Research and development-oriented organization productivity trend prediction method, device and equipment

Abstract

The invention relates to the technical field of large language model fusion, and discloses a research and development-oriented organization productivity trend prediction method, device and equipment. The method comprises the following steps: acquiring multi-source work occupation data of research and development personnel; mapping the multi-source work occupancy data to a unified time axis to obtain a multi-source nominal occupancy bitmap; fusing the multi-source nominal occupation bitmaps to obtain a fused nominal occupation bitmap; carrying out optimization solution on the weight, the occupancy judgment threshold value and the time migration parameter of each data source; correcting the fused nominal occupancy bitmap based on the weight and the time migration parameter of each data source obtained through optimization solution; processing according to a continuous concentration block constraint condition to obtain a fact occupation bitmap of the research and development personnel; and predicting the productivity of the organization according to the fact occupation bitmap of each research and development person in the organization. According to the invention, for research and development work, the reasonability of effective occupancy estimation is improved, so that the productivity prediction accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of large language model fusion, and particularly relates to a research and development-oriented organization productivity trend prediction method, device and equipment. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

[0003] For research and development tasks, such as software development, managers usually rely on project plans, work reports, meetings / calendars and the like to generate personnel occupation boards to determine the input of individuals and teams. However, the above data all have systematic noise and conflicts, for example, project plans are nominal commitments, often deviating from reality; work report data often has post-reporting, template copying and the like; meetings / calendars may contain repeated appointments, robot meetings, cross-time zone offsets and temporary insertions and the like. If the three types of data sources are directly added or a simple threshold judgment is made to reflect the real time occupation of personnel or teams in a period of time, the accuracy is poor.

[0004] In addition, the software development industry belongs to knowledge work, unlike manufacturing, and there is not a linear relationship between output and time input, and only using time accumulation to estimate productivity will systematically overestimate deliverable capacity. SUMMARY

[0005] Therefore, the present application provides a research and development-oriented organization productivity trend prediction method, device and equipment to estimate the effective occupation of research and development work and predict the organization productivity.

[0006] One aspect of the present application provides a research and development-oriented organization productivity trend prediction method, comprising the following steps: Obtaining multi-source work occupation data of research and development personnel; Mapping the multi-source work occupation data to a unified time axis and discretizing according to a set time slice to obtain a multi-source nominal occupation bitmap; Fusing the multi-source nominal occupation bitmap according to the initial weight of each data source to obtain a fused nominal occupation bitmap; Obtaining a true value segment, taking the minimum difference between the true value segment and an occupation judgment result based on the fused nominal occupation bitmap as an optimization objective, constructing an objective function, and optimizing and solving the weight of each data source, the occupation judgment threshold, and the time offset parameter; wherein the occupation judgment result is determined according to the fused nominal occupation bitmap corrected by the time offset parameter and the occupation judgment threshold; Correcting the fused nominal occupation bitmap based on the weight and time offset parameter of each data source obtained by optimization and solving; According to the continuous focus block constraint condition, the corrected fusion nominal occupation bitmap is processed to obtain the fact occupation bitmap of the R&D personnel. According to the fact occupation bitmap of each R&D personnel in the organization, the organization productivity is predicted.

[0007] In some embodiments, the initial weight determination method of each data source is: For each R&D personnel, a plurality of true value segments are obtained for each of the three types of data sources, and each true value segment represents the real occupation state of a time slice; For each of the three types of data sources, consistency detection is performed on the plurality of real segments and the original bitmap, and the initial weights of the three types of data sources are determined according to the consistency.

[0008] In some embodiments, after obtaining the fusion nominal occupation bitmap, smoothing processing is further performed.

[0009] In some embodiments, the objective function uses cross-entropy to measure the difference between the true value segment and the occupation determination result based on the fusion nominal occupation bitmap.

[0010] In some embodiments, the objective function further includes a time smoothing constraint, which is represented by the weight of each data source, the occupation determination threshold, and the difference sum of squares between adjacent time slices with a time offset parameter.

