Radar air defense reconnaissance effectiveness evaluation method based on deep learning
By building a multi-dimensional radar air defense reconnaissance efficiency evaluation system and deep neural network, the problems of high computational complexity and subjective factors in the existing technology are solved, and the radar air defense reconnaissance efficiency evaluation with high accuracy and fault tolerance are achieved.
Patent Information
- Application Number
- CN202210299565.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-25
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2042-03-25
AI Technical Summary
The effectiveness evaluation method of existing radar air defense reconnaissance systems has the problems of high computational complexity, great influence of subjective factors, and poor fault tolerance, especially in large data volume and complex environments, which are difficult to achieve accurate evaluation.
Build a radar air defense reconnaissance efficiency evaluation system with 4 evaluation dimensions, 13 evaluation indicators and 42 evaluation factors, and use deep neural networks to perform multi-dimensional evaluation, through equipment credibility, battlefield survivability, signal detection ability and signal processing ability The subnet is parallel output, combined with the adaptively adjusted learning rate for training, and realizes quantitative evaluation.
It reduces the computational complexity, avoids the influence of subjective factors, improves the accuracy and fault tolerance of evaluation, and can better extract the data characteristics and quantitative evaluation results of radar air defense reconnaissance efficiency.
Smart Images

Figure CN114662392B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of radar technology, and in particular relates to a radar air defense reconnaissance effectiveness evaluation method based on deep learning. Background Art
[0002] In the complex and ever-changing electromagnetic environment of air defense, efficient situational awareness can determine victory or defeat. Radar, the battlefield's clairvoyant, not only effectively detects enemy activity and foresees danger, but also seizes opportunities and suppresses the enemy. Its air defense reconnaissance capability, to a certain extent, reflects battlefield situational awareness, making its effective assessment crucial.
[0003] Research on the effectiveness evaluation of radar reconnaissance systems in air defense operations has made some progress. Sun Yangchao and his colleagues, based on the U.S. Industry Weapon Systems Effectiveness Advisory Committee (WSEIAC) model, employed the ADC method to comprehensively evaluate the overall performance of radar systems based on three key factors: availability, reliability, and capacity. They also evaluated the effectiveness of a specific type of reconnaissance radar using a practical example. However, the ADC method is not very applicable and suffers from low efficiency when multiple factors are involved.
[0004] Long Wenwu and his colleagues used the analytic hierarchy process (AHP) based on grey theory to evaluate radar operational effectiveness. They established an index system for evaluating the operational effectiveness of multi-function radars, encompassing seven aspects: detection, tracking, reconnaissance, identification, positioning, jamming, and attack. Using real-world simulations with specific data, they validated the effectiveness and practicality of the evaluation model. However, the AHP method involves multiple qualitative analysis components, making weighting difficult to determine and limiting its error tolerance.
[0005] Du Zhen et al. applied fuzzy comprehensive evaluation theory to establish a radar reconnaissance equipment operational effectiveness evaluation architecture based on fuzzy comprehensive evaluation. They conducted a comprehensive evaluation of five aspects: detection performance, signal interception capability, signal sorting and identification capability, parameter measurement capability, and reliability. They also demonstrated the application of this method to radar reconnaissance equipment effectiveness evaluation using practical examples. However, fuzzy comprehensive evaluation methods are significantly influenced by subjective factors. When the weight vector and the fuzzy matrix do not match, the results can become overly ambiguous or even fail to evaluate.
[0006] Qu Wentao and others introduced cloud theory, combining qualitative and quantitative assessments to propose a method for evaluating the reconnaissance effectiveness of airborne radar reconnaissance equipment based on the cloud center of gravity. They established an evaluation index system, derived a verbal evaluation value for system effectiveness, quantified it, and ultimately obtained evaluation results, providing a basis for improving the effectiveness of airborne radar reconnaissance equipment. However, the cloud center of gravity evaluation method is still subject to subjective factors to a certain extent, and the evaluation calculation process is not completely objective.
