RU2813682C1 - Device for calculating degree of confidence in reconnaissance tool - Google Patents

Abstract

FIELD: computer engineering.

SUBSTANCE: effect is achieved through a device for calculating the degree of trust in reconnaissance means in real time under changing weather and operational-tactical conditions, which contains an input parameters processing unit that divides and transmits the values of input parameters to units for processing physiographic and hydrometeorological data, operational-tactical situation and enemy counteraction data, a computing unit that calculates the degree of confidence in the reconnaissance tool taking into account the coefficients of the artificial neural network tuning unit, an output signal generation unit that translates the calculation results into a storage and processing format, and a results storage unit.

EFFECT: calculating the degree of trust in reconnaissance means in real time under changing weather and operational-tactical conditions.

4 cl, 1 dwg, 1 tbl

Description

The invention relates to weapons and military equipment, namely to intelligence information processing devices based on specific mathematical models.

There are known devices based on mathematical models used in relation to information security.

There is a known system for automated fact analysis (see [1] Russian Federation patent for invention No. 2656583 “System for automated fact analysis”, IPC G06F 15/00, publ. 06/05/2018), which is an information security technology that allows you to track the activity of company employees on social networks, forums, blogs outside the company’s security perimeter and analyze the received data to prevent potential threats. The goal is achieved by creating an automated fact analysis system that provides verification and analysis of behavioral risks of social media users, has flexible configuration of monitoring criteria and increased efficiency in detecting threats emanating from the actions of employees on social media. Threat levels have the following categories: informational level, warning level, or critical level.

There is a known method that describes assessing the reliability of the reported information based on a structured interview (see [2] Russian Federation patent for invention No. 2438558 “Method for assessing the reliability of the reported information based on the analysis of the dynamics of the parameters of nonverbal and verbal components of expressive speech”, IPC A61B 5/00 15 /00, publ. 01/10/2012), characterized by the fact that the parameters of the verbal and nonverbal components of expressive speech are recorded when answering blocks of questions. The dynamics of the selected parameters is studied with the definition of the most informative, grouped on the basis of factor analysis, the trends of changes characteristic of the subject under study are identified for each parameter and/or group of parameters of the verbal and non-verbal components of expressive speech, based on which a conclusion is made about the falsity/truth of the reported information by comparing quantitative parameters of reactions to questions of the test block with quantitative parameters of reactions to questions of neutral and/or control blocks of questions.

There is a known method that describes increasing the reliability of assessing the trustworthiness of an individual consumer of financial services by constructing a psychological and social portrait of the individual (see [3] Russian Federation patent for invention No. 2473967 “System for assessing the trustworthiness of an individual consumer of financial services”, IPC G06Q 90/00 , published 10/20/2012). The trustworthiness assessment system contains a set of interacting means for data entry, analysis of entered data, collection of information on social networks, collection of data on network interactions, survey of an individual’s counterparties on social networks, calculation of a numerically expressed and standardized assessment of the trustworthiness of a consumer of financial services. The system also includes a database of individuals, a component of psychological and social testing of an individual - a prospective consumer of financial services in the form of a question and answer; component of social environment analysis. And also a component for forming a profile of the psychological and social qualities of an individual, designed to receive information from the outputs of the input data analysis component, the data collection component on network interactions, the social environment analysis component, the analysis component of network interactions between individuals, the survey component of an individual’s counterparties on social networks, and component of psychological and social testing.

The patent of the Russian Federation on quality control of the operator of an industrial vehicle (see [4] patent of the Russian Federation for invention No. 2643459 “Quality control of the operator of an industrial vehicle”, IPC G06Q 10/06, published 02/08/2017) describes a method for combining indicators performing the work of an industrial vehicle operator. The technical result is the creation of tools that ensure the collection of information about labor resources and assessment of the quality of their work. To do this, the overall quality of the workforce is analyzed, quantified and presented using the computing unit of the analysis system, which performs enterprise data analysis. The analysis system presents key information about the work performance of the workforce through a range of hardware devices to inform various users. The analytics system links a customizable work performance profile to a variety of work performance metrics. Each job performance measure represents a job performance measurement system that defines an aspect of the job duties performed by the associated work unit. Quantitative assessments are combined into an overall quantitative assessment of the work performance profile. To calculate quantitative estimates, data from multiple work areas is considered by collecting and analyzing data from various systems.

In the field of weapons and military equipment, there are methodological approaches to assessing the degree of trust in reconnaissance means, which make it possible to obtain static assessments of trust that do not take into account the dynamically changing conditions of reconnaissance.

The above mathematical methods cannot be used to assess the degree of trust in information sources in real time.

