An AI-based park investment process data monitoring method and system

Abstract

This invention relates to the field of structured data query technology, specifically to an AI-based data monitoring method and system for the investment promotion process in industrial parks. The method includes the following steps: collecting logs to construct a time-series event sequence; calculating funding intervals and penalty scores to correct resource attributes; calculating attenuation weights based on interaction intervals to construct spatiotemporal potential energy constraints; traversing path combinations to select priority models and calculating deviations; identifying the degree of violation at abnormal nodes; and generating investment promotion behavior monitoring alerts. In this invention, by traversing path combinations and accumulating potential energy scores to select priority paths, and calculating the deviation between nominal state and deduced path scores to generate logical deviation indicators, the invention achieves in-depth mining of hidden violations or abnormal progress nodes throughout the entire investment promotion chain. This effectively solves the problem of insufficient identification accuracy in traditional monitoring under multi-dimensional variable dynamic coupling scenarios, improving the timeliness and accuracy of risk warnings during the investment promotion process.

Description

Technical Field

[0001] This invention relates to the field of structured data query technology, and in particular to an AI-based method and system for monitoring data during the investment promotion process in industrial parks. Background Technology

[0002] The field of structured data query technology mainly involves a technical system for targeted retrieval and extraction of standardized data stored in relational databases or two-dimensional tables. Core aspects of this field include index building, query parsing, and data matching. It systematically utilizes structured query languages ​​to perform conditional filtering and sorting operations on datasets containing fields such as text, numerical values, and timestamps, thereby obtaining data records that conform to specific logic. Traditional data monitoring methods for industrial park investment promotion refer to the technical aspects of real-time tracking and verification of data such as basic enterprise information, negotiation progress records, and contract signing status generated during industrial park investment promotion activities. Traditional industrial park investment promotion data monitoring relies on manual entry of investment logs into spreadsheets or basic databases. Business personnel execute query commands containing specific field matching rules, such as retrieving unupdated records within a specific time period, or confirming the current stage of a project by comparing it with preset investment promotion stage identifiers, and judging whether the investment promotion progress meets expectations based on single-dimensional numerical thresholds.

[0003] Existing investment promotion monitoring relies on manual data entry and static rule matching. The timeliness and accuracy of the data are limited by human operation delays and input errors. Simply relying on field comparison or single-dimensional numerical threshold judgment makes it difficult to deeply correlate the dynamic coupling relationship between multi-dimensional variables such as negotiation time, budget and carrier matching degree. When facing complex investment promotion scenarios, it is unable to accurately capture hidden risks such as budget deviation or cycle anomalies. This results in rigid monitoring logic and a lack of overall control over the spatiotemporal evolution of the entire process. As a result, the identification of abnormal states is lagging and prone to misjudgment and omission, which cannot meet the requirements of refined management for real-time risk warning. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides an AI-based data monitoring method for the investment promotion process in industrial parks, comprising the following steps: S1: Collect the original investment promotion interaction logs of the park's investment promotion management, arrange them according to timestamps, extract text, industry, budget amount, carrier and price, and construct a time-series investment promotion event sequence; S2: Call the time-series investment promotion event sequence, calculate the funding gap between the budget amount and the price and compare it with the tolerance limit. For nodes that exceed the tolerance limit, generate negative penalty scores and add them to the initial score to generate a resource attribute matching correction sequence. S3: Based on the resource attribute matching correction sequence, calculate the interaction interval duration and compare it with the upper limit of the standard time interval obtained. For event pairs that exceed the upper limit of the standard time interval, calculate the attenuation weight and multiply it into the total transfer score to construct a spatiotemporal potential dynamic constraint sequence. S4: Call the spatiotemporal potential energy dynamic constraint sequence, filter the model deduction path with priority in score, obtain the nominal state path and calculate the score deviation from the model deduction path, and generate the investment promotion path logic deviation index. S5: Based on the deviation index of the investment promotion path logic, filter the project numbers that exceed the preset abnormal alarm limit, extract the abnormal node and carrier information, identify the type and severity of violations, and generate an abnormal investment promotion behavior monitoring alarm.

[0005] As a further aspect of the present invention, the nominal state path and the model deduction path are traversed respectively to identify multiple nodes included in the path and the edges connecting multiple nodes, and to extract the basic potential energy attributes of nodes and edges in the spatiotemporal potential energy dynamic constraint sequence. The potential energy scores of nodes and edges in the nominal state path are summed to generate a nominal total potential energy value. Simultaneously, the potential energy scores of nodes and edges in the model deduction path are summed to generate a standard total potential energy value.

[0006] As a further aspect of the present invention, the time-series investment promotion event sequence includes ordered event nodes, timestamps corresponding to the nodes, and extracted structured attribute feature data; the resource attribute matching correction sequence includes funding matching interval values, generated negative penalty scores, and corrected initial node scores; the spatiotemporal potential energy dynamic constraint sequence includes interaction interval duration values, calculated time decay weight coefficients, and constrained state transition scores; the investment promotion path logic deviation index includes the cumulative total score of the model deduction path, the cumulative total score of the nominal state path, and the path score deviation degree between the two; and the abnormal investment promotion behavior monitoring alarm includes the index number of the abnormal project, a description of the identified violation type, and a risk severity level.

[0007] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Collect the original investment promotion interaction logs of the park's investment promotion management, traverse the entire data and parse the absolute timestamp value of each log record, use the unique project number index to lock multiple interaction records associated with the same investment promotion project, perform ascending sorting operation on multiple interaction records according to the absolute timestamp value, and generate a time-ordered investment promotion log set. S102: Call the time-ordered investment promotion log set, extract the unstructured text information describing the negotiation content, the industry category and budget amount of the intended enterprise, and the type and rental price of the promotion carrier from the log entries, convert the unstructured text information into a multi-dimensional semantic vector, and perform standardized encoding processing on the numerical attributes to obtain the structured feature data of investment promotion elements. S103: Based on the structured feature data of the investment promotion elements, obtain the time index information of the corresponding log entries, perform chain concatenation operation on the feature data belonging to the same investment promotion project according to the time dimension, establish a two-way mapping relationship between feature data and time nodes, assign business status labels and attribute weight values ​​to sequence nodes, reconstruct the discrete feature data, and generate a time-series investment promotion event sequence.

