A data center-based recruitment and selection index multi-system integration and data collaboration method

Abstract

This invention relates to the field of bidding and procurement process management technology, specifically a method for multi-system integration and data collaboration of bidding and procurement indicators based on a data platform. The method includes: establishing an indicator monitoring network that matches the bidding and procurement process; continuously collecting indicator snapshots during the observation period, identifying indicator deviation events through comparison and triggering early warnings; after an early warning, performing status backtracking on the monitoring nodes and extracting background parameter data within the time window before the event; inputting the background parameters into a linkage evaluation model to calculate the comprehensive impact factor of the background parameters; calculating the linkage rectification intensity value accordingly; dynamically revising the relevant threshold parameters in the original bidding and procurement process rule base based on this intensity value, generating and replacing them with target rectification constraint values; after completing the parameter replacement, restarting the process status monitoring; if no more indicator deviation events are triggered, the closed-loop processing is completed.

Description

Technical Field

[0001] This invention relates to the field of bidding and procurement process management technology, and in particular to a method for multi-system integration and data collaboration of bidding and procurement indicators based on a data platform. Background Technology

[0002] In the field of digital management of bidding and procurement, monitoring and issuing early warnings for key business indicators is a common practice. Existing technologies typically monitor single indicators based on preset static thresholds, such as setting fixed warning lines for price deviations or supplier response times. Once the monitored data exceeds the threshold, the system triggers an alarm, notifying managers for manual verification and intervention. This conventional approach has significant limitations when dealing with complex and dynamic bidding and procurement processes. Existing solutions artificially separate early warnings from subsequent corrective actions; early warnings serve only as a signal of a problem, while the corrective action relies entirely on managers' experience for manual judgment and decision-making, lacking systematic guidance and support. Furthermore, the threshold parameters used to trigger early warnings are usually pre-set and fixed in the long term, unable to self-adjust based on changes in the actual operating environment and the accumulation of historical anomaly patterns, resulting in poor system adaptability. When anomalies are caused by a combination of background factors, simple threshold-exceeding alarms fail to reveal the associated conditions and root causes of the anomaly, making accurate attribution difficult for managers. Static thresholds are prone to false alarms or missed alarms; frequent invalid alarms lead to early warning fatigue, while genuine risks may be overlooked due to improperly set thresholds. How to enable the early warning system to automatically correlate and analyze multi-dimensional background information before and after the anomaly occurs after detecting a problem, and autonomously drive the iterative optimization of process control rules based on the analysis results, so as to achieve a closed loop from problem identification to system self-adjustment, rather than remaining at the level of passive alarm, is a key problem that existing technologies have not yet effectively solved. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for multi-system integration and data collaboration of procurement indicators based on a data platform.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a method for multi-system integration and data collaboration of procurement indicators based on a data middle platform, comprising:

[0005] Establish an indicator monitoring network that matches the bidding and procurement process, the indicator monitoring network consisting of several monitoring nodes;

[0006] During the preset observation period, quantitative information reflecting the process status is continuously collected through the indicator monitoring network, and each collection is defined as an indicator snapshot.

[0007] The system compares consecutive snapshots of the indicator. When the difference between the snapshot and the adjacent historical snapshot exceeds the preset tolerance boundary, an indicator deviation event is identified, and an early warning procedure is triggered.

[0008] After the early warning program is initiated, the status of the monitoring nodes in the indicator monitoring network is traced back to extract several background parameter data that are strongly related to the process within the time window before the indicator deviation event occurs.

[0009] The extracted background parameter data is input into a preset linkage evaluation model, and the linkage evaluation model calculates the comprehensive impact factor of the background parameters.

[0010] The intensity value of joint rectification is calculated based on the comprehensive impact factors of the background parameters.

[0011] Based on the aforementioned linkage rectification intensity value, the relevant threshold parameters in the original bidding and procurement process rule base are dynamically revised to generate target rectification constraint values, which are then used to replace the original threshold parameters.

[0012] After the parameter replacement is completed, a new round of process status monitoring is restarted. If the aforementioned indicator deviation event is no longer triggered, the closed-loop processing is completed.

[0013] As a further aspect of the present invention, the step of continuously collecting quantitative information reflecting the process status through the indicator monitoring network includes:

[0014] Define a set of indicators, which shall contain at least raw values ​​for three dimensions: cost, cycle, and compliance.

[0015] For each data collection, the rate of change of the original value of each dimension relative to its initial value at the start of the observation period is calculated, resulting in a set of rate of change sequences.

[0016] All data within the rate of change sequence are normalized, and the processed data in the three dimensions are linearly combined according to preset weights. The resulting values ​​constitute the index snapshot.

[0017] As a further aspect of the present invention, the specific process for identifying a single indicator deviation event includes:

[0018] Obtain a snapshot of the indicator generated at the current moment, and denote it as the current snapshot;

[0019] By tracing back several consecutive collection periods, a set of historical snapshots can be obtained;

[0020] The difference between the current snapshot and each snapshot in the historical snapshot set is calculated to obtain a set of snapshot difference values.

[0021] Calculate the mean and standard deviation of the snapshot difference set;

[0022] The current snapshot is compared with the average value of the historical snapshot set, and the deviation is calculated.

[0023] If the deviation is greater than the product of the standard deviation of the snapshot difference value set and the preset sensitivity coefficient, then the index deviation event is determined to have occurred.

[0024] As a further aspect of the present invention, the process of performing status backtracking on the monitoring nodes within the indicator monitoring network and extracting several background parameter data strongly correlated with the process within the time window before the occurrence of the indicator deviation event is as follows:

[0025] Based on the time point when the deviation event of the aforementioned indicator was identified, a backtracking window of a fixed time length is set forward.

