Method for constructing adaptive adjustment model of innovation project index based on multi-dimensional evaluation

Abstract

The present application relates to the technical field of data processing, and particularly relates to an innovation project index self-adaptive adjustment model construction method based on multi-dimensional evaluation, comprising: training a project intervention model of each index set by using the participation weight of the index set, and artificially regulating the project by using the intervention value output by the project intervention model and an intervention threshold value; updating the intervention threshold value according to the difference between the evaluation participation degree of the intervention set and the correction index set and the regulation effect, and updating the participation weight of the correction index set according to the difference between the evaluation participation degree of the reference index set except the intervention set and the correction index set and the regulation effect; and artificially regulating the project again by using the updated participation weight and the updated intervention threshold value. The present application establishes a dynamic adjustment model for adjusting and innovating the project again to adapt to the changing user demand and development trend.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to an innovation project index self-adaptive adjustment model construction method based on multi-dimensional evaluation. BACKGROUND

[0002] With the increasing and changing demand for electricity, and with the large-scale access of new energy and the promotion of power marketization, it is necessary to innovate in the aspects of stability mechanism and operation characteristics of high-proportion new energy, high-proportion power electronic equipment, and high-proportion external power access to AC / DC hybrid power grids, and to promote the high-quality development of power grids. In the construction of innovative power system projects, breakthroughs need to be made in key technical fields of new-type power system construction, including but not limited to energy storage technology, flexible DC power transmission technology, distributed power generation and microgrid technology, power electronic technology, and Internet of Things monitoring technology. The development of these technologies is of great significance for improving the regulation capacity of power grids, optimizing power resource allocation, and reducing carbon emissions.

[0003] However, during the implementation of innovative power system projects, project evaluation is needed to ensure the continuous optimization of the innovation management system and to form a virtuous cycle. However, conventional evaluation methods (such as expert evaluation methods) are difficult to establish a multi-dimensional evaluation system and cannot establish a dynamic adjustment mechanism to adapt to changing market and technological trends, resulting in the inability to timely and accurately optimize and improve and innovate again in response to project pain points during project implementation and operation, and the difficulty in improving satisfaction and achieving expected economic benefits and building technical barriers. SUMMARY

[0004] To solve the above problems, the present application provides an innovation project index self-adaptive adjustment model construction method based on multi-dimensional evaluation.

[0005] The innovation project index self-adaptive adjustment model construction method based on multi-dimensional evaluation of the present application adopts the following technical scheme:

[0006] One embodiment of the present application provides an innovation project index self-adaptive adjustment model construction method based on multi-dimensional evaluation, which includes the following steps:

[0007] The multi-index data of the project and the comprehensive evaluation result are obtained, all the index data are divided into a plurality of index sets, a project intervention model of each index set is trained by using a participation weight of the index set, the participation weight is positively correlated with an evaluation participation degree of each index set, and the evaluation participation degree is obtained according to a correlation between the index data in each index set and the comprehensive evaluation result; an index set whose intervention value output by the project intervention model is less than an intervention threshold value is recorded as an intervention set, and the project is artificially regulated according to the intervention set; the index set obtained before and after the regulation of the project is recorded as a reference index set and a modified index set respectively, and a regulation effect is obtained according to a change in the evaluation participation degree between the modified index set and the reference index set; the intervention threshold value is updated according to a difference in the evaluation participation degree between the intervention set and the modified index set and the regulation effect, and the participation weight of the modified index set is updated according to a difference in the evaluation participation degree between the reference index set and the modified index set except the intervention set and the regulation effect; the project intervention model of each modified index set is trained by using the updated participation weight, and the modified index set whose intervention value output by the project intervention model is less than the updated intervention threshold value is recorded as the intervention set again, and the project is artificially regulated according to the intervention set obtained again.

[0008] Preferably, the step of dividing all the index data into a plurality of index sets comprises the following specific steps:

[0009] In a time period T before the current time, time series data composed of all the comprehensive evaluation results is obtained and recorded as first time series data; time series data composed of all the values of any index data is recorded as second time series data; and an absolute value of a Pearson correlation coefficient between the second time series data and the first time series data is recorded as an evaluation result correlation of any index data.

[0010] The correlation degree between the second time series data corresponding to any two index data and the first time series data is obtained, and all the index data are divided into a plurality of index sets according to the evaluation result correlation and the correlation degree.

[0011] The time period T is a preset time period.

[0012] Preferably, the step of dividing all the index data into a plurality of index sets according to the evaluation result correlation and the correlation degree comprises the following specific steps:

[0013] The product of the correlation degree and the average of the evaluation result correlations of any two index data is recorded as a metric distance of any two index data, and all the index data are clustered by using the average K-Means according to the metric distance of any two index data to obtain a plurality of categories, and all the index data in each category are recorded as an index set.

[0014] Preferably, the participation weight of the index set is used to train the project intervention model of each index set, and the specific steps include the following:

[0015] In the time period between the two adjacent times of obtaining the comprehensive evaluation result, the values of all index data in each index set form a sample, and the comprehensive evaluation result obtained at the later time is recorded as the label of the sample;

[0016] In the time period T before the current time, all samples and labels corresponding to each index set form a data set;

[0017] Each data set corresponding to each index set corresponds to a project intervention model, and the output result of each project intervention model is recorded as the intervention value of each index set; the loss function used in the training of all project intervention models is:

[0018]

[0019] Wherein, S represents the loss function, w i represents the participation weight of the i-th index set, F i represents the intervention value of the i-th index set, F0 represents the label corresponding to the sample input into the project intervention model, K represents the number of index sets; and the time period T is a preset time period.