[0007] Qi Zongfeng and his colleagues developed a radar reconnaissance system combat capability assessment method based on a deep belief network. They analyzed the effectiveness assessment process in detail and verified the rationality and correctness of the method through simulation based on the radar reconnaissance system combat capability assessment object. However, when the input data volume is large, the deep belief network requires training many weight parameters, resulting in high computational complexity and difficulty in fully utilizing the advantages of large data samples. Summary of the Invention
[0008] The purpose of the present invention is to overcome the above-mentioned technical problems and propose a radar air defense reconnaissance effectiveness evaluation method based on deep learning to achieve a multi-dimensional evaluation of the air defense combat effectiveness of the radar reconnaissance system.
[0009] To achieve the above object, the technical solution of the present invention includes the following steps:
[0010] (1) Construct a radar air defense reconnaissance effectiveness evaluation system consisting of four evaluation dimensions, 13 evaluation indicators, and 42 evaluation factors;
[0011] (2) Construct a deep neural network. The deep neural network consists of four subnets: equipment credibility assessment subnet, battlefield survivability assessment subnet, signal detection capability assessment subnet, and signal processing capability assessment subnet. The subnets output in parallel and are then cascaded to form the overall combat effectiveness assessment subnet.
[0012] (3) Training deep neural networks, including:
[0013] 3a) Collect evaluation factors and calculate evaluation indicators;
[0014] 3b) The standard deviation method was used to normalize the 13 evaluation indicators and to assign values and interpret the four evaluation dimensions;
[0015] 3c) Using the backpropagation algorithm to train each subnet layer by layer, the training ends when the loss function is less than a preset threshold or the maximum number of training times is reached; otherwise, the weight parameters of each subnet are adjusted along the gradient descent direction of the loss function, the learning rate is adaptively adjusted, and each subnet is retrained;
[0016] (4) The normalized data to be evaluated is fed into the trained deep neural network to output the radar air defense reconnaissance effectiveness evaluation results.
[0017] Compared with the prior art, the present invention has the following advantages:
[0018] The present invention constructs a radar air defense reconnaissance effectiveness evaluation system including 4 evaluation dimensions, 13 evaluation indicators and 42 evaluation factors, and adopts a deep neural network based on adaptive adjustment of learning rate. By utilizing the generalization ability of the deep neural network, the computational complexity is reduced, and the characteristics of the radar air defense reconnaissance effectiveness evaluation data can be fully extracted, and the evaluation indicators and evaluation results can be quantified. It not only avoids the influence of subjective factors, but also has good fault tolerance, thereby greatly improving the accuracy of the air defense combat effectiveness evaluation of the radar reconnaissance system. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 is a flow chart of the present invention;
[0020] Figure 2 The radar air defense reconnaissance effectiveness evaluation system of the present invention;
[0021] Figure 3 It is the deep neural network of the present invention. DETAILED DESCRIPTION
[0022] In order to make the purpose, technical solutions and advantages of the present invention clearer and more specific, the present invention will be described in further detail below with reference to the accompanying drawings. Figure 1 As shown, the specific steps of the present invention are as follows:
[0023] Step 1: Build a radar air defense reconnaissance effectiveness evaluation system.
[0024] From the perspective of focusing on measuring the safety, reliability and mission execution of active radar systems, the radar air defense reconnaissance effectiveness evaluation index system is divided into four evaluation dimensions: equipment credibility, battlefield survivability, signal detection capability, and signal processing capability. The four evaluation dimensions cover different contents and aspects, which are described in detail as follows:
[0025] a) Equipment credibility is a measure of the ability of radar reconnaissance equipment to be in good condition when it starts working at a certain moment. It can be divided into three indicators: ① equipment reliability, ② fault detection capability, and ③ fault repair capability.
[0026] b) Battlefield survivability is used to characterize the safety performance of reconnaissance equipment during combat, and can be divided into four indicators: ① battlefield electromagnetic interference resistance, ② battlefield interference discrimination capability, ③ battlefield concealment capability, and ④ battlefield mobility capability.
[0027] c) Signal detection capability is used to characterize the ability of radar reconnaissance equipment to detect and discover signals during combat, which can be divided into three indicators: ① signal detection range, ② signal resolution capability, and ③ signal detection frequency domain.
