Research on a method of completeness index based on complex model

The complete distance E_X is the target solution parameter function X_at and the current solution parameter function X_t. Therefore, from the perspective of physical meaning, E_X is mostly caused by the difference in data from the real world to the model, and this difference is often very obvious.

Taking the actual problem of crowd movement as an example, measuring crowd pressure is the main task of the social emergency system. Crowd pressure is determined by crowd flow rate. It uses crowd movement presentation media as input for multi-source information, physical modelling of roads and road use. The ratio of the width to the number of recognised people is used as the basic characteristic of the crowd flow rate so as to determine the current road crowd pressure characteristics and possible emergency challenges, as shown in formula (1). (1) Challenge(T)=F(Crowd,T)F(Width) Challenge\left(T \right) = {{F\left({Crowd,T} \right)} \over {F\left({Width} \right)}}

However, the real crowd is composed of participants with height and weight, and they are affected by personal movement characteristics during the actual exercise. This is quite different from the motion model that is simply abstracted as a mass point during the simulation environment and simulation object based on the mass point. Once the logic model is formed, it is difficult for the system to improve its fidelity. This is a self-constraint, which is determined by the self-closed loop characteristics of the information source, simulation environment and simulation object as shown in formula (2). (2) Challenge(T)=∑iTF(Personi,State(Personi))F(Width) Challenge\left(T \right) = \mathop {\sum\limits_i^T}{{F\left({Person_{i},State\left({Person_{i}} \right)} \right)} \over {F\left({Width} \right)}}

This process of abstracting from the prototype system of the real environment to the typical simulation system based on the mass point has produced cumulative errors, although some researchers try to use complete context logic, rich constraint functions and more detailed simulation object structures. There is an improvement, but this error still exists, that is, E_X has to be calculated.

Since X_at is difficult to obtain and describe in practical problems, such as the actual size of an object, X_t can be obtained through a series of measurement and acquisition methods, but E_X at this time cannot be directly calculated.

The algorithm sets a transfer function X_at ∈ x, X_t ∈ x, and its inverse function F(x), then the calculation formula XE¯ \overline {{X_E}} (3) can be obtained indirectly. (3) XE¯=F(f(Xat))−F(f(Xt)) \overline {{X_E}} = F\left({f\left({{X_{at}}} \right)} \right) - \;F\left({f\left({{X_t}} \right)} \right)

For E_at, the physical meaning in the real world is clear. For the same problem, the results are obtained by observing the real world and the simulation model, but the physical meaning is not limited to this. Take the observation orbit data and system as an example to calculate Eat¯ \overline {{E_{at}}} .

Obtain the trajectory of the output elements in the t time interval in the simulation system synchronised with the orbital running time interval of S_i to derive the simulation data SEA and the three-dimensional space coordinates of the two static objects corresponding to each keyframe of S_i, and use the distance formula between the two points to calculate the output distances R₁and R₂ of the output elements and two static objects in each key frame of S_i and the spatial distances S₁ and S₂ in the t time interval in the simulation system. According to a certain distance ratio verification condition, the simulation data and output elements are correspondingly compared data.

Define R₁ : S₁ as a₁ and R₂ : S₂ as a₂, then in the synchronisation environment, time interval t = {t₁, t₂,...,t_m} there are {a_1t} and {a_2t}, where {a_1t} has the form (4), {a_2t} has the form (5). (4) {a1t|R1t:S1t=a1t} \{{a_{1t}}|{R_{1t}}:{S_{1t}} = {a_{1t}}\} (5) {a2t|R2t:S2t=a2t} \{{a_{2t}}|{R_{2t}}:{S_{2t}} = {a_{2t}}\}

If the effects of REA and SEA in the simulation system are the same in the orbital time interval of S_i, the ratios of {a_1t} and {a_2t} in the orbital time interval t = {t₁, t₂,...,t_m} are equal, as shown in Figure 1.

The ratio between {a_1t} and {a_2t} at a specific moment is not equal to other moments. There is an error between the real spatial position of the output element at that moment and the simulation data in the simulation system, so that in the trajectory and simulation of the output element within the orbital time interval, there are errors in the data.