[0026] Within the backtracking window, the original operation logs, resource allocation records, and approval flow timestamps associated with all dimensions in the indicator set are retrieved from the process database. The original operation logs, resource allocation records, and approval flow timestamps are collectively referred to as the initial background data.

[0027] The initial background data is cleaned by removing invalid and duplicate records to form a clean background dataset.

[0028] Feature extraction is performed on the clean background dataset to extract statistical features including operation frequency, resource consumption fluctuation, and process delay time. These statistical features constitute the background parameter data used for linkage evaluation.

[0029] As a further aspect of the present invention, the process by which the linkage evaluation model calculates the comprehensive influence factor of the background parameters includes:

[0030] The operation frequency, resource consumption fluctuation, and process delay time in the background parameter data are compared with the preset benchmark operation frequency, benchmark resource consumption, and benchmark process time, respectively, and the operation deviation coefficient, resource deviation coefficient, and time deviation coefficient are calculated.

[0031] Obtain a historical rectification case library, retrieve historical cases in the historical rectification case library that are similar to the current indicator deviation event type, and extract the historical background parameter data corresponding to the historical cases;

[0032] Calculate the cosine similarity between the current background parameter data and the retrieved historical background parameter data in each dimension to obtain the historical similarity;

[0033] The operation deviation coefficient, resource deviation coefficient, duration deviation coefficient, and historical similarity are input into a pre-trained multilayer perceptron network.

[0034] The multilayer perceptron network outputs a scalar value between zero and one, which is the comprehensive influence factor of the background parameters.

[0035] As a further aspect of the present invention, the method for calculating the linkage rectification intensity value is as follows:

[0036] Establish an intensity calculation function, the input of which is the comprehensive influence factor of the background parameter and the deviation degree in the index deviation event;

[0037] The basic intensity component is obtained by multiplying the background parameter comprehensive influence factor by a preset intensity base.

[0038] Multiply the deviation by another preset adjustment coefficient to obtain the adjustment intensity component;

[0039] The basic strength component and the adjusted strength component are added together, and the sum is mapped by a Sigmoid function to restrict its value range to between zero and one. The result after mapping is the linkage rectification strength value.

[0040] As a further aspect of the present invention, the process of dynamically revising the relevant threshold parameters in the original bidding and procurement process rule base is as follows:

[0041] Based on the dimension associated with the deviation event of the indicator, locate the specific threshold parameter entries in the original bidding and procurement process rule base that need to be adjusted, and mark them as target thresholds to be revised;

[0042] Obtain the current value of the target threshold to be revised;

[0043] Calculate the difference between the linked rectification intensity value and the value one to obtain the rectification buffer coefficient;

[0044] Multiply the current value of the target threshold to be revised by the rectification buffer coefficient to obtain a revision increment;

[0045] Based on the nature of the deviation event of the indicator, it is determined whether the target threshold to be revised should be tightened or relaxed.

[0046] If the adjustment is tightened, the current value of the target threshold to be revised is subtracted from the revision increment to obtain the target rectification constraint value;

[0047] If the adjustment is to be relaxed, the current value of the target threshold to be revised is added to the revision increment to obtain the target rectification constraint value.

[0048] As a further aspect of the present invention, the method for determining whether to tighten or loosen the target threshold to be revised based on the nature of the indicator deviation event is as follows:

[0049] When the deviation of the aforementioned indicators manifests as increased costs, extended cycles, or increased risk of violations, it is determined that the relevant thresholds need to be tightened and adjusted.

[0050] When the deviation of the aforementioned indicators manifests as an unexpected decrease in cost, an abnormally shortened cycle, or an improper skipping of process steps, it is determined that the relevant thresholds need to be relaxed and adjusted in order to correct existing monitoring blind spots or rule loopholes.

[0051] The judgment logic is built into the system's decision engine in the form of rules.

[0052] As a further aspect of the present invention, the process of restarting a new round of process status monitoring also includes a verification step:

[0053] After the target rectification constraint value takes effect, a verification and monitoring cycle is initiated;

[0054] During the verification and monitoring period, new index snapshots will continue to be generated at the original collection frequency.

[0055] The newly generated indicator snapshot is compared with the set of historical snapshots before the indicator deviation event occurs using the same deviation calculation.

[0056] If the calculated deviation does not exceed the judgment condition for triggering the deviation event of the indicator throughout the entire verification and monitoring cycle, the rectification is deemed effective and the closed loop is completed.

[0057] If a new indicator deviation event is triggered again during the verification and monitoring period, it will be determined that the rectification is not completely effective, and a new round of analysis and revision process will be initiated based on the latest situation.

[0058] As a further aspect of the present invention, when initiating a new round of analysis and revision process based on the latest situation, information inheritance processing is required:

[0059] The indicator deviation events generated in the previous round of processing, the corresponding background parameter data, the calculated linkage rectification intensity value, and the finally generated target rectification constraint value are stored as a complete case package in the historical rectification case library.

[0060] When performing background parameter data similarity matching for new indicator deviation events, all case packages in the historical rectification case library can be called to assist the judgment of the linkage evaluation model.

[0061] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0062] After the early warning program is activated, the status of monitoring nodes within the indicator monitoring network is automatically backtracked to extract background parameter data strongly related to the process within a specific time window before the indicator deviation event occurs. This step breaks the limitation of traditional monitoring that only focuses on instantaneous abnormal data points, extending the analytical perspective to the process context that led to the anomaly. By systematically capturing background information such as process status, operation sequence, resource load, and market environment fluctuations before the event occurs, a correlation graph of abnormal events can be constructed. The technical effect is that by placing isolated anomalies in the specific environment in which they occur for analysis, the tracing of the root cause of anomalies changes from speculation to data-driven attribution, significantly improving the accuracy and interpretability of early warnings. It can distinguish between accidental deviations caused by random fluctuations and substantial risks caused by systemic and correlated factors, providing specific and multi-dimensional decision-making basis for subsequent rectification decisions and avoiding the omissions and inefficiencies of manual backtracking.

[0063] The extracted background parameter data is input into a preset linkage evaluation model to calculate the comprehensive impact factor of the background parameters, thereby generating a linkage rectification intensity value. Based on this value, the relevant threshold parameters in the original process rule base are dynamically revised. This mechanism achieves automated connection from "diagnosis" to "treatment." Its core technology lies in the fact that the intensity of rectification actions is not subjectively determined by experience, but quantitatively driven by the results of in-depth analysis of abnormal events in the early stage. The effect is reflected in the system's ability to learn and optimize itself based on historical operational feedback. Each abnormal event processed and its analysis results are transformed into a calibration of the system's control parameters (thresholds). This enables the monitoring network to dynamically adapt to changes in business processes, changes in market conditions, and the evolution of internal operating modes. Through continuous and targeted revision of parameters, the system can proactively strengthen exposed weaknesses, fundamentally reducing the probability of recurrence of similar abnormal events, thus making the entire early warning and management system a dynamic and evolving closed-loop intelligence, rather than a static and passive regulatory tool. Attached Figure Description

[0064] Figure 1 The flowchart shows the multi-system integration and data collaboration method for procurement indicators based on a data middle platform as described in this invention.

[0065] Figure 2 A flowchart of the background parameter data extraction method;

[0066] Figure 3 This is a flowchart of the dynamic revision process for threshold parameters. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0068] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0069] See Figure 1 This invention proposes a method for multi-system integration and data collaboration of procurement indicators based on a data platform. The overall implementation scheme is as follows: An indicator monitoring network matching the bidding and procurement process is established, consisting of several monitoring nodes deployed at key stages of the process. During a preset observation period, these monitoring nodes continuously collect quantitative information reflecting the process status, with each collection defined as an indicator snapshot. The continuously generated indicator snapshots are compared; when the difference between an indicator snapshot and an adjacent historical snapshot exceeds a preset tolerance boundary, it is identified as an indicator deviation event and an early warning procedure is triggered. After the early warning procedure is activated, the status of the monitoring nodes within the indicator monitoring network is backtracked to extract several background parameter data strongly correlated with the process within a specific time window before the indicator deviation event. The extracted background parameter data is input into a preset linkage evaluation model, which calculates the comprehensive impact factor of the background parameters. Based on this comprehensive impact factor, the linkage rectification intensity value is further calculated. Based on the linkage rectification intensity value, the relevant threshold parameters in the original bidding and procurement process rule base are dynamically revised to generate target rectification constraint values, which replace the original threshold parameters. After the parameter replacement is completed, a new round of process status monitoring is restarted. If no further indicator deviation events are triggered, it signifies the end of a complete closed-loop process.

[0070] In one embodiment of the present invention, when continuously collecting quantitative information reflecting the process status through an indicator monitoring network, an indicator set is first defined, which includes raw values ​​of at least three dimensions: cost, cycle, and compliance. For each collection, the system calculates the rate of change of the raw value of each dimension relative to its initial value at the start of the observation period, thereby obtaining a set of rate of change sequences. All data within this rate of change sequence are normalized, and the processed data of the three dimensions are linearly combined according to preset weights, and the resulting values ​​constitute an indicator snapshot. The specific process for identifying indicator deviation events includes: obtaining the indicator snapshot generated at the current moment and recording it as the current snapshot; tracing back several consecutive collection cycles to obtain a set of historical snapshots; performing a difference operation between the current snapshot and each snapshot in the historical snapshot set to obtain a set of snapshot difference values; calculating the mean and standard deviation of the snapshot difference value set; comparing the current snapshot with the mean of the historical snapshot set and calculating its deviation; if the calculated deviation is greater than the product of the standard deviation of the snapshot difference value set and a preset sensitivity coefficient, then an indicator deviation event is determined to have occurred.

[0071] In practical implementation, the indicator set is defined to include at least the raw values ​​of cost, cycle, and compliance dimensions. These raw values ​​are continuously collected from monitoring nodes in the bidding and procurement process. For each collection operation, the rate of change of the raw value of the cost dimension relative to its initial value at the start of the observation period is calculated; the rate of change of the raw value of the cycle dimension relative to its initial value at the start of the observation period is calculated; and the rate of change of the raw value of the compliance dimension relative to its initial value at the start of the observation period is calculated, thus obtaining a set of rate of change sequences containing the three rate of change values. All data within the rate of change sequence are normalized, and the processed cost dimension data, cycle dimension data, and compliance dimension data are linearly combined according to preset weights. The resulting values ​​constitute an indicator snapshot, which can be expressed by the following formula:

[0072] ;

[0073] in Representative indicator snapshot, Preset weights representing the cost dimension The normalized value representing the rate of change in the cost dimension. Preset weights representing the period dimension, The normalized value representing the rate of change in the periodic dimension. Preset weights representing compliance dimensions Normalized value representing the rate of change in compliance dimensions.