[0028] d) Signal processing capability is used to characterize the radar reconnaissance system's ability to process and analyze received pulse signals, which can be divided into three indicators: ① target recognition capability, ② signal recognition accuracy, and ③ signal stable tracking capability.
[0029] Each evaluation indicator can be quantified by multiple evaluation factors, as shown below:
[0030] a)① Equipment reliability The evaluation factors include mission duration T, radar reconnaissance equipment mean time between failures t, mean fault detection time T R and mean time to repair T r ② Fault detection capability is measured using the mean fault detection time T R The evaluation factors include the mean fault detection time T R ③Fault repair capability is measured by the mean fault repair time T r Indicates that the evaluation factors include the mean time to repair a fault T r ;
[0031] b)① The battlefield electromagnetic interference resistance is expressed by the anti-interference coefficient of radar reconnaissance equipment, that is, The evaluation factors include the receiver sensitivity S of the radar reconnaissance equipment under interference-free conditions or laboratory conditions. r , receiver sensitivity S under interference conditions r ';② battlefield anti-interference discrimination capability denoted by κ1+κ2+κ3+κ4, where the evaluation factors include the probability PA of the radar reconnaissance equipment being able to correctly identify the target signal under interference-free conditions and the probability PB of the radar reconnaissance equipment being able to correctly identify the target signal under interference conditions; ③ battlefield concealment capability is denoted by κ1+κ2+κ3+κ4, where the evaluation factors include the weighting coefficient κ1 of the radar reconnaissance system's control working time and frequency, the weighting coefficient κ2 of the complexity of the emission waveform, the weighting coefficient κ3 of the deception of the added decoy, and the weighting coefficient κ4 of the radar maneuverability; ④ battlefield maneuverability is denoted by It indicates that the evaluation factors include the maneuvering speed υ of the radar reconnaissance equipment under experimental conditions and the maneuvering speed V of the radar reconnaissance equipment in a battlefield environment.
[0032] c)① Signal detection range is the maximum distance that radar reconnaissance equipment can operate, expressed as min(R max , R s ) indicates that K is the Boltzmann constant, usually 1.33×10 -23 J / K, R0 is the radius of the earth, usually 6370 km, and the evaluation factors include the radar's transmission power P t , the gain G of the radar transmitting antenna t, radar emission wavelength λ, effective reflection area of the reconnaissance target σ, receiver noise temperature T0, receiver bandwidth Δf r , noise factor F n , system loss factor L, radar minimum detection signal-to-noise ratio (S N ) min , radar reconnaissance equipment antenna height H a and the height of the reconnaissance target H t ;②Signal resolution capability Indicates that the evaluation factor includes the frequency resolution f s and azimuth resolution θ s ;③The signal detection frequency domain is expressed by the frequency coverage coefficient, that is, The evaluation factors include the frequency range f of the device, the instantaneous operating bandwidth Δf of the device, and the operating frequency band f of the radar receiver. r and the instantaneous operating bandwidth f of the radar receiver r ;
[0033] d)① The target recognition capability is expressed by the correct recognition probability of the signal, that is, The evaluation factor includes the number of correctly identified signals N ζ and the number of signals N reflected from the target γ ;②Signal recognition accuracy Indicates that the evaluation factor includes the frequency resolution E fs , azimuth resolution Signal pulse width and the signal arrival time error ③ Signal stable tracking capability uses the three-dimensional azimuth error E after integrating multiple tracks of the target x 、E y 、E z To express, the evaluation factor includes the tracking target horizontal coordinate error E x , tracking target vertical coordinate error E y and the vertical coordinate error E of the tracking target z .
[0034] Step 2: Construct a deep neural network.
[0035] like Figure 3 As shown in the figure, the designed deep neural network consists of five subnets. First, the equipment credibility assessment subnet, battlefield survivability assessment subnet, signal detection capability assessment subnet, and signal processing capability assessment subnet output are parallelized, and then the overall combat effectiveness assessment subnet is cascaded. The specific functions and structure are described as follows:
[0036] a) Equipment credibility assessment subnet
[0037] The device credibility evaluation subnet mainly evaluates the device credibility. It has three input nodes, namely device reliability, fault detection capability, and fault repair capability; and one output node, namely device credibility.