[0008] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Call the time-series investment promotion event sequence, traverse multiple event nodes in the time-series investment promotion event sequence, read the budget amount value of the intended enterprise and the rental price value of the promotion carrier associated with the node, perform numerical subtraction operation on the budget amount value of the intended enterprise and the rental price value of the promotion carrier under the same node, obtain the absolute value, quantify the funding matching distance, and generate a set of supply and demand funding matching differences. S202: Based on the set of supply and demand funding mismatches, obtain a preset funding matching tolerance threshold, perform a numerical comparison operation between the supply and demand funding mismatch and the funding matching tolerance threshold, filter out the difference items where the supply and demand funding mismatch is greater than the funding matching tolerance threshold, calculate a negative penalty value proportional to the excess of the difference item, and generate a non-fit node penalty mapping table. S203: Based on the non-adaptive node penalty mapping table, obtain the initial probability evaluation scores of multiple event nodes, accumulate the corresponding negative penalty values ​​of the nodes recorded in the non-adaptive node penalty mapping table to the initial probability evaluation scores, update the state probability distribution parameters of the nodes based on the calculated new scores, and generate a resource attribute matching correction sequence.

[0009] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Traverse the investment promotion event nodes within the resource attribute matching and correction sequence, filter adjacent node pairs with flow relationship, extract the absolute timestamp attribute markers of the subsequent node and the preceding node, perform differential operation on each pair of adjacent nodes, obtain the interval duration value, and generate a set of adjacent event time domain differences. S302: Based on the set of time-domain differences between adjacent events, retrieve the database of successful investment cases in the park and extract the standard time distribution data of similar state transitions. According to the principle of normal distribution, calculate the upper limit of the standard transition time covering the core interval. Perform a numerical comparison operation between the multiple differences in the set and the upper limit of the standard transition time. For long-tail delay terms whose values ​​exceed the upper limit, calculate the penalty coefficient and generate an irregular interval attenuation coefficient table. S303: According to the irregular interval decay coefficient table, obtain the initial flow score defined in the preset state transition probability matrix. For the delayed event pairs recorded in the irregular interval decay coefficient table, perform a weighted multiplication operation on the initial flow score using the penalty coefficient to keep the original score of the non-delayed event pairs constant. Perform logical association between the calculated edge potential energy score and the existing node potential energy score in the sequence to generate a spatiotemporal potential energy dynamic constraint sequence.

[0010] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Call the spatiotemporal potential energy dynamic constraint sequence, construct the state transition topology grid, traverse the node connection path combinations inside the state transition topology grid, integrate the resource matching correction term and time decay weight term in the sequence, accumulate the node potential energy score and edge potential energy score for the node connection path combination, perform global optimization operation, filter the node sequence with priority in the accumulated total score, record the theoretical optimal state transition trajectory and peak potential energy total score, and generate the optimal path set for model deduction; S402: Based on the model, deduce the optimal path set, extract the investment promotion progress log filled in by business personnel, parse the status change record, reconstruct the business flow link, map the business flow link to the probability evaluation space defined by the spatiotemporal potential energy dynamic constraint sequence, and calculate the nominal state path potential energy value. S403: Based on the nominal state path potential energy value, retrieve the peak potential energy total score recorded in the optimal path set derived by the model, perform numerical subtraction operation between the nominal state path potential energy value and the peak potential energy total score to obtain the score difference, perform standardization processing on the score difference, calculate the deviation magnitude, and generate the investment promotion path logical deviation index.

[0011] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Call the investment promotion path logical deviation indicator, obtain the preset business compliance anomaly alarm limit, perform a numerical comparison operation between the investment promotion path logical deviation indicator and the anomaly alarm limit, filter target project records with deviations greater than the anomaly alarm limit, identify the associated unique identity index number and perform serialization encapsulation, and generate an abnormal project index number list. S502: Based on the list of abnormal project index numbers, backtrack and retrieve the time-series investment promotion event sequence, locate the coordinates of the business nodes that caused the logical deviation, extract the industry category, budget amount, and type and rental price of the associated intended enterprise, and perform an aggregation operation on the business node coordinates and the extracted attribute data to generate a multi-dimensional feature set of abnormal nodes. S503: Based on the multi-dimensional feature set of the abnormal nodes, call the diagnostic model for investment promotion violations, input the aggregated feature data into the model to perform feature matching, identify the type of violation, calculate the quantitative rating coefficient of the risk severity, integrate the project number, violation type and severity rating data, and generate an abnormal investment promotion behavior monitoring alarm.

[0012] As a further aspect of the present invention, the multidimensional feature set of the abnormal node is analyzed, and the budget amount value and the rental price value are subjected to maximum and minimum normalization processing to construct a numerical deviation feature vector. Simultaneously, a one-hot encoding method is used to digitally transform the industry category of the target enterprise and the type of the promotion carrier, and construct semantic attribute feature vectors.

