Enterprise innovation ability evaluation and prediction device and method based on neural network

Abstract

The invention provides an enterprise innovation capability evaluation and prediction device and method based on a neural network, and is based on a three-level index system for evaluating enterprise innovation capability. Determining an enterprise type i according to the category to which the to-be-evaluated enterprise belongs, and quantifying the second-level indexes of the to-be-evaluated enterprise through the KAN neural network and the normalized data of all the third-level indexes forming the second-level indexes; and determining the quantized value of the first-level index according to the proportion of the radar map area defined by the quantized values of all the second-level indexes of the enterprise to be evaluated relative to the maximum possible area to serve as the quantized value of the enterprise innovation ability, and evaluating the innovation ability of the enterprise to be evaluated according to the quantized value. Training a bidirectional long-short-term memory network model combined with an attention mechanism through the enterprise innovation ability of the to-be-evaluated enterprise over the years; and a Dropout method is combined to predict the future enterprise innovation capability of the to-be-evaluated enterprise, and a 95% confidence interval is constructed, so that the future innovation development trend of the to-be-evaluated enterprise is predicted.

Description

Technical Field

[0001] The present invention belongs to the technical fields of big data analysis, enterprise digital transformation applications, etc., and specifically relates to an enterprise innovation capability evaluation and prediction device and method based on a neural network. Background Art

[0002] Enterprise innovation is the core driving force for the development of modern enterprises, and is also the key for enterprises to maintain their competitive advantages and adapt to the ever-changing market environment. Most of the existing enterprise innovation capability assessment programs have limitations, mainly reflected in the single assessment indicators and limited data sources, which cannot fully and accurately reflect the real innovation capabilities of enterprises. In addition, many enterprises lack future predictions of their own innovation capabilities, resulting in a lack of foresight and scientificity in strategic planning and resource allocation, and a lack of scientific and objective guidance for the current development of enterprises and their own status.

[0003] At present, with the development of big data industry, artificial intelligence and machine learning industry, related products are emerging in an endless stream. However, in terms of enterprise innovation evaluation, there is no reliable model or technology that uses relevant technologies to develop and solve the above enterprise innovation evaluation and prediction. Summary of the invention

[0004] Therefore, the present invention evaluates and predicts the innovation capability of enterprises by introducing data mining technology and neural network technology, and improves the accuracy and comprehensiveness of the evaluation by mining potential patterns and relationships from a large amount of heterogeneous data. At the same time, the powerful predictive ability of neural networks can help enterprises predict future innovation trends, provide strong support for decision-making, and thus promote the sustainable and healthy development of enterprises.

[0005] In the solution provided by the present invention, the combination of KAN neural network, bidirectional long short-term memory network with attention mechanism and Dropout can help enterprises to more accurately and effectively evaluate the current enterprise innovation capabilities and make reasonable predictions on future innovation development trends.

[0006] The technical solution specifically adopted by the present invention to solve the technical problem is:

[0007] A device for evaluating and predicting enterprise innovation capability based on a neural network, based on a computer system, comprising:

[0008] A three-level indicator system for assessing the innovation capabilities of enterprises;

[0009] Enterprise category database D established according to enterprise categories i , used to store the normalized third-level indicator data of various categories of enterprises;

[0010] The secondary evaluation module is used to determine the enterprise type i according to the category to which the enterprise to be evaluated belongs, and quantify the secondary indicators of the enterprise to be evaluated through the KAN neural network and the data normalized by all the third-level indicators that constitute the secondary indicators;

[0011] The first-level evaluation module determines the quantitative value of the first-level indicator by the ratio of the radar chart area enclosed by the quantitative values ​​of all the second-level indicators of the enterprise to be evaluated to the maximum possible area, which is used as the quantitative value of the enterprise's innovation ability, and evaluates the innovation ability of the enterprise to be evaluated accordingly;

[0012] In addition, the prediction module combines the bidirectional long short-term memory network model with the attention mechanism through the corporate innovation capability training of the company to be evaluated over the years; and predicts the future corporate innovation capability of the company to be evaluated through the trained bidirectional long short-term memory network model and the Dropout method and constructs a 95% confidence interval, thereby predicting the future innovation development trend of the company to be evaluated.

