A cloud platform auditing method based on collaborative auditing

Abstract

The application discloses a cloud platform auditing method based on cooperative auditing, relates to the technical field of cloud platform auditing, and significantly improves the cooperative efficiency and data reliability of building engineering safety auditing through multi-dimensional technology fusion; aiming at the non-standard document processing problem, semantic depth analysis and pixel-level detection technology are adopted to accurately identify the implicit hierarchical relationship of unstructured Excel, and the workload of manual adjustment is reduced; the operation fingerprint sequence in the offline environment and the Merkel tree storage mechanism guarantee the data integrity when multiple people modify simultaneously, and key information loss is avoided; meanwhile, based on the media file identification system of BIM coordinates, combined with GeoHash coding and time stamp time correction, accurate space-time positioning and rapid retrieval of engineering photos are realized.

Description

Technical Field

[0001] This invention relates to the field of cloud platform auditing technology, and in particular to a cloud platform auditing method based on collaborative auditing. Background Technology

[0002] In the field of safety auditing of construction projects, collaborative work among multiple parties has become the norm. Supervision units, construction parties, design institutes and other institutions need to conduct joint inspections of hundreds of acceptance nodes such as scaffolding erection and concrete pouring. The audit checklist contains thousands of technical specification clauses. Currently, mainstream cloud platform auditing systems generally adopt a microservice architecture, support functions such as mobile on-site photo evidence collection and automatic synchronization after offline editing, and some systems integrate BIM model visualization comparison modules.

[0003] Existing technologies often struggle with unstructured audit materials, such as Excel lists submitted by construction companies, which frequently use multiple methods to divide chapters and levels, including merging cells, indentation spaces, and color coding. This leads to frequent errors in the cloud platform's automatic recognition module. For example, it might misclassify "3.2.1 Foundation Pit Support" as a fourth-level sub-item, causing delays in the acceptance process. Although some solutions introduce template matching algorithms based on regular expressions, they are not effective for non-standard documents commonly found in the engineering industry, requiring supervising engineers to spend a significant amount of time manually adjusting the hierarchical structure. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] This invention provides a cloud platform auditing method based on collaborative auditing to solve the problems of traditional auditing relying on paper circulation, such as inefficient multi-role collaboration, chaotic project hierarchy management, easy loss of offline data, and difficulty in identifying non-standard documents.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a cloud platform auditing method based on collaborative auditing, which includes:

[0008] Step S1: The platform operations administrator establishes a project hierarchy tree in the cloud backend. The hierarchy tree includes a parent project node and at least two levels of child project nodes. The child project nodes automatically inherit the parent project's audit list template and member configuration.

[0009] Step S2: Receive multi-format review documents uploaded by users, perform semantic deep parsing on unstructured Excel documents, construct a document hierarchical structure tree based on cell text features and indentation pixel detection, and generate an interactive review list with topological relationships.

[0010] Step S3: The mobile terminal captures on-site audit data in offline mode and generates an operation fingerprint sequence for parallel modifications of the same audit item. The fingerprint sequence includes the operator identifier, timestamp, and content hash value.

[0011] Step S4: When the network is restored, conflicting modifications are automatically merged according to the topological relationship of the fingerprint sequence, and all operation trajectories are preserved to form a version evolution map;

[0012] Step S5: The media files captured on site are used to generate a unique identifier by the triplet of project number-shooting location coordinates-timestamp, and then associated with the structured data node of the corresponding review item.

[0013] Step S6: Dynamically adjust the expert resource allocation weights based on reinforcement learning, and generate the optimal assignment scheme according to the urgency of the review items, the expert domain matching degree, and the current load status.

[0014] As a preferred embodiment of the cloud platform auditing method based on collaborative auditing described in this invention, the semantic deep parsing in step S2 specifically includes:

[0015] Extract text features from cells in an Excel document and identify feature words containing "Chapter X" and "Section X" as hierarchical anchors;

[0016] The left boundary indentation of cells is obtained by pixel-level detection, and a mapping relationship between indentation and layer depth is established. The OpenCV edge detection algorithm is used to identify the left boundary of cells, and the relative indentation is calculated by the horizontal projection difference between adjacent cells, establishing a mapping rule that each pixel corresponds to 0.1 layers.

[0017] Detect visual dividing lines formed by consecutive blank cells and use them as boundary markers for sibling nodes;

[0018] Perform visual continuity analysis on tables spanning multiple pages and achieve automatic splicing through header feature matching.