[0020] Preferably, the regulation effect is obtained according to the change of the evaluation participation degree between the modified index set and the reference index set, and the specific steps include the following:

[0021] All reference index sets are matched with all modified index sets one by one, the difference between the evaluation participation degree of the modified index set and the evaluation participation degree of the matched reference index set is obtained, recorded as the first difference of the modified index set, and the first difference of all modified index sets is linearly normalized. The mean value of the first difference of all modified index sets after linear normalization is recorded as B1;

[0022] The difference between the current comprehensive evaluation result and the last comprehensive evaluation result is recorded as the second difference, and the regulation effect is positively correlated with B1 and also positively correlated with the second difference.

[0023] Preferably, the intervention threshold is updated according to the difference between the evaluation participation degree of the intervention set and the modified index set and the regulation effect, and the specific steps include the following:

[0024] The intervention set is included in all reference indicator sets, the evaluation participation degree of the modified indicator set matched with the intervention set is denoted as C1, the evaluation participation degree of the intervention set is denoted as C2, and the ratio of (C2-C1) to max(C2,C1) is denoted as w2; max() represents the maximum value function;

[0025] The intervention threshold of the modified indicator set matched with the intervention set is (1-w2×q)×T1, T1 represents the intervention threshold corresponding to the intervention set; q is a regulation effect mask factor, which is obtained from the regulation effect;

[0026] For other modified indicator sets except the modified indicator set matched with the intervention set, the intervention threshold corresponding to the other modified indicator set is equal to the intervention threshold of the reference indicator set matched with the other modified indicator set.

[0027] Preferably, the intervention weight of the modified indicator set is updated according to the difference between the evaluation participation degree of the reference indicator set and the modified indicator set except the intervention set and the regulation effect, and the specific steps include the following:

[0028] All reference indicator sets include the intervention set, and the reference indicator set except the intervention set is denoted as a first set,

[0029] The evaluation participation degree of the first set is denoted as D1, the evaluation participation degree of the modified indicator set matched with the first set is denoted as D2, and |D1-D2| and max(D1,D2) are denoted as the weight correction factor w3 of the modified indicator set matched with the first set;

[0030] The first weight of the modified indicator set matched with the first set is denoted as (1+w3×q)×P, and P represents the intervention weight of the modified indicator set matched with the first set;

[0031] For other modified indicator sets except the modified indicator set matched with the first set, the first weight of the other modified indicator set is equal to the intervention weight of the other modified indicator set;

[0032] The first weights of all modified indicator sets are normalized, and the normalized result is denoted as the modified intervention weight of the modified indicator set;

[0033] q is a regulation effect mask factor, which is obtained from the regulation effect; max() represents the maximum value function.

[0034] Preferably, the specific acquisition steps of the regulation effect mask factor are as follows:

[0035] When the regulation effect is greater than a preset first threshold, q=0, and when the regulation effect is less than or equal to the preset first threshold, q=1.

[0036] Preferably, the evaluation participation degree is obtained from the correlation between the index data in each index set and the comprehensive evaluation result, and the specific steps include the following:

[0037] For any one index set, the second time series data corresponding to all index data in the index set and the first time series data are subjected to PCA dimension reduction to one dimension, and the variance of all dimension reduction results is obtained, which is recorded as the evaluation participation degree of each index set.

[0038] The method for obtaining the participation weight of each index set includes:

[0039] The evaluation participation degrees of all index sets are normalized, and the normalized evaluation participation degree is recorded as the participation weight of each index set.

[0040] The specific steps for obtaining the correlation degree between the second time series data corresponding to any two index data and the first time series data include the following:

[0041] The second time series data corresponding to any two index data and the first time series data are subjected to dimension reduction to one dimension by using the PCA algorithm, and the variance of all dimension reduction results is obtained, which is recorded as the correlation degree between the first time series data and any two index data.

[0042] Preferably, the one-to-one matching between all reference index sets and all modified index sets includes the following specific steps:

[0043] The KM algorithm is used to perform one-to-one matching between all reference index sets and all modified index sets, and the matched index sets have the largest intersection ratio.

[0044] The technical scheme of the present application has the following advantages:

[0045] The present application trains the project intervention model of each index set by using the participation weight of the index set, and the index set whose intervention value output by the project intervention model is less than the intervention threshold is recorded as the intervention set, and the project is artificially regulated according to the intervention set; in this process, the project intervention model is obtained by using the index sets that are correlated and have a significant influence on the evaluation result, and the project intervention model is used to evaluate the innovation project from specific dimensions, different index sets have different participation weights, so that the project intervention model can pay attention to the index data that are more correlated and influential to the evaluation result when evaluating the innovation project from specific dimensions, and avoid the problem that the evaluation result (i.e. intervention value) is unreliable when irrelevant index data cannot evaluate the pain points of the innovation project. The project is artificially regulated by the intervention set, so that the innovation project can be adjusted and re-innovated in view of the pain points of the innovation project and the insufficient evaluation when running.