[0011] In some embodiments, processing the corrected fusion nominal occupation bitmap according to the continuous focus block constraint condition includes: According to the occupation determination threshold, the occupation state of each time slice in the corrected fusion nominal occupation bitmap is preliminarily determined; Identifying time slices that are continuously in the occupation state, and retaining segments with a duration longer than a minimum continuous duration; Judging the interval duration between adjacent retained segments, and if the interval duration is less than a set threshold, merging the adjacent retained segments; Based on morphological opening / closing operation, optimizing the morphology of the focus block; Mapping the processed continuous focus block back to the original time axis to obtain the fact occupation bitmap.

[0012] In some embodiments, according to the maximum productivity of the R&D personnel in the organization within a unit period, the productivity conversion coefficients of different project stages, and the fact occupation bitmap of each R&D personnel in the organization, the organization productivity is predicted.

[0013] In some embodiments, based on the obtained fact occupation bitmap, one or more of the effective occupation rate, the focus utilization rate, the fragmentation index, and the source consistency score of the R&D personnel in the organization within a certain period are calculated, and the working state of the organization is comprehensively evaluated.

[0014] The second aspect of the present application provides a research and development-oriented organization productivity trend prediction device, comprising: A data source acquisition module configured to acquire multi-source work occupancy data of the research and development personnel; A data preprocessing module configured to map the multi-source work occupancy data to a unified time axis and discretize according to a set time slice to obtain multi-source nominal occupancy bitmaps; A nominal occupancy determination module configured to fuse the multi-source nominal occupancy bitmaps according to initial weights of each data source to obtain fused nominal occupancy bitmaps; A correction parameter solving module configured to acquire true value segments, take the minimum difference between the true value segments and an occupancy determination result based on the fused nominal occupancy bitmaps as an optimization target, construct an objective function, and optimize and solve the weight of each data source, an occupancy determination threshold, and a time offset parameter; wherein the occupancy determination result is determined according to the fused nominal occupancy bitmaps corrected by the time offset parameter and the occupancy determination threshold; A nominal occupancy correction module configured to correct the fused nominal occupancy bitmaps based on the weight of each data source and the time offset parameter obtained by optimization and solving; A factual occupancy determination module configured to process the corrected fused nominal occupancy bitmaps according to a continuous focus block constraint condition to obtain a factual occupancy bitmap of the research and development personnel; An organization productivity prediction module configured to predict the organization productivity according to the factual occupancy bitmap of each research and development personnel in the organization.

[0015] The third aspect of the present application provides an electronic device comprising a processor and a memory, wherein the memory stores computer instructions, and when the computer instructions are executed by the processor, the electronic device executes the method.

[0016] One or more technical solutions above determine nominal occupancy based on occupancy data of different data sources, and solve dynamic optimization of weights, thresholds and time offsets through an objective function, correct deviations in the nominal occupancy bitmap, improve the reliability of the nominal occupancy bitmap, and then determine effective focus time relying on a continuous focus block constraint in combination with the specific scenario of research and development, so as to obtain effective work input of the research and development personnel, which is used for productivity prediction of a team, and solves the misjudgment problem of traditional time length accumulation statistics. BRIEF DESCRIPTION OF DRAWINGS

[0017] The drawings accompanying the specification of the present application form a part thereof, serve to provide further understanding of the present application, and together with the exemplary embodiments of the present application and their description, serve to explain the present application, and do not constitute improper limitations on the present application.

[0018] Figure 1A flow chart of a method for predicting R&D-oriented organization productivity trend is shown in an example embodiment of the present application. Figure 2 A structural block diagram of an apparatus for predicting R&D-oriented organization productivity trend is shown in an example embodiment of the present application. DETAILED DESCRIPTION

[0019] Embodiments of the present application will be described in more detail with reference to the drawings. Although some embodiments of the present application are shown in the drawings, it is understood that the present application can be embodied in various forms and should not be interpreted as being limited to the embodiments set forth herein, but rather the embodiments are provided to more thoroughly and completely understand the present application. It is understood that the drawings and embodiments of the present application are for exemplary purposes only and are not intended to limit the scope of protection of the present application.

[0020] In the description of embodiments of the present application, the term "comprising" and its conjugations should be understood to encompass the meanings of "including but not limited to", i.e., open-ended. The term "based on" should be understood as "based at least in part on".