[0013] Furthermore, the three-level indicator system consists of 1 first-level indicator, 5 second-level indicators and 15 third-level indicators; among them, the first-level indicator is the enterprise innovation capability I, the first-level indicator I is composed of 5 second-level indicators: management innovation I1, innovation foundation I2, innovation input I3, innovation output I4 and marketing innovation I5, and the second-level indicator I1 is composed of the total number of managers I 11 , the proportion of managers with a master's degree or above 12 Ratio of senior managers to 13 There are three third-level indicators in total, and the second-level indicator I2 is composed of the number of existing authorized patents I 21 、The number of existing software copyrights 22 and the number of registered trademarksI 23 There are three third-level indicators in total, and the second-level indicator I3 is composed of the R&D investment amount I 31 , R&D investment as a percentage of operating income 32 and the proportion of R&D personnel 33 There are 3 third-level indicators in total, and the second-level indicator I4 is composed of the annual patent proportion I 41 、Per capita patent shareI 42 and invention patents 43 There are three third-level indicators in total, and the second-level indicator I5 is composed of the ratio between input and output I 51 、Operating Profit I 52 and Net Profit I 53 It consists of 3 third-level indicators.

[0014] Furthermore, the specific method of the normalization process is: for the third-level index I 12 ,I 13 ,I 21 ,I 22 ,I 23 ,I31 ,I 32 ,I 33 ,I 41 ,I 42 ,I 43 ,I 52 and I 53 , the normalized value m=1,2,…,M; for the third-level index I 11 and I 51 Normalized value m=1,2,…,M, where x c It represents the original data of a third-level indicator of the enterprise to be evaluated before normalization, min(x m ) represents the minimum value of the third-level indicator in the enterprise category, max(x m ) represents the maximum value of the third-level indicator in the enterprise category, M represents the database D i The total number of enterprises in .

[0015] Furthermore, the specific method of quantifying the secondary indicators of the enterprise to be evaluated by using the KAN neural network and the data normalized by all the third-level indicators constituting the secondary indicators is as follows:

[0016] Database D i The j-th secondary index of the m-th enterprise is j=1,2,…,5, where k=1,2,3 represents database D i The kth normalized third-level indicator that constitutes the jth second-level indicator of the enterprise;

[0017] The neural network KAN consists of an input layer, a hidden layer and an output layer;

[0018] The process of calculating the secondary index I1 through the neural network KAN based on the Kolmogorov-Arnold theorem is as follows: At the input layer, the input vector received is m=1,2,…,M; In the hidden layer, first, for each input variable k=1,2,3 and index q, q=1,2,…,7, define the nonlinear transformation function through two fully connected layers and ReLU activation function in, and is the weight matrix and bias vector of the first fully connected layer, and is the weight matrix and bias vector of the second fully connected layer. Secondly, for each q, all The output of Finally, for each S qThe nonlinear transformation function is defined by two fully connected layers and the ReLU activation function. in, and is the weight matrix and bias vector of the first fully connected layer, and is the weight matrix and bias vector of the second fully connected layer; randomly initialize the KAN network parameters In the output layer, the quantitative value of the secondary index I1 is expressed as Among them, Sig(·) is the Sigmoid activation function, which is used to convert The value of is mapped to [0,1]; other secondary indicators use the same calculation process;

[0019] The loss function used by the KAN neural network is expressed as Where M represents the database D i The total number of enterprises in .

[0020] Furthermore, the method of training the KAN neural network is to iteratively adjust the parameters based on the back propagation algorithm. When the set number of iterations is reached, the network training ends and the network parameters at this time are the network parameters that minimize the loss function Loss1.

[0021] Furthermore, the specific method for determining the quantitative value of the primary indicator by the ratio of the radar chart area enclosed by the quantitative values ​​of all secondary indicators of the enterprise to be evaluated to the maximum possible area is as follows: Database D i The first-level index of the mth enterprise is in, j=1,2,…,5 represents database D i The quantitative value of the jth secondary indicator of the enterprise in the radar chart is in, The maximum possible area of ​​a radar chart is The quantitative value of the enterprise's first-level indicator is

[0022]

[0023] Furthermore, the specific method of combining the bidirectional long short-term memory network model with the attention mechanism through the enterprise innovation ability training of the enterprise to be evaluated over the years is as follows:

[0024] The quantitative values ​​of the enterprise innovation capabilities of the enterprises to be evaluated are sorted from small to large according to the year, and the data set data = {v t1 ,v t2 ,…,v tn}, where v tn It represents the quantitative value of the innovation capability of the enterprise to be evaluated in the year tn;

