CN111832911A - Underwater combat effectiveness evaluation method based on neural network algorithm - Google Patents

Underwater combat effectiveness evaluation method based on neural network algorithm Download PDF

Info

Publication number
CN111832911A
CN111832911A CN202010591561.9A CN202010591561A CN111832911A CN 111832911 A CN111832911 A CN 111832911A CN 202010591561 A CN202010591561 A CN 202010591561A CN 111832911 A CN111832911 A CN 111832911A
Authority
CN
China
Prior art keywords
combat
matrix
neural network
equipment
task
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010591561.9A
Other languages
Chinese (zh)
Inventor
马雪飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Harbin Engineering University
Original Assignee
Harbin Engineering University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Harbin Engineering University filed Critical Harbin Engineering University
Priority to CN202010591561.9A priority Critical patent/CN111832911A/en
Publication of CN111832911A publication Critical patent/CN111832911A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Development Economics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Educational Administration (AREA)
  • Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Strategic Management (AREA)
  • Game Theory and Decision Science (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an underwater combat effectiveness evaluation method based on a neural network algorithm. Step 1: the combat parameters take values within a specified range, and a plurality of groups of data are respectively generated for corresponding combat systems according to actual requirements; step 2: solving an availability matrix, a credibility matrix and a capability matrix by a WSEIAC method according to the groups of data in the step 1; and step 3: calculating and solving the fighting effect value according to the step 2 and the corresponding fighting system; and 4, step 4: determining the number of neurons in an input layer, the range of the number of neurons in a hidden layer and the output of a neural network according to a BP algorithm principle; and 5: and (4) evaluating and analyzing the combat effectiveness based on the BP algorithm principle in the step 4 to determine an optimal value. The invention aims to solve the problem of insufficient capacity of the traditional evaluation method in processing mass data.