[0074] In some embodiments, the process of identifying indicator deviation events is initiated by acquiring an indicator snapshot generated at the current moment and recording it as the current snapshot. Several consecutive acquisition cycles are traced back to obtain a set of historical snapshots, which contains multiple indicator snapshots generated consecutively before the current snapshot. The current snapshot is then compared with each indicator snapshot in the historical snapshot set to obtain a set of snapshot difference values, where each value represents the numerical difference between the current snapshot and a historical snapshot. The mean and standard deviation of the snapshot difference value set are calculated, and the current snapshot is compared with the mean of the historical snapshot set to calculate the deviation of the current snapshot. Optionally, the deviation can be calculated based on the absolute or relative difference between the current snapshot value and the mean of the historical snapshot set. If the calculated deviation is greater than the product of the standard deviation of the snapshot difference value set and a preset sensitivity coefficient, an indicator deviation event is determined to have occurred.

[0075] In practical implementation, the number of historical snapshots selected can be adjusted according to the monitoring frequency and process stability to ensure the reliability of statistical representation. It is understood that a preset sensitivity coefficient is used to adjust the sensitivity of the warning; a higher sensitivity coefficient makes the system more likely to trigger warnings for smaller fluctuations. In some embodiments, the difference calculation can use absolute value calculation to avoid directional influence. When calculating the standard deviation of the snapshot difference value set, whether to use the sample standard deviation formula or the population standard deviation formula depends on the amount of historical data. Optionally, normalization can use the min-max normalization method or the Z-score standardization method; the specific method selection is based on the pre-configured data distribution characteristics. It is understood that the linear combination weights of the indicator snapshots reflect the relative importance of different dimensions in the overall state assessment; these weights are pre-set through domain knowledge or historical data analysis. In practical implementation, the calculation of the rate of change ensures the consistency of the time base; the initial values ​​at the start of the observation period are fixed and recorded at the beginning of the monitoring cycle.

[0076] See Figure 2 In one embodiment of the present invention, the status of monitoring nodes within the indicator monitoring network is backtracked to extract several background parameter data strongly correlated with the process within a time window prior to the occurrence of an indicator deviation event. This process uses the time point when the indicator deviation event is identified as a baseline and sets a backtracking window of a fixed length. Within this backtracking window, original operation logs, resource allocation records, and approval flow timestamps associated with all dimensions in the indicator set are retrieved from the process database; these data are collectively referred to as initial background data. The initial background data is cleaned, removing invalid and duplicate records to form a clean background dataset. Feature extraction is performed on this clean background dataset to extract statistical features including operation frequency, resource usage fluctuations, and process delay duration; these statistical features constitute the background parameter data used for linkage evaluation.

[0077] In practice, the process of backtracking the status of monitoring nodes within the indicator monitoring network is initiated based on the time point when the indicator deviation event is identified. A backtracking window of a fixed length is set forward, the length of which is predefined based on the typical cycle of the bidding and procurement process and historical data analysis. Within the backtracking window, the original operation logs, resource allocation records, and approval flow timestamps associated with all dimensions in the indicator set are retrieved from the process database. The original operation logs record user operation behaviors, the resource allocation records the allocation of computing, storage, and human resources, and the approval flow timestamps mark the initiation and completion times of process steps. This data is collectively referred to as the initial background data. The initial background data is then cleaned to remove invalid and duplicate records. Invalid records include entries with incomplete log information, resource records with negative values ​​or exceeding reasonable ranges, and entries with logically incorrect timestamps. Duplicate records refer to redundant entries that are identical within the same time granularity. After cleaning, a clean background dataset is formed. Feature extraction is performed on a clean background dataset to extract statistical features including operation frequency, resource usage fluctuation, and process delay duration. Operation frequency is obtained by calculating the number of operation logs per unit time. Resource usage fluctuation is obtained by calculating the variance or standard deviation of resource allocation records within the backtracking window. Process delay duration is obtained by calculating the difference between adjacent approval flow timestamps and statistically analyzing their distribution characteristics. These statistical features constitute the background parameter data used for linkage evaluation.

[0078] In some embodiments, the fixed time length of the backtracking window can be set to an integer multiple of the standard cycle of the bidding and procurement process. In specific implementation, when retrieving initial background data from the process database, a correlation query is performed based on the unique identifier of the monitoring node and the key fields of the data table. Optionally, the process of cleaning invalid records can be implemented by setting rule filters, which preset the legal value range and format constraints of various types of data. It is understood that removing duplicate records helps to avoid repeated calculations of the same event during subsequent feature extraction, thereby ensuring the accuracy of background parameter data. In some embodiments, the calculation of resource consumption fluctuations can focus on specific types of resources, such as computing resources or manpower input for specific positions. The statistics of operation frequency can be further subdivided into the frequency of different operation types, such as the frequency of approval operations, modification operations, and query operations. Optionally, the statistical characteristics of process dwell time can include the maximum value, minimum value, average value, or median value, and the feature extraction formula can be expressed as:

[0079] ;

[0080] in, This represents the average characteristic of process dwell time. This represents the number of complete process steps identified in a clean background dataset. Representing the The start timestamp of each step Representing the The end timestamp of each step. In practice, the background parameter data produced by the feature extraction step is organized in the form of a structured vector or list for use by the linkage evaluation model. It can be understood that the start and end points of the backtracking window form a closed interval on the timeline, and the system ensures that the timestamps of all extracted initial background data fall within this interval.