[0025] Construct a Bi-LSTM+Attention model consisting of five layers: input layer, Bi-LSTM layer, attention layer, Dropout layer, and output layer. In the input layer, the time series data received is n=1,2,…,N-5, where N is the total number of data points contained in the data set data. are the inputs of time steps 1 to 5 respectively; in the Bi-LSTM layer, the hidden state of the forward LSTM is represented as where v (p) is the input of the pth time step, and are the hidden state and cell state of the forward LSTM at the (p-1)th time step, respectively. The hidden state of the backward LSTM is expressed as where v (p) is the input of the pth time step, and are the hidden state and cell state of the backward LSTM at the (p+1)th time step, respectively. The forward hidden state and the backward hidden state are concatenated into a bidirectional hidden state vector H p , In the attention layer, the attention score at each time step p is denoted as e p =tanh(W p H p +b p ), where W p and b p is the weight matrix and bias vector of the attention layer at the pth time step, tanh(·) is the hyperbolic tangent activation function, and the attention weight is expressed as The context vector is represented as In the Dropout layer, 10% of the neurons from the attention layer are randomly selected and discarded, expressed as O'=Dropout(O,0.1), where 0.1 is the proportion of neurons discarded; in the output layer, the output of the model is expressed as V tn+5 =W out O'+b out , where V tn+5 To use time series data n Compared with the predicted value of enterprise innovation capability obtained by Bi-LSTM+Attention model in year (tn+5), W out and b out is the weight matrix and bias vector of the output layer;

[0026] The loss function used by the Bi-LSTM+Attention model is expressed as Among them, v tn+5 is the true value of the enterprise innovation capability in year (tn+5), N is the total number of data points contained in the data set data;

[0027] The Bi-LSTM+Attention model is trained by iteratively updating the Bi-LSTM+Attention model parameters based on the back-propagation algorithm. When the set number of iterations is reached, the network training ends, and the network parameters at this time are the network parameters that minimize the loss function Loss2.

[0028] Furthermore, the specific method of predicting the future enterprise innovation capability of the enterprise to be evaluated and constructing a 95% confidence interval by using the trained bidirectional long short-term memory network model and the Dropout method is as follows: Using the trained Bi-LSTM+Attention model and time series data n=N-4When performing time series prediction, enable Dropout and perform T random forward propagations to obtain a series of predicted values n=N-4; the mean of T prediction values ​​is expressed as The standard deviation of the T-time prediction value is expressed as The confidence interval is expressed as The mean μ of T predicted values ​​is used as the quantitative value of the enterprise innovation capability of the enterprise to be evaluated in the next year.

[0029] And, a method for evaluating and predicting enterprise innovation capability based on a neural network, based on the above device, after collecting all the three-level indicator data of various types of enterprises and normalizing them, inputting and improving the enterprise category database, the secondary evaluation module and the primary evaluation module are called successively, firstly, the enterprise type i is determined according to the category to which the enterprise to be evaluated belongs, and the quantitative value of the secondary indicator of the enterprise to be evaluated is calculated through the KAN neural network and the normalized data of all the three-level indicators that constitute the secondary indicator; then the quantitative value of the primary indicator is determined by the proportion of the radar chart area enclosed by the quantitative values ​​of all the secondary indicators of the enterprise to be evaluated relative to the maximum possible area, and the innovation capability of the enterprise is evaluated accordingly.

[0030] Furthermore, based on the enterprise category database, the prediction module is called, and first, the enterprise innovation ability of the enterprise to be evaluated over the years is trained in combination with a bidirectional long short-term memory network model with an attention mechanism; then, the future enterprise innovation ability of the enterprise to be evaluated is predicted through the trained bidirectional long short-term memory network model and the Dropout method, and a 95% confidence interval is constructed, thereby predicting the future innovation development trend of the enterprise to be evaluated.

[0031] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the enterprise innovation capability evaluation and prediction method based on a bidirectional long short-term memory network are implemented as described above.

[0032] A non-transitory computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the enterprise innovation capability evaluation and prediction method based on a neural network as described above.