Ideally, if there is no error in the simulation deduction of the output element characterisation state, that is, the effects of RPA and SPA are the same, and the respective ratios of {a_1t} and {a_2t} are equal. The respective ratios of {a_1t} and {a_2t} are evaluated as r₁and r₂ r2Eat¯ {r_2}\overline {{E_{at}}} , reference value, at this time r₁=r₂=1, Eat¯ \overline {{E_{at}}} is 0. If the ratio of and {a_2t} at a certain time is not equal to other time, define the ratio of {a_1t} and {a_2t} at that time to the ratio of other time as r_1t and, r_2t and the ratio r_t of r_1t and r_2t is the time 1−Eat¯ 1 - \overline {{E_{at}}} .

The simulation model is constructed and compared with the actual performance results, the presentation process of the output elements is simulated, and the presentation process of the target problem is recorded at the same time.

In this research, a simulation system and peripheral components capable of observing environmental changes are constructed. It can model and store the physical characteristics, position and trajectory parameters of the scene environment. The control simulation of the system can obtain the physical trajectory and output simulation data.

The simulation model relies on the characteristics of the elements established by the simulation pipeline to simulate the characterisation state of the elements. For example, in the visualisation system, such as colour, position, volume characterisation, the continuous performance of the element characterisation at different times is selected. The characterisation state and vector characteristics in the trajectory running time interval.

Fig. 2

According to the interpolation process at the continuous time, the characterisation process of the element is constructed, in which the three-dimensional space coordinates and vector characteristics of the entity element in the complex environment at a specific time t are shown in formula (6). (6) seait(xit,yit,zit,rxit,ryit,rzit) sea_{{it}}\left({{x_{it}},{y_{it}},{z_{it}},r{x_{it}},r{y_{it}},r{z_{it}}} \right)

At this time, the set of three-dimensional space coordinates and vector features of all entity elements i={1, 2, 3,..., n} is shown in formula (7). (7) SEAt={sea1t,sea2t,…,seait,…,seant} SEA_{t} = \left\{{sea_{{1t}},sea_{{2t}}, \ldots,sea_{{it}}, \ldots,sea_{{nt}}} \right\}

In the running time interval o t = {t₁, t₂, t₃,...,t_m} f the entity elements in a period of time, the representation state position data set SEA of all the entity elements entity t travelling along the trajectory is given as in formula (8). (8) SEA={SEAt1,SEAt2,SEAt3,…,SEAtm} SEA = \left\{{SEA_{{t_1}},SEA_{{t_2}},SEA_{{t_3}}, \ldots,SEA_{{t_m}}} \right\}

In the data acquisition stage, by comparing the simulation data with the real source orbit data obtained by processing the imaging presentation media captured by the presentation media monitoring, it is determined that the data modelling corresponding to the simulation is required.

In the data comparison stage, the two-dimensional position coordinates of a single entity element rea defined at a specific time t on the medium orbit are rea_it (x_it, y_it) and the two-dimensional position coordinate set oREAf all entities presents the key to the medium. The data i = {1, 2, 3,...,n} in the frame is shown in formula (9). (9) REAt={rea1t,rea2t,…,reait,…,reant} REA_{t} = \left\{{rea_{1t},rea_{2t}, \ldots,rea_{it}, \ldots,rea_{nt}} \right\}

In the orbit time period t = {t₁, t₂, t₃,...,t_m}, when all the physical elements i are running along the orbit, the representation state position data set REA in the key frame sequence of the presentation medium is expressed by formula (10). (10) REA={REAt1,REAt2,REAt3,…,REAtm} REA = \left\{{REA_{{t_1}},REA_{{t_2}},REA_{{t_3}}, \ldots,REA_{{t_m}}} \right\}

In the calculation stage, the position of the simulated entity group is compared with the position of the characterising entity group, and the formalisation of Eat¯ \overline {{E_{at}}} is obtained, as shown in formula (11). (11) XE¯=1tm−t1∑i=t1tmREAtiSEAti \overline {{X_E}} = {1 \over {{t_m} - {t_1}}}\mathop {\sum\limits_{\;i = {t_1}}^{{t_m}}}\limits_ {{REA_{{t_i}}} \over {SEA_{{t_i}}}}

The actual problem trajectory tracking and simulation process is a typical complex actual problem. Use this as an experimental environment to verify the effect of the completeness distance on actual problems. The video data of the real orbit system and the simulation vector data of the virtual simulation system are input into the complex model, and the completeness distance E_X is used to describe the difference between the two systems, thus proving the effectiveness of E_X in describing the fidelity of the system.