[0021] For R&D tasks such as ERP software development, R&D personnel are the most core assets. However, since software R&D is a complex intellectual activity, there is not a linear relationship between output and time input, for example, an 8-hour period frequently interrupted is worth much less than a 4-hour period of continuous focus. Existing mainstream project management tools often equate time "quantity" with time "quality" when measuring input, ignoring the non-linear relationship between output and time input, which can lead to resource mismatch and distorted commitment when planning resources and making project decisions.

[0022] Existing R&D input-related data sources include project plans, employee work reports, meeting calendars, etc. However, these data sources all have deviations from reality. For example, project plans are usually made at the time of project approval, which is seriously out of touch with the actual work of project implementation; due to the fact that R&D work includes a large number of temporary technical support, internal discussion and other implicit tasks, employee work reports generally have lag and estimation behavior, resulting in inaccurate data and systematic underestimation of actual work load of employees; meetings or invitations recorded in the calendar may not be actually attended due to time conflicts, etc. The problems existing in these data sources make it difficult to determine the actual time occupation of R&D personnel.

[0023] To solve the above problems, one or more embodiments of the present application map different data sources to a unified timeline, perform weighted fusion, and preliminarily obtain a nominal occupancy bitmap representing "nominal busyness"; then solve the dynamic weights of different data sources, the occupancy determination threshold, and the time offset in combination with the objective function, wherein the weights of different data sources are used to measure the credibility of different data sources, the occupancy determination threshold is used to determine that "busyness" is a real occupancy, and the time offset is used to correct the inaccuracy of the time stamp, and thus a corrected and relatively credible nominal occupancy bitmap is obtained. Finally, the continuous focus block constraint is introduced to filter the effective occupancy and obtain the "effective focus time" of the R&D personnel that is truly invested and can produce value. Thus, the accuracy of the R&D personnel work investment estimation is improved, and thus a decision basis is provided for the organization's capacity prediction.

[0024] Figure 1 A flowchart of an R&D-oriented organization capacity trend prediction method provided by an exemplary embodiment of the present application is shown, and the method comprises the following steps: S1, obtaining multi-source work occupancy data of R&D personnel; S2, mapping the multi-source work occupancy data to a unified timeline and discretizing according to a set time slice to obtain a multi-source nominal occupancy bitmap; S3, fusing the multi-source nominal occupancy bitmap according to the initial weight of each data source to obtain a fused nominal occupancy bitmap; S4, obtaining a true value segment, taking the minimum difference between the true value segment and the occupancy determination result based on the fused nominal occupancy bitmap as the optimization target, constructing an objective function, and optimizing and solving the weight, the occupancy determination threshold, and the time offset parameter of each data source; wherein the occupancy determination result is determined according to the fused nominal occupancy bitmap corrected by the time offset parameter and the occupancy determination threshold; S5, correcting the fused nominal occupancy bitmap based on the weight and the time offset parameter of each data source obtained by the optimization and solution; S6, processing the corrected fused nominal occupancy bitmap according to the continuous focus block constraint condition to obtain a fact occupancy bitmap of the R&D personnel; S7, predicting the organization capacity according to the fact occupancy bitmap of each R&D personnel in the organization.

[0025] The above method guarantees the comprehensiveness of the data source through multi-source data access, solves the time asynchronization problem through time axis discretization, and on this basis, the multi-source data is weighted and fused based on the weight of different data sources, the dynamic optimization weight, threshold and time offset are solved combined with the target function, the deviation in the nominal occupation bitmap is corrected, the data reliability is improved, and then combined with the specific scene of R&D, relying on the continuous focus block constraint, the effective focus time is accurately screened, the fact occupation bitmap is determined, and the effective work input of the R&D personnel is obtained, and then used for team productivity prediction, solving the misjudgment problem of traditional time length accumulation statistics.

[0026] In step S1, the data sources of the multi-source work occupation data include plan, work report, meeting / calendar and other multi-source data in the software R&D scene. The plan data mainly includes project plan data, including task decomposition, plan start and end time, and the corresponding responsible person of each subtask. The work report data is used to record the completed work of software R&D, mainly the submission log, which includes work content, work start and end time, and completion person. The meeting / calendar data is used to record the collaboration occupation time in the software R&D process. The meeting data contains check-in time, actual attendance time, attendance list, actual start / end time, etc. The calendar data provides meeting reservation information, including meeting reservation time, attendees, etc.