[0019] As a preferred embodiment of the cloud platform auditing method based on collaborative auditing described in this invention, the operation fingerprint sequence generation method in step S3 includes:

[0020] Differential encoding is used to perform text modification operations to generate an operation instruction set for the smallest editing unit; the insertion operation is defined as INS(position,content), the deletion operation as DEL(position,length), and the replacement operation as REP(position,old_content,new_content), where position is located using two-dimensional row and column coordinates;

[0021] Calculate the SHA-256 hash value associated with the content for each editing unit;

[0022] The operation instruction set, hash value, device fingerprint, and GPS positioning data are encapsulated into an atomic operation package;

[0023] A Merkle tree structure is used to organize atomic operation packages, forming an immutable operation chain.

[0024] As a preferred embodiment of the cloud platform review method based on collaborative review described in this invention, the media file identifier generation method in step S5 includes:

[0025] Extract the 3D spatial coordinates of the shooting points from the engineering BIM model and convert them into the WGS84 geographic coordinate system;

[0026] The captured timestamp is encoded as a combination of Unix time and millisecond precision; when the device clock deviation is detected to exceed ±5 minutes, the NTP time synchronization service is automatically triggered and the clock correction log is recorded.

[0027] The GeoHash algorithm is used to reduce the dimensionality of spatial coordinates and generate a 9-bit feature string;

[0028] The media file base name is formed by combining the project number, the GeoHash string, and the timestamp.

[0029] As a preferred embodiment of the cloud platform auditing method based on collaborative auditing described in this invention, step S5, which involves using the GeoHash algorithm to perform dimensionality reduction encoding of spatial coordinates, includes:

[0030] The 3D spatial coordinates extracted from the engineering BIM model are converted into WGS84 geographic coordinates, and then the resulting latitude and longitude coordinates are encoded using the GeoHash algorithm for dimensionality reduction.

[0031] As a preferred embodiment of the cloud platform auditing method based on collaborative auditing described in this invention, the dimensionality reduction coding step in step S5 specifically comprises:

[0032] Define the Earth ellipsoid parameters and calculate the eccentricity and semi-minor axis:

[0033] b1 = a1(1-f1),

[0034] Where a1 represents the semi-major axis of the Earth's ellipsoid, and f1 represents the flattening of the ellipsoid. b1 represents the square of the first eccentricity, and b1 represents the semi-minor axis of the ellipsoid. Represents the square of the second eccentricity;

[0035] Convert the BIM 3D coordinates (U, V, W) to longitude λ and latitude.

[0036] λ = arctan 2(V,U),

[0037]

[0038] Where U, V, W represent the spatial coordinate components extracted from the BIM model, λ represents longitude, p represents the horizontal distance projected onto the equatorial plane, and θ represents the Bowring auxiliary angle. Here, arctan 2(y,x) represents latitude, and arctan 2(y,x) represents the arctangent function with quadrant information.

[0039] right GeoHash encoding involves the following steps:

[0040] a) Initialize the interval:

[0041]

[0042] b) Perform binary search to generate a bit sequence:

[0043]

[0044] in, Indicates the initial lower and upper limits of longitude. Indicates the initial lower and upper limits of latitude, L (k) B (k) This represents the longitude and latitude positions after the k-th binary search, where k represents the binary search round index, k = 0, 1, ..., K-1, and K here is related to the final character length.

[0045] c) Perform alternating arrangements and grouping:

[0046]

[0047] Among them, g i s represents the interleaved bit sequence. j This represents the decimal index of the j-th group after conversion from 5 binary bits, where j represents the group index, and there are a total of 9 groups.

[0048] d) Mapped to a Base32 string:

[0049] GeoHash=encode32(s0)|…|encode32(s8),

[0050] Here, encode32(·) means mapping the decimal index from 0 to 31 to the Base32 character table, and | means string concatenation, finally resulting in a 9-bit GeoHash string, completing the dimensionality reduction encoding.

[0051] As a preferred embodiment of the cloud platform auditing method based on collaborative auditing described in this invention, the expert assignment scheme optimization method in step S6 includes:

[0052] Construct a three-dimensional feature vector, which includes the urgency score of the review item, the historical accuracy rate of expert review, and the current amount of tasks to be processed; the urgency score is calculated by weighting three sub-items: the time slack before project milestones, the impact coefficient of design changes, and the supervision level of the regulatory department, with a weight ratio of 4:3:3, which can be adjusted as needed;

[0053] Design a dual-objective optimization function to simultaneously minimize task processing latency and expert load imbalance.