Description

Underwater combat effectiveness evaluation method based on neural network algorithm
Technical Field
The invention belongs to the technical field of underwater battles; in particular to an underwater combat effectiveness evaluation method based on a neural network algorithm.
Background
China is a large ocean country and has a long coastline and a vast sea area. With the progress of science and technology and the development of society, the construction of ocean power is becoming a great importance. The construction of the ocean franchise puts higher requirements on the ocean operational capacity. After finding the enemy ship, the submarine can attack the enemy ship by using the torpedo, and the fighting efficiency needs to be pre-estimated before action is carried out, so that the fighting scheme is adjusted by combining a decision algorithm, and the best fighting effect is achieved. After the combat action is finished, the effectiveness of the combat action also needs to be evaluated, so as to provide a basis for the design scheme and the use rule of weapons and the establishment of a fighting scheme.
The battle effectiveness evaluation is the basis of command decision, and before the battle action is implemented, the effectiveness of the action needs to be estimated, and the final high-efficiency rule and scheme are adopted to implement the action by combining with decision information. After the combat action is completed, the performance of the action is analyzed, so as to provide a basis for optimizing the action scheme and using the weapon rules. The conventional countermeasures include a soft killer weapon and a hard killer weapon, and how to correctly use various countermeasures under different combat conditions to optimize the countermeasures, so that the effectiveness of different countermeasures needs to be analyzed. In a commonly used performance analysis method, either a monte carlo method requires modeling for a large number of simulations on the premise of determining the law of each variable, or an expert evaluation method requires time and energy consumption of an expert for analysis. These approaches either add time complexity or introduce artifacts. The traditional evaluation method has a little insufficient capability in processing massive data, and in a data-driven current environment, the development of a neural network technology provides a new direction for solving the problem, and the excellent autonomous learning capability and high-precision nonlinear approximation capability of the neural network technology enable the neural network technology to be widely applied to military.
Disclosure of Invention
The invention provides an underwater combat effectiveness evaluation method based on a neural network algorithm, and aims to solve the problem of insufficient capacity of a traditional evaluation method in processing mass data.
The invention is realized by the following technical scheme:
an underwater combat effectiveness evaluation method based on a neural network algorithm comprises the following steps,
step 1: the combat parameters take values within a specified range, and a plurality of groups of data are generated for a combat system according to actual requirements;
step 2: respectively solving an availability matrix, a credibility matrix and a capability matrix in the WSEIAC efficiency evaluation model by a WSEIAC method according to the groups of data in the step 1;
and step 3: calculating and solving the fighting effect value according to the availability matrix, the credibility matrix and the capability matrix in the step 2 and the matrix of the fighting system;
and 4, step 4: determining the number of neurons in an input layer, the range of the number of neurons in an implied layer and the output of a neural network according to a neural network algorithm;
and 5: and (4) carrying out operational effectiveness evaluation analysis based on the neural network algorithm in the step (4), and in the performance evaluation analysis, integrally researching the influence of different values of the momentum factor on the operational effectiveness in a serial training mode, a centralized training mode, and hidden layer node number selection problems of the neural network, so as to determine an optimal value.
Further, the step 1 is specifically to calculate the corresponding combat effectiveness value of the multiple groups of data generated by the combat system by using a WSEIAC method, and use the value as the expected value.
Further, the step 2WSEIAC method represents three components of the model with probabilities: availability, credibility and capability of equipment contained in a combat system, the model within the WSEIAC method may be formulated as:
Figure BDA0002556297990000023
in the formula:
Figure BDA0002556297990000021
for the efficiency matrix of the combat system, ATThe matrix is an availability matrix of the combat system, D is a credibility matrix of the combat system, and C is a capacity matrix of the combat system;
when only one purpose is needed for executing the task, namely, the target is completely destroyed or loses the fighting capacity, the efficiency evaluation model is simplified as follows:
ES＝A*[D]*C (2)
availability in the WAEIAC Performance evaluation model, the system population is divided into N components and expressed in terms of steady-state availability
Figure BDA0002556297990000022
In the formula, MTTRkMean repair time, MTBF, of the kth component of an installation during the standby phase of the execution of a taskkMean time between failures of the kth component of the equipment during the standby period of the task;
if there is not enough time for the equipment to repair while in the failed state, the availability matrix at this time may be represented as:
AT＝(a1,a2) (4)
in the formula: a is1For the probability that the equipment system is in a normal working state at the beginning of the execution of a task, a2To provide a probability that the equipment system will be in a fault state at the beginning of the execution of a task,
and has the following components:
a1+a2＝1 (5)
Figure BDA0002556297990000031
Figure BDA0002556297990000032
in the formula: lambda [ alpha ]sFailure rate of standby period, mu, for equipment system to perform taskssStandby repair rate, MTTR, for equipment system to perform taskssMean repair time, MTBF, for equipment systems during standby periods for performing taskssMean time between failures of the equipment system during standby periods of performing tasks.
Further, the WAEIAC performance evaluation model is divided into two states, namely a normal state and a fault state, and the availability matrix can be expressed as
Figure BDA0002556297990000033