The technical result of the claimed invention is the creation of a device for calculating the degree of trust in reconnaissance means in real time under changing weather and operational-tactical conditions. The invention makes it possible to process a set of factors necessary for assessing the degree of trust, sequentially determining the importance of each factor in given hydrometeorological and operational-tactical conditions and, on their basis, calculating the degree of trust in reconnaissance means when clarifying the class of a detected enemy troop object (ETO).

Among the factors influencing the degree of trust in a reconnaissance device that has detected an air defense system, the most significant are physical-geographical and hydrometeorological factors, factors of the operational-tactical situation, as well as factors of the enemy’s counteraction to technical reconnaissance means.

An approximate list of factors and their values used to dynamically assess the degree of confidence in a reconnaissance tool that has detected airborne weapons is presented in Table 1.

The task of calculating the degree of trust in a reconnaissance tool is a linear regression task with polynomial coefficients, estimating the input vector of parameters of reconnaissance conditions (for example, 18 numbers according to Table 1) and giving the output a vector with estimates of the degrees of trust in a specific reconnaissance tool in relation to the recognition of a specific class OBVP, and is solved using neural network modeling technology in the presence of an artificial neural network (ANN) trained using the supervised learning method.

“Supervised learning” ANN assumes that for each input vector of variables (see Table 1) the target output is known. The ANN is trained and then tested on a given number of examples of labeled data. Each of the examples is a one-to-one set (vector) of values of the ANN input variables and the corresponding values of the ANN output variables.

ANN training takes place at an automated workstation (AWS) that is not associated with a device for calculating the degree of trust in the reconnaissance tool. The results of calculations obtained at the workstation in the form of coefficients are recorded in the device for calculating the degree of confidence before it begins to operate.

The device for calculating the degree of confidence in a reconnaissance means consists of a block for processing input parameters, a block for processing physical-geographical and hydrometeorological data, a block for processing operational-tactical situation data, a block for processing data on enemy counteraction, a computing block, an ANN tuning block, a block for generating an output signal and results storage block.

Input parameters in the form of a vector of linguistic variables (see Table 1), characterizing physical-geographical, hydrometeorological factors, factors of the operational-tactical situation and the enemy’s counteraction to technical reconnaissance means, describing the current situation, are sent from an external device to the input of the input parameters processing unit . In the block for processing input parameters, variables are divided into physical-geographical and hydrometeorological factors, factors of the operational-tactical situation and factors of enemy counteraction to reconnaissance means, which are transmitted respectively to the block for processing physical-geographical and hydrometeorological data, the block for processing data of the operational-tactical situation and the block processing data on enemy counteraction, where preprocessing of factors is carried out by converting linguistic variables into a binary representation (0 and 1), and for categorical variables factorization is performed, which consists of the following: each term of the factor is assigned a numerical value, after which it is assigned a binary vector , consisting of zeros except for the position corresponding to the desired number (for example, if there are 3 terms, the numbers 1, 2 and 3 are assigned to them, and the vectors (0,0,1), (0,1,0) and ( 1,0,0)). Information from the physical-geographical and hydrometeorological data processing unit, the operational-tactical situation data processing unit and the enemy countermeasures data processing unit is transmitted to the computing unit, which calculates the degree of confidence in the reconnaissance tool, taking into account the coefficients of the INS tuning unit obtained when training the INS on A workstation not associated with a device for calculating the degree of confidence in the reconnaissance tool, and recorded in the INS settings block. Calculations of the computing unit are transferred to the output signal generation unit, where the calculation result is interpreted, translated into a storage and processing format, as well as the response of the task of calculating the degree of confidence in reconnaissance means is transmitted to the results storage unit and to an external executive body for making tactical decisions.

A diagram of the device for calculating the degree of confidence in the reconnaissance tool is shown in Figure 1, where:

1 - Block for processing input parameters (Bl.In.);

2 - Unit for processing physical-geographical and hydrometeorological data (Bl.FGD);

3 - Operational-tactical situation data processing unit (OTO Unit);

4 - Data processing unit on enemy countermeasures (Bl.PP);

5 - Computing block (Calc.Bl.);

6 - INS tuning unit (BL.INS);

7 - Output signal generation block (Bl.Out.);

8 - Results storage block (Bl.Shr.).