[0054] An improved Hungarian algorithm is used to solve for the optimal assignment matrix, and a preemptive scheduling strategy is set for urgent tasks.

[0055] As a preferred embodiment of the cloud platform auditing method based on collaborative auditing described in this invention, in step S6, during the process of setting a preemptive scheduling strategy for urgent tasks, a task-expert matching weight matrix is ​​constructed based on the urgency of the audit item, the expert domain matching degree, and the current load. Based on this, a preemptive scheduling strategy is integrated, and an improved Hungarian algorithm is used to solve for the optimal assignment matrix. Specifically:

[0056] Define the task set and expert set:

[0057]

[0058] Where T represents the set of all m unassigned review items, and E represents the set of all n experts.

[0059] Construct the basic weight matrix W based on the parameters:

[0060] W ij =αu i +β(1-a ij )+γl j ,

[0061] Among them, W ij This indicates that entry t will be... i Assigned to expert e j The overall cost, where α, β, and γ represent urgency, domain mismatch, and load weighting coefficients, respectively, u i Indicates entry t i Urgency rating, a ijExpert e j For entry t i Historical accuracy rate in professional fields, l j Expert e j Current number of tasks to be processed;

[0062] Let the set of emergency entries be E. T :

[0063] E T ={i∣u i ≥U th},

[0064] Define a pre-emption bias constant P that satisfies:

[0065]

[0066] Construct the pre-occupancy weight matrix

[0067]

[0068] Among them, U th This represents the urgency threshold, and P represents the priority bias during preemptive scheduling.

[0069] Under the premise of non-negative cost, Add constant offset D:

[0070]

[0071] Among them, C ij Given the final input cost matrix of the Hungarian algorithm, D guarantees that all C ij ≥0;

[0072] Solving the binary assignment matrix X = (x) using an improved Hungarian algorithm ij ):

[0073]

[0074] Constraints:

[0075] Where, x ij =1 indicates that entry t i Assigned to expert e j Constraint 1 ensures that each item is processed by exactly one expert, and constraint 2 ensures that each expert processes only one item at a time.

[0076] As a preferred embodiment of the cloud platform auditing method based on collaborative auditing described in this invention, the method further includes a version difference visualization step.

[0077] A difference tracking layer is embedded in the authorized summary version, which records the modification trajectory of each audited item;

[0078] A semantically based three-color annotation strategy is used to identify text differences: red indicates changes in key parameters, yellow indicates the addition of supplementary explanations, and blue indicates adjustments to logical relationships. Changes in key parameters refer to modifications involving mechanical performance indicators or material specifications. Adjustments to logical relationships refer to changes in the complex calculation formulas of construction procedures or acceptance conditions.

[0079] It generates dynamic comparison views for differences in structured data, and supports two-way linkage comparison of allocation tables and progress Gantt charts.

[0080] As a preferred embodiment of the cloud platform auditing method based on collaborative auditing described in this invention, the permission inheritance method in the hierarchical tree includes:

[0081] Set an independent permission switch for each sub-project node, and the switch can override the permission configuration inherited from the parent node;

[0082] When a sub-project enables independent permissions, a member capability profile library for that node is automatically generated.

[0083] The capability profile database includes members' historical response speed, approval rate, and professional field tags;

[0084] Dynamically allocate audit task load thresholds to sub-project nodes based on the capability profile library.

[0085] The beneficial effects of this invention are as follows: Through multi-dimensional technology integration, this invention significantly improves the collaborative efficiency and data reliability of construction engineering safety audits; addressing the challenges of non-standard document processing, it employs semantic deep analysis and pixel-level detection technology to accurately identify the hidden hierarchical relationships in unstructured Excel files, reducing the workload of manual adjustments; the operation fingerprint sequence and Merkle tree evidence storage mechanism in offline environments ensure data integrity during concurrent modifications by multiple users, preventing the loss of critical information; simultaneously, based on a BIM coordinate-based media file identification system, combined with GeoHash encoding and timestamp synchronization, it achieves accurate spatiotemporal positioning and rapid retrieval of engineering photos; the introduction of an expert resource dynamic scheduling model, through three-dimensional feature fusion and an improved Hungarian algorithm, optimizes the rationality of task allocation, balancing emergency response and load balancing; and the adoption of version evolution maps and semantic annotation strategies intuitively presents differences in the audit trajectory, strengthening process traceability capabilities. Attached Figure Description

[0086] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0087] Figure 1 This is a flowchart illustrating the cloud platform audit method based on collaborative auditing in Example 1. Detailed Implementation

[0088] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0089] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0090] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0091] Example 1, referring to Figure 1 This embodiment provides a cloud platform auditing method based on collaborative auditing, including the following steps:

[0092] Step S1: The platform operations administrator creates a project hierarchy tree in the cloud backend. The hierarchy tree contains a parent project node and at least two levels of child project nodes. The child project nodes automatically inherit the parent project's audit list template and member configuration.

[0093] The permission inheritance methods in a hierarchical tree include:

[0094] Set an independent permission switch for each sub-project node; the switch can override the permission configuration inherited from the parent node.

[0095] When a sub-project enables independent permissions, a member capability profile library for that node is automatically generated.

[0096] The competency profile database includes members’ historical response speed, approval rate, and professional field tags; Response speed = Σ(task processing time) / number of valid tasks, where tasks that have not been processed for more than 24 hours are counted as 24 hours, and abnormally interrupted tasks are not included in the statistical sample.

[0097] Dynamically allocate audit task load thresholds to sub-project nodes based on the capability profile library;

[0098] Step S2: Receive multi-format review documents uploaded by users, perform semantic deep parsing on unstructured Excel documents, construct a document hierarchical structure tree based on cell text features and indentation pixel detection, and generate an interactive review list with topological relationships.

[0099] The semantic deep parsing in step S2 specifically includes:

[0100] Extract text features from cells in an Excel document and identify feature words containing "Chapter X" and "Section X" as hierarchical anchors;

[0101] The left boundary indentation of cells is obtained by pixel-level detection, and a mapping relationship between indentation and layer depth is established. The OpenCV edge detection algorithm is used to identify the left boundary of cells, and the relative indentation is calculated by the horizontal projection difference between adjacent cells, establishing a mapping rule that each pixel corresponds to 0.1 layers.

[0102] Detect visual dividing lines formed by consecutive blank cells and use them as boundary markers for sibling nodes;

[0103] Perform visual continuity analysis on tables spanning multiple pages and automatically stitch them together by matching header features;

[0104] Step S3: The mobile terminal captures on-site audit data in offline mode, and generates an operation fingerprint sequence for parallel modifications of the same audit item. The fingerprint sequence includes the operator identifier, timestamp and content hash value.

[0105] The method for generating the operation fingerprint sequence in step S3 includes:

[0106] Differential encoding is used to perform text modification operations to generate an operation instruction set for the smallest editing unit; the insertion operation is defined as INS(position,content), the deletion operation as DEL(position,length), and the replacement operation as REP(position,old_content,new_content), where position is located using two-dimensional row and column coordinates;

[0107] Calculate the SHA-256 hash value associated with the content for each editing unit;

[0108] The operation instruction set, hash value, device fingerprint, and GPS positioning data are encapsulated into an atomic operation package;

[0109] A Merkle tree structure is used to organize atomic operation packages, forming an immutable operation chain;

[0110] In step S3, the methods for calculating the SHA-256 hash value related to the content for each editing unit include:

[0111] The content is bit-encoded and padded to obtain the padded message M′:

[0112] M′=X|1|0 k |[l] 64 ,

[0113] Where X represents editing unit C j The bit sequence after UTF-8 encoding, where k represents the number of zeros padded, l represents the bit length of X, 64 represents the big-endian binary length, [l] 64 The symbol represents the 64-bit big-endian binary representation of 'l', and | represents the concatenation operation.

[0114] According to the filling rules:

[0115] l+1+k+64≡0 (mod 512),

[0116] Where 512 represents the bit length of each message packet;

[0117] The padded message M′ is split into several 512-bit blocks, and the hash state is initialized.

[0118] M′=M (1) |M (2) |…|M (N) ,

[0119] Among them, M (i) This represents the i-th 512-bit group, and N represents the total number of groups;

[0120] Each group is further divided into 16 32-bit words:

[0121] in, This represents the 0th 32-bit word of the i-th group;

[0122] Initialize the hash vector:

[0123]

[0124] in, This represents the 0th component of the initial hash state vector, where 0 represents the component index, which ranges from 0 to 7.