In the formula (d)11The probability that the equipment system is in a normal state at the starting moment of executing the task and is also in a normal state in the process; d12The probability that the equipment system is in a normal state at the starting moment of executing the task but is in a fault state in the process; d21Probability that the equipment system is in a fault state at the starting moment of executing the task, but is in a normal state in the process; d22The probability that the equipment system is in a fault state at the beginning of executing the task and is also in a fault state in the process,
in the WAEIAC performance evaluation model, one is a repairable equipment system, and the other is a non-repairable equipment system;
that is, if the equipment is in a failure state, the possibility that it is repaired is 0;
for non-repairable equipment systems, there are:
Figure BDA0002556297990000034
Figure BDA0002556297990000035
d21＝0 (11)
d22＝1-d21＝1 (12)
namely, the method comprises the following steps:
Figure BDA0002556297990000036
furthermore, the capability matrix in the WAEIAC performance evaluation model is that the equipment system has different values in different states,
the whole equipment system serves a task, the equipment system is considered as a whole, only a normal state and a fault state exist in the process of executing the task, and the capacity matrix can be expressed as follows:
Figure BDA0002556297990000041
in the formula: c. C1Probability of completing task in normal state for equipment in the process of executing task, c2It is easy to derive a measure of the ability of the equipment to be in a fault state and to complete a task, with a value of 0.
Further, the step 4 specifically includes the following steps:
step 4.1: the number of the input layer neurons is x, x influencing factors are selected as input, and for a combat system, the x influencing factors are respectively a system motion rate, an angle observation value, a shooting advance angle and a distance;
step 4.2: the range formula for determining the number of hidden layer neurons of the neural network algorithm is,
Figure BDA0002556297990000042
wherein n is the number of neurons in the hidden layer, n0Number of neurons in input layer, n1The number of neurons in the output layer is shown, and a is an integer between 1 and 10;
step 4.3: the output of the neural network is a predicted value of the combat effectiveness value of the combat system, is a real number between 0 and 1, and therefore the number of output layer neurons is set to 1.
Further, the step 5: the battle effectiveness evaluation analysis based on the neural network algorithm specifically comprises,
for the parallel training mode, the cost function is defined as:
Figure BDA0002556297990000043
in the formula: e.g. of the typej(n) error between output value of output neuron and expected value in training of nth training sample, mLIs the number of output layer neurons, N is the number of training set samples,
the gradient descent algorithm expression is:
Figure BDA0002556297990000044
in the formula, eta is the learning rate, alpha is the momentum factor, and the value range is [0,1 ].
The invention has the beneficial effects that:
after analyzing the influence factors of the combat effectiveness, constructing a BP-based effectiveness evaluation model, evaluating the combat effectiveness by using a neural network algorithm, and discussing the influence of each factor on the performance of the neural network algorithm; the potential rules of the samples can be learned by the overall neural network algorithm, the trained neural network model can avoid the interference of human factors, the problems of overlarge data volume and low speed are solved, and the battle effectiveness is quickly and accurately estimated. The method can fully utilize huge data in the field of marine combat for years, establish a model from bottom data to high-level results directly, and provide basis and support for the formulation of use methods, combat rules and schemes of various weapons by utilizing a large amount of data. When the influence factors change, the estimation of the combat effectiveness can be realized only by utilizing new data for training and adjusting parameters, complex intermediate process modeling is not needed, and manpower and material resources are greatly saved.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is a diagram of an artificial neuron model according to the present invention.
FIG. 3 is a schematic diagram of the topology of the neural network of the present invention.
FIG. 4 is a flow chart of the algorithm of the neural network of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
An underwater combat effectiveness evaluation method based on a neural network algorithm comprises the following steps,
step 1: the combat parameters take values within a specified range, and a plurality of groups of data are generated for a combat system according to actual requirements;
step 2: respectively solving an availability matrix, a credibility matrix and a capability matrix in the WSEIAC efficiency evaluation model by a WSEIAC method according to the groups of data in the step 1;
and step 3: calculating and solving the fighting effect value according to the availability matrix, the credibility matrix and the capability matrix in the step 2 and the matrix of the fighting system;
and 4, step 4: determining the number of neurons in an input layer, the range of the number of neurons in a hidden layer and the output of a neural network according to the principle of a neural network algorithm;
and 5: and (4) carrying out operational effectiveness evaluation analysis based on the neural network algorithm principle in the step (4), wherein in the performance evaluation analysis, a serial training mode and a centralized training mode are integrally researched, the selection problem of the number of nodes of the hidden layer of the neural network is solved, and the influence of different momentum factor values on the operational effectiveness is determined, so that the optimal value is determined.
Further, the step 1 is specifically a method for determining how to determine that each system simulation parameter should take a value within the indicated range, the values of the parameters need to be within the range, and some parameters need to satisfy a certain condition relationship at the same time, and a WSEIAC method is applied to generate multiple sets of data for the combat system to calculate the corresponding combat effectiveness value as an expected value.
Further, the step 2WSEIAC method represents three components of the model with probabilities: availability, trustworthiness and capability of equipment, the model within the WSEIAC method may be formulated as:
Figure BDA0002556297990000065
in the formula:
Figure BDA0002556297990000061
for the efficiency matrix of the combat system, ATThe matrix is an availability matrix of the combat system, D is a credibility matrix of the system, and C is a capacity matrix of the combat system;
for the vast majority of weaponry, however, the purpose of performing the task is only one: the target is destroyed completely or loses its fighting ability, so the efficiency evaluation model can be simplified as follows:
ES＝A*[D]*C (2)
availability in the waiiac performance evaluation model describes a measure of the different states that an equipment system may be in when it begins to perform a task. Therefore, each element of the matrix comprises the probability of all possible states of the equipment at the moment of starting to execute the task; dividing the system into N components, expressed in terms of steady-state availability, i.e.
Figure BDA0002556297990000062
The whole system as a whole takes into account that the time to actual action after the command of battle is given is very short, in which the MTTRkMean repair time, MTBF, of the kth component of an installation during the standby phase of the execution of a taskkThe mean time between failures of the kth component of an equipment in the standby period of performing a task, if there is insufficient time for the equipment to repair while in the failed state, and therefore there is no transition between the various states, the availability matrix at this time can be expressed as:
AT＝(a1,a2) (4)
in the formula: a is1Probability of the equipment system being in a normal working state at the beginning of executing a task (i.e. availability A of the equipment system)s)，a2Probability of equipment system being in fault state at the beginning of executing task (i.e. unavailability of equipment system)
And has the following components:
a1+a2＝1 (5)
Figure BDA0002556297990000063
Figure BDA0002556297990000064
in the formula: lambda [ alpha ]sFailure rate of standby period, mu, for equipment system to perform taskssStandby repair rate, MTTR, for equipment system to perform taskssMean repair time, MTBF, for equipment systems during standby periods for performing taskssMean time between failures of the equipment system during standby periods of performing tasks.
Further, in the WAEIAC performance evaluation model, the credibility matrix describes the state of the equipment during executing the task, and when the equipment is actually used, the system is divided into two states, namely a normal state and a fault state, and the availability matrix can be expressed as
Figure BDA0002556297990000071
In the formula (d)11The probability that the equipment system is in a normal state at the starting moment of executing the task and is also in a normal state in the process; d12The probability that the equipment system is in a normal state at the starting moment of executing the task but is in a fault state in the process; d21Probability that the equipment system is in a fault state at the starting moment of executing the task, but is in a normal state in the process; d22The probability that the equipment system is in a fault state at the beginning of executing the task and is also in a fault state in the process,
in the WAEIAC performance evaluation model, the equipment system can be divided into two types for calculation by the step, namely, a repairable equipment system and a non-repairable equipment system;
many arming systems are non-serviceable during the performance of a mission, typically with percussion weaponry, for which the time from receiving a command to executing the command is short, and if in a fault state prior to executing the command, there is no time to repair it, only to keep the fault state unchanged. That is, if the equipment is in a failure state, the possibility that it is repaired is 0;
for non-repairable equipment systems, there are:
Figure BDA0002556297990000072
Figure BDA0002556297990000073
d21＝0 (11)
d22＝1-d21＝1 (12)
namely, the method comprises the following steps:
Figure BDA0002556297990000074
further, the capability matrix in the waiiac performance evaluation model describes the capability of the equipment to complete a given task, and is usually expressed by using the probability of the equipment to complete the task, and the equipment system has different capabilities of completing the task when in different states, specifically in the capability matrix, that is, the equipment system has different values in different states,
in practical applications, the whole equipment system serves a task, the equipment system as a whole is considered, and only a normal state and a fault state exist in the process of executing the task, and under the condition, the capacity matrix can be expressed as:
Figure BDA0002556297990000081
in the formula: c. C1Probability of completing task in normal state for equipment in the process of executing task, c2It is easy to derive a measure of the ability of the equipment to be in a fault state and to complete a task, with a value of 0.
The performance of the algorithm is represented by the error between the estimated value and the expected value of the neural network model, and the smaller the error is, the closer the estimated value and the expected value are, and the better the performance of the algorithm is.
In the modeling implementation process of the neural network, the numerical range of each variable is changed greatly, if normalization is not carried out, the change scales of each variable are different when the gradient calculation of the neural network is carried out, a zigzag trend is formed, and convergence can be realized through a plurality of iterations.
There are different methods for data normalization, and this time, a linear normalization method is used to perform normalization processing on each kind of data, and the normalization formula is as follows:
Figure BDA0002556297990000082
the neural network algorithm is a simulation of a working mechanism of a human brain, artificial neurons can be obtained after biological neurons are abstracted, just as the brain is formed by connecting a plurality of neurons, the neural network is also formed by connecting neuron nodes, one node is one neuron, and the structure of the artificial neurons is as shown in 1:
the BP (Back propagation Algorithm) neural network is a fully-connected network, an input signal reaches an output neuron after being acted by a weight, a bias and an activation function, the output of the neuron only affects the next neuron connected with the neuron, and meanwhile, the neuron is only affected by the output of the neuron of the upper layer connected with the neuron. In the BP neural network, errors propagate forward along the opposite road to the input signal, thereby affecting the weights of the output layer and the hidden layer and the weights of the input layer and the hidden layer. With the back propagation of the error for one time, the weight of each connection layer is changed for one time, so that the output is changed, and finally the output value of the neural network model is closer to the expected value, thereby achieving the purpose of using the algorithm. It should be noted that, although in the BP neural network algorithm, the error signal is generated by the last layer of neurons, i.e. output neurons, and propagates backward layer by layer, it is not a feedback network, and there is only one signal channel in the whole neural network, and only the propagation directions of the function signal and the error signal are different, which is different from the feedback network essentially.