[0046] Further, the application updates and corrects the intervention threshold and participation weight according to the change of the evaluation participation degree of the index set before and after the regulation, avoids the loss of the original correlation of the index data monitored by the innovation project after the regulation, and further avoids the following problems: after the comprehensive evaluation result is obtained, the participation weight of each index set cannot be accurately obtained, and then the project intervention model cannot be accurately trained, which affects the next regulation and continuously affects all subsequent regulation processes. Through the above process, the innovation project can be dynamically adjusted to adapt to the changing market demand and technology development trend when continuously running. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0048] Figure 1 The step flow chart of the innovation project index self-adaptive adjustment model construction method based on multi-dimensional evaluation provided by an embodiment of the present application. DETAILED DESCRIPTION

[0049] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the following will combine the drawings and the preferred embodiments to specifically describe the innovation project index self-adaptive adjustment model construction method based on multi-dimensional evaluation according to the present application, its specific implementation, structure, characteristics and effects. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0051] The following will specifically describe the specific scheme of the innovation project index self-adaptive adjustment model construction method based on multi-dimensional evaluation provided by the present application in combination with the drawings.

[0052] Please refer to Figure 1 which shows the step flow chart of the innovation project index self-adaptive adjustment model construction method based on multi-dimensional evaluation provided by an embodiment of the present application. The method comprises the following steps:

[0053] Step S001, obtaining multiple index data of the project and a comprehensive evaluation result.

[0054] In the construction of an innovative power system project (referred to as an innovative project), multiple index data are often used to monitor and evaluate the innovative project from multiple dimensions or aspects, and the evaluation result is used for improvement, adjustment or re-innovation, so as to improve the digital level of power grid research, and make the innovative project better and more sustainable to provide high-level, convenient, and intelligent results and transformation services.

[0055] In this embodiment, the multiple index data of the innovative project include load fluctuation, fault early warning capability, rapid recovery capability under fault condition, clean energy access ratio, renewable energy access ratio, and energy storage device input amount of the power grid in the innovative project.

[0056] As an example, the method for obtaining the load fluctuation includes:

[0057] In a normal working state of the power grid (for example, in a non-power failure state), the power flowing through each busbar in the power grid is obtained,

[0058] The difference between the maximum value and the minimum value of the output power of each busbar in each hour is recorded as the fluctuation amplitude of each busbar. In one day, the average value of all fluctuation amplitudes of all busbars in the power grid is recorded as the load fluctuation of the power grid.

[0059] In other embodiments, each busbar in the above method is replaced by each power distribution cabinet or each user's electricity meter. In other embodiments, the power of other power equipment can also be collected to calculate the load fluctuation of the power grid. This embodiment is not limited specifically.

[0060] As an example, the method for obtaining the fault early warning capability includes:

[0061] The number of fault maintenance times N0 of the power grid in each day is obtained, wherein the number of times that the fault is warned before the fault occurs (for example, warned by an Internet of Things detection device) is recorded as N1. The ratio of N1 to (N0+1) is recorded as the fault early warning capability. (N0+1) is to avoid the denominator equal to 0.

[0062] As an example, the method for obtaining the rapid recovery capability under fault condition includes:

[0063] The time from the fault occurrence to the power grid returning to steady state is recorded as t, and the ratio of t to H is recorded as the rapid recovery capability, wherein H represents a human preset fault level, the more serious the fault, the larger H is, in the embodiment, when the fault causes 1% of the users to be unable to use electricity normally, H is set to 1, when the fault causes 10% of the users to be unable to use electricity normally, H is set to 2, when the fault causes 30% of the users to be unable to use electricity normally, H is set to 3, and when the fault causes more than 30% of the users to be unable to use electricity normally, H is set to 4. In other embodiments, H can be set according to the specific power supply situation of the power grid, and the embodiment is not specifically limited.

[0064] As an example, the method for obtaining the clean energy access ratio, the renewable energy access ratio, and the input amount of the energy storage device includes:

[0065] The ratio of the total amount of clean energy generated in a day to the amount of electricity generated by all power sources in the power grid is recorded as the clean energy access ratio, and the ratio of the total amount of renewable energy generated in a day to the amount of electricity generated by all power sources in the power grid is recorded as the renewable energy access ratio.

[0066] The average of the charging amount and the discharging amount of all energy storage devices in a day to the amount of electricity generated by all power sources in the power grid is recorded as the utilization rate of the energy storage device, and the product of the utilization rate of the energy storage device and the total capacity of the energy storage device is recorded as the input amount of the energy storage device.

[0067] In more specific implementations, more index data can be introduced, such as the processing amount of the Internet of Things detection device or the control device in the power grid, the carbon dioxide emission amount, the fund input amount for regular inspection or regular maintenance of the power grid, and the like, which will not be described one by one in the embodiment.

[0068] The above-mentioned multiple index data come from multiple dimensions or multiple aspects of the innovation project, and are used for multi-dimensional monitoring of the innovation project, such as monitoring from the dimensions of power grid load, power grid fault, power grid steady state recovery, and the like. Only when the innovation input (such as the innovation input of power grid technology, the innovation input of funds and management, the innovation input of talents and thinking concept, and the like) is good in each of the above-mentioned dimensions, can the many challenges in the construction of the power grid be solved innovatively.

[0069] However, in the implementation process of the innovation project, there are increasing and changing demands, and it is difficult to solve all problems at once or in a short period of time. As described above, the innovation project often needs to be continuously monitored and evaluated, and the project needs to be optimized, so that the innovation project can solve many problems and gradually adapt to changing demands.

[0070] The innovation project is evaluated periodically in the embodiment to obtain a comprehensive evaluation result of the project. In the embodiment, the evaluation is performed every other month to obtain the comprehensive evaluation result. In other embodiments, the evaluation can be performed every other week or every half year to obtain the comprehensive evaluation result. The embodiment is not specifically limited.

[0071] The comprehensive evaluation result is used to describe whether the innovation project can achieve the expected purpose. For example, whether the innovation project still operates as expected when facing various problems or whether the innovation project can meet the growing and changing user demand. The smaller the comprehensive evaluation result is, the more the innovation project cannot achieve the expected purpose. The larger the comprehensive evaluation result is, the more the innovation project can achieve the expected purpose or exceed the expected purpose.