This experiment studies the completeness of the target problem of the complex model using domain theory to prove several conditions for the completeness of the complex model, one of which is the expression of the distance E_X caused by the difference of the data; the distance can be passed through the transfer function f (x), which indirectly calculates the complete distance between simulation systems and can also test the morphological difference between the real environment and the simulation system, which can be used in the selection process of knowledge and the improvement process of the cognitive network model.

This experiment constructs the first set of experimental Table 1 to test the performance of the blue square model P₁ and the blue square model P₂ in different mid-markets and its completeness distance index E_X.

The experiment uses the Random Go nine-way Go system to compare the distance E_X between the complex model models P₁ and P₂. The blue model P₁ uses UCT and Monte Carlo algorithm for knowledge selection, and the blue model P₂ uses a 3×3 pattern recognition move algorithm for knowledge selection. At the same time, select the static game record of StoneBase 1000 mid-games for the distance E_X constraint comparison process, including the opening scenes F_0–4 (200), F_5–9 (200), F_10–14 (200), F_15–19 (200) and F_20–24 (200), etc. The difference of the completeness distance index under different middle game steps.

This experiment constructs the second set of experimental Tables 2 Test track simulation and observation system distance E_X, in the state of continuous 300 frames, the consistency of the movement of each entity. The study uses orbit simulation and observation system observation distance E_X to compare the distance E_X between the simulation system and the real world. This target problem uses the observation module to obtain the current state information in the vision and compares the actual simulation information of the simulation system to obtain the incomplete target distance.

This experiment constructs the third set of experimental Table 3 using the complex model cognitive network to generate video and real video for observation distance E_X. The research uses a cognitive network to generate video and real video to observe the distance E_X, and observe the completeness of the generated video target from the perspective of key feature tracking.

The experimental result in Table 4 shows that the placement order of the model is inconsistent with the placement order of StoneBase. This inconsistency is one of the characteristics of the difference between the learning network and the natural person. However, as the number of moves increases, the completeness of the distance keeps E_X shrinking. This shows that facing the placement process of the middle board of Go, the complex model and the cognition of natural persons gradually converge, indicating the effectiveness of the value evaluation function.

In addition, it is found that the algorithm based on pattern matching is more random when there are a few moves, that is, the distance E_X of P₂ is much larger than P₁, and as the moves increase, the stability of the distance E_X increases, indicating that the pattern structure is good. In the case of dense chess pieces, it is easier to match the pattern. The experiment also shows that the completeness distance E_X can effectively discover the differences between the models and lay the foundation for the improvement of the models.

Experimental results in Tables 5 and 6 are the synchronisation position gap between a certain track system simulation data and video acquisition data. Through the data, it can be found that the completeness distance difference of different entities is different. The value X_E for each entity that does not exceed 1% is relatively stable, which shows that the system simulation process has high completeness. From another perspective, the distance X_E may be caused by the error of the lens shake, the optical effect of the collecting lens and the signal noise. This shows that any complete system is a high degree of coordination between software and hardware, and neither is indispensable.

Table of experimental results 7. Figure of experimental results 5.1 The observation distance E_X between the complex model cognitive network generated video and the real video. This experiment uses 600 video framed images of the Cityscape dataset to generate three sets of generated data up to 300 frames.

From the overall effect of distance E_X, the average E_X value of the video generated by the cognitive network does not exceed 2.1%, where Create_1 is the average E_X value of the video generated when the 100th frame is held, Create_2 is the average E_X value of the video generated when the 200th frame is held, Create_3 is the average E_X value of the generated video when the 250th frame is maintained and Create_4 is the average E_X value of the generated video when the 275th frame is maintained. At a lower completeness distance, it indicates that the generated data is successful.

The experiment shows that as time increases, the distance E_X index of the three sets of data increases, and the two show a positive proportional relationship. This represents an increase in the difference between the generated data and the original data and also represents the creativity of the cognitive network. Experiments show that the generated data 3 greatly increases in the distance E_X, which is caused by the generation of bad data. It also proves that the cognitive network still needs someone to participate and guide. These experiments all prove that the distance E_X is an accurate completeness index.

Fig. 3