The topological structure of the BP neural network is shown in FIG. 2, and the algorithm flow chart of the BP neural network is shown in FIG. 3.
The operational effectiveness analysis model based on the BP neural network is characterized in that factors influencing operational effectiveness of an operational system are extracted and then used as input variables of the neural network, the effectiveness value is used as output variables of the neural network, and the neural network is repeatedly trained, so that the output value of the neural network can infinitely approximate to a real effectiveness value. The error between the model estimated value and the expected value reflects the performance of the algorithm, and the smaller the error is, the better the performance of the algorithm is.
Further, the step 4 specifically includes the following steps:
step 4.1: the number of the input layer neurons is x, x influencing factors are selected as input, and for a combat system, the x influencing factors are respectively a system motion rate, an angle observation value, a shooting advance angle and a distance;
step 4.2: the range formula for determining the number of hidden layer neurons of the neural network algorithm is,
Figure BDA0002556297990000091
wherein n is the number of neurons in the hidden layer, n0Number of neurons in input layer, n1The number of neurons in the output layer is shown, and a is an integer between 1 and 10;
step 4.3: the output of the neural network is a predicted value of the combat effectiveness value of the combat system, is a real number between 0 and 1, and therefore the number of output layer neurons is set to 1.
Up to this point, the operational effectiveness analysis model based on the BP neural network is established, and is of an x × n × 1 structure.
Further, the step 5: the battle effectiveness evaluation analysis based on the neural network algorithm specifically comprises,
impact analysis of different network training modes
For the training mode of the neural network algorithm, a serial training mode and a centralized training mode are commonTwo kinds. In the serial training mode, the neural network learns the input of each training sample and the corresponding output of each training sample and adjusts the synaptic weights once. Assuming that the training sample set is { (x (n)),
Figure BDA0002556297990000092
in the serial training mode, if the training sequence is ((x (1), d (1)), (x (2), d (2)), (x (n)), and d (n)), it indicates that for the network, the training sample (x (1), d (1)) and its expectation are presented to the network for the first time, and the adjustment of the synaptic weights and thresholds is completed for one time, the training sample (x (2), d (2)) and its expectation are presented to the network for the second time, and the adjustment of the synaptic weights and thresholds for the second time is completed until the last training sample and its expectation in the sample set (x (n), d (n)) are presented to the network, and the final adjustment of the synaptic weights and thresholds is completed. If the network performance reaches the requirement in advance, the network training is stopped; if the requirement is not met, the training is carried out again until the training process reaches the stop condition. In practical applications, however, the samples in the training sample set are often disorderly.
The parallel training mode is different from the serial training mode in that the synapse weight and the threshold are adjusted once when one training sample is input and expected, and the parallel training mode is that the synapse weight and the threshold are updated after all samples trained once in a training set are trained.
For the parallel training mode, the cost function is defined as:
Figure BDA0002556297990000101
in the formula: e.g. of the typej(n) error between output value of output neuron and expected value in training of nth training sample, mLIs the number of output layer neurons, N is the number of training set samples,
compared with the serial training mode, the centralized training mode adopts the real gradient searching method with the minimum systematic error, and the parallelization processing is easier. Different from a centralized training mode, although the training sample randomization of the serial training mode is a theoretical condition of algorithm convergence, the randomization of the training sample is difficult, but the randomization of the training sample makes the search of the algorithm when updating the weight and the threshold have randomness, so that the possibility that the back propagation algorithm falls into local minimum is reduced.
Impact analysis of hidden layer node number
For the selection problem of the number of hidden layer nodes of the neural network, if the number of hidden layer neuron nodes is too small, the neural network cannot establish a complex mapping relation, and the rules in the sample are fully learned; if the number of the hidden layer neuron nodes is too large, not only the training time of the neural network is increased, but also an overfitting problem is generated, so that the error of the network on a training sample is small, but the error on other samples is large. When designing a network, other parameters are comprehensively considered, and an optimal combination is selected, which is often a parameter combination for making the structure of the network more compact on the premise of meeting the precision requirement.
Analysis of the influence of momentum factors
When the weight and the threshold are corrected by the gradient descent algorithm, the correction is performed in a negative gradient mode, but the accumulated experience of the previous moment is not considered, particularly the gradient direction of the previous moment, so that the learning process is oscillated, and the situation is improved by adding the momentum term, wherein the expression is as follows:
Figure BDA0002556297990000102
in the formula, η is the learning rate, α is the momentum factor, and the value range is [0,1 ], and when the momentum factor is 0, the traditional gradient descent method is used.
The addition of the momentum term is equivalent to a damping term, so that the oscillation trend in the learning process is slowed down, and the convergence performance is improved. As the momentum factor becomes larger, the convergence speed of the error descending curve becomes faster, and the increase of the momentum factor also reduces the mean square error of the verification set. In practical application, the optimal combination needs to be selected by combining other parameters.