[0013] An AI-based data monitoring system for the investment promotion process in industrial parks, comprising: The investment promotion interaction sequence construction module collects the original investment promotion interaction logs of the park's investment promotion management, performs sorting operations on the original investment promotion interaction logs based on the timestamp values, extracts text, industry, budget amount, carrier and price fields, and constructs a time-series investment promotion event sequence. The resource attribute matching and correction module calls the time-series investment promotion event sequence, calculates the funding gap between the budget amount and the price and compares it with the obtained tolerance limit. For nodes that exceed the tolerance limit, a negative penalty score is generated and added to the initial score to generate the resource attribute matching and correction sequence. The dynamic constraint module calculates the interaction interval duration based on the resource attribute matching correction sequence and compares it with the upper limit of the standard time interval. For event pairs that exceed the upper limit of the standard time interval, it calculates the attenuation weight and multiplies it into the total transfer score to construct a spatiotemporal potential dynamic constraint sequence. The investment promotion path identification module calls the spatiotemporal potential energy dynamic constraint sequence, traverses the path combination and accumulates the node and edge potential energy scores, filters the model deduction path with priority in score, obtains the nominal state path and calculates the score deviation from the model deduction path, and generates the investment promotion path logical deviation index. The anomaly monitoring module, based on the deviation indicators of the investment promotion path logic, filters project numbers that exceed the preset anomaly alarm limits, extracts anomaly node and carrier information, calls the decision tree model to identify the type and severity of violations, and generates anomaly investment promotion behavior monitoring alarms.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a time-series event sequence is constructed and a negative penalty is generated by calculating the funding interval. The attenuation weight is calculated by combining the interaction interval to construct a dynamic constraint of spatiotemporal potential energy. This accurately quantifies the combined impact of funding suitability and time urgency on the investment promotion progress. The potential energy score is accumulated by combining traversal paths to select priority paths. The deviation between the nominal state and the deduced path score is calculated to generate a logical deviation index. This enables in-depth mining of hidden violations or abnormal progress nodes in the entire investment promotion chain, effectively solving the problem of insufficient identification accuracy of traditional monitoring in multi-dimensional variable dynamic coupling scenarios, and improving the timeliness and accuracy of risk warning in the investment promotion process. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides an AI-based data monitoring method for the investment promotion process in industrial parks, comprising the following steps: S1: Collect the original investment promotion interaction logs of the park's investment promotion management, perform time-series sorting operation on the records of the same investment promotion project based on absolute timestamp information, extract text information describing the negotiation content, industry category of the interested enterprise, budget amount value, and type and rental price value of the promotion carrier, and construct a time-series investment promotion event sequence. S2: Call the time-series investment promotion event sequence, calculate the funding matching gap between the enterprise budget amount and the carrier rental price value in the time-series investment promotion event sequence node, perform a numerical comparison operation between the funding matching gap and the preset tolerance limit, generate negative penalty scores for nodes that exceed the preset tolerance limit and add them to the initial state score, and generate a resource attribute matching correction sequence. S3: Based on the resource attribute matching correction sequence, calculate the interaction interval duration of adjacent event nodes, retrieve the standard flow time distribution range from the successful case data, compare the interaction interval duration with the upper limit of the standard flow time distribution range, filter event pairs that exceed the upper limit of the standard flow time, calculate the time decay weight and multiply it into the total path transfer score, and construct a spatiotemporal potential energy dynamic constraint sequence. S4: Call the spatiotemporal potential energy dynamic constraint sequence, traverse the state transition path combination and accumulate the node potential energy score and edge potential energy score, select the path with the highest accumulated score as the model deduction path, obtain the nominal state path filled in by the business personnel, and calculate the numerical deviation from the total score of the model deduction path to generate the investment promotion path logic deviation index. S5: Based on the deviation indicators of the investment promotion path logic and the preset abnormal alarm limit, perform a numerical comparison operation, filter the investment promotion project numbers that exceed the preset abnormal alarm limit, extract the corresponding abnormal node information and related enterprise carrier information, identify the type and severity of violations, and generate an abnormal investment promotion behavior monitoring alarm. The time-series investment promotion event sequence includes ordered event nodes, timestamps corresponding to the nodes, and extracted structured attribute feature data. The resource attribute matching correction sequence includes the funding matching interval value, the generated negative penalty score, and the corrected initial node score. The spatiotemporal potential energy dynamic constraint sequence includes the interaction interval duration value, the calculated time decay weight coefficient, and the constrained state transition score. The investment promotion path logic deviation index includes the cumulative total score of the model deduction path, the cumulative total score of the nominal state path, and the path score deviation degree between the two. The abnormal investment promotion behavior monitoring alarm includes the index number of the abnormal project, the description of the identified violation type, and the risk severity level.

[0020] Please see Figure 2 The specific steps of S1 are as follows: S101: Collect the original investment promotion interaction logs of the park's investment promotion management, traverse the entire data and parse the absolute timestamp value of each log record, use the unique project number index to lock multiple interaction records associated with the same investment promotion project, perform ascending sorting operation on multiple interaction records according to the absolute timestamp value, and generate a time-ordered investment promotion log set. Establish a communication connection with the park's distributed Hadoop cluster log server, send SQL data query commands, and batch retrieve the original investment promotion interaction log data stored in HDFS files. This log data is encapsulated in JSON format. The parsing engine traverses the entire dataset, locates the timestamp field in the JSON structure, and stores it in Unix timestamp format (e.g., ...). The log entries are converted to a standard date and time format. Using the `project_id` field as a unique index key, a hash mapping operation is performed to aggregate hundreds of interaction records associated with the same investment project, scattered across different storage blocks, into a single memory cache. Within the cache, based on the parsed absolute timestamp values, a quicksort algorithm is used to select a reference time point. Timestamps earlier than the reference point are placed in the left partition, and timestamps later than the reference point are placed in the right partition. This process is recursively executed until the interaction records are sorted in ascending order according to their chronological order. For example, for investment project number PJ2024001, three log entries were collected, with absolute timestamps of... , and After processing by the sorting algorithm, the generated sequence order is as follows: , , , and construct a time-ordered collection of investment promotion logs.

[0021] S102: Call the time-ordered investment promotion log set, extract the unstructured text information describing the negotiation content, the industry category and budget amount of the intended enterprise, and the type and rental price of the promotion carrier from the log entries, convert the unstructured text information into a multi-dimensional semantic vector, and perform standardized encoding processing on the numerical attributes to obtain the structured feature data of investment promotion elements. Regular expressions are used to extract unstructured text information such as meeting minutes and customer feedback from the `content` field of log entries. Simultaneously, the `metadata` field is parsed to obtain the industry category code (e.g., national standard industry classification code), budget amount (unit: RMB 10,000 / year) of the target company, and the property type (e.g., Grade A office building, standard factory building) and rental price (unit: RMB / square meter / day) of the promotional vehicle. This unstructured text information is then input into a pre-trained BERT bidirectional encoder representation model for processing. This model internally consists of an input embedding layer, 12 Transformer encoder layers, and an output layer. The input embedding layer segments the text into tokens and adds positional encoding. Each Transformer layer uses a multi-head self-attention mechanism (containing 12 attention heads, with a hidden layer dimension of...). It consists of a deep neural network (using the GELU activation function) and a feedforward neural network. Residual connections and layer normalization are used between layers. After feature extraction by the 12-layer deep network, the output layer outputs... The CLS vector, representing the multidimensional semantic vector of the text, is encoded using Z-score normalization logic for budget and rental price values, and the historical average value of these values ​​is obtained. with standard deviation , set the current value Subtract the mean and then divide by the standard deviation, i.e., execute the formula. Eliminating the influence of dimensions, if the budget amount of a prospective enterprise is... The historical average budget was 10,000 yuan. 10,000 yuan, standard deviation is 10,000 yuan, then minus get Divide by The standardized feature values ​​are obtained as follows: By combining semantic vectors with standardized numerical values, structured feature data of investment attraction elements can be obtained.