[0072] As an example, the method for obtaining the comprehensive evaluation result of the project includes:

[0073] The satisfaction of each project is investigated by expert investigation. For example, the satisfaction is divided into 0, 1, 2, …, 9, a total of ten levels. The higher the satisfaction is, the more satisfied the expert is with the innovation project. The mean value of the satisfaction of all experts is taken as the comprehensive evaluation result.

[0074] As another example, the method for obtaining the comprehensive evaluation result of the project includes:

[0075] Considering that the ultimate purpose of the innovation project is to enable the power grid users to use electricity safely, stably and continuously, in the embodiment, the total power provided to all power grid users when the power grid operates safely, stably and continuously since the innovation project is put into operation is recorded as A1, and the product of the number of users affected and the duration of all faults when the power grid fails is recorded as A2. The ratio of A1 to A2 is taken as the comprehensive evaluation result. The larger the ratio is, the more the innovation project can enable the power grid users to use electricity safely, stably and continuously.

[0076] In some embodiments, the difference between the total fund income and the total fund investment of the innovation project is recorded as the comprehensive evaluation result. The comprehensive evaluation result describes whether the innovation project can achieve the expected purpose (or exceed the expected purpose) from the perspective of economic benefits. However, considering that the economic income brought by the innovation project can also be affected by the macro-control of the social environment, the comprehensive evaluation result in these embodiments is not reliable and can be used as a secondary implementation manner of all embodiments of the application.

[0077] In other embodiments, the comprehensive evaluation result can be obtained from the perspectives of the advancement of the technical project, the number of intellectual property rights, and the number of technical problems solved.

[0078] In some embodiments, the comprehensive evaluation results obtained in all the above embodiments are weighted and summed to obtain more comprehensive and reliable comprehensive evaluation results, wherein the weights required for the weighted sum are preset, for example, the weights corresponding to the comprehensive evaluation results calculated above are 0.4, 0.3, 0.05, 0.05, 0.1, and 0.1, respectively; in other embodiments, the weights can be set to other values, and the present embodiment is not limited in this regard.

[0079] In step S002, all the index data are divided into a plurality of index sets according to the correlation between the different index data and the comprehensive evaluation results, and the participation weights of each index set are obtained according to the correlation between all the index data in each index set and the comprehensive evaluation results.

[0080] The monitoring results of the innovation project (i.e., all the index data) are correlated with the evaluation results of the innovation project (i.e., the comprehensive evaluation results of the project), for example, when the power dispatching scheme of the power grid is accurate and reliable, the load balance of the power grid can be ensured, and significant fluctuations in the load of the power grid can be avoided, which will make the index data of the load fluctuation in step S001 smaller, and the smaller the load fluctuation is, the more likely the innovation project will achieve the expected purpose (i.e., the larger the comprehensive evaluation results are). For another example, when the energy storage devices in the power grid are fully used, the safe and stable operation of the power grid can be maintained, which will make the index data of the input amount of the energy storage devices monitored in step S001 larger, and make the innovation project achieve the expected purpose (i.e., the larger the comprehensive evaluation results are).

[0081] Further, it is considered that making innovation and improvement in only some aspects of the power grid (for example, using only the innovative power dispatching scheme to improve or solve the load problem of the power grid, or using only the innovative energy storage device input scheme to solve the problem of insufficient input amount of the energy storage devices) may not make the innovation project significantly approach the expected purpose (i.e., the comprehensive evaluation results may not significantly increase).

[0082] In the present embodiment, all the index data are divided into a plurality of index sets according to the correlation between the different index data and the comprehensive evaluation results. Each index set includes a plurality of index data, and further adjustment and innovation of the innovation project from the dimensions corresponding to these index data can significantly change the comprehensive evaluation results (i.e., can make the innovation project further achieve the expected purpose).

[0083] As an example, all the index data are divided into a plurality of index sets according to the correlation between the different index data and the comprehensive evaluation results, including the following methods:

[0084] In the time period T before the current time, time series data of all comprehensive evaluation results is obtained, and is recorded as first time series data. For any index data, time series data of all values of the index data is recorded as second time series data of the index data.

[0085] The time period T before the current time in the embodiment is two years. In other embodiments, the time period T before the current time can be defined as other time periods, for example, all time since the innovation project is put into operation. In the embodiment, if the innovation project is put into operation for less than two years, the time period T before the current time is all time since the innovation project is put into operation.

[0086] When the length of the second time series data is not equal to the length of the first time series data, the DTW algorithm is used to match the second time series data with the first time series data in the embodiment, so that the length of the second time series data is equal to the length of the first time series data. The absolute value of the Pearson correlation coefficient of the second time series data and the first time series data is recorded as the evaluation result correlation of any index data.

[0087] Further, the second time series data corresponding to any two index data is obtained, and the second time series data corresponding to the two index data and the first time series data are respectively regarded as three high-dimensional vectors. The PCA algorithm is used to reduce the three high-dimensional vectors to one dimension, and the variance of all reduced results is obtained, which is recorded as the correlation degree of any two index data and the first time series data. The greater the variance, the greater the common information contained between any two index data and the first time series data, and the greater the correlation.

[0088] The product of the correlation degree and the mean of the evaluation result correlation of any two index data is recorded as d, and exp(-d) is recorded as the metric distance of any two index data. According to the metric distance of any two index data, mean K-Means clustering is performed on all index data, and K categories are obtained. In the embodiment, K is set to one third (rounded up) of the number of index data, and the purpose is to have about three index data in each category. In other embodiments, K can be set to other values, and the embodiment is not specifically limited.