Claims (7)

1. A battle effectiveness evaluation method based on a neural network algorithm is characterized by comprising the following steps,
step 1: the combat parameters take values within a specified range, and a plurality of groups of data are generated for a combat system according to actual requirements;
step 2: respectively solving an availability matrix, a credibility matrix and a capability matrix in the WSEIAC efficiency evaluation model by a WSEIAC method according to the groups of data in the step 1;
and step 3: calculating and solving the fighting effect value according to the availability matrix, the credibility matrix and the capability matrix in the step 2 and the matrix of the fighting system;
and 4, step 4: determining the number of neurons in an input layer, the range of the number of neurons in an implied layer and the output of a neural network according to a neural network algorithm;
and 5: and (4) carrying out operational effectiveness evaluation analysis based on the neural network algorithm in the step (4), and in the performance evaluation analysis, integrally researching the influence of different values of the momentum factor on the operational effectiveness in a serial training mode, a centralized training mode, and hidden layer node number selection problems of the neural network, so as to determine an optimal value.
2. The method for evaluating the effectiveness of underwater combat based on the neural network algorithm as claimed in claim 1, wherein the step 1 is specifically to apply a WSEIAC method to a plurality of groups of data generated by a combat system to obtain the corresponding effectiveness value as the expected value.
3. The underwater combat effectiveness evaluation method based on the neural network algorithm as recited in claim 1, wherein the step 2WSEIAC method represents three components of a WSEIAC effectiveness evaluation model by probability: availability, credibility and capability of equipment contained in a combat system, the model within the WSEIAC method may be formulated as:
Figure FDA0002556297980000011
in the formula:
Figure FDA0002556297980000012
for the efficiency matrix of the combat system, ATThe matrix is an availability matrix of the combat system, D is a credibility matrix of the combat system, and C is a capacity matrix of the combat system;
when only one purpose is needed for executing the task, namely, the target is completely destroyed or loses the fighting capacity, the efficiency evaluation model is simplified as follows:
ES＝A*[D]*C (2)
availability in the WAEIAC Performance evaluation model, the system population is divided into N components and expressed in terms of steady-state availability
Figure FDA0002556297980000013
In the formula, MTTRkMean repair time, MTBF, of the kth component of an installation during the standby phase of the execution of a taskkMean time between failures of the kth component of the equipment during the standby period of the task;
if there is not enough time for the equipment to repair while in the failed state, the availability matrix at this time is represented as:
AT＝(a1,a2) (4)
in the formula: a is1For the probability that the equipment system is in a normal working state at the beginning of the execution of a task, a2To provide a probability that the equipment system will be in a fault state at the beginning of the execution of a task,
and has the following components:
a1+a2＝1 (5)
Figure FDA0002556297980000021
Figure FDA0002556297980000022
in the formula: lambda [ alpha ]sFailure rate of standby period, mu, for equipment system to perform taskssStandby repair rate, MTTR, for equipment system to perform taskssMean repair time, MTBF, for equipment systems during standby periods for performing taskssMean time between failures of the equipment system during standby periods of performing tasks.
4. The underwater combat effectiveness evaluation method based on the neural network algorithm as claimed in claim 3, wherein the WAEIAC effectiveness evaluation model, the system is divided into two states, namely a normal state and a fault state, the availability matrix is expressed as,
Figure FDA0002556297980000023
in the formula (d)11The probability that the equipment system is in a normal state at the starting moment of executing the task and is in the normal state in the process at the same time; d12The probability that the equipment system is in a normal state at the starting moment of executing the task but is in a fault state in the process; d21Probability that the equipment system is in a fault state at the starting moment of executing the task, but is in a normal state in the process; d22The probability of the equipment system being in a fault state at the start of its task execution, and in the fault state in the process,
in the WAEIAC performance evaluation model, one is a repairable equipment system, and the other is a non-repairable equipment system;
that is, if the equipment is in a failure state, the possibility that it is repaired is 0;
for non-repairable equipment systems, there are:
Figure FDA0002556297980000024
Figure FDA0002556297980000025
d21＝0 (11)
d22＝1-d21＝1 (12)
namely, the method comprises the following steps:
Figure FDA0002556297980000031
5. the method as claimed in claim 3, wherein the capability matrix of the WAEIAC performance evaluation model is different values of the equipment system under different states,
the entire equipment system serves for a task, considering the equipment system as a whole, and when performing the task, there are only normal and fault states, the capability matrix can be expressed as,
Figure FDA0002556297980000032
in the formula: c. C1Probability of completing task in normal state for equipment in the process of executing task, c2It is easy to derive a measure of the ability of the equipment to be in a fault state and to complete a task, with a value of 0.
6. The method for evaluating the effectiveness of underwater combat based on the neural network algorithm as claimed in claim 1, wherein the step 4 comprises the following steps:
step 4.1: the number of the input layer neurons is x, x influencing factors are selected as input, and for a combat system, the x influencing factors are respectively a system motion rate, an angle observation value, a shooting advance angle and a distance;
step 4.2: the range formula for determining the number of cryptic neurons of the BP algorithm is,
Figure FDA0002556297980000033
wherein n is the number of neurons in the hidden layer, n0Number of neurons in input layer, n1The number of neurons in the output layer is shown, and a is an integer between 1 and 10;
step 4.3: the output of the neural network is a predicted value of the combat effectiveness value of the combat system, is a real number between 0 and 1, and therefore the number of output layer neurons is set to 1.
7. The method for evaluating the effectiveness of underwater combat based on the neural network algorithm as claimed in claim, wherein the step 5: the tactical performance evaluation analysis based on the BP algorithm is specifically,
for the parallel training mode, the cost function is defined as:
Figure FDA0002556297980000034
in the formula: e.g. of the typej(n) error between output value of output neuron and expected value in training of nth training sample, mLIs the number of output layer neurons, N is the number of training set samples,
the gradient descent algorithm expression is:
Figure FDA0002556297980000041
in the formula, eta is the learning rate, alpha is the momentum factor, and the value range is [0,1 ].
CN202010591561.9A 2020-06-24 2020-06-24 Underwater combat effectiveness evaluation method based on neural network algorithm Pending CN111832911A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010591561.9A CN111832911A (en) 2020-06-24 2020-06-24 Underwater combat effectiveness evaluation method based on neural network algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010591561.9A CN111832911A (en) 2020-06-24 2020-06-24 Underwater combat effectiveness evaluation method based on neural network algorithm