[0038] b) Battlefield Survivability Assessment Subnet
[0039] The battlefield survivability assessment subnet mainly assesses battlefield survivability. It has four input nodes, namely battlefield electromagnetic interference immunity, battlefield interference discrimination capability, battlefield concealment capability, and battlefield mobility capability; and one output node, namely battlefield survivability.
[0040] c) Signal detection capability assessment subnet
[0041] The signal detection capability evaluation subnet mainly evaluates the signal detection capability. It has three input nodes, namely signal detection range, signal resolution capability and signal detection frequency domain; and one output node, which is signal detection capability.
[0042] d) Signal processing capability evaluation subnet
[0043] The signal processing capability evaluation subnet mainly evaluates the signal processing capability. It has three input nodes, namely target recognition capability, signal recognition accuracy, and signal stable tracking capability; and one output node, which is the signal processing capability.
[0044] e) Overall combat effectiveness assessment subnet
[0045] The overall combat effectiveness evaluation subnet mainly evaluates the overall combat effectiveness of the system. It has four input nodes, namely equipment credibility, battlefield survivability, signal detection capability and signal processing capability; there is one output node, which is the overall combat effectiveness of the system.
[0046] It should be noted that the number of hidden layer nodes in each subnet is determined by the following formula:
[0047]
[0048] Among them, N s Indicates the number of hidden layer nodes in the subnet, N x is the number of input layer nodes of the subnet, N y is the number of nodes in the output layer of the subnet, and m is usually an integer between 0 and 10.
[0049] The number of hidden layer nodes, the number of hidden layers (selected in the range of 1-10), the activation function, and the training function of each subnet are determined by comparing the loss function value and the number of training times in turn. The specific method is as follows:
[0050] a) Compare the loss function values. The smaller the loss function value, the better. For example, the loss function value obtained by the activation function Sigmoid function is Esigmoid , the loss function value obtained by the activation function tanh function is E tanh , if E sigmoid <E tanh , then choose Sigmoid function as the activation function; if E sigmoid =E tanh , then go to step b); if E sigmoid >E tanh , then choose tanh function as the activation function.
[0051] b) Compare the number of training times. The smaller the number of training times, the better. For example, when using the activation function Sigmoid function, the number of training times is L sigmoid When using the activation function tanh function, the number of training times is L tanh , if L sigmoid <L tanh , then choose Sigmoid function as the activation function; if L sigmoid =L tanh , then either Sigmoid function or tanh function can be selected as activation function; if L sigmoid >L tanh , then choose tanh function as the activation function.
[0052] Step 3: Train the deep neural network.
[0053] 3.1) Data Preprocessing
[0054] Using the standard deviation method Figure 2 The 13 evaluation indicators are normalized, namely
[0055]
[0056] Among them, x' represents the evaluation index after normalization, x represents the evaluation index before normalization, represents the mean of the evaluation index, and s represents the standard deviation of the evaluation index.
[0057] right Figure 2 The four evaluation dimensions are assigned and interpreted as follows:
[0058] a) Equipment credibility
[0059] When the output is [0, 0.4], it indicates that the device reliability is extremely low; when it is (0.4, 0.6], it indicates that the device reliability is low; when it is (0.6, 0.8], it indicates that the device reliability is high; when it is (0.8, 1], it indicates that the device reliability is extremely high.
[0060] b) Battlefield survivability
[0061] When the output is [0, 0.4], it indicates that the battlefield survivability is extremely weak; when it is (0.4, 0.6], it indicates that the battlefield survivability is relatively weak; when it is (0.6, 0.8], it indicates that the battlefield survivability is relatively strong; when it is (0.8, 1], it indicates that the battlefield survivability is extremely strong.
[0062] c) Signal detection capability
[0063] When the output is [0, 0.4], it indicates that the signal detection capability is extremely weak; when it is (0.4, 0.6], it indicates that the signal detection capability is relatively weak; when it is (0.6, 0.8], it indicates that the signal detection capability is relatively strong; when it is (0.8, 1], it indicates that the signal detection capability is extremely strong.