[0089] All index data in the same category have a relatively obvious correlation with the comprehensive evaluation result (that is, all index data in the same category have a smaller metric distance), for example, a category includes load fluctuation, rapid recovery capability and other index data, and then through balanced adjustment and steady-state recovery control of the power grid load (for example, intelligent power dispatching), the comprehensive evaluation result can be increased (that is, the innovative project can be obviously close to the expected purpose, for example, the satisfaction or the power grid stability can be improved in a short time, or the satisfaction or the power grid stability can be improved with less time cost or capital cost).

[0090] The index data in different categories do not have a relatively obvious correlation with the comprehensive evaluation result, for example, two categories respectively include rapid recovery capability and carbon dioxide emission, and then the steady-state recovery control of the power grid load is performed while the carbon dioxide emission is controlled, and the comprehensive evaluation result cannot be increased (that is, the innovative project cannot be obviously close to the expected purpose, for example, the satisfaction or the power grid stability cannot be improved in a short time, or the satisfaction or the power grid stability can be improved to a certain extent with a large time cost or capital cost).

[0091] All index data in each category is recorded as an index set, and then K index sets are obtained in total.

[0092] For any index set, the second time sequence data corresponding to all index data in the index set and the first time sequence data are subjected to PCA dimension reduction to one dimension, and the variance of all dimension reduction results is recorded as the evaluation participation degree of each index set. The greater the evaluation participation degree of each index set, the greater the correlation between the index data in each index set and the comprehensive evaluation result, and the more the index set can be used for evaluation of the innovative project.

[0093] The evaluation participation degrees of all index sets are normalized, and the normalized evaluation participation degrees are recorded as the participation weights of each index set. In this embodiment, the softmax method is used for normalization processing, and the purpose is to make the sum of the participation weights of all index sets equal to 1. The greater the participation weight of each index set, the more the index data in each index set can be used for evaluation of the innovative project.

[0094] Step S003, training the project intervention model of each index set by using the participation weight of the index set.

[0095] Whenever the comprehensive evaluation result of an innovation project is obtained, a project intervention model is trained using all the index data in each index set, and multiple index sets correspond to multiple project intervention models. The project intervention model used in this embodiment is a fully connected neural network model (for example, a fully connected neural network model with 4 hidden layers). In other embodiments, an LSTM neural network model can also be used, and this embodiment is not specifically limited. The fully connected neural network model and the LSTM neural network model are well-known technologies, and this embodiment will not be described in detail.

[0096] The method for obtaining the data set used to train each project intervention model is as follows:

[0097] In this embodiment, the comprehensive evaluation result of the project is obtained regularly, and the comprehensive evaluation result is obtained multiple times as the innovation project runs. In the time period from the last time the comprehensive evaluation result is obtained to the next time the comprehensive evaluation result is obtained (i.e., in the time period between the two adjacent times when the comprehensive evaluation result is obtained), each index data corresponds to multiple values obtained, and these values form a time sequence. The time sequence obtained by all the index data in an index set is referred to as a sample, and the comprehensive evaluation result obtained at the next time is referred to as the label of the sample.

[0098] In summary, in a time period (i.e., in the time period between the two adjacent times when the comprehensive evaluation result is obtained), each index set corresponds to a sample and a label.

[0099] In the time period T before the current time, all the samples and labels obtained by each index set form a data set. Each index set corresponds to a data set, and each data set corresponds to a project intervention model. After each sample in each data set is input into the project intervention model, the result output by the project intervention model is referred to as the intervention value of each index set.

[0100] In the training process, each time the training is performed, a sample is selected from each data set corresponding to all the index sets (the selected samples are all in the same time period, and the corresponding labels are the same). These samples are input into the respective project intervention models, and the intervention values output by the respective project intervention models are obtained. The loss function S used in the training is:

[0101]

[0102] where w i represents the participation weight of the i-th index set, F i represents the intervention value of the i-th index set, F0 represents the label corresponding to the sample input into the project intervention model, and K represents the number of index sets.

[0103] In the training process, the parameter updating method of the project intervention model of the embodiment is the stochastic gradient descent algorithm. With the progress of the training process, all samples in the data set corresponding to all indicator sets are input into the project intervention model in turn, and the parameters of all project intervention models are continuously updated through the above loss function, so that the loss function S gradually decreases until the training of all project intervention models is completed.

[0104] It should be noted that, given the data set and the loss function S, training the project intervention model is a known technique, and the embodiment will not repeat the specific training details.

[0105] For the intervention values output by all trained project intervention models, the intervention values are weighted and summed using the participation weight of each indicator set, which equals or approaches the comprehensive evaluation result of the innovative project, indicating that all trained project intervention models can evaluate the innovative project (for evaluating whether the innovative project achieves the expected purpose, such as whether it improves satisfaction, power grid safety and stability, whether it improves economic benefits, overcomes technical barriers, and whether it produces intellectual property rights, etc.).

[0106] For each project intervention model, the project intervention model can evaluate the innovative project from certain dimensions or aspects, such as the indicator data of load fluctuation and rapid recovery capability in the indicator set. The project intervention model trained by the indicator set can evaluate the innovative project from two aspects of balanced adjustment of power grid load and steady-state recovery control.

[0107] Step S004, the indicator set whose intervention value output by the project intervention model is less than the intervention threshold is recorded as the intervention set, and the project is artificially controlled according to the intervention set.

[0108] After the above comprehensive evaluation result of the innovative project is obtained, the project intervention model corresponding to each indicator set can be trained.

[0109] After the current comprehensive evaluation result of the innovative project is obtained, all indicator data (i.e. one sample described in step S003) in each indicator set are input into the trained project intervention model within a period of time from the last time the comprehensive evaluation result is obtained to the current time, the project intervention model outputs the intervention value of each indicator set, and an intervention threshold is set for each indicator set. The indicator set whose intervention value is less than the intervention threshold is recorded as the intervention set, and the intervention value of the intervention set is smaller, indicating that the evaluation result cannot be obtained from the dimensions corresponding to the indicator data in the intervention set, and further innovation optimization and artificial intervention control of the innovative project are needed from these dimensions.

[0110] For example, the intervention value corresponding to the index set of load fluctuation and rapid recovery capability is less than the intervention threshold (e.g., less than the intervention threshold due to an increase in users or an increase in power consumption of factory enterprise users), indicating that the power grid cannot obtain a high evaluation result in terms of load fluctuation suppression and rapid recovery of power grid stability (e.g., cannot obtain a high satisfaction or cannot ensure the safe and stable power supply of the power grid), and thus the innovation project needs to be artificially regulated from the two aspects or two dimensions of load fluctuation and rapid recovery capability, such as optimizing the power dispatching scheme, changing the power grid structure, and investing more technical personnel.

[0111] Step S005: The participation weight of the intervention threshold is updated according to the index set after regulation, and the project intervention model is trained again and the project is artificially regulated again.

[0112] After the innovation project is regulated, the index data is continuously collected (the index data is also continuously collected during regulation), and then the comprehensive evaluation result of the innovation project is obtained again. After the comprehensive evaluation result is obtained again, a plurality of index sets are obtained again according to step S002, denoted as modified index sets, and the evaluation participation degree and the participation weight of each modified index set are obtained.

[0113] The index set obtained after the last comprehensive evaluation result is obtained is denoted as a reference index set. All reference index sets are matched with all modified index sets. The matching method is that the intersection union ratio of the reference index set and the modified index set is denoted as the matching degree between the index sets, and the intersection union ratio refers to the ratio of the number of index data in the intersection set to the number of index data in the union set.

[0114] According to the matching degree between the index sets, all reference index sets and all modified index sets are matched one by one using the KM algorithm in this embodiment. The matched index sets have the largest matching degree. The difference between the evaluation participation degree of the modified index set and the evaluation participation degree of the matched reference index set is denoted as the first difference of the modified index set. The first differences of all modified index sets are linearly normalized. The mean value of the first differences of all modified index sets after linear normalization is denoted as B1.

[0115] The difference between the current comprehensive evaluation result and the last comprehensive evaluation result is denoted as a second difference, the maximum value between the current comprehensive evaluation result and the last comprehensive evaluation result is denoted as a first maximum value, and the ratio of the second difference to the first maximum value is denoted as B2. The product of B1 and B2 is denoted as the regulation effect. When B2 is less than 0, B2 is set to 0.

[0116] The smaller the regulation effect is, the more obvious the growth trend of the comprehensive evaluation result is not, and the correlation between the index data in the index set and the comprehensive evaluation result does not increase significantly. On the one hand, it shows that the innovation project after regulation does not further achieve the expected goal (such as improving satisfaction, power grid safety and stability, and improving economic benefit, overcoming technical barriers, and outputting intellectual property, etc.). On the other hand, it shows that the innovation project is not reasonably evaluated from the correct dimension, which leads to the inability to further innovate and improve the pain points of the innovation project, and further leads to the loss of the original correlation of the index data (or the index set) monitored by the innovation project after regulation (or the inability to further enhance the correlation between the index data and the comprehensive evaluation result), which further leads to the inability to accurately obtain the participation weight of each index set after obtaining the previous comprehensive evaluation result, and further cannot accurately train the project intervention model in the next step, which will affect the next regulation and all subsequent regulation processes.

[0117] Therefore, before the next regulation, the embodiment needs to avoid a series of problems caused by the small regulation effect.

[0118] For all reference index sets, the evaluation participation degree of the reference index set adjusted (that is, the intervention set obtained in step S004) and the modified index set matched with the intervention set is denoted as C1, the evaluation participation degree of the intervention set is denoted as C2, and the ratio of (C2-C1) to max(C2,C1) is denoted as w2. max() represents the maximum value function.

[0119] The intervention threshold of the modified index set matched with the intervention set is (1-w2×q)×T1, and T1 represents the intervention threshold corresponding to the intervention set.

[0120] For other modified index sets other than the modified index set matched with the intervention set, the intervention threshold corresponding to the other modified index set is equal to the intervention threshold of the reference index set matched with the other modified index set.

[0121] The larger w2 is, the more significantly the correlation between the index sets after adjustment is improved. At this time, the intervention threshold of the matched index set of the target set is reduced, so that the index set can be adjusted when the index set has a smaller intervention value, so that the innovation project can be regulated or innovated and optimized from the dimension with clear correlation or the aspect with clear mutual influence, and the innovation project can achieve the expected purpose in time and effectively.

[0122] The smaller w2 is, the more significantly the correlation of the index set after adjustment decreases. In this case, the intervention threshold of the matching index set of the target set is increased, so that the index set is adjusted when it has a larger intervention value, thereby avoiding the situation that irrelevant dimensions cannot effectively and timely achieve the expected purpose (e.g., cannot timely improve satisfaction or improve power grid safety and stability and economic benefits) when participating in adjustment and re-innovation.