Publications (1)

Publication Number Publication Date
CN111832911A true CN111832911A (en) 2020-10-27

Family

ID=72898089

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010591561.9A Pending CN111832911A (en) 2020-06-24 2020-06-24 Underwater combat effectiveness evaluation method based on neural network algorithm

Country Status (1)

Country Link
CN (1) CN111832911A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112859807A (en) * 2021-01-10 2021-05-28 西北工业大学 Underwater vehicle collaborative search efficiency evaluation method based on situation simulation and Monte Carlo
CN112948163A (en) * 2021-03-26 2021-06-11 中国航空无线电电子研究所 Method for evaluating influence of equipment on functional fault based on BP neural network
CN113435780A (en) * 2021-07-14 2021-09-24 北京信息科技大学 Emergency communication perception equipment system efficiency evaluation method based on neural network

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104063613A (en) * 2014-07-02 2014-09-24 东南大学 Multi-mode bus network effectiveness evaluation method
CN105976098A (en) * 2016-04-28 2016-09-28 大连理工大学 BP neural network-based college and university teaching quality evaluation method
CN109614729A (en) * 2018-12-18 2019-04-12 上海机电工程研究所 A kind of equipment Efficacy assessment result rapid analysis method based on meta-model
CN110544011A (en) * 2019-07-31 2019-12-06 北京航空航天大学 Intelligent system combat effectiveness evaluation and optimization method
CN110619148A (en) * 2019-08-13 2019-12-27 上海机电工程研究所 Equipment ADC (analog to digital converter) efficiency evaluation method based on interval gray number