[0027] After obtaining the multi-source data, standardization processing is also performed: for each R&D personnel, the work content and occupation time period are extracted according to different data sources, and the multi-source work occupation data is obtained.

[0028] In step S2, the standardized data is mapped to a unified time axis, and the time axis is discretized according to the set time slice, so as to represent the multi-source work occupation data in a unified time frame, and obtain a multi-source nominal occupation bitmap. The nominal occupation bitmap of each data source represents the nominal occupation of each time slice on the continuous time slice. If the occupation state is occupied, the value is 1, otherwise the value is 0. Exemplarily, the nominal occupation bitmaps of the plan, work report, meeting / calendar three types of data sources are represented as .

[0029] In step S3, the nominal occupation bitmaps of multiple data sources are weighted and fused according to the time slice to obtain a fused nominal occupation bitmap.

[0030]

[0031] Wherein, represents the fused nominal occupation bitmap of the R&D personnel r, r is the R&D personnel identifier, t represents the time slice, 、 、 respectively represent the initial weights of the three data sources of plan, report work, and meeting, which are obtained based on the consistency check of the true value fragments; 、 、 respectively represent the nominal occupancy bitmaps of the three data sources of plan, report work, and meeting.

[0032] In some embodiments, the initial weights of each data source are determined according to the reliability of each data source in the multi-source work occupancy data. Specifically, for each R&D personnel, a plurality of true value fragments are obtained for the three types of data sources respectively, each true value fragment representing the true occupancy state of a time slice, which is used to determine the reliability of the data source. For example, through audit sampling, 2 hours are randomly selected each week, and according to the work communication records and other information, it is confirmed whether the R&D personnel is in effective work during the time period; the time slice of effective work is obtained through the time anchor of code submission / building log; the work authenticity of a certain time period is confirmed through stage review information. For each R&D personnel, a plurality of true value fragments and original bitmaps are subjected to consistency detection for the three types of data sources respectively, and the reliability weights of the three types of data sources are determined according to the consistency, and the higher the consistency, the greater the reliability weight. For example: in the last 20 true value fragments of a certain employee, the consistency rate of report work with true value is 70%, the consistency rate of meeting is 40%, and the consistency rate of plan is 50%, and the initial weight w is converted to 0.7:0.4:0.5.

[0033] Part of the data has occasional exceptions. It can be understood that when there is a repeated appointment of a meeting, a single-point report work (only a certain time slice has a report work record, and there is no association before and after), and a second-level dot error (such as report work accurate to seconds but actual work calculated by minutes), it will cause the data to have jitter, repeated marking or missing count. Therefore, after obtaining the fused nominal occupancy bitmap, smoothing processing is performed within a limited window to suppress burrs.

[0034]

[0035] wherein, represents the smoothed fused nominal occupancy bitmap of the R&D personnel r, represents the smoothing weight, which can be a Gaussian weight or a uniform weight, and satisfies , , represents the time offset, is the smoothing half window. Exemplarily, when m=2, a local time window containing t -2、 t -1、 t 、 t +1、 t +2 time slices is used for smoothing operation.

[0036] In step S4, a target function is constructed with the minimum difference between the true value segment and the occupation determination result based on the fused nominal occupation bitmap as the optimization objective, wherein the occupation determination result is determined according to the fused nominal occupation bitmap corrected by the time offset parameter and the occupation determination threshold.

[0037] The time offset parameter is used to correct the alignment error of data and real time. Specifically, independent time offset parameters are respectively set for different data sources to reflect the systematic differences of data sources such as work reports, meetings, plans, etc., and to solve the problems of work lag, meeting advance, etc. In the initialization stage, the time offset parameter can be uniformly initialized as the average alignment error of multiple data sources, for example, by comparing the work report records and the code submission time, the meeting timestamp and the actual attendance check-in time to obtain the time alignment error of multiple data sources, or based on the time alignment error of the true value segment and the original signal, taking the average difference as the initial time offset parameter. When the time alignment error is counted, a safety window is also set to limit the maximum range of allowed time offset adjustment (for example, ± 30 minutes) to avoid "mismatching" a morning meeting to the afternoon, and the safety window can be determined by experience statistics or data distribution.