[0064] d) Signal processing capabilities
[0065] When the output is [0, 0.4], it indicates that the signal processing capability is very weak; when it is (0.4, 0.6], it indicates that the signal processing capability is relatively weak; when it is (0.6, 0.8], it indicates that the signal processing capability is relatively strong; when it is (0.8, 1], it indicates that the signal processing capability is extremely strong.
[0066] e) Overall combat effectiveness
[0067] When the output is [0, 0.4], it means that the overall combat effectiveness of the radar reconnaissance equipment is very poor; when it is (0.4, 0.6], it means that the overall combat effectiveness of the radar reconnaissance equipment is average; when it is (0.6, 0.8], it means that the overall combat effectiveness of the radar reconnaissance equipment is good; when it is (0.8, 1], it means that the overall combat effectiveness of the radar reconnaissance equipment is very good.
[0068] 3.2) Training subnet
[0069] The back-propagation algorithm is used to train each sub-network layer by layer. When the loss function is less than the preset threshold or the maximum number of training times is reached, the training ends. Otherwise, the weight parameters and learning rate of each sub-network are adjusted along the gradient descent direction of the loss function, and the training is repeated. During the training process, the loss function is determined by the following formula:
[0070]
[0071] Among them, E(t) is the loss function value at t training times, y'(t) is the sample output value at t training times, and y(t) is the subnet output value at t training times.
[0072] It should be noted that each subnet adopts an adaptive learning rate during training, which is determined by the following formula:
[0073]
[0074] Among them, λ(t+1) is the learning rate during the t+1 training, λ(t) is the learning rate during the t training, E(t) is the loss function value during the t training, and E(t-1) is the loss function value during the t-1 training.
[0075] Step 4: Evaluate the effectiveness of radar air defense reconnaissance.
[0076] The normalized data to be evaluated is fed into the trained deep neural network to output the radar air defense reconnaissance effectiveness evaluation results.
[0077] The effect of the present invention can be further illustrated by simulation:
[0078] 1. Simulation Conditions: Consider a reconnaissance radar system composed of various ground-to-air and air-to-ground radars, capable of detecting aircraft and mobile equipment at ranges up to 100 kilometers. Evaluate the radar's operational effectiveness by combining operational data from the radar with data from the targets being detected.
[0079] 2. Simulation content: The radar air defense reconnaissance effectiveness evaluation method based on deep learning of the present invention and the hierarchical analysis method are used to simulate and compare the radar air defense reconnaissance effectiveness. The results are shown in Table 1.
[0080] Table 1
[0081]
[0082] The simulation results show that after adding a certain amount of noise, the error of the results obtained by the AHP evaluation method is greater than 0.01, and the greater the noise, the greater the error. The error of the evaluation results obtained using the deep learning-based radar air defense reconnaissance effectiveness evaluation method remains around 0.001, indicating that the deep learning-based radar air defense reconnaissance effectiveness evaluation method is much more fault-tolerant than the AHP method.
Claims
1. A radar air defense reconnaissance effectiveness evaluation method based on deep learning, characterized in that: The following steps are involved: (1) Construct a radar air defense reconnaissance effectiveness evaluation system that includes four evaluation dimensions, 13 evaluation indicators, and 42 evaluation factors; (2) Construct a deep neural network. The deep neural network consists of four subnets: equipment credibility assessment subnet, battlefield survivability assessment subnet, signal detection capability assessment subnet, and signal processing capability assessment subnet. The subnets are output in parallel and then cascaded to form an overall combat effectiveness assessment subnet. (3) Training deep neural networks, including: 3a) Collect evaluation factors and calculate evaluation indicators; 3b) The standard deviation method was used to normalize the 13 evaluation indicators and the 4 evaluation dimensions were assigned values and interpreted; 3c) Using the back propagation algorithm to train each subnet of the deep neural network layer by layer, the training ends when the loss function is less than the preset threshold or reaches the maximum number of training times; otherwise, the