[0022] S103: Based on the structured feature data of investment promotion elements, obtain the time index information of the corresponding log entries, perform chain concatenation operation on the feature data belonging to the same investment promotion project according to the time dimension, establish a two-way mapping relationship between feature data and time nodes, assign business status labels and attribute weight values ​​to sequence nodes, reconstruct the discrete feature data, and generate a time-series investment promotion event sequence. The original index table of the log set is retrieved, and the absolute timestamp corresponding to each feature data is extracted as time index information. A doubly linked list data structure is allocated in memory. Taking investment projects as units, feature data is encapsulated into node objects according to the chronological order of the time index. Each node object contains a predecessor pointer pointing to the feature data at the previous time point and a successor pointer pointing to the feature data at the next time point, thus performing a chain concatenation operation. When establishing a bidirectional mapping relationship, feature data is located not only by timestamp but also by using feature vectors to inversely index to specific time nodes. Based on the business rule base, each node in the sequence is assigned a business status label (such as initial contact, on-site viewing, price negotiation) and an attribute weight value. The attribute weight value is set based on the information entropy of the feature data; the lower the information entropy, the higher the weight. The originally discretely stored text vectors and numerical features are reorganized into a tightly coupled temporal structure. For example, the weight of the "initial contact" node is set to... The weight of the "price negotiation" node is set to... The generated time-series investment promotion event sequence is in the form of: .

[0023] Please see Figure 3 The specific steps of S2 are as follows: S201: Call the time-series investment promotion event sequence, traverse multiple event nodes in the time-series investment promotion event sequence, read the budget amount value of the intended enterprise and the rental price value of the promotion carrier associated with the node, perform numerical subtraction operation on the budget amount value of the intended enterprise and the rental price value of the promotion carrier under the same node, obtain the absolute value, quantify the funding matching distance, and generate a set of supply and demand funding matching differences. The iterator iterates through each event node object stored in the sequence, accesses the node data area, and reads the stored values ​​of the intended enterprise's budget amount (standardized actual amount, unit: yuan / square meter / day) and the recommended vehicle rental price (unit: yuan / square meter / day). It then performs a numerical difference logic, using the intended enterprise's budget amount as the minuend and the recommended vehicle rental price as the subtrahend, and applies an absolute value function to the result to quantify the matching distance between the supply and demand sides of funds. For example, at a certain negotiation node, the intended enterprise's budget amount is... The rental price for the promotional space is [amount] yuan / square meter / day. Yuan / square meter / day, perform subtraction. get After taking the absolute value, we get If the budget in another node is The price is The difference is The absolute value is still This process imports the calculation results of the nodes into a list, generating a set of supply and demand funding matching differences, as shown in Table 1, which displays the funding data collection and calculation results of some nodes.

[0024] Table 1: Sample Table for Calculating Funding Matching Difference S202: Based on the set of supply and demand funding mismatches, obtain the preset funding matching tolerance threshold, perform a numerical comparison operation between the supply and demand funding mismatch and the funding matching tolerance threshold, filter out the difference items where the supply and demand funding mismatch is greater than the funding matching tolerance threshold, calculate the negative penalty value proportional to the excess of the difference item, and generate a non-fit node penalty mapping table. The system reads the preset funding matching tolerance threshold from the configuration parameter file. This threshold is based on statistical analysis of successful contract signing cases within the park, calculating the average maximum deviation between budget and price in these successful cases, and setting the funding matching tolerance threshold accordingly. The system iterates through the set of supply and demand funding mismatches, comparing each mismatch with a funding tolerance threshold. If the mismatch is less than or equal to the threshold, it's considered a match; otherwise, it's filtered as a difference. For each filtered difference, the excess is calculated by subtracting the threshold from the mismatch, dividing by the threshold, and then applying a linear penalty function, multiplying the excess by a preset penalty factor (e.g., [missing information]). The negative penalty value is obtained. Taking node N003 in Table 1 as an example, the adaptation difference is... The threshold is , Calculate the penalty: First, calculate the excess portion. The excess range is Set the penalty factor as The negative penalty value is The result is , combine the node ID with the calculated Store in a hash table to generate a penalty mapping table for non-fit nodes.

[0025] S203: Based on the non-adaptive node penalty mapping table, obtain the initial probability evaluation scores of multiple event nodes, accumulate the corresponding negative penalty values ​​of the nodes recorded in the non-adaptive node penalty mapping table to the initial probability evaluation scores, update the node's state probability distribution parameters based on the calculated new scores, and generate a resource attribute matching correction sequence. Extract the initial probability assessment score (e.g., the range of values) calculated based on the conversion rate. It iterates through the non-fit node penalty mapping table, locks the node object to be corrected according to the key-value pair matching principle, reads the negative penalty value recorded in the mapping table, performs a numerical subtraction update operation, subtracts the corresponding negative penalty value from the initial probability evaluation score, and if the subtraction result is lower than the preset minimum probability limit (e.g., ... If the initial probability assessment score of a node is 0, it is forcibly adjusted to the minimum lower bound. Subsequently, based on the calculated new score, the state probability distribution parameters of the node in the current state are updated using Softmax normalization logic. For example, if the initial probability assessment score of a node is 0, the node's initial probability assessment score is 0. The negative penalty value found in the mapping table is: Then execute The updated score is This operation reduces the weight of nodes with poor funding matching in subsequent path deduction and generates a resource attribute matching correction sequence.