[0033] Compared with the prior art, the present invention and its preferred embodiments have at least the following advantages:

[0034] 1. By constructing a three-level indicator system, the third-level indicators are used to calculate the second-level indicators, and then the second-level indicators are used to calculate the first-level indicators (i.e., enterprise innovation capabilities). The hierarchical indicator system improves the accuracy and reliability of the evaluation, facilitates in-depth analysis of specific factors that affect the innovation capabilities of enterprises, and helps enterprises accurately identify improvement directions. All three-level indicators are quantifiable indicators with specific values, which avoids subjective judgments, can reflect more objective and real situations, help eliminate human bias, and ensure the fairness and consistency of the evaluation results. In addition, the three-level indicator system also has good flexibility and scalability, is suitable for different types of enterprises, improves the comparability and practical value of the evaluation results, and fills the gaps in existing technologies in the quantitative evaluation of enterprise innovation capabilities.

[0035] 2. Introduce neural network technology in the field of artificial intelligence, especially the bidirectional long short-term memory network and Dropout method combined with the attention mechanism, into the field of enterprise management. Through in-depth analysis of the innovation ability scores of the evaluated enterprises over the years, accurate prediction of the future enterprise innovation ability is achieved, with a value interval of [0,1]. The present invention constructs a prediction model with a 95% confidence interval, clarifies the upper and lower limits of the prediction results, and makes the prediction results more robust and credible. The construction of this confidence interval effectively reduces the uncertainty in the prediction process, helping enterprise decision makers to better understand the possible fluctuation range of innovation ability, so as to make more scientific and reasonable decisions in strategic planning and resource allocation. By providing accurate innovation ability predictions and reliable confidence intervals, it helps enterprises to formulate more forward-looking and competitive innovation strategies, which is of great significance to promoting the development of enterprise management technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0037] Figure 1 It is a flowchart for constructing and implementing the solution of an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the features and advantages of the present invention more clearly understood, the following embodiments are specifically described in detail as follows:

[0039] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0041] like Figure 1 As shown, this embodiment introduces in detail the design and construction process of the proposed enterprise innovation capability evaluation and prediction device based on neural network and the corresponding method, including the following steps:

[0042] 1) According to the enterprise category, enterprises are divided into 20 types, such as construction enterprises, mining enterprises, manufacturing enterprises, etc., and a database is established for each enterprise category. i ,i=1,2,…,20.

[0043] 2) Construct a three-level indicator system for evaluating the innovation capabilities of enterprises.

[0044] In this embodiment, the three-level indicator system consists of 1 first-level indicator, 5 second-level indicators and 15 third-level indicators. Among them, the first-level indicator is the enterprise innovation capability (I), the first-level indicator I is composed of 5 second-level indicators: management innovation (I1), innovation foundation (I2), innovation input (I3), innovation output (I4) and marketing innovation (I5), and the second-level indicator I1 is composed of the total number of managers (I 11 ), the proportion of managers with a master's degree or above (I 12 ) and the proportion of senior managers (I 13 ) consists of three third-level indicators, and the second-level indicator I2 consists of the number of existing authorized patents (I 21 ), the number of existing software copies (I 22 ) and the number of registered trademarks (I 23 ) consists of three third-level indicators, and the second-level indicator I3 consists of the amount of R&D investment (I 31 ), R&D investment as a percentage of operating income (I 32 ) and the proportion of R&D personnel (I 33)3 third-level indicators, the second-level indicator I4 consists of the annual patent proportion (I 41 ), patent share per capita (I 42 ) and the proportion of invention patents (I 43 )3 third-level indicators, and the second-level indicator I5 consists of the ratio between input and output (I 51 )、Operating Profit(I 52 ) and net profit (I 53 ) consists of three third-level indicators.

[0045] 3) Collect and normalize all the three-level indicator data of various types of enterprises, input and improve the above 20 databases D i ,i=1,2,…,20.

[0046] In this embodiment, for the third-level indicator I 12 ,I 13 ,I 21 ,I 22 ,I 23 ,I 31 ,I 32 ,I 33 ,I 41 ,I 42 ,I 43 ,I 52 and I 53 These positive indicators, the normalized values m=1,2,…,M; for the third-level index I 11 and I 51 These negative indicators, the normalized values m=1,2,…,M, where x c It represents the original data of a third-level indicator of the enterprise to be evaluated before normalization, min(x m ) represents the minimum value of the third-level indicator in the enterprise category, max(x m ) represents the maximum value of the third-level indicator in the enterprise category, M represents the database D i The total number of enterprises in .

[0047] 4) Determine the enterprise type i according to the category to which the enterprise to be evaluated belongs, and quantify the secondary indicators of the enterprise to be evaluated through the KAN neural network and the normalized data of all the third-level indicators that constitute the secondary indicators.