[0125] Perform message expansion and 64 rounds of compression on each group, and update the hash state;

[0126] To expand the message, the formula is:

[0127]

[0128] Among them, W t Let σ0(x) and σ1(x) represent the value of the message expansion sequence in round t, and let σ0(x) and σ1(x) represent the lowercase sigma function, where σ0(x) = ROTR. 7 (x)⊕ROTR 18 (x)⊕SHR 3 (x),

[0129] σ1(x)=ROTR 17 (x)⊕ROTR 19 (x)⊕SHR 10 (x),

[0130] 'x' represents the 32-bit input word, 7, 18, 3, 17, 19, and 10 represent the number of bits for the circular right shift or logical right shift, '⊕' represents the bitwise XOR operation, and 'ROTR' represents the input word. n SHR represents a circular right shift of n bits. n This indicates a logical right shift by n bits;

[0131] Round function compression:

[0132] T1=h+Σ1(e)+Ch(e,f,g)+K t +W t ,

[0133] T2=Σ0(a)+Maj(a,b,c),

[0134] Where a, b, c, d, e, f, g, h represent the 8 components of the working register.

[0135] Σ0(x)=ROTR 2 (x)⊕ROTR 13 (x)⊕ROTR 22 (x),

[0136] Σ1(x)=ROTR 6 (x)⊕ROTR 11 (x)⊕ROTR 25 (x),

[0137]

[0138] Maj(x,y,z)=(x∧y)⊕(x∧z)⊕(y∧z),

[0139] K t T1 and T2 represent constants in round t, where t represents the current round index, and T1 and T2 represent temporary calculation variables.

[0140] Update register state:

[0141] (a,b,c,d,e,f,g,h)←(T1+T2,a,b,c,d+T1,e,f,g),

[0142] All additions are modulo 2. 32 Operations;

[0143] Combine the outputs of all groups to form the final hash;

[0144] Specifically, this process strictly follows the SHA256 standard to perform padding, grouping, message expansion, and multiple rounds of compression operations on the content of editing units of arbitrary length, ensuring that the same content is always mapped to a unique 256-bit hash value. The encoding and padding steps ensure that the data is aligned to 512-bit blocks. The message expansion function achieves high diffusion through circular right shift and XOR mixing. The compression function outputs an irreversible digest after dozens of rounds of nonlinear operations. The final accumulation and merging cause the verification states of each group to be gradually superimposed.

[0145] This method is collision-resistant and pre-image-resistant, and can generate reliable fingerprint sequences in an offline environment for subsequent collision merging and evolution tracking.

[0146] Step S4: When the network is restored, conflicting modifications are automatically merged according to the topological relationship of the fingerprint sequence, and all operation trajectories are preserved to form a version evolution map;

[0147] Step S5: The media files captured on site are used to generate a unique identifier by the triplet of project number-shooting location coordinates-timestamp, and then associated with the structured data node of the corresponding review item.

[0148] The media file identifier generation method in step S5 includes:

[0149] Extract the 3D spatial coordinates of the shooting points from the engineering BIM model and convert them into the WGS84 geographic coordinate system;

[0150] The captured timestamp is encoded as a combination of Unix time and millisecond precision; when the device clock deviation is detected to exceed ±5 minutes, the NTP time synchronization service is automatically triggered and the clock correction log is recorded.

[0151] The GeoHash algorithm is used to reduce the dimensionality of spatial coordinates and generate a 9-bit feature string;

[0152] Combine the project number, GeoHash string, and timestamp to form the base name of the media file;

[0153] Step S5, which involves using the GeoHash algorithm to perform dimensionality reduction encoding of spatial coordinates, includes:

[0154] The three-dimensional spatial coordinates extracted from the engineering BIM model are converted into WGS84 geographic coordinates, and then the resulting latitude and longitude are encoded using the GeoHash algorithm for dimensionality reduction.

[0155] The dimensionality reduction encoding step in step S5 is as follows:

[0156] Define the Earth ellipsoid parameters and calculate the eccentricity and semi-minor axis:

[0157] b1 = a1(1-f1),

[0158] Where a1 represents the semi-major axis of the Earth's ellipsoid, and f1 represents the flattening of the ellipsoid. b1 represents the square of the first eccentricity, and b1 represents the semi-minor axis of the ellipsoid. Represents the square of the second eccentricity;

[0159] Convert the BIM 3D coordinates (U, V, W) to longitude λ and latitude.

[0160] λ = arctan 2(V,U),

[0161]

[0162] Where U, V, W represent the spatial coordinate components extracted from the BIM model, λ represents longitude, p represents the horizontal distance projected onto the equatorial plane, and θ represents the Bowring auxiliary angle. Here, arctan 2(y,x) represents latitude, and arctan 2(y,x) represents the arctangent function with quadrant information.