[0026] Please see Figure 4 The specific steps of S3 are as follows: S301: Traverse the investment promotion event nodes within the resource attribute matching and correction sequence, filter adjacent node pairs with flow relationship, extract the absolute timestamp attribute markers of the subsequent node and the preceding node, perform differential operation on each pair of adjacent nodes, obtain the interval duration value, and generate a set of adjacent event time domain differences. Identify adjacent node pairs in the sequence connected by "predecessor-successor" pointers. Define the node preceding the previous node on the timeline as the predecessor node and the node following the previous node as the successor node. Read the start timestamp of the successor node and the end timestamp of the predecessor node (in hours), perform a difference operation, and subtract the end timestamp of the predecessor node from the start timestamp of the successor node to obtain the interval between the two events. For example, if the end timestamp of the predecessor node corresponds to January 1st at 10:00 AM, and the start timestamp of the successor node is January 3rd at 2:00 PM, the difference between the two is... For each hour, the time interval values ​​of each pair of adjacent nodes are stored in an array in order to generate a set of temporal differences between adjacent events.

[0027] S302: Based on the set of time-domain differences between adjacent events, retrieve the database of successful investment cases in the park and extract the standard time distribution data of similar state transitions. According to the principle of normal distribution, calculate the upper limit of the standard transition time covering the core interval. Perform numerical comparison operation between multiple differences in the set and the upper limit of the standard transition time. For long-tail delay terms whose values ​​exceed the upper limit, calculate the penalty coefficient and generate a table of irregular interval attenuation coefficients. Filter standard circulation data that are in the same industry category and the same investment promotion stage as the current project, and calculate the average circulation interval of similar events based on the statistical normal distribution principle. ) and standard deviation ( The logic for setting the upper limit of the standard circulation time covering the core area is: the average value plus 3 times the standard deviation, i.e. The system compares the duration of each interval in the set of temporal differences between adjacent events with the upper limit of the standard turnaround time. If the interval duration is greater than the upper limit, it is identified as a long-tailed delay term. For long-tailed delay terms, a penalty coefficient is calculated. The calculation logic is: the upper limit of the standard turnaround time is divided by the actual interval duration, and the result is used as the attenuation coefficient (within a certain range). If the average turnaround time from "initial contact" to "on-site property viewing" in the database is... Hours, standard deviation is The upper limit is for hours. ,Right now If the interval duration calculated in step S301 is 1 hour, At that time, because Therefore, a penalty calculation is triggered, with a penalty coefficient of 1. The calculation result is approximately The coefficient is associated with the node pair index to generate an irregular interval decay coefficient table.

[0028] S303: Based on the irregular interval decay coefficient table, obtain the initial flow score defined in the preset state transition probability matrix. For the delayed event pairs recorded in the irregular interval decay coefficient table, perform a weighted multiplication operation on the initial flow score using the penalty coefficient, keep the original score of the non-delayed event pairs constant, and perform logical association between the calculated edge potential energy score and the existing node potential energy score in the sequence to generate a spatiotemporal potential energy dynamic constraint sequence. Extract the initial transition score (e.g., base score) of the defined corresponding state transition. The process iterates through the irregular interval decay coefficient table, locates event pairs with delays, reads the corresponding penalty coefficients, performs weighted multiplication, multiplies the initial flow score by the penalty coefficient, and obtains the corrected edge potential score. For normal flow event pairs not recorded in the table, their initial flow scores remain unchanged. The calculated edge potential score (representing the resistance or driving force of state transition) is logically associated with the existing node potential scores in the sequence (probability scores derived from S203). The association method is to use the node potential score as the vertex weight and the edge potential score as the edge weight, constructing a weighted directed graph structure. If the initial flow score is... The penalty coefficient is The corrected edge potential energy score is This score, together with the node's own score, constitutes a time-constrained spatiotemporal field, generating a dynamic constrained sequence of spatiotemporal potential energy.

[0029] Please see Figure 5 The specific steps of S4 are as follows: S401: Call the spatiotemporal potential energy dynamic constraint sequence, construct the state transition topology grid, traverse the node connection path combinations inside the state transition topology grid, integrate the resource matching correction term and time decay weight term in the sequence, accumulate the node potential energy score and edge potential energy score for the node connection path combination, perform global optimization operation, filter the node sequence with priority in the accumulated total score, record the theoretical optimal state transition trajectory and peak potential energy total score, and generate the optimal path set for model deduction; In the grid, vertices represent the state of investment promotion events, and edges represent flow relationships. A dynamic programming or depth-first search (DFS) variant algorithm is used to traverse feasible node connection path combinations within the grid. During traversal, resource matching correction terms (node ​​scores) and time decay weight terms (edge ​​scores) are integrated into the sequence. An accumulator is set up, and for each path combination, the potential energy scores of the nodes and edges along the path are accumulated hop-by-hop. A global optimization operation is performed, comparing the cumulative total scores of feasible paths, selecting the node sequence with the highest cumulative total score, recording this sequence as the theoretically optimal state flow trajectory, and saving its corresponding peak potential energy score. For example, the algorithm compares path A (total score...) ) and path B (total score) ), determine that path A is optimal, and record the total peak potential energy. Generate a set of optimal paths by model deduction.

[0030] S402: Based on the model-derived optimal path set, extract the investment promotion progress logs submitted by business personnel, parse the status change records, reconstruct the business flow chain, and map the business flow chain to the probability evaluation space defined by the spatiotemporal potential energy dynamic constraint sequence, using the formula: ; The nominal state path potential energy value is calculated; in, Represents the nominal state path potential value. Represents the total number of nodes. Representing the first in the artificial link The basic potential energy score of each node Representing the The mapping probability coefficients of each node in the probability evaluation space This represents the preset dimensionless dynamic impedance adjustment factor. This represents the average potential energy score of nodes in the optimal path set derived by the model. Representing the The duration of the flow of each node, Represents the maximum spatiotemporal span value defined by the dynamic constraint sequence of spatiotemporal potential energy; Extract the actual business development progress logs filled in by business personnel from the CRM, parse the status change records (such as from "intent" to "signed"), reconstruct the business flow chain in chronological order, map this manually generated chain to the probability evaluation space defined by the spatiotemporal potential energy dynamic constraint sequence, and call the nominal state path potential energy value calculation logic, that is: obtain the first... The potential energy base score of the node, the first The mapping probability coefficients of each node in the probability evaluation space and the preset dimensionless dynamic impedance adjustment factor are multiplied together to obtain the basic potential energy of a single node, thus obtaining the first node's potential energy. The time normalization factor is obtained by multiplying the average potential energy score of nodes by the flow duration, the maximum spatiotemporal span defined by the dynamic constraint sequence of spatiotemporal potential energy, and the average potential energy score of nodes in the optimal path set derived by the model. The basic potential energy of a single node is then divided by the time normalization factor to obtain the weighted potential energy of that single node. The weighted potential energies of all nodes on the link are summed, and the sum is divided by the total number of nodes to calculate the nominal state path potential energy value. It is assumed that the artificial link includes... There are 10 nodes (the total number of nodes is 100). ), for the first Each node has a potential energy fundamental score of [value]. The mapping probability coefficient is The dimensionless dynamic impedance adjustment factor is set to (Based on experimental measurements, this factor is used to regulate external environmental resistance, and its value range is...) Then the numerator part is operated as follows: The result is The denominator is the average potential energy score of the nodes in the optimal path set derived by the model. The duration of the flow at this node is The maximum spatiotemporal span is [value missing]. Then the denominator operation is: The result is The single-node weighted potential energy is The result is approximately For the link Repeat this process for each node and sum them up, assuming the total sum is... The nominal state path potential energy is then... The result is The nominal state path potential energy value is obtained.

[0031] S403: Based on the nominal state path potential energy value, retrieve the peak potential energy total score recorded in the optimal path set derived by the model, perform numerical subtraction operation between the nominal state path potential energy value and the peak potential energy total score to obtain the score difference, perform standardization processing on the score difference, calculate the deviation magnitude, and generate the investment promotion path logical deviation index. Retrieve the peak potential energy total score recorded in the optimal path set derived from the model (the total score in S401 needs to be converted to a value of the same magnitude using the same normalization logic as S402, or the unnormalized total score in S402 can be directly used for comparison; here it is assumed that the units have been unified), perform a numerical subtraction operation, subtract the nominal state path potential energy value from the peak potential energy total score, and obtain the score difference. If the difference is negative, it means that the artificial path is better than the model path, and it is set to... Standardize the score difference by dividing it by the total peak potential energy score, and calculate the deviation (in percentage form). If the total peak potential energy score is converted to... The nominal state path potential energy calculated by S402 is... The score difference is The deviation range is The result is ,Right now This indicator quantifies the logical gap between actual operation and theoretical optimal solution, generating a logical deviation indicator for investment promotion path.

[0032] Please see Figure 6 The specific steps of S5 are as follows: S501: Call the business development path logic deviation indicator, obtain the preset business compliance anomaly alarm limit, perform a numerical comparison operation between the business development path logic deviation indicator and the anomaly alarm limit, filter target project records with deviations greater than the anomaly alarm limit, identify the associated unique identity index number and perform serialization encapsulation, and generate a list of abnormal project index numbers. Read the preset business compliance anomaly alarm limit, which is set based on audit data, for example, set to... (Right now The deviation of the investment promotion path logic from the indicator (such as the degree of deviation). ) and abnormal alarm limits ( Perform a numerical comparison operation. If the indicator value is greater than the limit ( The system determines that the project has an abnormal risk, filters out the target project records, identifies the unique identity index number (ProjectID) associated with the record, serializes and encapsulates it into a list format, and generates a list of abnormal project index numbers.

[0033] S502: Based on the list of abnormal project index numbers, backtrack and retrieve the time-series investment promotion event sequence, locate the coordinates of the business nodes that caused the logical deviation, extract the industry category, budget amount, and type and rental price of the related intended enterprises, and perform aggregation operation on the business node coordinates and the extracted attribute data to generate a multi-dimensional feature set of abnormal nodes. Using ProjectID as the key, the time-series investment promotion event sequence generated by S103 is retrieved back. By comparing the local differences between the actual node scores and the theoretically optimal node scores, the coordinates of the specific business node that causes the largest logical deviation are located (e.g., locating the "rent approval" node). The industry category of the intended enterprise associated with this node (e.g., "Internet finance") and the budget amount (e.g., ...) are extracted. The text appears to be a mix of unrelated phrases and sentences, making it difficult to translate coherently. It includes fragments like "yuan / ㎡ / day" (referred to as "yuan / ㎡ / day"), "promotional carrier type" (e.g., "research land"), "rental price value" (e.g., "yuan / ㎡ / day"), "promotional carrier type" (e.g., "research land"), and "rental price value". A more accurate translation would require the full context of (RMB / ㎡ / day), perform aggregation operations on the business node coordinates (time, stage) and the extracted attribute data to construct a feature vector containing multiple fields, and generate a multi-dimensional feature set of abnormal nodes.

[0034] S503: Based on the multi-dimensional feature set of abnormal nodes, call the diagnostic model for investment promotion violations, input the aggregated feature data into the model to perform feature matching, identify the type of violation, calculate the quantitative rating coefficient of risk severity, integrate project number, violation type and severity rating data, and generate an abnormal investment promotion behavior monitoring alarm; The model for diagnosing investment promotion violations is invoked. This model is built upon a deep convolutional neural network (CNN). The input layer receives a multi-dimensional feature set of aggregated abnormal nodes (which needs to be pre-reconstructed into matrix form to adapt to convolution operations); the first layer is a convolutional layer, containing... indivual The first layer uses a convolutional kernel with ReLU activation to extract local correlations between features; the second layer is a max pooling layer, using... The window is used for dimensionality reduction; the third layer is a fully connected layer, containing... The model has 10 neurons; the output layer is a Softmax classification layer that outputs the probability of each violation type. Feature data is input into the model to perform feature matching and forward propagation calculations to identify violation types (such as "low-price profit transfer," "process reversal," and "fake customers"). Simultaneously, based on the probability values ​​of the output layer, a quantitative rating coefficient for the severity of risk is calculated (e.g., probability). (Corresponding to high-risk levels), integrating project numbers, identified violation types, and severity rating data, generates abnormal investment promotion behavior monitoring alerts. For example, after the input feature set is processed by the model, the output layer's probability in the "low-price transfer of benefits" category is... Regarding "process reversal" The model determined the violation type to be "transferring benefits at a low price," and based on... The probability value is used to classify the risk rating as "Level 1 Severe" and output an alarm: "Project PJ2024001 has the risk of transferring benefits at a low price during the rent approval stage. Risk level: Level 1". Table 2 shows an example of the model diagnosis output.