[0048] In this embodiment, the secondary indicators of the enterprise to be evaluated are quantified by the KAN neural network and the data normalized by all the third-level indicators constituting the secondary indicators. The specific method is as follows:

[0049] Database D i The j-th secondary index of the m-th enterprise is j=1,2,…,5, where k=1,2,3 represents database D i The kth normalized third-level indicator that constitutes the jth second-level indicator of the enterprise;

[0050] The neural network KAN consists of an input layer, a hidden layer and an output layer;

[0051] The process of constructing a neural network KAN based on the Kolmogorov-Arnold theorem to calculate the secondary index I1 is as follows: In the input layer, the input vector received is m=1,2,…,M; In the hidden layer, first, for each input variable k=1,2,3 and index q(q=1,2,…,7), define the nonlinear transformation function through two fully connected layers and ReLU activation function in, and is the weight matrix and bias vector of the first fully connected layer, and is the weight matrix and bias vector of the second fully connected layer. Secondly, for each q, all The output of Finally, for each S q The nonlinear transformation function is defined by two fully connected layers and the ReLU activation function. in, and is the weight matrix and bias vector of the first fully connected layer, and is the weight matrix and bias vector of the second fully connected layer; randomly initialize the KAN network parameters In the output layer, the quantitative value of the secondary index I1 is expressed as Among them, Sig(·) is the Sigmoid activation function, which is used to convert The value of is mapped to [0,1]; other secondary indicators use the same calculation process;

[0052] The loss function used by the KAN neural network is expressed as Where M represents the database D i The total number of enterprises in

[0053] Training the KAN neural network is based on the back propagation algorithm by iteratively adjusting the parameters When the set number of iterations is reached, the network training ends and the network parameters at this time are the network parameters that minimize the loss function Loss1.

[0054] 5) The quantitative value of the primary indicator is determined by the ratio of the area of ​​the radar chart enclosed by the quantitative values ​​of all secondary indicators of the enterprise to be evaluated to the maximum possible area, which is used as the quantitative value of the enterprise's innovation capability, and the innovation capability of the enterprise to be evaluated is evaluated accordingly.

[0055] In this embodiment, the specific method for determining the quantitative value of the primary indicator by the ratio of the area of ​​the radar chart enclosed by the quantitative values ​​of all secondary indicators of the enterprise to be evaluated to the maximum possible area is as follows:

[0056] Database D i The first-level index of the mth enterprise is in, j=1,2,…,5 represents database D i The quantitative value of the jth secondary indicator of the enterprise in the radar chart is in, The maximum possible area of ​​a radar chart is The quantitative value of the enterprise's first-level indicator is

[0057] 6) A bidirectional long short-term memory network model is built by combining the corporate innovation capability training of the company to be evaluated over the years with the attention mechanism.

[0058] In this embodiment, the Bi-LSTM+Attention model is trained based on the enterprise innovation capabilities of the enterprise to be evaluated over the years. The specific method is as follows:

[0059] The quantitative values ​​of the enterprise innovation capabilities of the enterprises to be evaluated are sorted from small to large according to the year, and the data set data = {v t1 ,v t2 ,…,v tn}, where v tn It represents the quantitative value of the innovation capability of the enterprise to be evaluated in the year tn;

[0060] Construct a Bi-LSTM+Attention model consisting of five layers: input layer, Bi-LSTM layer, attention layer, Dropout layer, and output layer. In the input layer, the time series data received is n=1,2,…,N-5, where N is the total number of data points contained in the data set data. are the inputs of time steps 1 to 5 respectively; in the Bi-LSTM layer, the hidden state of the forward LSTM (from time steps 1 to 5) is represented as where v (p) is the input of the pth time step, and are the hidden state and cell state of the forward LSTM at the (p-1)th time step, and the hidden state of the backward LSTM (from time step 5 to 1) is expressed as where v (p) is the input of the pth time step, and are the hidden state and cell state of the backward LSTM at the (p+1)th time step, respectively. The forward hidden state and the backward hidden state are concatenated into a bidirectional hidden state vector H p , In the attention layer, the attention score at each time step p is denoted as e p =tanh(W p H p +b p ), where W p and b p is the weight matrix and bias vector of the attention layer at the pth time step, tanh(·) is the hyperbolic tangent activation function, and the attention weight is expressed as The context vector is represented as In the Dropout layer, 10% of the neurons from the attention layer will be randomly selected and discarded. This process can be expressed as O'=Dropout(O,0.1), where 0.1 is the proportion of neurons discarded. In the output layer, the output of the model is expressed as V tn+5 =W out O'+b out , where V tn+5 To use time series data n Compared with the predicted value of enterprise innovation capability obtained by Bi-LSTM+Attention model in year (tn+5), W out and b out is the weight matrix and bias vector of the output layer;

[0061] The loss function used by the Bi-LSTM+Attention model is expressed as Among them, v tn+5 is the true value of the enterprise innovation capability in year (tn+5), N is the total number of data points contained in the data set data;

[0062] Training the Bi-LSTM+Attention model means updating the Bi-LSTM+Attention model parameters through iteration based on the back propagation algorithm. When the set number of iterations is reached, the network training ends. The network parameters at this time are the network parameters that minimize the loss function Loss2.

[0063] 7) The future innovation capability of the enterprise to be evaluated is predicted through the trained Bi-LSTM+Attention model and Dropout method, and a 95% confidence interval is constructed to predict the future innovation development trend of the enterprise to be evaluated.

[0064] In this embodiment, the future enterprise innovation capability of the enterprise to be evaluated is predicted by the trained Bi-LSTM+Attention model and the Dropout method, and a 95% confidence interval is constructed. The specific method is as follows:

[0065] Use the trained Bi-LSTM+Attention model and time series data n=N-4When performing time series prediction, enable Dropout and perform T random forward propagations to obtain a series of predicted values n=N-4; the mean of T prediction values ​​is expressed as The standard deviation of the T-time prediction value is expressed as The confidence interval is expressed as The mean μ of T predicted values ​​is taken as the enterprise innovation capability of the enterprise to be evaluated in the next year.

[0066] Through the above design, the enterprise innovation capability evaluation and prediction device and method based on neural network of the present invention are constructed, which can realize the evaluation and prediction of enterprise innovation capability. The constructed three-level indicator system not only improves the accuracy and reliability of the evaluation, but also facilitates the in-depth analysis of the specific factors affecting the enterprise innovation capability, which helps the enterprise to accurately locate the direction of improvement; all three-level indicators are quantifiable indicators with specific values, which can reflect a more objective and real situation, help eliminate human bias, and ensure the fairness and consistency of the evaluation results; and the three-level indicator system has good flexibility and scalability, which is suitable for different types of enterprises, and improves the comparability and practical value of the evaluation results; the algorithm analyzes the innovation capability scores of enterprises over the years and realizes the accurate prediction of future innovation capability, with a value range of [0,1]. By constructing a prediction model with a 95% confidence interval, the upper and lower limits of the prediction results are clarified, and the robustness and credibility of the prediction are improved. The confidence interval reduces the uncertainty in the prediction, helps enterprise decision makers make more scientific judgments in strategic planning and resource allocation, and is of great significance to promoting the development of enterprise management technology.

[0067] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0068] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0069] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0070] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0071] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0072] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

[0073] The present invention is not limited to the above-mentioned optimal implementation mode. Anyone can derive other various forms of enterprise innovation capability assessment and prediction devices and methods based on neural networks under the inspiration of the present invention. All equivalent changes and modifications made according to the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. A device for evaluating and predicting enterprise innovation capability based on neural network, characterized in that: Computer-based systems, including: A three-level indicator system for assessing the innovation capabilities of enterprises; Establish enterprise category database according to enterprise categories i , used to store the normalized third-level indicator data of various categories of enterprises; The secondary evaluation module is used to determine the enterprise type i according to the category to which the enterprise to be evaluated belongs, and quantify the secondary indicators of the enterprise to be evaluated through the KAN neural network and the data normalized by all the third-level indicators that constitute the secondary indicators; The first-level evaluation module determines the quantitative value of the first-level indicator by the ratio of the radar chart area enclosed by the quantitative values ​​of all the second-level indicators of the enterprise to be evaluated to the maximum possible area, which is used as the quantitative value of the enterprise's innovation ability, and evaluates the innovation ability of the enterprise to be evaluated accordingly; In addition, the prediction module combines the bidirectional long short-term memory network model with the attention mechanism through the corporate innovation capability training of the company to be evaluated over the years; and predicts the future corporate innovation capability of the company to be evaluated through the trained bidirectional long short-term memory network model and the Dropout method and constructs a 95% confidence interval, thereby predicting the future innovation development trend of the company to be evaluated.

2. The enterprise innovation capability evaluation and prediction device based on neural network according to claim 1 is characterized by: The three-level indicator system consists of 1 first-level indicator, 5 second-level indicators and 15 third-level indicators; among them, the first-level indicator is the enterprise innovation capability I, the first-level indicator I is composed of 5 second-level indicators: management innovation I1, innovation foundation I2, innovation input I3, innovation output I4 and marketing innovation I5, and the second-level indicator I1 is composed of the total number of managers I 11 , the proportion of managers with a master's degree or above 12 Ratio of senior managers to 13 There are three third-level indicators in total. The second-level indicator I2 is composed of the number of existing authorized patents I 21 、The number of existing software copyrights 22 and the number of registered trademarksI 23 There are 3 third-level indicators in total, and the second-level indicator I3 is composed of the R&D investment amount I 31 , R&D investment as a percentage of operating income 32 and the proportion of R&D personnel 33 There are 3 third-level indicators in total, and the second-level indicator I4 is composed of the annual patent proportion I 41 、Per capita patent shareI 42 and invention patents 43 There are three third-level indicators in total, and the second-level indicator I5 is composed of the ratio between input and output I 51 、Operating Profit I 52 and Net Profit I 53 It consists of 3 third-level indicators.

3. The enterprise innovation capability evaluation and prediction device based on neural network according to claim 2 is characterized by: The specific method of the normalization process is: for the three-level index I 12 ,I 13 ,I 21 ,I 22 ,I 23 ,I 31 ,I 32 ,I 33 ,I 41 ,I 42 ,I 43 ,I 52 and I 53 , the normalized value For level 3 indicator I 11 and I 51 Normalized value Among them, x c It represents the original data of a third-level indicator of the enterprise to be evaluated before normalization, min(x m ) represents the minimum value of the third-level indicator in the enterprise category, max(x m ) represents the maximum value of the third-level indicator in the enterprise category, M represents the database D i The total number of enterprises in .

4. The enterprise innovation capability evaluation and prediction device based on neural network according to claim 2 is characterized by: The specific method of quantifying the secondary indicators of the enterprise to be evaluated by using the KAN neural network and the data normalized by all the third-level indicators constituting the secondary indicators is as follows: Database D i The j-th secondary index of the m-th enterprise is in, Represents database D i The kth normalized third-level indicator that constitutes the jth second-level indicator of the enterprise; The neural network KAN consists of an input layer, a hidden layer and an output layer; The process of calculating the secondary index I1 through the neural network KAN based on the Kolmogorov-Arnold theorem is as follows: At the input layer, the input vector received is In the hidden layer, first, for each input variable and index q, q = 1, 2, ..., 7, define the nonlinear transformation function through two fully connected layers and ReLU activation function in, and is the weight matrix and bias vector of the first fully connected layer, and is the weight matrix and bias vector of the second fully connected layer. Secondly, for each q, all The output of Finally, for each S q The nonlinear transformation function is defined by two fully connected layers and the ReLU activation function. in, and is the weight matrix and bias vector of the first fully connected layer, and is the weight matrix and bias vector of the second fully connected layer; randomly initialize the KAN neural network parameters In the output layer, the quantitative value of the secondary index I1 is expressed as Among them, Sig(·) is the Sigmoid activation function, which is used to convert The value of is mapped to [0,1]; other secondary indicators use the same calculation process; The loss function used by the KAN neural network is expressed as Where M represents the database D i The total number of enterprises in .

5. The enterprise innovation capability evaluation and prediction device based on neural network according to claim 4 is characterized by: The method of training KAN neural network is to iterate the parameters based on the back propagation algorithm. When the set number of iterations is reached, the network training ends and the network parameters at this time are the network parameters that minimize the loss function Loss1.

6. The enterprise innovation capability evaluation and prediction device based on neural network according to claim 4 is characterized by: The specific method for determining the quantitative value of the primary indicator by the ratio of the radar chart area enclosed by the quantitative values ​​of all secondary indicators of the enterprise to be evaluated to the maximum possible area is as follows: Database D i The first-level index of the mth enterprise is in, Represents database D i The quantitative value of the jth secondary indicator of the enterprise in the radar chart is in, The maximum possible area of ​​a radar chart is The quantitative value of the enterprise's first-level indicator is 7. The enterprise innovation capability evaluation and prediction device based on neural network according to claim 6 is characterized by: The specific method of combining the bidirectional long short-term memory network model with the attention mechanism through the enterprise innovation ability training of the enterprise to be evaluated over the years is as follows: The quantitative values ​​of the enterprise innovation capabilities of the enterprises to be evaluated are sorted from small to large according to the year, and the data set data = {v t1 ,v t2 ,…,v tn }, where v tn It represents the quantitative value of the innovation capability of the enterprise to be evaluated in the year tn; Construct a Bi-LSTM+Attention model consisting of five layers: input layer, Bi-LSTM layer, attention layer, Dropout layer, and output layer. In the input layer, the time series data received is Among them, N is the total number of data points contained in the data set data, are the inputs of time steps 1 to 5 respectively; in the Bi-LSTM layer, the hidden state of the forward LSTM is represented as where v (p) is the input of the pth time step, and are the hidden state and cell state of the forward LSTM at the (p-1)th time step, respectively. The hidden state of the backward LSTM is expressed as where v (p) is the input of the pth time step, and are the hidden state and cell state of the backward LSTM at the (p+1)th time step, respectively. The forward hidden state and the backward hidden state are concatenated into a bidirectional hidden state vector H p , In the attention layer, the attention score at each time step p is denoted as e p =tanh(W p H p +b p ), where W p and b p is the weight matrix and bias vector of the attention layer at the pth time step, tanh(·) is the hyperbolic tangent activation function, and the attention weight is expressed as The context vector is represented as In the Dropout layer, 10% of the neurons from the attention layer are randomly selected and discarded, expressed as O'=Dropout(O,0.1), where 0.1 is the proportion of neurons discarded; in the output layer, the output of the model is expressed as V tn+5 =W out O'+b out , where V tn+5 To use time series data n Compared with the predicted value of enterprise innovation capability obtained by Bi-LSTM+Attention model in year (tn+5), W out and b out is the weight matrix and bias vector of the output layer; The loss function used by the Bi-LSTM+Attention model is expressed as Among them, v tn+5 is the true value of the enterprise innovation capability in year (tn+5), and N is the total number of data points contained in the data set data; The Bi-LSTM+Attention model is trained by iteratively updating the Bi-LSTM+Attention model parameters based on the back-propagation algorithm. When the set number of iterations is reached, the network training ends, and the network parameters at this time are the network parameters that minimize the loss function Loss2.

8. The enterprise innovation capability evaluation and prediction device based on neural network according to claim 7 is characterized by: The specific method of predicting the future enterprise innovation capability of the enterprise to be evaluated and constructing a 95% confidence interval by using the trained bidirectional long short-term memory network model and the Dropout method is as follows: Using the trained Bi-LSTM+Attention model and time series data When making time series predictions, enable Dropout and perform T random forward propagations to obtain a series of predicted values. The mean of the T-times predicted values ​​is expressed as The standard deviation of the T-time prediction value is expressed as The confidence interval is expressed as The mean μ of T predicted values ​​is used as the quantitative value of the enterprise innovation capability of the enterprise to be evaluated in the next year.

9. A method for evaluating and predicting enterprise innovation capability based on a neural network, based on the enterprise innovation capability evaluation and prediction device based on a neural network according to any one of claims 1 to 8, characterized in that: After collecting and normalizing all the third-level indicator data of various types of enterprises, inputting and improving the enterprise category database, the second-level evaluation module and the first-level evaluation module are called successively. First, the enterprise type i is determined according to the category to which the enterprise to be evaluated belongs, and the quantitative value of the second-level indicator of the enterprise to be evaluated is calculated through the KAN neural network and the normalized data of all the third-level indicators that constitute the second-level indicator; then the quantitative value of the first-level indicator is determined by the proportion of the radar chart area enclosed by the quantitative values ​​of all the second-level indicators of the enterprise to be evaluated relative to the maximum possible area, and the innovation ability of the enterprise is evaluated accordingly.

10. The method for evaluating and predicting enterprise innovation capability based on neural network according to claim 9 is characterized in that: Based on the enterprise category database, the prediction module is called. First, the enterprise innovation ability of the enterprise to be evaluated over the years is trained in combination with a bidirectional long short-term memory network model with an attention mechanism. Then, the future enterprise innovation ability of the enterprise to be evaluated is predicted through the trained bidirectional long short-term memory network model and the Dropout method, and a 95% confidence interval is constructed, thereby predicting the future innovation development trend of the enterprise to be evaluated.

Images