Description of the structure and connections of the device

The input of the input parameters processing unit (1) of the device for calculating the degree of trust in the intelligence tool is connected to an external device, from where input parameters are supplied in the form of a vector of linguistic variables (see Table 1). In the input parameters processing block (1), variables are divided into physical-geographical and hydrometeorological factors, operational-tactical situation factors and factors of enemy countermeasures to reconnaissance means, which are transmitted to the inputs of the physical-geographical and hydrometeorological data processing block (2), data processing block operational-tactical situation (3) and data processing unit on enemy counteraction (4), respectively. The outputs of the physical-geographical and hydrometeorological data processing unit (2), the operational-tactical situation data processing unit (3) and the enemy countermeasures data processing unit (4), as well as the output of the ANN tuning unit are connected to the input of the computing unit (5), which carries out calculating the degree of trust in the reconnaissance tool, the output of which is connected to the input of the output signal generation block (7), which provides the answer to the task of calculating the degree of trust in the reconnaissance tool to the input of the results storage block (8), as well as to the external executive body.

The principle of training an ANN device for calculating the degree of trust in an intelligence tool

On an automated workstation that is not connected to the device for calculating the degree of confidence in the reconnaissance tool, an ANN is formed in advance by polynomizing the vector of variables to the 2nd degree. A training sample is formed, which is a vector of the form (x[1],…x[n],x[1]*x[2],…x[1]*x[n],… x[i]*x[j ],…x[1]^2,…x[n]^2), containing all possible combinations of vector components, including terms of the 1st and 2nd degree, as well as cross terms (x[i]*x[j]) .

Next, the weighting coefficients of the ANN with polynomial coefficients are selected to minimize the error functional:

where k=(1,..7) - reconnaissance vehicle type, j=(1,..10) - OBVP class.

- error function corresponding to the k-th (k = (1,..7)) reconnaissance means,

in relation to class j recognition (j=(1,..10));

- degree of confidence, calculated in accordance with methods [5, 6];

- current response of the model;

m is the number of training examples.

The ANN taking into account terms of the 2nd degree was chosen, on the one hand, based on the maximum accuracy of the algorithm achieved on the training set, and, on the other hand, based on the clarity and ease of implementation of the linear regression algorithm.

The ANN training process consists of sequential input of vectors from the training set, calculation of the error functional and the process of adjusting the coefficients W _kj [i] for each term (i=(1..N), k=(1,..7), j=(1,..10)), where N is the total number of polynomial terms of linear regression.

Coefficient updating stops when the prediction accuracy exceeds 95%. For this purpose, the criterion of early stopping of training is used, which makes it possible not to perform a large number of iterations (unknown in advance before starting training), limiting itself to a test sample of vectors.

The algorithm is considered to have converged when it reaches 95% accuracy on a test sample of input examples. The test sample size is 25% of the training sample size.

For each vector of input parameters, a polynomial vector of 192 numbers is formed, which is associated with weighting coefficients W[i], i=1..192 to evaluate each of (K*J)=70 numbers corresponding to the estimated degrees of confidence in intelligence means.

The obtained coefficients are recorded in the device for calculating the degree of confidence in the reconnaissance tool (in the ANN tuning unit (6)) before its operation begins.

Operating principle of a device for calculating the degree of trust in an intelligence tool

To process input parameters in the form of a vector of linguistic variables (see Tab. 1), characterizing physical-geographical and hydrometeorological factors, factors of the operational-tactical situation and the enemy’s counteraction to technical reconnaissance means, describing the current situation, in the input parameters processing block (1) formalization and transfer of factor values are carried out to the inputs of the physical-geographical and hydrometeorological data processing unit (2), the operational-tactical situation data processing unit (3) and the enemy counteraction data processing unit (4), respectively, to translate linguistic variables into a binary representation (0 and 1), and for categorical variables factorization is carried out, which consists of the following: each term of the factor is assigned a numerical value, after which it is assigned a binary vector consisting of zeros except for the position corresponding to the desired number (for example, if there are 3 terms, they are assigned to correspondence to the numbers 1, 2 and 3, and to the numbers - the vectors (0,0,1), (0,1,0) and (1,0,0)). The received data is fed to the input of the computing unit (5), which implements the ANN algorithm taking into account the weighting coefficients of the ANN tuning unit (6), where the degree of confidence is calculated for all reconnaissance means in these conditions according to the following formula:

where k=(1,..7) is the type of reconnaissance means, j=(1,..10) is the class of the enemy object.

The calculation results are fed to the output signal generation unit (7), where the calculation result is interpreted, translated into a storage and processing format, as well as the response of the task for calculating the degree of trust in reconnaissance means is transmitted to the results storage unit (8) and to an external executive body for processing by the operator and making a decision on further actions based on the response received.