[0123] q is a regulation effect mask factor. In this embodiment, when the regulation effect is greater than a preset first threshold th1, q = 0; and when the regulation effect is less than or equal to th1, q = 1.

[0124] In this embodiment, th1 = 0.4 is taken as an example for description, and th1 can be set to other values in other embodiments, which are not specifically limited in this embodiment.

[0125] In other embodiments, q = exp(-x), where x represents the regulation effect, and the smaller the regulation effect is, the greater the adjustment amplitude of T1 is, and exp() represents an exponential function with a natural constant as the base.

[0126] Further, for the reference index set that is not adjusted (i.e., the reference index set whose intervention value is greater than or equal to the intervention threshold, or the reference index set outside the intervention set), denoted as a first set, the evaluation participation degree D1 of the first set is obtained, the evaluation participation degree of the modified index set matched with the first set is denoted as D2, and |D1-D2| and max(D1, D2) are denoted as the weight correction factor w3 of the modified index set matched with the first set.

[0127] The first weight of the modified index set matched with the first set is denoted as (1+w3×q)×P, and P represents the participation weight of the modified index set matched with the first set.

[0128] For other modified index sets other than the modified index set matched with the first set, the first weight of the other modified index set is equal to the participation weight of the other modified index set.

[0129] The first weights of all modified index sets are normalized, and the normalized result is denoted as the modified participation weight of the modified index set.

[0130] The index set (i.e., the first set) that is not adjusted in the above process changes significantly after the innovation project regulation, which indicates that the project intervention model training corresponding to the first set is inaccurate. In this case, the participation weight of the index set (i.e., the modified index set) obtained after regulation is corrected to obtain a modified participation weight, so that the subsequently trained project intervention model training is more accurate.

[0131] At this point, the above-mentioned intervention threshold is corrected and the participation weight is corrected to avoid a series of subsequent problems caused by too small control effect.

[0132] Further, the project intervention model of each indicator set is retrained by using the corrected participation weight, the intervention set is reacquired by using the project intervention model, and the innovative project is regulated again by using the intervention set. In this way, the innovative project is continuously regulated.

[0133] In the above-mentioned embodiment, with the operation of the innovative project, the innovative project is continuously evaluated (i.e., the comprehensive evaluation result is obtained), the participation weight of the indicator set is dynamically and adaptively corrected, so that the project intervention model is more accurate and reliable, and the intervention threshold is dynamically corrected, so that the innovative project can be reasonably and accurately controlled and innovated again according to the project intervention model. The correction process of the participation weight and the intervention threshold is based on the last regulation of the innovative project. With the operation of the innovative project, the number of users used by the innovative project gradually increases, and the user demand changes, so that the innovative project can achieve the expected purpose (e.g., improving satisfaction, power grid safety and stability, and improving economic benefits, overcoming technical barriers, and outputting intellectual property rights).

[0134] It should be noted that at the initial stage of the operation of the innovative project (e.g., the first half year of the operation of the innovative project), more indicator data cannot be collected, so that a project intervention model with high precision cannot be trained. At this time, when the innovative project is regulated according to the above-mentioned embodiment, expert verification is required to avoid errors in the regulation of the innovative project, causing economic losses or safety accidents.

[0135] In addition, the method for obtaining the initial value of the intervention threshold in the embodiment includes:

[0136] After obtaining the comprehensive evaluation result for the first time, the comprehensive evaluation result is denoted as Q, the participation weight of each indicator set is obtained, and Q is proportionally distributed by using the participation weight to obtain the initial evaluation of each indicator set. The initial evaluation satisfies: the initial evaluation of all indicator sets is weighted and summed by using the participation weight of each indicator set, and the weighted summation result is equal to Q. The th2x100% of the initial evaluation of each indicator set is taken as the initial value of the intervention threshold of each indicator set.

[0137] The embodiment takes th2=60 as an example for description, and other embodiments can be set to other values, and the embodiment is not specifically limited.