[0081] In one embodiment of the present invention, the process of calculating the comprehensive impact factor of background parameters in the linkage evaluation model includes: comparing the operation frequency, resource consumption fluctuation, and process delay duration in the background parameter data with preset benchmark operation frequency, benchmark resource consumption, and benchmark process duration, respectively, to calculate the operation deviation coefficient, resource deviation coefficient, and duration deviation coefficient. Historical cases similar to the current indicator deviation event type are retrieved from the historical rectification case library, and the corresponding historical background parameter data are extracted. The cosine similarity between the current background parameter data and the retrieved historical background parameter data in each dimension is calculated to obtain the historical similarity. The operation deviation coefficient, resource deviation coefficient, duration deviation coefficient, and historical similarity are input into a pre-trained multilayer perceptron network. The multilayer perceptron network outputs a scalar value between zero and one, which is the comprehensive impact factor of background parameters. The method for calculating the linkage rectification intensity value is as follows: an intensity calculation function is established, the input of which is the comprehensive impact factor of background parameters and the deviation degree in the indicator deviation event. The comprehensive impact factor of background parameters is multiplied by a preset intensity base to obtain the basic intensity component. Multiply the deviation by another preset adjustment coefficient to obtain the adjustment intensity component. Add the base intensity component and the adjustment intensity component, and pass the sum through a Sigmoid function to restrict its value range to between zero and one. The mapped result is the linkage rectification intensity value.

[0082] In practical implementation, the process of calculating the comprehensive impact factor of background parameters in the linkage evaluation model is initiated by comparing the operation frequency in the background parameter data with a preset benchmark operation frequency. The ratio or difference between the operation frequency and the benchmark operation frequency is calculated to obtain the operation deviation coefficient. The resource deviation coefficient is calculated by comparing the resource usage fluctuation in the background parameter data with a preset benchmark resource usage. The duration deviation coefficient is calculated by comparing the process delay time in the background parameter data with a preset benchmark process time. A historical rectification case library is retrieved, and historical cases similar to the current indicator deviation event type are searched within the library. The corresponding historical cases are then extracted. The historical background parameter data is used to calculate the cosine similarity between the current background parameter data and the retrieved historical background parameter data in each dimension. The cosine similarity calculation is based on the dot product and magnitude of feature vectors such as operation frequency, resource consumption fluctuation, and process delay time. The operation deviation coefficient, resource deviation coefficient, time deviation coefficient, and historical similarity are input into a pre-trained multilayer perceptron network. The multilayer perceptron network contains an input layer, at least one hidden layer, and an output layer. The input layer receives four input features, and the output layer outputs a scalar value between zero and one through an activation function. This scalar value is the comprehensive influence factor of the background parameters.

[0083] In some embodiments, the operation deviation coefficient, resource deviation coefficient, and duration deviation coefficient can be calculated in the form of percentage deviation or absolute deviation. The preset baseline operation frequency, baseline resource usage, and baseline process duration are determined based on the statistical average or median of historical normal process data. Optionally, the retrieval of the historical rectification case library is based on the dimension type and deviation direction of the indicator deviation event for initial similarity screening. It can be understood that cosine similarity calculation treats background parameter data as points in a multi-dimensional space, and evaluates similarity by measuring the cosine value of the angle between vectors. The pre-training of the multilayer perceptron network is completed using historical case data with labeled background parameter comprehensive influence factors, and the training objective is to minimize the error between the predicted value and the true value.

[0084] In practical implementation, the method for calculating the intensity value of coordinated rectification is based on establishing an intensity calculation function. The input of the intensity calculation function is the background parameter comprehensive influence factor and the deviation degree in the indicator deviation event. The background parameter comprehensive influence factor is multiplied by a preset intensity base to obtain the basic intensity component. The deviation degree is multiplied by another preset adjustment coefficient to obtain the adjustment intensity component. The basic intensity component and the adjustment intensity component are added together, and the sum is mapped by a Sigmoid function. The Sigmoid function maps any real number to the range between zero and one. The result after mapping is the coordinated rectification intensity value. The formula for calculating the coordinated rectification intensity value can be defined as follows:

[0085] ;

[0086] in This represents the intensity value of joint rectification efforts. The representative background parameter comprehensive influence factor, Represents the degree of deviation. Represents the preset strength base. This represents the preset adjustment coefficient. This represents an exponential function.

[0087] In some embodiments, the preset intensity base and adjustment coefficient are calibrated by analyzing the correlation between historical rectification effects and the comprehensive influence factors and deviations of background parameters. Optionally, the sigmoid function can be replaced with other monotonically increasing functions with similar saturation characteristics, such as the hyperbolic tangent function, which undergoes a linear transformation to adjust its range. The addition of the base intensity component and the adjusted intensity component ensures that the two are of comparable magnitude before numerical calculation, and standardization is performed if necessary. It can be understood that the linked rectification intensity value is limited to between zero and one to facilitate subsequent normalization, understanding, and application of the rectification intensity.

[0088] See Figure 3 In one embodiment of the present invention, the process of dynamically revising the relevant threshold parameters in the original bidding and procurement process rule base is as follows: Based on the dimension associated with the indicator deviation event, locate the specific threshold parameter entry in the original bidding and procurement process rule base that needs to be adjusted, and mark it as the target threshold to be revised. Obtain the current value of the target threshold to be revised. Calculate the difference between the linkage rectification intensity value and the value one to obtain the rectification buffer coefficient. Multiply the current value of the target threshold to be revised by the rectification buffer coefficient to obtain a revision increment. Based on the nature of the indicator deviation event, determine whether the target threshold to be revised should be tightened or relaxed. If it is determined to be a tightening adjustment, subtract the revision increment from the current value of the target threshold to be revised to obtain the target rectification constraint value. If it is determined to be a relaxing adjustment, add the revision increment to the current value of the target threshold to be revised to obtain the target rectification constraint value. The method for determining whether to tighten or relax the adjustment is as follows: when the indicator deviation event manifests as increased cost, extended cycle, or increased risk of violation, it is determined that the relevant threshold needs to be tightened. When a deviation from the target metrics manifests as an unexpected decrease in cost, an abnormally shortened cycle, or an improper skipping of process steps, it is determined that the relevant thresholds need to be relaxed or adjusted to correct any monitoring blind spots or rule loopholes. This judgment logic is built into the system's decision engine as a rule.

[0089] In practical implementation, the process of dynamically revising relevant threshold parameters in the original bidding and procurement process rule base is initiated based on the dimension associated with the indicator deviation event. The system locates the specific threshold parameter entries in the original bidding and procurement process rule base that need to be adjusted and marks them as target thresholds to be revised. Target thresholds to be revised can include specific numerical parameters such as the maximum budget limit, the shortest announcement period, and the minimum number of bidders. The current value of the target threshold to be revised is obtained, directly read from the configuration file of the process rule base. The difference between the linkage rectification intensity value and the value one is calculated to obtain the rectification buffer coefficient, which ranges from zero to one. The current value of the target threshold to be revised is multiplied by the rectification buffer coefficient to obtain a revision increment. Based on the nature of the indicator deviation event, the system determines whether the target threshold to be revised should be tightened or relaxed. If it is determined to be a tightening adjustment, the current value of the target threshold to be revised is subtracted from the revision increment to obtain the target rectification constraint value. If it is determined to be a relaxing adjustment, the current value of the target threshold to be revised is added to the revision increment to obtain the target rectification constraint value. The formula for calculating the target rectification constraint value is:

[0090] ;

[0091] in This represents the target rectification constraint value. This represents the current value of the target threshold to be revised. The plus sign represents the intensity of joint rectification; the minus sign is used to relax the adjustment and the plus sign is used to tighten the adjustment.

[0092] In some embodiments, the calculation of the revision increment ensures that the adjustment magnitude is negatively correlated with the linkage rectification intensity value. A higher linkage rectification intensity value indicates a greater background impact, resulting in a smaller rectification buffer coefficient, a smaller revision increment, and a more moderate adjustment. Optionally, the target rectification constraint value undergoes a rationality check before replacing the original threshold parameter, such as checking whether it exceeds the system's allowed global upper and lower limits. It can be understood that locating the target threshold to be revised is achieved by establishing a mapping relationship between the indicator deviation event type and specific rule parameter entries in the rule base.

[0093] The method for determining whether to tighten or loosen the target threshold based on the nature of the deviation event is built into the rule logic of the system decision engine. In practice, when the deviation event manifests as increased costs, extended cycles, or increased risk of violations, the system decision engine determines that the relevant thresholds need to be tightened, such as lowering the maximum budget limit, extending the minimum review time, or increasing the number of compliance checkpoints. When the deviation event manifests as unexpectedly lower costs, abnormally shortened cycles, or improperly skipped process steps, the system decision engine determines that the relevant thresholds need to be loosened, such as appropriately increasing the budget limit to correct the supplier shortage problem caused by an excessively low budget, or shortening the time limit of redundant steps to correct existing monitoring blind spots or rule loopholes. The judgment logic is predefined in the form of "IF-THEN" rules. Table 1 below shows a simplified rule mapping example:

[0094] Table 1: Mapping Table of Indicator Deviation Nature and Threshold Adjustment Direction

[0095]

[0096] In some embodiments, the nature of an indicator deviation event is determined by both the specific indicator dimension that triggered it and the direction of deviation. The direction of deviation is determined by comparing the current snapshot value with the historical average. Optionally, the system decision engine's rule base supports administrators in dynamically maintaining and expanding it based on policy changes. It is understood that the purpose of relaxing or adjusting thresholds is to correct process anomalies or data distortions caused by overly strict rules, ensuring the sensitivity of the monitoring network.

[0097] In one embodiment of the present invention, the process of restarting a new round of process status monitoring also includes a verification step. After the target rectification constraint value takes effect, a verification monitoring cycle is initiated. During this verification monitoring cycle, new indicator snapshots are generated at the original collection frequency. The newly generated indicator snapshots are compared with the historical snapshot set before the indicator deviation event occurs using the same deviation calculation. If the calculated deviation does not exceed the judgment condition for triggering an indicator deviation event during the entire verification monitoring cycle, the rectification is deemed effective, and the closed loop is completed. If a new indicator deviation event is triggered again during the verification monitoring cycle, the rectification is deemed not fully effective, and a new round of analysis and revision process starting from the latest situation is initiated. When a new round of analysis and revision process is initiated, information inheritance processing is required: the indicator deviation event generated in the previous processing, the corresponding background parameter data, the calculated linkage rectification intensity value, and the finally generated target rectification constraint value are stored as a complete case package in the historical rectification case library. When performing background parameter data similarity matching for new indicator deviation events in the future, all case packages in the historical rectification case library can be called to assist the judgment of the linkage evaluation model.

[0098] In practice, restarting a new round of process status monitoring begins after the target rectification constraints take effect, initiating a verification step. The system starts a verification monitoring cycle, the duration of which can be set to a fixed multiple of the original observation period or an independent preset duration. Within the verification monitoring cycle, new indicator snapshots are generated at the original collection frequency. The generation method for these new indicator snapshots is completely consistent with the initial monitoring phase, calculating the rate of change in cost, cycle, and compliance dimensions based on the latest process data and then weighting the results. The newly generated indicator snapshots are compared with the historical snapshot set before the indicator deviation event using the same deviation calculation algorithm. The new snapshot is compared with the average of the historical snapshot set, and the standard deviation and sensitivity coefficient of the historical snapshot set are also referenced. If, throughout the entire verification monitoring cycle, the deviation calculated for each newly generated indicator snapshot does not exceed the criteria for triggering an indicator deviation event, the system determines that the rectification is effective, and the closed-loop processing is complete. If, during any calculation within the verification and monitoring period, the deviation of the newly generated indicator snapshot exceeds the criteria for triggering an indicator deviation event, the system determines that the rectification is not fully effective and will initiate a new round of analysis and revision process based on the latest situation. This new round of analysis and revision process includes the complete steps of identifying new indicator deviation events, performing status backtracking, linkage assessment, and threshold revision.

[0099] In some embodiments, the length of the verification monitoring period can be adjusted based on the complexity of the bidding and procurement activities; a longer verification monitoring period can be set for procurement projects with long cycles. In specific implementations, when calculating the deviation of a new indicator snapshot, the historical snapshot set referenced can be consistent with the set used when the alarm was first triggered, to ensure consistency in the judgment benchmark. Optionally, multiple checkpoints can be set within the verification monitoring period, and batch deviation assessments can be performed on the cumulatively generated indicator snapshots at each checkpoint. It is understood that the criterion for determining effective rectification is that the monitoring status at all times within the verification monitoring period has returned to normal, and no new warnings have been triggered.

[0100] In practical implementation, when initiating a new round of analysis and revision process based on the latest situation, information inheritance processing is required. The system encapsulates the indicator deviation events generated in the previous round of processing, the corresponding background parameter data, the calculated linkage rectification intensity value, and the finally generated target rectification constraint value as a complete case package. The complete case package is stored in the historical rectification case library, and the stored information includes the feature vector of the indicator deviation event, the background parameter data vector, the linkage rectification intensity value, the threshold parameter values ​​before and after revision, and the revision direction. When performing background parameter data similarity matching for new indicator deviation events in the future, all case packages in the historical rectification case library can be called. The background parameter data vector in the historical case package will serve as a candidate set to assist the step of calculating historical similarity in the linkage evaluation model. The similarity calculation can be expressed as:

[0101] ;

[0102] in This represents the cosine similarity between the current background parameter data vector and the background parameter data vector of a historical case. Represents the current background parameter data vector. The vector represents the background parameters of historical cases. The dot operator represents the dot product of vectors, and the double vertical bar symbol represents the magnitude of the vector.

[0103] In some embodiments, information inheritance ensures that the process data of each closed-loop attempt, regardless of success or failure, is accumulated into system knowledge. Optionally, the stored case packages are appended with timestamps and process context tags to facilitate efficient retrieval and filtering in the historical rectification case library. In specific implementations, when a new indicator deviation event shares certain similarities in background parameters with multiple historical cases, the linkage evaluation model can select the top few historical cases with the highest similarity for calculation reference. It can be understood that through continuous information inheritance and case accumulation, the historical rectification case library is continuously enriched, and the linkage evaluation model's assessment of the comprehensive impact factors of background parameters will become more accurate.

[0104] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for multi-system integration and data collaboration of procurement indicators based on a data middle platform, characterized in that, include: Establish an indicator monitoring network that matches the bidding and procurement process, the indicator monitoring network consisting of several monitoring nodes; During the preset observation period, quantitative information reflecting the process status is continuously collected through the indicator monitoring network, and each collection is defined as an indicator snapshot. The system compares consecutive snapshots of the indicator. When the difference between the snapshot and the adjacent historical snapshot exceeds the preset tolerance boundary, an indicator deviation event is identified, and an early warning procedure is triggered. After the early warning program is initiated, the status of the monitoring nodes in the indicator monitoring network is traced back to extract several background parameter data that are strongly related to the process within the time window before the indicator deviation event occurs. The extracted background parameter data is input into a preset linkage evaluation model, and the comprehensive impact factor of the background parameters is calculated through the linkage evaluation model. The intensity value of joint rectification is calculated based on the comprehensive impact factors of the background parameters. Based on the aforementioned linkage rectification intensity value, the relevant threshold parameters in the original bidding and procurement process rule base are dynamically revised to generate target rectification constraint values, which are then used to replace the original threshold parameters. After the parameter replacement is completed, a new round of process status monitoring is restarted. If the aforementioned indicator deviation event is no longer triggered, the closed-loop processing is completed.

2. The method for multi-system integration and data collaboration of procurement indicators based on a data middle platform according to claim 1, characterized in that, The continuous collection of quantitative information reflecting the process status through the indicator monitoring network includes: Define a set of indicators, which shall contain at least raw values ​​for three dimensions: cost, cycle, and compliance. For each data collection, the rate of change of the original value of each dimension relative to its initial value at the start of the observation period is calculated, resulting in a set of rate of change sequences. All data within the rate of change sequence are normalized, and the processed data in the three dimensions are linearly combined according to preset weights. The resulting values ​​constitute the index snapshot.

3. The method for multi-system integration and data collaboration of procurement indicators based on a data middle platform according to claim 2, characterized in that, The specific process for identifying a single indicator deviation event includes: Obtain a snapshot of the indicator generated at the current moment, and denote it as the current snapshot; By tracing back several consecutive collection periods, a set of historical snapshots can be obtained; The difference between the current snapshot and each snapshot in the historical snapshot set is calculated to obtain a set of snapshot difference values. Calculate the mean and standard deviation of the snapshot difference set; The current snapshot is compared with the average value of the historical snapshot set, and the deviation is calculated. If the deviation is greater than the product of the standard deviation of the snapshot difference value set and the preset sensitivity coefficient, then the index deviation event is determined to have occurred.

4. The method for multi-system integration and data collaboration of procurement indicators based on a data middle platform according to claim 3, characterized in that, The process of performing status backtracking on the monitoring nodes within the indicator monitoring network and extracting several background parameter data strongly correlated with the process within the time window before the occurrence of the indicator deviation event is as follows: Based on the time point when the deviation event of the aforementioned indicator was identified, a backtracking window of a fixed time length is set forward. Within the backtracking window, the original operation logs, resource allocation records, and approval flow timestamps associated with all dimensions in the indicator set are retrieved from the process database. The original operation logs, resource allocation records, and approval flow timestamps are collectively referred to as the initial background data. The initial background data is cleaned by removing invalid and duplicate records to form a clean background dataset. Feature extraction is performed on the clean background dataset to extract statistical features including operation frequency, resource consumption fluctuation, and process delay time. These statistical features constitute the background parameter data used for linkage evaluation.

5. The method for multi-system integration and data collaboration of procurement indicators based on a data middle platform according to claim 4, characterized in that, The process by which the linkage assessment model calculates the comprehensive impact factor of background parameters includes: The operation frequency, resource consumption fluctuation, and process delay time in the background parameter data are compared with the preset benchmark operation frequency, benchmark resource consumption, and benchmark process time, respectively, and the operation deviation coefficient, resource deviation coefficient, and time deviation coefficient are calculated. Obtain a historical rectification case library, retrieve historical cases in the historical rectification case library that are similar to the current indicator deviation event type, and extract the historical background parameter data corresponding to the historical cases; Calculate the cosine similarity between the current background parameter data and the retrieved historical background parameter data in each dimension to obtain the historical similarity; The operation deviation coefficient, resource deviation coefficient, duration deviation coefficient, and historical similarity are input into a pre-trained multilayer perceptron network. The multilayer perceptron network outputs a scalar value between zero and one, which is the comprehensive influence factor of the background parameters.

6. The method for multi-system integration and data collaboration of procurement indicators based on a data middle platform according to claim 5, characterized in that, The method for calculating the intensity value of joint rectification is as follows: Establish an intensity calculation function, the input of which is the comprehensive influence factor of the background parameter and the deviation degree in the index deviation event; The basic intensity component is obtained by multiplying the background parameter comprehensive influence factor by a preset intensity base. Multiply the deviation by another preset adjustment coefficient to obtain the adjustment intensity component; The basic strength component and the adjusted strength component are added together, and the sum is mapped by a Sigmoid function to restrict its value range to between zero and one. The result after mapping is the linkage rectification strength value.

7. A method for multi-system integration and data collaboration of procurement indicators based on a data middle platform as described in claim 6, characterized in that, The process of dynamically revising the relevant threshold parameters in the original bidding and procurement process rule base is as follows: Based on the dimension associated with the deviation event of the indicator, locate the specific threshold parameter entries in the original bidding and procurement process rule base that need to be adjusted, and mark them as target thresholds to be revised; Obtain the current value of the target threshold to be revised; Calculate the difference between the linked rectification intensity value and the value one to obtain the rectification buffer coefficient; Multiply the current value of the target threshold to be revised by the rectification buffer coefficient to obtain a revision increment; Based on the nature of the deviation event of the indicator, the target threshold to be revised is tightened or relaxed. If the adjustment is tightened, the current value of the target threshold to be revised is subtracted from the revision increment to obtain the target rectification constraint value; If the adjustment is to be relaxed, the current value of the target threshold to be revised is added to the revision increment to obtain the target rectification constraint value.

8. A method for multi-system integration and data collaboration of procurement indicators based on a data middle platform according to claim 7, characterized in that, Based on the nature of the deviation event of the indicator, the method for tightening or relaxing the target threshold to be revised is as follows: When the deviation of the aforementioned indicators manifests as increased costs, extended cycles, or increased risk of violations, it is determined that the relevant thresholds need to be tightened and adjusted. When the deviation of the aforementioned indicators manifests as an unexpected decrease in cost, an abnormally shortened cycle, or an improper skipping of process steps, it is determined that the relevant thresholds need to be relaxed and adjusted in order to correct existing monitoring blind spots or rule loopholes. The judgment logic is built into the system's decision engine in the form of rules.

9. A method for multi-system integration and data collaboration of procurement indicators based on a data middle platform according to claim 1, characterized in that, The process of restarting a new round of process status monitoring also includes a verification step: After the target rectification constraint value takes effect, a verification and monitoring cycle is initiated; During the verification and monitoring period, new index snapshots will continue to be generated at the original collection frequency. The newly generated indicator snapshot is compared with the set of historical snapshots before the indicator deviation event occurs using the same deviation calculation. If the calculated deviation does not exceed the judgment condition for triggering the deviation event of the indicator throughout the entire verification and monitoring cycle, the rectification is deemed effective and the closed loop is completed. If a new indicator deviation event is triggered again during the verification and monitoring period, it will be determined that the rectification is not completely effective, and a new round of analysis and revision process will be initiated based on the latest situation.

10. A method for multi-system integration and data collaboration of procurement indicators based on a data middle platform according to claim 9, characterized in that, When initiating a new round of analysis and revision process based on the latest situation, information inheritance processing is required: The indicator deviation events generated in the previous round of processing, the corresponding background parameter data, the calculated linkage rectification intensity value, and the finally generated target rectification constraint value are stored as a complete case package in the historical rectification case library. When performing background parameter data similarity matching for new indicator deviation events, all case packages in the historical rectification case library can be called to assist the judgment of the linkage evaluation model.

Images