The results obtained in relation to the speed of calculation and the performance of the device for calculating the degree of trust in the reconnaissance tool confirm the validity of the approach taken and the effectiveness of the use of machine learning methods in systems for automated processing of operational information, as well as the device architecture described above. Moreover, the device architecture can be implemented in the form of a storage device using complementary metal-oxide-semiconductor (CMOS) digital circuits, which has the advantage of easy design and construction. They are based on existing logic elements and fully reflect advances in digital circuits over the past decades [7]. The weighting coefficients W _kj [i], corresponding to each element of the vector of values of the input variables of the ANN, can be implemented using digital memory cells. Effective implementation of ANN weights is critical to maintaining the accuracy and predictive power of the algorithm. Summation of neuron states can be easily implemented using multipliers and adders.

The claimed device can be used to improve the system for complex processing of intelligence information.

BIBLIOGRAPHICAL LIST

1. Patent of the Russian Federation No. 2656583 G06F 15/00; G06Q 50/01 V.V. Ermakov. Automated fact analysis system.

2. Patent of the Russian Federation No. 2438558 A61B 5/00 L.N. Ivanov. A method for assessing the reliability of the reported information based on an analysis of the dynamics of the parameters of nonverbal and verbal components of expressive speech.

3. Patent of the Russian Federation No. 2473967 G06Q 90/00 G06F 17/40 M.E. Trukhin. A system for assessing the trustworthiness of an individual - a consumer of financial services.

4. Patent of the Russian Federation No. 2643459 G06Q 10/06 S.Sh. DE OLIVEIRA. Quality control of the operator of an industrial vehicle.

5. Theoretical foundations of artillery reconnaissance planning. I.N. Fomin, - St. Petersburg: Military Artillery University, 2000 - 96 p.

6. Fundamentals of mathematical modeling of the processes of functioning of artillery reconnaissance forces and means. A.S. Vorobyov, - St. Petersburg: Mikhailovskaya Artillery Academy, 1997 - 39 p.

7. S. Jung and S.S. Kim. Hardware implementation of real-time neural network controller with DSP and FPGA for nonlinear systems "Industrial Electronics, IEEE Transactions, vol. 54" pp. 265-271, 2007.

Claims (10)

1. A device for calculating the degree of confidence in a reconnaissance tool, containing a block for processing input parameters, a block for processing physical-geographical and hydrometeorological data, a block for processing operational-tactical situation data, a block for processing data on enemy counteraction, a computing block, a block for setting up an artificial neural network, a block generating an output signal and a block for storing results, and a block for processing input parameters, to the input of which a vector of variables is supplied from an external device, representing the values of physical-geographical and hydrometeorological factors, factors of the operational-tactical situation and the enemy’s counteraction to technical reconnaissance means that characterize the current situation , separates and transmits the corresponding variables to the inputs of the physical-geographical and hydrometeorological data processing unit, the operational-tactical situation data processing unit and the enemy countermeasures data processing unit, in which the factor values are preprocessed by converting them into a binary or factorized representation and at the outputs of which the obtained numerical variables are combined and fed to the input of the computational unit, which calculates the degree of confidence in the reconnaissance tool, taking into account the coefficients of the artificial neural network tuning unit, obtained and recorded in the tuning unit when training the artificial neural network using the “supervised learning” method at an automated workstation, not associated with the device for calculating the degree of confidence in the reconnaissance tool, but from the output of the computing device, the calculation result is sent to the input of the output signal generation unit, where the calculation result is interpreted, translated into a storage and processing format, as well as transmitting the answer to the task of calculating the degree of confidence in reconnaissance means to the results storage unit and to an external executive body for making tactical decisions in real time. 2. The device according to claim 1, characterized in that at an automated workstation not associated with a device for calculating the degree of trust in an intelligence tool, an artificial neural network is formed by polynomizing to the 2nd degree a vector of variables, as a result of which a vector of the form ( x[1],...x[n],x[1]*x[2],...x[1]*x[n],...x[i]*x[j],...x[1]^2 ,…x[n]^2), containing all possible combinations of vector components, including terms of the 1st and 2nd degree, as well as cross terms (x[i]*x[j]), and then the weighting coefficients of the artificial neural network are selected with polynomial coefficients minimizing the error functional: Where: - error function corresponding to the k-th (k=(1,..7)) reconnaissance means, in relation to the recognition of class j (j=(1,..10)); - degree of confidence calculated in accordance with the methods; - current response of the model. m is the number of training examples. 3. The device according to claim 1, characterized in that the process of training an artificial neural network consists of sequentially feeding vectors from the training set to the input, calculating the error functional and adjusting the coefficients Wkj[i] for each term (i=(1.. N), k=(1,..7), j=(1,..10)), where N is the total number of polynomial terms of linear regression, and updating of coefficients stops when the prediction accuracy on the test sample becomes more than 95%. 4. The device according to claim 1, characterized in that the computing unit can be implemented using a complementary structure of metal-oxide-semiconductor-digital circuits.