weight parameters of each subnet are adjusted along the gradient descent direction of the loss function, the learning rate is adaptively adjusted, and each subnet is retrained; (4) The normalized data to be evaluated is fed into the trained deep neural network to output the radar air defense reconnaissance effectiveness evaluation results; The four evaluation dimensions in step (1) refer to equipment credibility, battlefield survivability, signal detection capability and signal processing capability. The four evaluation dimensions involve a total of 13 evaluation indicators, specifically: Equipment credibility involves three evaluation indicators: equipment reliability, fault detection capability, and fault repair capability; Battlefield survivability involves four evaluation indicators: battlefield electromagnetic interference resistance, battlefield interference discrimination, battlefield concealment, and battlefield mobility. Signal detection capability involves three evaluation indicators: signal detection range, signal resolution, and signal detection frequency domain; Signal processing capability involves three evaluation indicators: target recognition capability, signal recognition accuracy, and signal stable tracking capability; The 13 evaluation indicators in step (1) are quantitatively expressed by 42 evaluation factors, as follows: Equipment reliability Indicates that T is the mission duration T, t is the mean time between failures of radar reconnaissance equipment, T R is the average fault detection time of radar reconnaissance equipment, T r The mean time to repair failures of radar reconnaissance equipment; Fault detection capability Average fault detection time T of radar reconnaissance equipment R express; Fault repair capability: Average fault repair time T of radar reconnaissance equipment r express; Battlefield electromagnetic interference resistance Indicates that S r The receiver sensitivity S of the radar reconnaissance equipment under interference-free conditions or laboratory conditions r , S r ′ is the receiver sensitivity S under interference conditions r ′; Battlefield anti-interference discrimination capability Indicates that PA is the probability that the radar reconnaissance equipment can correctly identify the target signal under non-interference conditions, and PB is the probability that the radar reconnaissance equipment can correctly identify the target signal under interference conditions; The battlefield concealment capability is expressed by κ1+κ2+κ3+κ4, where κ1 is the weighted coefficient of the radar reconnaissance system control working time and frequency, κ2 is the weighted coefficient of the complexity of the emission waveform, κ3 is the weighted coefficient of the deception of the added bait, and κ4 is the weighted coefficient of the radar mobility. Battlefield mobility Indicates that υ is the maneuvering speed of the radar reconnaissance equipment under experimental conditions, and V is the maneuvering speed of the radar reconnaissance equipment under battlefield conditions; Signal detection range is expressed as min(R max , R s ) indicates that K is the Boltzmann constant, which is 1.33×10 -23 J / K, R0 is the radius of the earth, take 6370Km, P t is the radar transmission power, G t is the gain of the radar transmitting antenna, λ is the radar transmitting wavelength, σ is the effective reflection area of the reconnaissance target, T0 is the receiver noise temperature, Δf r is the receiver bandwidth, F n is the noise coefficient, L is the system loss factor, (S N ) min is the minimum detection signal-to-noise ratio of the radar, H a is the antenna height of radar reconnaissance equipment, H t To detect the target height; Signal resolution It means that f s is the frequency resolution, θ s is the azimuth resolution; The signal detection frequency domain is expressed by the frequency coverage factor, that is, f is the frequency range of the device, Δf is the instantaneous working bandwidth of the device, and f r is the radar receiver operating frequency band, Δf r is the instantaneous working bandwidth of the radar receiver; Target recognition capability Indicates that N ζ is the number of correctly identified signals, N γ is the number of all signals reflected from the target; Signal recognition accuracy express, is the frequency resolution, is the azimuth resolution, is the signal pulse width, is the error of the echo signal arrival time; The signal stability tracking capability is based on the three-dimensional position error E after integrating multiple tracks of the target. x 、E y 、E z To indicate that E x is the horizontal coordinate error of the tracking target, E y is the tracking target ordinate error, E z is the vertical coordinate error of the tracking target.