[0035] Table 2: Output Results of the Violation Diagnosis Model Please see Figure 7 An AI-based data monitoring system for the investment promotion process in industrial parks, comprising: The investment promotion interaction sequence construction module collects the original investment promotion interaction logs of the park's investment promotion management, performs sorting operations on the original investment promotion interaction logs based on the timestamp values, extracts text, industry, budget amount, carrier and price fields, and constructs a time-series investment promotion event sequence. The resource attribute matching and correction module calls the time-series investment promotion event sequence, calculates the funding gap between the budget amount and the price and compares it with the obtained tolerance limit. For nodes that exceed the tolerance limit, a negative penalty score is generated and added to the initial score to generate the resource attribute matching and correction sequence. The dynamic constraint module calculates the interaction interval duration based on the resource attribute matching correction sequence and compares it with the upper limit of the standard time interval. For event pairs that exceed the upper limit of the standard time interval, it calculates the decay weight and multiplies it into the total transfer score to construct a spatiotemporal potential dynamic constraint sequence. The investment promotion path identification module calls the spatiotemporal potential energy dynamic constraint sequence, traverses the path combination and accumulates the node and edge potential energy scores, filters the model deduction path with priority in score, obtains the nominal state path and calculates the score deviation from the model deduction path, and generates the investment promotion path logical deviation index. The anomaly monitoring module filters project numbers that exceed preset anomaly alarm limits based on deviation indicators from the investment promotion path logic, extracts information on anomaly nodes and carriers, calls a decision tree model to identify the type and severity of violations, and generates anomaly investment promotion behavior monitoring alarms.

[0036] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A data monitoring method for the investment promotion process in industrial parks based on AI, characterized in that, Includes the following steps: S1: Collect the original investment promotion interaction logs of the park's investment promotion management, arrange them according to timestamps, extract text, industry, budget amount, carrier and price, and construct a time-series investment promotion event sequence; S2: Call the time-series investment promotion event sequence, calculate the funding gap between the budget amount and the price and compare it with the tolerance limit. For nodes that exceed the tolerance limit, generate negative penalty scores and add them to the initial score to generate a resource attribute matching correction sequence. S3: Based on the resource attribute matching correction sequence, calculate the interaction interval duration and compare it with the upper limit of the standard time interval obtained. For event pairs that exceed the upper limit of the standard time interval, calculate the attenuation weight and multiply it into the total transfer score to construct a spatiotemporal potential dynamic constraint sequence. S4: Call the spatiotemporal potential energy dynamic constraint sequence, filter the model deduction path with priority in score, obtain the nominal state path and calculate the score deviation from the model deduction path, and generate the investment promotion path logic deviation index. S5: Based on the deviation index of the investment promotion path logic, filter the project numbers that exceed the preset abnormal alarm limit, extract the abnormal node and carrier information, identify the type and severity of violations, and generate an abnormal investment promotion behavior monitoring alarm.

2. The AI-based data monitoring method for investment promotion in industrial parks according to claim 1, characterized in that, The nominal state path and the model deduction path are traversed separately to identify multiple nodes included in the path and the edges connecting multiple nodes, and the basic potential energy attributes of nodes and edges in the spatiotemporal potential energy dynamic constraint sequence are extracted. The potential energy scores of nodes and edges in the nominal state path are summed to generate a nominal total potential energy value. Simultaneously, the potential energy scores of nodes and edges in the model deduction path are summed to generate a standard total potential energy value.

3. The AI-based data monitoring method for the investment promotion process in industrial parks according to claim 1, characterized in that, The time-series investment promotion event sequence includes ordered event nodes, timestamps corresponding to the nodes, and extracted structured attribute feature data. The resource attribute matching correction sequence includes funding matching interval values, generated negative penalty scores, and corrected initial node scores. The spatiotemporal potential dynamic constraint sequence includes interaction interval duration values, calculated time decay weight coefficients, and constrained state transition scores. The investment promotion path logic deviation index includes the cumulative total score of the model deduction path, the cumulative total score of the nominal state path, and the path score deviation degree between the two. The abnormal investment promotion behavior monitoring alarm includes the index number of the abnormal project, the description of the identified violation type, and the risk severity level.

4. The AI-based data monitoring method for investment promotion in industrial parks according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Collect the original investment promotion interaction logs of the park's investment promotion management, traverse the entire data and parse the absolute timestamp value of each log record, use the unique project number index to lock multiple interaction records associated with the same investment promotion project, perform ascending sorting operation on multiple interaction records according to the absolute timestamp value, and generate a time-ordered investment promotion log set. S102: Call the time-ordered investment promotion log set, extract the unstructured text information describing the negotiation content, the industry category and budget amount of the intended enterprise, and the type and rental price of the promotion carrier from the log entries, convert the unstructured text information into a multi-dimensional semantic vector, and perform standardized encoding processing on the numerical attributes to obtain the structured feature data of investment promotion elements. S103: Based on the structured feature data of the investment promotion elements, obtain the time index information of the corresponding log entries, perform chain concatenation operation on the feature data belonging to the same investment promotion project according to the time dimension, establish a two-way mapping relationship between feature data and time nodes, assign business status labels and attribute weight values ​​to sequence nodes, reconstruct the discrete feature data, and generate a time-series investment promotion event sequence.

5. The AI-based data monitoring method for investment promotion in industrial parks according to claim 4, characterized in that, The specific steps of S2 are as follows: S201: Call the time-series investment promotion event sequence, traverse multiple event nodes in the time-series investment promotion event sequence, read the budget amount value of the intended enterprise and the rental price value of the promotion carrier associated with the node, perform numerical subtraction operation on the budget amount value of the intended enterprise and the rental price value of the promotion carrier under the same node, obtain the absolute value, quantify the funding matching distance, and generate a set of supply and demand funding matching differences. S202: Based on the set of supply and demand funding mismatches, obtain a preset funding matching tolerance threshold, perform a numerical comparison operation between the supply and demand funding mismatch and the funding matching tolerance threshold, filter out the difference items where the supply and demand funding mismatch is greater than the funding matching tolerance threshold, calculate a negative penalty value proportional to the excess of the difference item, and generate a non-fit node penalty mapping table. S203: Based on the non-adaptive node penalty mapping table, obtain the initial probability evaluation scores of multiple event nodes, accumulate the corresponding negative penalty values ​​of the nodes recorded in the non-adaptive node penalty mapping table to the initial probability evaluation scores, update the state probability distribution parameters of the nodes based on the calculated new scores, and generate a resource attribute matching correction sequence.