[0163] right GeoHash encoding involves the following steps:

[0164] a) Initialize the interval:

[0165]

[0166] b) Perform binary search to generate a bit sequence:

[0167]

[0168] in, Indicates the initial lower and upper limits of longitude. Indicates the initial lower and upper limits of latitude, L (k) B (k) This represents the longitude and latitude positions after the k-th binary search, where k represents the binary search round index, k = 0, 1, ..., K-1, and K here is related to the final character length.

[0169] c) Perform alternating arrangements and grouping:

[0170]

[0171] Among them, g i s represents the interleaved bit sequence. j This represents the decimal index of the j-th group after conversion from 5 binary bits, where j represents the group index, and there are a total of 9 groups.

[0172] d) Mapped to a Base32 string:

[0173] GeoHash=encode32(s0)|…|encode32(s8),

[0174] Here, encode32(·) means mapping the decimal index from 0 to 31 to the Base32 character table, and | means string concatenation, finally resulting in a 9-bit GeoHash string, completing the dimensionality reduction encoding;

[0175] Specifically, the Bowring method is used to accurately project the 3D coordinates of the BIM model into latitude and longitude in the WGS84 coordinate system, ensuring high-precision geographic location acquisition. Subsequently, the GeoHash algorithm effectively converts the two-dimensional continuous values ​​through iterative binary search and positional interleaving. It converts the data into a fixed-length discrete character sequence, achieving spatial dimensionality reduction and regional encoding. The 9-bit GeoHash divides the Earth's surface into approximately 5m × 5m grid units, facilitating indexing and spatial similarity search. This method maps high-dimensional spatial locations into concise strings, providing an efficient and scalable solution for the accurate location and retrieval of media files.

[0176] Step S6: Dynamically adjust the expert resource allocation weights based on reinforcement learning, and generate the optimal assignment scheme according to the urgency of the review items, the expert domain matching degree, and the current load status.

[0177] The expert assignment scheme optimization method in step S6 includes:

[0178] Construct a three-dimensional feature vector, which includes the urgency score of the review item, the historical accuracy rate of expert review, and the current amount of tasks to be processed; the urgency score is calculated by weighting three sub-items: the time slack before project milestones, the impact coefficient of design changes, and the supervision level of the regulatory department, with a weight ratio of 4:3:3, which can be adjusted as needed;

[0179] Design a dual-objective optimization function to simultaneously minimize task processing latency and expert load imbalance.

[0180] An improved Hungarian algorithm is used to solve for the optimal assignment matrix, and a preemptive scheduling strategy is set for urgent tasks.

[0181] In step S6, during the process of setting a preemptive scheduling strategy for urgent tasks, a task-expert matching weight matrix is ​​constructed based on the urgency of the review items, the expert domain matching degree, and the current load. The preemptive scheduling strategy is then integrated into this matrix, and the optimal assignment matrix is ​​solved using an improved Hungarian algorithm. Specifically:

[0182] Define the task set and expert set:

[0183]

[0184] Where T represents the set of all m unassigned review items, and E represents the set of all n experts.

[0185] Construct the basic weight matrix based on the parameters:

[0186] W ij =αu i +β(1-a ij )+γl j ,

[0187] Among them, W ij This indicates that entry t will be... i Assigned to expert e j The overall cost, where α, β, and γ represent urgency, domain mismatch, and load weighting coefficients, respectively, u i Indicates entry t i Urgency rating, a ij Expert e j For entry t i Historical accuracy rate in professional fields, l j Expert e j Current number of tasks to be processed;

[0188] Let the set of emergency entries be E. T :

[0189] E T ={i∣u i ≥U th},

[0190] Define a pre-emption bias constant P that satisfies:

[0191]

[0192] Construct the pre-occupancy weight matrix

[0193]

[0194] Among them, U th This represents the urgency threshold, and P represents the priority bias during preemptive scheduling.

[0195] Under the premise of non-negative cost, Add constant offset D:

[0196]

[0197] Among them, C ij Given the final input cost matrix of the Hungarian algorithm, D guarantees that all C ij ≥0;

[0198] Solving the binary assignment matrix X = (x) using an improved Hungarian algorithm ij ):

[0199]

[0200] Constraints:

[0201] Where, x ij =1 indicates that entry t i Assigned to expert e j Constraint 1 ensures that each item is processed by exactly one expert, and constraint 2 ensures that each expert processes only one item at a time.

[0202] Specifically, the task-expert matching problem is reduced to a weighted bipartite graph minimum assignment problem. By introducing a weight matrix W, the three-dimensional features of urgency, domain matching, and expert load are systematically integrated, overcoming the shortcomings of traditional single cost measurement. For high-priority urgent items, the cost matrix is ​​dynamically adjusted through preemption bias P, and the preemption requirement is seamlessly integrated into the Hungarian framework. In addition, the constant offset D ensures the algorithm's stable support for arbitrary values, greatly improving the solution efficiency and robustness. The final optimal assignment matrix ensures that all items are accurately allocated while giving priority to urgent items, effectively shortening the scheduling latency of high-priority tasks, and taking into account the load balancing and professional matching among experts, thus meeting the dual requirements of real-time performance and fairness.

[0203] The cloud platform's review methods also include methods for visualizing version differences:

[0204] Embed a difference tracking layer in the authorized summary version; the layer records the modification trajectory of each review item.

[0205] A semantically based three-color annotation strategy is used to identify text differences: red indicates changes in key parameters, yellow indicates the addition of supplementary explanations, and blue indicates adjustments to logical relationships. Changes in key parameters refer to modifications involving mechanical performance indicators or material specifications. Adjustments to logical relationships refer to changes in the complex calculation formulas of construction procedures or acceptance conditions.

[0206] It generates dynamic comparison views for differences in structured data, and supports two-way linkage comparison of allocation tables and progress Gantt charts.

[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, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A cloud platform auditing method based on collaborative auditing, characterized in that, include, Step S1: The platform operations administrator establishes a project hierarchy tree in the cloud backend. The hierarchy tree includes a parent project node and at least two levels of child project nodes. The child project nodes automatically inherit the parent project's audit list template and member configuration. Step S2: Receive multi-format review documents uploaded by users, perform semantic deep parsing on unstructured Excel documents, construct a document hierarchical structure tree based on cell text features and indentation pixel detection, and generate an interactive review list with topological relationships. The semantic deep parsing specifically includes: Extract text features from cells in an Excel document and identify feature words containing "Chapter X" and "Section X" as hierarchical anchors; The left boundary indentation of cells is obtained by pixel-level detection, and a mapping relationship between indentation and layer depth is established. The OpenCV edge detection algorithm is used to identify the left boundary of cells, and the relative indentation is calculated by the horizontal projection difference between adjacent cells, establishing a mapping rule that each pixel corresponds to 0.1 layers. Detect visual dividing lines formed by consecutive blank cells and use them as boundary markers for sibling nodes; Perform visual continuity analysis on tables spanning multiple pages and automatically stitch them together by matching header features; Step S3: The mobile terminal captures on-site audit data in offline mode and generates an operation fingerprint sequence for parallel modifications of the same audit item. The fingerprint sequence includes the operator identifier, timestamp, and content hash value. Step S4: When the network is restored, conflicting modifications are automatically merged according to the topological relationship of the fingerprint sequence, and all operation trajectories are preserved to form a version evolution map; Step S5: The media files captured on site are used to generate a unique identifier by the triplet of project number-shooting location coordinates-timestamp, and then associated with the structured data node of the corresponding review item. Step S6: Dynamically adjust the expert resource allocation weights based on reinforcement learning, and generate the optimal assignment scheme according to the urgency of the review items, the expert domain matching degree, and the current load status.

2. The cloud platform auditing method based on collaborative auditing as described in claim 1, characterized in that, The method for generating the operation fingerprint sequence in step S3 includes: Differential encoding is applied to text modification operations to generate a set of operation instructions for the smallest editing unit; Calculate the SHA-256 hash value associated with the content for each editing unit; The operation instruction set, hash value, device fingerprint, and GPS positioning data are encapsulated into an atomic operation package; A Merkle tree structure is used to organize atomic operation packages, forming an immutable operation chain.

3. The cloud platform auditing method based on collaborative auditing as described in claim 1, characterized in that, The media file identifier generation method in step S5 includes: Extract the 3D spatial coordinates of the shooting points from the engineering BIM model and convert them into the WGS84 geographic coordinate system; Encode the shooting timestamp as a combination of Unix time and millisecond precision; The GeoHash algorithm is used to reduce the dimensionality of spatial coordinates and generate a 9-bit feature string; The media file base name is formed by combining the project number, the GeoHash string, and the timestamp.

4. The cloud platform auditing method based on collaborative auditing as described in claim 3, characterized in that, Step S5, which involves using the GeoHash algorithm to perform dimensionality reduction encoding of spatial coordinates, includes: The 3D spatial coordinates extracted from the engineering BIM model are converted into WGS84 geographic coordinates, and then the resulting latitude and longitude coordinates are encoded using the GeoHash algorithm for dimensionality reduction.