[0138] The above-mentioned is only a preferred embodiment of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for constructing an adaptive adjustment model of innovation project indicators based on multi-dimensional evaluation, characterized in that, The method comprises the following steps: Obtaining multiple index data of the project and a comprehensive evaluation result, wherein the multiple index data comprises load fluctuation of a power grid, fault early warning capability, rapid recovery capability under fault condition, clean energy access proportion, renewable energy access proportion, and input amount of energy storage equipment in the innovation project; Dividing all the index data into a plurality of index sets, training a project intervention model of each index set by using a participation weight of the index set, wherein the participation weight is positively correlated with an evaluation participation degree of each index set, the evaluation participation degree is obtained from a correlation between index data in each index set and the comprehensive evaluation result, an index set with an intervention value output by the project intervention model being less than an intervention threshold is recorded as an intervention set, and the project is artificially regulated according to the intervention set; the index set obtained before and after the project regulation is recorded as a reference index set and a modified index set respectively, and a regulation effect is obtained from a change in the evaluation participation degree between the modified index set and the reference index set; the intervention threshold is updated according to a difference in the evaluation participation degree between the intervention set and the modified index set and the regulation effect, and the participation weight of the modified index set is updated according to a difference in the evaluation participation degree between the reference index set and the modified index set and the regulation effect except for the intervention set; the project intervention model of each modified index set is trained by using the updated participation weight, the modified index set with an intervention value output by the project intervention model being less than an updated intervention threshold is recorded as the intervention set again, and the project is artificially regulated according to the intervention set obtained again; The regulation effect obtained from the change in the evaluation participation degree between the modified index set and the reference index set comprises the following specific steps: All the reference index sets and all the modified index sets are matched one by one, a difference between the evaluation participation degree of the modified index set and the evaluation participation degree of the matched reference index set is obtained, and the difference is recorded as a first difference of the modified index set; linear normalization processing is performed on the first difference of all the modified index sets, and a mean value of the first difference of all the modified index sets after the linear normalization processing is recorded as B1; A difference between a current comprehensive evaluation result and a previous comprehensive evaluation result is recorded as a second difference, the regulation effect is positively correlated with B1, and the regulation effect is also positively correlated with the second difference. 2.The method of claim 1, wherein, The division of all the index data into a plurality of index sets comprises the following specific steps: In a time period T before the current time, time series data constituted by all the comprehensive evaluation results is obtained and recorded as first time series data; time series data constituted by all the values of any index data is recorded as second time series data; an absolute value of a Pearson correlation coefficient between the second time series data and the first time series data is recorded as an evaluation result correlation of any index data; The correlation degree between the first time series data and the second time series data corresponding to any two index data is obtained, and all the index data are divided into a plurality of index sets according to the evaluation result correlation and the correlation degree; The time period T is a preset time period. 3.The method of claim 2, wherein, According to the evaluation result relevance and the correlation degree, all the index data are divided into several index sets, and the specific steps include the following: The product of the correlation degree and the average of the evaluation result relevance of any two index data is recorded as the metric distance of any two index data, and the metric distance of any two index data is used to perform mean K-Means clustering on all the index data to obtain several categories, and all the index data in each category are recorded as an index set. 4.The method of claim 1, wherein, The specific steps of training the project intervention model of each index set by using the participation weight of the index set include the following: In the time period between adjacent two times of obtaining the comprehensive evaluation result, the values of all the index data in each index set form a sample, and the comprehensive evaluation result obtained at the later time is recorded as the label of the sample; In the time period T before the current time, all the samples and labels obtained by each index set form a data set; Each data set corresponding to each index set is used to train a project intervention model, and the output result of each project intervention model is recorded as the intervention value of each index set; and the loss function used when training all the project intervention models is: wherein, represents a loss function, represents a participation weight of the i-th indicator set, represents an intervention value of the i-th indicator set, represents a label corresponding to a sample input into the intervention model, K represents the number of indicator sets; and the time period T is a preset time period. 5.The method of claim 1, wherein, The specific steps of updating the intervention threshold according to the difference between the evaluation participation degree of the intervention set and the correction index set and the regulation effect include the following: All the reference index sets include the intervention set, the evaluation participation degree of the correction index set matched with the intervention set is recorded as C1, the evaluation participation degree of the intervention set is recorded as C2, and the ratio of (C2-C1) to max(C2,C1) is recorded as w2; max() represents the maximum value function; The intervention threshold of the correction index set matched with the intervention set is (1-w2×q)×T1, T1 represents the intervention threshold corresponding to the intervention set; and q is a regulation effect mask factor, which is obtained from the regulation effect; For other correction index sets except the correction index set matched with the intervention set, the intervention threshold corresponding to the other correction index set is equal to the intervention threshold of the reference index set matched with the other correction index set. 6.The method of claim 1, wherein, The specific steps of updating the participation weight of the correction index set according to the difference between the evaluation participation degree of the reference index set except the intervention set and the correction index set and the regulation effect include the following: All the reference index sets include the intervention set, and the reference index set except the intervention set is recorded as a first set, The evaluation participation degree of the first set is recorded as D1, the evaluation participation degree of the correction index set matched with the first set is recorded as D2, and |D1-D2| and max(D1,D2) are recorded as the weight correction factor w3 of the correction index set matched with the first set; The first weight of the correction index set matched with the first set is recorded as (1+w3×q)×P, and P represents the participation weight of the correction index set matched with the first set; For other correction index sets except the correction index set matched with the first set, the first weight of the other correction index set is equal to the participation weight of the other correction index set. The first weight of all the modified indicator sets is normalized, and the normalized result is recorded as a modified participation weight of the modified indicator set; q is a regulation effect mask factor, obtained from the regulation effect; max() represents a maximum value function.

7. The method according to claim 5 or 6, wherein, The specific acquisition steps of the regulation effect mask factor are as follows: When the regulation effect is greater than a preset first threshold, q=0; when the regulation effect is less than or equal to the preset first threshold, q=1. 8.The method of claim 2, wherein, The specific steps of obtaining the evaluation participation degree from the correlation between the indicator data in each indicator set and the comprehensive evaluation result are as follows: For any one indicator set, the second time series data corresponding to all the indicator data in the indicator set and the first time series data are subjected to PCA dimension reduction together, reduced to one dimension, and the variance of all the reduced results is obtained, recorded as the evaluation participation degree of each indicator set; The method for obtaining the participation weight of each indicator set comprises: The evaluation participation degrees of all the indicator sets are normalized, and the normalized evaluation participation degree is recorded as the participation weight of each indicator set; The specific steps of obtaining the correlation degree between the second time series data corresponding to any two indicator data and the first time series data are as follows: The second time series data corresponding to any two indicator data and the first time series data are subjected to dimension reduction together by using the PCA algorithm, reduced to one dimension, and the variance of all the reduced results is obtained, recorded as the correlation degree between the first time series data and any two indicator data. 9.The method of claim 1, wherein, The specific steps of one-to-one matching between all the reference indicator sets and all the modified indicator sets are as follows: All the reference indicator sets and all the modified indicator sets are matched one-to-one by using the KM algorithm, and the matched indicator sets have the largest intersection union ratio.

Images