[0038] The occupation determination threshold is used to determine the real occupation state of each time slice. Since the work modes of researchers are different in different stages (for example, there are more meetings in the early stage of iteration and more development in the later stage), the occupation determination threshold can be dynamically updated by sliding average or distribution adaptive method to reflect the differences in research modes in different stages. The occupation state is determined according to the difference between the value corresponding to a time slice in the fused nominal occupation bitmap and the occupation determination threshold. The cross entropy minimization form is as follows:

[0039] wherein CE represents cross entropy, represents the true value segment corresponding to time slice t, is a scaling coefficient for adjusting the sensitivity of determination, represents the weighted time offset parameter of each data source, represents the smoothed fused nominal occupation bitmap, is a Sigmoid function for mapping the value in the parentheses to 0-1 to obtain the probability of the fused nominal occupation being determined as real occupation after correction by the time offset parameter.

[0040]

[0041]

[0042] wherein, , , respectively represent the fusion weights of different data sources, representing the credibility of each data source in the final occupancy determination; 、 、 respectively represent the nominal occupancy bitmap of the plan, work report, and meeting three data sources; 、 、 respectively represent the time offset parameters of the plan, work report, and meeting three data sources.

[0043] In addition, a time smoothing constraint is introduced to ensure that the above three types of parameters change continuously over time, prohibit sudden changes, and avoid large fluctuations in parameters due to abnormal true value fragments on a certain day, thereby ensuring the stability of the calibration aperture. The cross-entropy minimization form with time smoothing regularization is as follows:

[0044] wherein, is a penalty coefficient, and the Smooth function calculates the time fluctuation degree of the parameter. The greater the fluctuation, the higher the penalty value, and the greater the total optimization target value; otherwise, the penalty value is low.

[0045] In some embodiments, the Smooth function can adopt the form of the sum of squares of adjacent time slice differences, for example:

[0046] wherein, and respectively represent the weight mean and time offset parameter mean of different data sources. Through the time smoothing constraint, the above three optimization parameters can be constrained to change continuously over time, avoiding large fluctuations in aperture parameters due to single-point anomalies.

[0047] The expectation maximization (EM) or equivalent time optimization is iterated until convergence. Since what is learned is "a function of aperture parameters over time", not the individual behavior itself, it is necessary to continuously correct the source bias (such as long-term lag work report) to maintain the stability and continuity of the aperture while ensuring accuracy.

[0048] In the above step S4, for different data sources such as plans, work reports, and meetings, the original time slices and true value fragments are aligned respectively, and systematic errors are counted; on this basis, the time-varying time offset parameters and the optimized weights of each data source are obtained by optimization.

[0049] In step S5, based on the weight and time offset parameter of each data source obtained by optimization solution, the fused nominal occupancy bitmap is corrected; specifically, according to the optimization weight of each data source, the nominal occupancy bitmaps of multiple data sources are re-weighted and fused according to the time slice to obtain the optimized fused nominal occupancy bitmap; the time labels of each data source are corrected according to the time offset parameter, for example, the report work data is advanced by several minutes as a whole, and the conference data is delayed by several minutes, to obtain the corrected fused nominal occupancy bitmap.

[0050] In step S6, the corrected fused nominal occupancy bitmap is processed according to the continuous focus block constraint condition, specifically including: S601, according to the occupancy judgment threshold, the occupancy state of each time slice in the corrected fused nominal occupancy bitmap is preliminarily judged; the formula is as follows:

[0051] Among them, is an indicator function, when the condition in the parentheses is satisfied, it means that the time slice is nominal busy, The value is 1.

[0052] S602, identify the time slices that are continuously in the occupancy state, and retain the segments whose length exceeds the minimum continuous time length, that is, only retain the segments whose continuous value is 1 and whose length exceeds the minimum continuous time length The minimum continuous time length can be adjusted according to personnel and task type, such as 30 minutes for developers and 20 minutes for testers), and the fragmented busy shorter than the time length is eliminated.

[0053] S603, judge the interval length between adjacent retained segments, if the interval length is less than the set threshold, merge the adjacent retained segments, that is, the merging processing of the adjacent busy segments with an interval less than the set threshold Solves the problem of splitting focus blocks caused by short-term interruptions (such as answering a phone call) and restores the real continuous working state.