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104063613A (en) * 2014-07-02 2014-09-24 东南大学 Multi-mode bus network effectiveness evaluation method
CN105976098A (en) * 2016-04-28 2016-09-28 大连理工大学 BP neural network-based college and university teaching quality evaluation method
CN109614729A (en) * 2018-12-18 2019-04-12 上海机电工程研究所 A kind of equipment Efficacy assessment result rapid analysis method based on meta-model
CN110544011A (en) * 2019-07-31 2019-12-06 北京航空航天大学 Intelligent system combat effectiveness evaluation and optimization method
CN110619148A (en) * 2019-08-13 2019-12-27 上海机电工程研究所 Equipment ADC (analog to digital converter) efficiency evaluation method based on interval gray number

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
张泽旭: "《神经网络控制与MATLAB仿真》", 31 July 2011 *
梅雄 等: "基于WSEIAC模型的雷达侦察系统效能评估模型研究", 《舰船电子对抗》 *
王小艺 等: "一种改进的系统效能评估方法及其应用", 《火力与指挥控制》 *
陈侠等: "基于APSO-BP神经网络的无人机空地作战效能评估研究", 《飞行力学》 *
陈侠等: "基于改进型小波神经网络的电子战无人机作战效能评估研究", 《电光与控制》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112859807A (en) * 2021-01-10 2021-05-28 西北工业大学 Underwater vehicle collaborative search efficiency evaluation method based on situation simulation and Monte Carlo
CN112859807B (en) * 2021-01-10 2022-03-22 西北工业大学 Underwater vehicle collaborative search efficiency evaluation method based on situation simulation and Monte Carlo
CN112948163A (en) * 2021-03-26 2021-06-11 中国航空无线电电子研究所 Method for evaluating influence of equipment on functional fault based on BP neural network
CN112948163B (en) * 2021-03-26 2023-09-19 中国航空无线电电子研究所 Method for evaluating influence of equipment on functional failure based on BP neural network
CN113435780A (en) * 2021-07-14 2021-09-24 北京信息科技大学 Emergency communication perception equipment system efficiency evaluation method based on neural network

Similar Documents

Publication Publication Date Title
CN111262858B (en) Network security situation prediction method based on SA _ SOA _ BP neural network
Kim et al. Length-adaptive transformer: Train once with length drop, use anytime with search
CN111832911A (en) Underwater combat effectiveness evaluation method based on neural network algorithm
Wang et al. A compact constraint incremental method for random weight networks and its application
CN112491818B (en) Power grid transmission line defense method based on multi-agent deep reinforcement learning
CN111553469A (en) Wireless sensor network data fusion method, device and storage medium
CN111260124A (en) Chaos time sequence prediction method based on attention mechanism deep learning
CN110544011A (en) Intelligent system combat effectiveness evaluation and optimization method
CN111709511A (en) Harris eagle optimization algorithm based on random unscented Sigma point variation
Chouikhi et al. Single-and multi-objective particle swarm optimization of reservoir structure in echo state network
CN113722980A (en) Ocean wave height prediction method, system, computer equipment, storage medium and terminal
CN116680540A (en) Wind power prediction method based on deep learning
CN115730743A (en) Battlefield combat trend prediction method based on deep neural network
Song et al. A comprehensive evaluation and comparison of enhanced learning methods
CN115470889A (en) Network-on-chip autonomous optimal mapping exploration system and method based on reinforcement learning
CN106055789B (en) Autonomous unmanned aerial vehicle formalized inference method based on weighted fuzzy Petri net
CN116340737A (en) Heterogeneous cluster zero communication target distribution method based on multi-agent reinforcement learning
CN116522747A (en) Two-stage optimized extrusion casting process parameter optimization design method
CN116165886A (en) Multi-sensor intelligent cooperative control method, device, equipment and medium
Yu et al. A reinforcement learning algorithm based on neural network for economic dispatch
Wu et al. Heterogeneous networks topology optimization based on simulated annealing algorithm
CN104537224A (en) Multi-state system reliability analysis method and system based on self-adaptive learning algorithm
Smith et al. Recurrent neural network ensembles for convergence prediction in surrogate-assisted evolutionary optimization
Rahman et al. Implementation of artificial neural network on regression analysis
CN113884936A (en) Lithium ion battery health state prediction method based on ISSA coupling DELM

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination