshold is... . Reference Figure 2This demonstrates that in existing technologies, when the LOF algorithm has a fixed neighborhood parameter K=5, P2 is marked as a false alarm because the K value is too small to find neighborhood data points in the sparse region, while P4 is marked as an anomaly and correctly diagnosed. This method leads to the false alarm of P2. (See reference...) Figure 3 This demonstrates that in the existing LOF algorithm, with a fixed neighborhood parameter K=100, P2 is marked as normal and the diagnosis is correct, but P4 is falsely flagged because the K value is too large, causing P4 to be overwhelmed by a large number of normal neighborhood data points. (Refer to...) Figure 4 This demonstrates the beneficial effects of the present invention. The adaptive neighborhood parameter K of P2 is adaptively set to K=100 (a large K value), with a score below 1.5, indicating a correct diagnosis. Similarly, the adaptive neighborhood parameter K of P4 is adaptively set to K=6 (a small K value), with a score above 1.5, also indicating a correct diagnosis. Thus, the present invention, through dynamic K values, simultaneously solves the false alarm problem of P2 and the false alarm problem of P4, achieving high-precision adaptive anomaly diagnosis.

[0086] The present invention also provides a facility condition anomaly diagnosis system based on robot inspection data. The intelligent detection system includes a processor and a memory. The memory stores computer program instructions. When the computer program instructions are executed by the processor, they implement all or part of the steps of the facility condition anomaly diagnosis method in this application. The processor may be a central processing unit (CPU), a graphics processing unit (GPU), an application-specific integrated circuit (ASIC), or a field-programmable gate array (FPGA). The memory may include volatile memory (such as RAM) and non-volatile memory (such as ROM or flash memory).

[0087] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for diagnosing facility status anomalies based on robot inspection data, characterized in that, include: Acquire historical inspection data points consisting of multi-dimensional operational data collected by the robot during facility inspections, and construct a historical normal dataset based on the historical inspection data points; Traverse the historical normal dataset, count the number of neighbors of each historical inspection data point based on the preset exploration radius to determine the local density, obtain the extreme value range of the local density in the historical normal dataset, and build a mapping model to map the local density to neighborhood parameters based on the extreme value range. For newly collected inspection data points by the robot, the real-time local density of the new inspection data points is calculated based on the historical normal dataset, and the adaptive neighborhood parameters of the new inspection data points are calculated using a mapping model. By calculating the local density of the new inspection data points relative to the historical normal dataset, the working environment in which they are located is located, and then the optimal neighborhood parameters of the new inspection data points are calculated using a mapping model. This allows the neighborhood range to be automatically expanded in low-density working conditions to suppress false alarms, and the neighborhood range to be automatically reduced in high-density working conditions to prevent missed alarms. Based on adaptive neighborhood parameters, a neighborhood set of new inspection data points is determined in the historical normal dataset. The local reachability density of the new inspection data points is calculated through the neighborhood set. The local anomaly factor of the new inspection data points is determined based on the relative difference between the local reachability density of the new inspection data points and the local reachability density of historical inspection data points in their neighborhood set. The mapping model satisfies: ； For new inspection data points Adaptive neighborhood parameters, The preset minimum neighborhood parameter, The preset maximum neighborhood parameter, For new inspection data points Real-time local density, and These represent the maximum and minimum values ​​within the extreme range of local density in the historical normal dataset, respectively. The rounding up symbol; Based on the comparison results of local anomaly factors of new inspection data points and preset anomaly thresholds, it is determined whether the facility status has become abnormal, so as to complete the facility status anomaly diagnosis.

2. The facility status anomaly diagnosis method based on robot inspection data according to claim 1, characterized in that, The determination of local density based on counting the number of neighbors for each historical inspection data point using a preset detection radius is performed in the following way: For any historical inspection data point in the historical normal dataset, calculate the Euclidean distance between the historical inspection data point and other historical inspection data points in the historical normal dataset (excluding itself); count the number of historical inspection data points whose Euclidean distance is less than or equal to the preset detection radius, and determine this number as the local density of the historical inspection data point.

3. The method for diagnosing facility status anomalies based on robot inspection data according to claim 1, characterized in that, Locally reachable density satisfies the following relationship: ;in, For new inspection data points Locally achievable density, and They are respectively and Adaptive neighborhood parameters, Based on Determined in historical normal datasets The neighborhood set, This represents the total number of historical inspection data points contained in this neighborhood set. For any historical inspection data point contained in this neighborhood set, for and The Euclidean distance between them Centralized historical normal data and The most recent Historical inspection data points and The Euclidean distance between them.

4. The facility status anomaly diagnosis method based on robot inspection data according to claim 3, characterized in that, Local anomaly factors satisfy the following relationship: ;in, For new inspection data points Local anomalous factors, Based on Determined in historical normal datasets The neighborhood set, This represents the total number of historical inspection data points contained in this neighborhood set. For any historical inspection data point contained in this neighborhood set, for Locally achievable density, for The locally achievable density.

5. The facility status anomaly diagnosis method based on robot inspection data according to claim 1, characterized in that, Based on adaptive neighborhood parameters, the neighborhood set of new inspection data points is determined in the historical normal dataset, including: Calculate the relationship between each historical inspection data point in the historical normal dataset and Determine the Euclidean distance between them. The Recent historical inspection data points and The distance between them is used as the target distance. All historical inspection data points whose Euclidean distance is less than or equal to the target distance are considered as the target distance. The neighborhood set.

6. The method for diagnosing facility status anomalies based on robot inspection data according to claim 1, characterized in that, The real-time local density of the new inspection data points is calculated based on the following method: Calculate the Euclidean distance between the new inspection data point and each historical inspection data point in the historical normal dataset, count the number of historical inspection data points whose Euclidean distance is less than or equal to the preset detection radius, and use this number as the real-time local density of the new inspection data point.

7. The method for diagnosing facility status anomalies based on robot inspection data according to claim 1, characterized in that, Based on the comparison results between local anomaly factors of new inspection data points and preset anomaly thresholds, determine whether the facility status has become abnormal, including: The local anomaly factor is compared with the preset anomaly threshold. If the local anomaly factor is greater than the preset anomaly threshold, the facility status is determined to be abnormal. If the local anomaly factor is not greater than the preset anomaly threshold, the facility status is determined to be normal.

8. The method for diagnosing facility status anomalies based on robot inspection data according to claim 1, characterized in that, Acquiring multi-dimensional operational data collected by the robot during facility inspection includes: periodically collecting operational parameters of key points of the facility using the robot according to a preset inspection route, wherein the operational parameters include at least one or more of temperature, pressure, frequency, and current.

9. A facility status anomaly diagnosis system based on robot inspection data, characterized in that, The system includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of a facility status anomaly diagnosis method based on robot inspection data as described in any one of claims 1-8.

Images