5. The cloud platform auditing method based on collaborative auditing as described in claim 4, characterized in that, The dimensionality reduction encoding step in step S5 is as follows: Define the Earth ellipsoid parameters and calculate the eccentricity and semi-minor axis: ， in, Represents the semi-major axis of the Earth's ellipsoid. This represents the flattening of the ellipsoid. This represents the square of the first eccentricity. Indicates the semi-minor axis of the ellipsoid. Represents the square of the second eccentricity; BIM 3D coordinates Convert to longitude with latitude : ， ， in, This represents the spatial coordinate components extracted from the BIM model. Indicates longitude. This represents the horizontal distance projected onto the equatorial plane. Indicates the Bowring auxiliary angle. Indicates latitude, Represents the arctangent function with quadrant information; right GeoHash encoding involves the following steps: a) Initialize the interval: ， b) Perform binary search to generate a bit sequence: ， ， in, Indicates the initial lower and upper limits of longitude. Indicates the initial lower and upper limits of latitude. , Indicates the first Longitude and latitude positions after the second bisection This represents a binary round index. Here, K is related to the final character length; c) Perform alternating arrangements and grouping: ， ， in, This represents the interleaved bit sequence. Indicates the first Group 5 decimal indices converted from binary. This represents a group index, with a total of 9 groups; d) Mapped to a Base32 string: ， in, This means mapping the decimal indices from 0 to 31 to the Base32 character table. This indicates string concatenation, ultimately resulting in a 9-character GeoHash string, completing the dimensionality reduction encoding.

6. The cloud platform auditing method based on collaborative auditing as described in claim 1, characterized in that, The expert assignment scheme optimization method in step S6 includes: Construct a three-dimensional feature vector, which includes the urgency score of the review item, the historical accuracy rate of expert reviews, and the current number of tasks to be processed; Design a dual-objective optimization function to simultaneously minimize task processing latency and expert load imbalance. An improved Hungarian algorithm is used to solve for the optimal assignment matrix, and a preemptive scheduling strategy is set for urgent tasks.

7. The cloud platform auditing method based on collaborative auditing as described in claim 6, characterized in that, In step S6, during the process of setting a preemptive scheduling strategy for urgent tasks, a task-expert matching weight matrix is ​​constructed based on the urgency of the review item, the expert domain matching degree, and the current load. The preemptive scheduling strategy is then integrated into this matrix, and the optimal assignment matrix is ​​solved using an improved Hungarian algorithm. Specifically: Define the task set and expert set: ， in, Indicates all A set of review items to be assigned. Indicates all A group of experts gathered. Construct the basic weight matrix based on the parameters. : ， in, Indicates the entry Assigned to experts The overall cost, These represent urgency, domain mismatch, and load weighting coefficients, respectively. Indicates an entry Urgency rating, Experts For the entry Historical accuracy rate of audits in professional fields Experts Current number of tasks to be processed; Let the set of emergency entries be... : ， Define the pre-emption bias constant. ,satisfy: ， Construct the pre-occupancy weight matrix : ， in, This indicates the threshold for determining urgency. This indicates the priority bias in preemptive scheduling. Under the premise of non-negative cost, Add constant offset : ， ， in, The final input cost matrix for the Hungarian algorithm is... Protect all ; Solving the binary assignment matrix using an improved Hungarian algorithm : ， ， in, Indicates the entry Assigned to experts Constraint 1 ensures that each item is processed by exactly one expert, and constraint 2 ensures that each expert processes only one item at a time.

8. The cloud platform auditing method based on collaborative auditing as described in claim 7, characterized in that, The method also includes a version difference visualization step: A difference tracking layer is embedded in the authorized summary version, which records the modification trajectory of each audited item; A semantically based three-color annotation strategy is used to identify text differences: red indicates changes in key parameters, yellow indicates the addition of supplementary explanations, and blue indicates adjustments to logical relationships. It generates dynamic comparison views for differences in structured data, and supports two-way linkage comparison of allocation tables and progress Gantt charts.

9. A cloud platform auditing method based on collaborative auditing as described in claim 8, characterized in that, The permission inheritance methods in the hierarchical tree include: Set an independent permission switch for each sub-project node. When the switch is enabled, it will override the permission configuration inherited from the parent node. When a sub-project enables independent permissions, a member capability profile library for that node is automatically generated. The capability profile database includes members' historical response speed, approval rate, and professional field tags; Dynamically allocate audit task load thresholds to sub-project nodes based on the capability profile library.

Images