2. The radar air defense reconnaissance effectiveness evaluation method based on deep learning according to claim 1 is characterized in that: The specific functions and structures of the equipment credibility assessment subnet, battlefield survivability assessment subnet, signal detection capability assessment subnet, signal processing capability assessment subnet and overall combat effectiveness assessment subnet in step (2) are described as follows: 4a) Equipment credibility evaluation subnet evaluates equipment credibility. It has three input nodes: equipment reliability, fault detection capability, and fault repair capability. The output node is equipment credibility. 4b) The battlefield survivability assessment subnet assesses battlefield survivability. It has four input nodes: battlefield electromagnetic interference resistance, battlefield interference discrimination, battlefield concealment, and battlefield mobility. The output node is battlefield survivability. 4c) The signal detection capability evaluation subnet evaluates the signal detection capability. It has three input nodes: signal detection range, signal resolution capability, and signal detection frequency domain. The output node is the signal detection capability. 4d) The signal processing capability evaluation subnet evaluates the signal processing capability. It has three input nodes: target recognition capability, signal recognition accuracy, and signal stable tracking capability. The output node is signal processing capability. 4e) The overall combat effectiveness evaluation subnet evaluates the overall combat effectiveness of the system. It has four input nodes: equipment credibility, battlefield survivability, signal detection capability and signal processing capability, and the output node is the overall combat effectiveness of the system.
3. The radar air defense reconnaissance effectiveness evaluation method based on deep learning according to claim 1 is characterized in that: The formula for normalizing the 13 evaluation indicators using the standard deviation method in step 3b) is: Among them, x' represents the evaluation index after normalization, x represents the evaluation index before normalization, It represents the mean of the evaluation index, and s represents the standard deviation of the evaluation index.
4. The method according to claim 1, characterized in that In step 3b), the four evaluation dimensions are assigned values and interpreted. The specific meanings of the assignments are as follows: 6a) When the device credibility output is [0,0.4], it means that the device credibility is extremely low; when the output is (0.4,0.6], it means that the device credibility is low; when the output is (0.6,0.8], it means that the device credibility is high; when the output is (0.8,1], it means that the device credibility is extremely high; 6b) When the battlefield survivability output is [0,0.4], it means that the battlefield survivability is extremely weak; when the output is (0.4,0.6], it means that the battlefield survivability is relatively weak; when the output is (0.6,0.8], it means that the battlefield survivability is relatively strong; when the output is (0.8,1], it means that the battlefield survivability is extremely strong; 6c) When the signal detection capability output is [0,0.4], it indicates that the signal detection capability is extremely weak; when the output is (0.4,0.6], it indicates that the signal detection capability is relatively weak; when the output is (0.6,0.8], it indicates that the signal detection capability is relatively strong; when the output is (0.8,1], it indicates that the signal detection capability is extremely strong; 6d) When the signal processing capability output is [0,0.4], it indicates that the signal processing capability is extremely weak; when the output is (0.4,0.6], it indicates that the signal processing capability is relatively weak; when the output is (0.6,0.8], it indicates that the signal processing capability is relatively strong; when the output is (0.8,1], it indicates that the signal processing capability is extremely strong; 6e) When the overall combat effectiveness output is [0,0.4], it means that the overall combat effectiveness of the radar reconnaissance equipment is very poor; when the output is (0.4,0.6], it means that the overall combat effectiveness of the radar reconnaissance equipment is average; when the output is (0.6,0.8], it means that the overall combat effectiveness of the radar reconnaissance equipment is good; when the output is (0.8,1], it means that the overall combat effectiveness of the radar reconnaissance equipment is very good.
5. The radar air defense reconnaissance effectiveness evaluation method based on deep learning according to claim 1 is characterized in that: The error function of the deep neural network in step 3c) is: Among them, E(t) is the loss function value at the tth training, y'(t) is the sample output value at the tth training, and y(t) is the subnet output value at the tth training.
6. The radar air defense reconnaissance effectiveness evaluation method based on deep learning according to claim 1 is characterized in that: The learning rate of the adaptively adjusted deep neural network in step 3c) is: Among them, λ(t+1) is the learning rate during the t+1 training, λ(t) is the learning rate during the t training, E(t) is the error function value during the t training, and E(t-1) is the error function value during the t-1 training.
7. The radar air defense reconnaissance effectiveness evaluation method based on deep learning according to claim 2 is characterized in that: The number of hidden layer nodes of each subnet is determined by the following formula: Among them, N s Represents the number of hidden layer nodes in the subnet, N x is the number of nodes in the input layer of the subnet, N y is the number of nodes in the output layer of the subnet, and m is an integer between 0 and 10.