[0054] S604, based on morphological opening / closing operation, optimize the morphology of the focus block and suppress the sawtooth of in-meeting report work and invalid conference. Among them, the in-meeting report work refers to the situation that the developer temporarily fills in or mistakenly fills in the report work record of other work during the participation of the conference, resulting in "burr" of the focus block in the time period, and the invalid conference refers to the situation that the developer is included in the conference invitation but does not actually participate, resulting in "small hole" of the voting score.

[0055] S605, remap the processed continuous focus block to the original time axis to obtain the fact occupancy bitmap, wherein the time slice corresponding to the continuous focus block corresponds to the fact occupancy state of occupancy, and the value is 1. The fact occupancy bitmap of the developer r is represented as where the set of focus blocks is denoted as Each interval represents a continuous fact-occupied time period, is the start time slice of the period, is the end time slice.

[0056] Software development is a knowledge-based work, and the stacking of scattered time ≠ effective output (such as 1 day is divided into 8 15-minute segments, which is far less productive than 2 continuous 2-hour focus time). The core role of the "continuous focus block constraint" is to filter out the effective focus time that can be converted into actual delivery (such as code, scheme) based on the stable nominal occupation bitmap, avoiding the overestimation of output capacity caused by fragmented busy.

[0057] In step S7, based on the fact-occupied bitmap, combined with the post upper limit and the stage efficiency, the actual productivity of the organization in a certain period is estimated. First, introduce the post upper limit , which represents the maximum output capacity of the R&D personnel r in a unit period; introduce the stage efficiency coefficient , which represents the productivity conversion coefficient of different project stages (requirements, coding, testing, etc.). Then, taking the fact-occupied as a constraint, the actual effective productivity of the organization in time slice t is calculated:

[0058] In the formula, is the effective capacity of the organization g in time slice t, is the post upper limit, which is the maximum workable time / output capacity of the post in a unit period, is the stage efficiency coefficient, which is dynamically adjusted with the project stage (such as requirements, coding, testing), and the value is 0-1.

[0059] Further, the actual productivity is compared with the demand arrival amount to obtain the evolution of the demand backlog:

[0060] In the formula, represents the demand backlog of period t+1 (i.e. the next period), represents the initial backlog of period t (this period), is the new demand arrival amount of period t, represents the effective capacity of the organization, i.e. the bearable productivity.

[0061] Based on this, the leadership can not only trace back the real input of the R&D personnel in the past period (fact-occupied), but also predict the bearable productivity and potential backlog risk in the future period.

[0062] On the basis of unified and reliable caliber, the short-term gap and congestion risk trend can be intuitively given in the form of supply and demand difference. Specifically, the demand backlog and the effective capacity of the organization in the continuous period are obtained, and the demand gap trend analysis and congestion risk trend analysis can be carried out. If the demand backlog continues to increase, it is a short-term capacity shortage warning. If the demand backlog continues to increase and exceeds the set threshold, it is a medium and long-term congestion risk warning. Decision makers can refer to and start relevant response plans.

[0063] In addition, based on the obtained fact occupation bitmap, the working state of the R&D personnel or organization (team) can also be comprehensively evaluated to avoid misjudgment caused by single time length accumulation. Exemplarily, on a daily, weekly or iteration scale, the following evaluation indicators are used for evaluation: effective occupation rate, focused utilization rate, fragmentation index, source consistency score.

[0064] Among them, the effective occupation rate is the proportion of the cumulative length of the focused block of the R&D personnel in a certain time period to the total length, which is used to reflect the true effective input degree of the personnel. The average effective occupation rate of multiple R&D personnel in a team can be used to reflect the true effective input of the team. If the average effective occupation rate of a certain team is low, it means that there is more fragmented work.

[0065] The focused utilization rate is the ratio of the focused block length of the R&D personnel in a certain time period to the total occupation length (or called nominal busy length) in the fact occupation bitmap, which is used to reflect the proportion of effective focus in nominal busy. If the proportion of effective focus is small, it means that there is more fragmented work.

[0066] The fragmentation index is the number of switches per unit time or the proportion of short blocks, which is used to quantify the frequency of interrupted work. Among them, the number of switches per unit time is the ratio of the number of attention switches in a certain period to the total working time. By traversing the fact occupation bitmap, when the occupation state changes from "occupied" to "unoccupied" and then to "occupied" (i.e. the value changes from 1 to 0 and then to 1), it is recorded as once attention switch. The short block proportion is the ratio of the total length of the fragmented occupation of the minimum continuous length in a certain period to the total occupation length in the fact occupation bitmap. The more the number of switches per unit time or the larger the proportion of short block occupation, the more fragmented work.