6. The AI-based data monitoring method for investment promotion in industrial parks according to claim 5, characterized in that, The specific steps for S3 are as follows: S301: Traverse the investment promotion event nodes within the resource attribute matching and correction sequence, filter adjacent node pairs with flow relationship, extract the absolute timestamp attribute markers of the subsequent node and the preceding node, perform differential operation on each pair of adjacent nodes, obtain the interval duration value, and generate a set of adjacent event time domain differences. S302: Based on the set of time-domain differences between adjacent events, retrieve the database of successful investment cases in the park and extract the standard time distribution data of similar state transitions. According to the principle of normal distribution, calculate the upper limit of the standard transition time covering the core interval. Perform a numerical comparison operation between the multiple differences in the set and the upper limit of the standard transition time. For long-tail delay terms whose values ​​exceed the upper limit, calculate the penalty coefficient and generate an irregular interval attenuation coefficient table. S303: According to the irregular interval decay coefficient table, obtain the initial flow score defined in the preset state transition probability matrix. For the delayed event pairs recorded in the irregular interval decay coefficient table, perform a weighted multiplication operation on the initial flow score using the penalty coefficient to keep the original score of the non-delayed event pairs constant. Perform logical association between the calculated edge potential energy score and the existing node potential energy score in the sequence to generate a spatiotemporal potential energy dynamic constraint sequence.

7. The AI-based data monitoring method for investment promotion in industrial parks according to claim 6, characterized in that, The specific steps of S4 are as follows: S401: Call the spatiotemporal potential energy dynamic constraint sequence, construct the state transition topology grid, traverse the node connection path combinations inside the state transition topology grid, integrate the resource matching correction term and time decay weight term in the sequence, accumulate the node potential energy score and edge potential energy score for the node connection path combination, perform global optimization operation, filter the node sequence with priority in the accumulated total score, record the theoretical optimal state transition trajectory and peak potential energy total score, and generate the optimal path set for model deduction; S402: Based on the model, deduce the optimal path set, extract the investment promotion progress log filled in by business personnel, parse the status change record, reconstruct the business flow link, map the business flow link to the probability evaluation space defined by the spatiotemporal potential energy dynamic constraint sequence, and calculate the nominal state path potential energy value. S403: Based on the nominal state path potential energy value, retrieve the peak potential energy total score recorded in the optimal path set derived by the model, perform numerical subtraction operation between the nominal state path potential energy value and the peak potential energy total score to obtain the score difference, perform standardization processing on the score difference, calculate the deviation magnitude, and generate the investment promotion path logical deviation index.

8. The AI-based data monitoring method for the investment promotion process in industrial parks according to claim 7, characterized in that, The specific steps of S5 are as follows: S501: Call the investment promotion path logical deviation indicator, obtain the preset business compliance anomaly alarm limit, perform a numerical comparison operation between the investment promotion path logical deviation indicator and the anomaly alarm limit, filter target project records with deviations greater than the anomaly alarm limit, identify the associated unique identity index number and perform serialization encapsulation, and generate an abnormal project index number list. S502: Based on the list of abnormal project index numbers, backtrack and retrieve the time-series investment promotion event sequence, locate the coordinates of the business nodes that caused the logical deviation, extract the industry category, budget amount, and type and rental price of the associated intended enterprise, and perform an aggregation operation on the business node coordinates and the extracted attribute data to generate a multi-dimensional feature set of abnormal nodes. S503: Based on the multi-dimensional feature set of the abnormal nodes, call the diagnostic model for investment promotion violations, input the aggregated feature data into the model to perform feature matching, identify the type of violation, calculate the quantitative rating coefficient of the risk severity, integrate the project number, violation type and severity rating data, and generate an abnormal investment promotion behavior monitoring alarm.

9. The AI-based data monitoring method for investment promotion in industrial parks according to claim 8, characterized in that, The multidimensional feature set of the abnormal nodes is analyzed, and the budget amount value and the rental price value are subjected to maximum and minimum normalization processing to construct the numerical deviation feature vector. Simultaneously, a one-hot encoding method is used to digitally transform the industry category of the target enterprise and the type of the promotion carrier, and construct semantic attribute feature vectors.

10. An AI-based data monitoring system for the investment promotion process in industrial parks, characterized in that: The system is used to implement the AI-based data monitoring method for the investment promotion process in industrial parks as described in any one of claims 1-9, and the system includes: The investment promotion interaction sequence construction module collects the original investment promotion interaction logs of the park's investment promotion management, performs sorting operations on the original investment promotion interaction logs based on the timestamp values, extracts text, industry, budget amount, carrier and price fields, and constructs a time-series investment promotion event sequence. The resource attribute matching and correction module calls the time-series investment promotion event sequence, calculates the funding gap between the budget amount and the price and compares it with the obtained tolerance limit. For nodes that exceed the tolerance limit, a negative penalty score is generated and added to the initial score to generate the resource attribute matching and correction sequence. The dynamic constraint module calculates the interaction interval duration based on the resource attribute matching correction sequence and compares it with the upper limit of the standard time interval. For event pairs that exceed the upper limit of the standard time interval, it calculates the attenuation weight and multiplies it into the total transfer score to construct a spatiotemporal potential dynamic constraint sequence. The investment promotion path identification module calls the spatiotemporal potential energy dynamic constraint sequence, traverses the path combination and accumulates the node and edge potential energy scores, filters the model deduction path with priority in score, obtains the nominal state path and calculates the score deviation from the model deduction path, and generates the investment promotion path logical deviation index. The anomaly monitoring module, based on the deviation indicators of the investment promotion path logic, filters project numbers that exceed the preset anomaly alarm limits, extracts anomaly node and carrier information, calls the decision tree model to identify the type and severity of violations, and generates anomaly investment promotion behavior monitoring alarms.

Images