[0067] In addition, according to the fact occupation bitmap and the multi-source work occupation data, the reliability of the fact occupation bitmap and the above-mentioned indicators can also be evaluated. Specifically, the source consistency score is used to measure. The source consistency score is used to measure the support degree of the effective occupation conclusion of different data sources such as plans, reports and meetings. Specifically, for each time slice corresponding to an effective focus block, the number of sources marked as occupied in different data sources is counted. The more the number of sources, the higher the source consistency and the higher the reliability, which can reflect the support degree of the above-mentioned multiple indicators.

[0068] Based on the above method, one or more embodiments of the present application also provide a research and development oriented organization productivity trend prediction device, comprising: a data source acquisition module configured to acquire multi-source work occupancy data of a research and development personnel; a data preprocessing module configured to map the multi-source work occupancy data to a unified time axis and discretize according to a set time slice to obtain a multi-source nominal occupancy bitmap; a nominal occupancy determination module configured to fuse the multi-source nominal occupancy bitmap according to an initial weight of each data source to obtain a fused nominal occupancy bitmap; a correction parameter solving module configured to acquire a true value segment, take the minimum difference between the true value segment and an occupancy determination result based on the fused nominal occupancy bitmap as an optimization target, construct an objective function, and optimize and solve a weight of each data source, an occupancy determination threshold, and a time offset parameter; wherein the occupancy determination result is determined according to the fused nominal occupancy bitmap corrected by the time offset parameter and the occupancy determination threshold; a nominal occupancy correction module configured to correct the fused nominal occupancy bitmap based on the weight of each data source and the time offset parameter obtained by the optimization and solving; a factual occupancy determination module configured to process the corrected fused nominal occupancy bitmap according to a continuous concentration block constraint condition to obtain a factual occupancy bitmap of the research and development personnel; and an organization productivity prediction module configured to predict organization productivity according to the factual occupancy bitmap of each research and development personnel in the organization.

[0069] One or more embodiments of the present application also provide an electronic device that can be used to implement the method in the above embodiments. The electronic device includes one or more processors, one or more memories coupled to the processors, and a communication module coupled to the processors.

[0070] The memory in the embodiments of the present application is used to store various types of data to support the execution of the methods as shown in Figure 2 .

[0071] It can be understood that the memory can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. The memory in the embodiments of the present application can store computer programs corresponding to each step in the method as shown in Figure 2 . The operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application program can include various application programs.

[0072] As an example, a processor can be an integrated circuit chip with signal processing capabilities, such as General Purpose Processors (GPPs), Digital Signal Processors (DSPs), or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or the like.

[0073] In particular, the processes described above with reference to the flow charts can be implemented as computer software programs. For example, embodiments of the present application include a computer program product which includes a computer program tangibly embodied on a computer readable medium, the computer program including instructions for execution by a processor to perform the processes described above with reference to the flow charts. Figure 2 The program code shown in the methods can be downloaded from the network through the communication part and installed, and / or installed from the detachable medium. When the computer program is executed by the central processing unit, various functions defined in the apparatus of the present application are executed.

[0074] wherein, Figure 2 The computer program instructions corresponding to the methods shown can also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer readable memory produce a product including instruction devices, which realize the functions specified in the flow Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0075] The above description is merely illustrative of the application, and is not intended to limit the application. Any modifications, equivalent replacements, improvements, and the like made within the spirit and principle of the application should be included in the protection scope of the application.

Claims

1. A method for predicting organizational capacity trends for R&D, characterized in that, Includes the following steps: Acquire multi-source workload data of R&D personnel; The multi-source workload occupancy data is mapped onto a unified time axis and discretized according to the set time slices to obtain a multi-source nominal occupancy bitmap; The multi-source nominal occupancy bitmap is merged based on the initial weight of each data source to obtain the merged nominal occupancy bitmap; Obtain a truth fragment, and construct an objective function with the goal of minimizing the difference between the truth fragment and the occupancy determination result based on the fused nominal occupancy bitmap. Optimize and solve for the weight, occupancy determination threshold, and time offset parameter of each data source. The occupancy determination result is determined based on the fused nominal occupancy bitmap and occupancy determination threshold corrected by the time offset parameter. Based on the weights and time offset parameters of each data source obtained from the optimization solution, the fused nominal bitmap is corrected; Based on the continuous focus block constraint, the corrected fusion nominal occupancy bitmap is processed to obtain the actual occupancy bitmap of the R&D personnel; Based on the actual bitmap of each R&D personnel within the organization, the organization's production capacity is predicted.

2. The method for predicting organizational capacity trends for R&D as described in claim 1, characterized in that, The initial weights for each data source are determined as follows: For each R&D personnel, multiple truth fragments are obtained for each of the three types of data sources. Each truth fragment represents the actual occupancy status of a time slice. For each of the three types of data sources, consistency checks are performed on multiple real fragments and the original bitmap, and the initial weights of the three types of data sources are determined based on the consistency.

3. The method for predicting organizational capacity trends for R&D as described in claim 1, characterized in that, After obtaining the nominal bitmap of the fusion, a smoothing process is also performed.

4. The method for predicting organizational capacity trends for R&D as described in claim 1, characterized in that, The objective function uses cross-entropy to measure the difference between the truth fragment and the occupancy determination result based on the fused nominal occupancy bitmap.

5. The method for predicting organizational capacity trends for R&D as described in claim 4, characterized in that, The objective function also includes time smoothing constraints, represented by the weight of each data source, the occupancy decision threshold, and the sum of squared differences between adjacent time slices for the time offset parameter.

6. The method for predicting organizational capacity trends for R&D as described in claim 1, characterized in that, Based on the continuous focus block constraint, the corrected fused nominal occupancy bitmap is processed as follows: Based on the occupancy determination threshold, the occupancy status of each time slice in the corrected fused nominal occupancy bitmap is initially determined; Identify time slices that are continuously occupied and retain segments whose duration exceeds the minimum continuous duration; Determine the time interval between adjacent reserved sections. If the time interval is less than a set threshold, merge the adjacent reserved sections. Optimize the shape of focus blocks based on morphological opening / closing operations; The processed continuous focus blocks are remapped onto the original timeline to obtain the fact occupancy bitmap.

7. The method for predicting organizational capacity trends for R&D as described in claim 6, characterized in that, Based on the maximum productivity of R&D personnel within the organization per unit cycle, the productivity conversion factor for different project stages, and the actual occupancy bitmap of each R&D personnel within the organization, the organization's productivity is predicted.

8. The method for predicting organizational capacity trends for R&D as described in claim 1, characterized in that, Based on the obtained factual occupancy bitmap, the organization's work status is comprehensively evaluated by statistically analyzing one or more of the following metrics for R&D personnel within a certain period: effective occupancy rate, focus utilization rate, fragmentation index, and source consistency score.

9. A device for predicting organizational capacity trends for R&D, characterized in that, include: The data source acquisition module is configured to acquire multi-source workload data of R&D personnel. The data preprocessing module is configured to map multi-source workload occupancy data onto a unified time axis and discretize it according to a set time slice to obtain a multi-source nominal workload bitmap. The nominal occupancy determination module is configured to: fuse the multi-source nominal occupancy bitmaps according to the initial weight of each data source to obtain a fused nominal occupancy bitmap; The correction parameter solving module is configured to: obtain a truth fragment, construct an objective function with the goal of minimizing the difference between the truth fragment and the occupancy determination result based on the fused nominal occupancy bitmap, and optimize the weight, occupancy determination threshold, and time offset parameter of each data source; wherein the occupancy determination result is determined based on the fused nominal occupancy bitmap and occupancy determination threshold corrected by the time offset parameter. The nominal occupancy correction module is configured to correct the fused nominal occupancy bitmap based on the weights and time offset parameters of each data source obtained from the optimization solution. The fact occupancy determination module is configured to: process the corrected fusion nominal occupancy bitmap according to the continuous focus block constraint conditions to obtain the fact occupancy bitmap of the R&D personnel; The organization capacity forecasting module is configured to forecast the organization's capacity based on the actual bitmap occupied by each R&D personnel within the organization.

10. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer instructions that, when executed by the processor, cause the electronic device to perform the method of any one of claims 1 to 8.

Images