The military-industrial scientific research system of the academic master

Perhaps it was due to the attempt when he was just reborn, or perhaps it was due to the fact that his management ability had improved from a blank LV0 to LV1.

In short, the subsequent formation of the research team and the process of assigning tasks to the 606 Institute and the 410 Factory were much smoother than Chang Haonan's own imagination.

After all, the abilities and experience of the engineers who can work in these places have been tested, which is much easier than bringing two undergraduates, Zhou Shuwan and Zhang Man.

When introducing calculation methods and experimental content, it is basically easy to understand.

So that night, the team responsible for the research on the forward sweep of the upper end of the first part of the moving blade and the positive bending of the lower end had already started working.

After assigning the last part of the task, Chang Haonan finally dragged his tired body out of the temporarily assigned office and prepared to go back to his residence in No. 601 to take a rest.

From the day the No. 01 prototype had an accident until now, he has not had a sound sleep.

There are still many problems left for him to solve, and many parts are not simple even with the help of the system. Even if the experimental part has been stripped out, the numerical simulation part alone cost him 50 points.

The main limitation is naturally the level of computers in this era.

The flow field complexity involved in each of the plans he gave is different. The swept blades have modifications with different sweep heights and sweep angles. The curved moving blades also have modifications with different bending heights and bending angles. Without going through the system, Before optimization, the number of automatically generated meshes for each blade was around 900,000.

Not to mention 1996, even in 2006, it is definitely a number that can make people's blood pressure immediately rise after hearing it.

And the pressure on this matter is greater than previous improvements to the wing design.

While the No. 04 prototype is being refitted with a new wing, the two prototypes No. 01 and No. 03 are still undergoing non-stop test flights, but certain adjustments are needed to the test subjects.

However, the engine, as the heart of the aircraft, had an accident as big as the last one, which almost resulted in the complete loss of the No. 01 prototype. It was obviously impossible to continue the test flight until the problem was solved.

Every day spent delaying this will cause immeasurable damage.

"call--"

Chang Haonan exhaled a breath of turbid air, feeling the already cool evening breeze at night in late August in the north, and couldn't help but quicken his pace.

Just as he was about to enter the dormitory building, the guard on duty at the door suddenly called him from behind.

"Regular worker!"

Although Chang Haonan is not a formal employee of Institute 601, most non-technical staff cannot figure out the more complicated twists and turns, so they are basically referred to as x work.

"Is there something wrong with this comrade?"

Chang Haonan stopped when he heard his name, turned around, and saw a security guard in uniform who looked about the same age as him running out of the guard box with something in his hand.

"This is a letter sent to you. Because I didn't write the specific dormitory number, I left it with the doorman."

"letter?"

Chang Haonan took the brown paper envelope from the other party in a daze, and could not see clearly what was written on it in the dim light.

Not many people knew about his work in Institute 601, and those who knew about it seemed to be able to contact him directly through Yang Fenghata.

"Excuse me, comrade."

"It doesn't matter. Just come to me and sign to confirm receipt."

A few minutes later.

With a soft click, Chang Haonan opened the dormitory door.

Usually after coming back exhausted, the first thing he does is to throw himself on the bed and lie down for half an hour without thinking.

But today, after a brief struggle when going upstairs, Chang Haonan decided that he still had to see what was going on with the letter.

Light the desk lamp.

Although several months have passed since his rebirth, and he has gradually become accustomed to this era without mobile Internet, seeing a letter with a stamp still makes him feel a strange feeling.

The sender column is filled with "Xu Yang, Institute of Computing Technology, Chinese Academy of Sciences."

The pen writing is very beautiful.

"No wonder……"

Chang Haonan said to himself while opening the letter.

We know that he works in the 601 office, but there is no way to contact him by phone and we need to send a letter. It seems that he is the only one.

The folded letter contained nearly 20 pages, and the thickness made him frown.

"Forget it, tear it all apart, let's take a look at what she wrote and then rest..."

Because he opened the letter so quickly, he didn't even have time to recall whether some bloody plot had flashed through his mind at that moment, before he saw the title at the top of the first page.

"Research on an extended Kalman filter algorithm based on improved differential evolution algorithm"

Abstract: The extended Kalman filter has been widely used in various fields such as engineering practice. However, this algorithm brings errors due to the assumption that the process noise is fixed, and the accuracy is not high. An improved EKF algorithm is studied, which mainly improves the filtering accuracy by assuming that the process noise is obtained by filtering results and observation results, and then using a differential evolution algorithm to optimize the obtained process noise variance...

The results show that the improved extended Kalman filter algorithm reduces the state estimation bias and obtains a relatively ideal filtering effect...

"hiss--"

"I kind of regret opening this letter..."

Chang Haonan rubbed his swollen temples, put the letter on the table, got up and left the bedroom.

He was going to make a cup of strong tea.

I'm afraid tonight's rest will be ruined again.

…

A few hours later.

Chang Haonan looked at the letter paper that had been marked with red and blue pencils and the notebook full of writing next to it, and drank the last sip of the bitter tea in the thermos cup.

It can only be said that Xu Yang is indeed a doctor who returned from UCB and personally participated in the development process of the differential evolution algorithm.

Based on the level of ability shown by the other party, it doesn't seem to be an exaggeration to say that she has a system similar to mine.

The ability to draw inferences from one example is quite powerful.

The content of this letter is not too complicated:

After seeing Chang Haonan's improvements to the differential evolution algorithm, Xu Yang proposed a new idea and wanted to discuss the feasibility and application prospects with the latter.

Simply put, filtering is the operation of filtering out specific frequency bands in signals. It is an important measure to suppress and prevent interference. It is almost an essential process for all behaviors that require the emission of electromagnetic signals.

In fact, the least squares method learned in high school mathematics is a classic filtering method, but since it only considers the relationship between measurement and state, the stability of this method is not good.

Later, the Kalman filter method was born for linear systems whose noise obeys Gaussian distribution. Due to its high efficiency and accuracy, it soon became one of the most widely used filtering methods in engineering.

However, because the original Kalman filtering method was only applicable to linear systems, many people later made certain expansions on this basis so that this well-received filtering algorithm can also be used in nonlinear systems.

The most typical one is the extended Kalman filter algorithm (EKF), which first uses Taylor series to expand the system equations and ignores second-order and higher-order terms above the second order. At this time, the nonlinear system is close to a linear form. , and then use the standard Kalman filter algorithm to estimate the state of the system.

Obviously, the calculation amount of the EKF algorithm is not large, and it is not difficult to implement. However, EKF is prone to serious oscillations and even divergence in strongly nonlinear systems.

By the mid-1990s, the engineering community had basically seen the limits of the EKF algorithm, and began to think about improving the EKF algorithm or creating a new Kalman filter algorithm.

The new algorithm proposed by Xu Yang uses an improved differential evolution algorithm to select the variance of the process noise before filtering, which greatly reduces the state estimation deviation and can rely on a slightly larger calculation amount than the traditional EKF algorithm. Achieve accuracy far beyond that of the EKF algorithm.

As an aerospace engineer, Chang Haonan's previous understanding of filtering algorithms was limited to the application level, and he did not have a deep understanding of the basic principles. However, if he only looked at the process of optimizing process noise through the differential evolution algorithm, it should be no problem.

He raised his head and looked at the horizon that was already turning fish belly white. The tiredness that had just been suppressed by the strong tea finally came over.

Chang Haonan decided to take a break and find a few books to study thoroughly before making a reply.

On the one hand, the optimization work on the blade shape design of the turbojet 14 compressor has just begun. Most of the work at this stage is completed by engineers from the 606 Institute. Moreover, because of the existence of the system, the code written by Chang Haonan does not require debugging as long as the idea is correct. This ability that envies countless programmers determines that the most time-consuming part for him is not programming, but waiting for calculation results, so there is Enough time to do it.

On the other hand, although the algorithm Xu Yang came up with has little effect on the current engine improvement, it is the basis for a series of technologies such as precision-guided weapon strapdown seekers, radar data links, and missile inertial navigation system calculations. They are all closely related to the Pili 11 semi-active radar-guided air-to-air missile, a copy of the A-bomb that Chang Haonan was previously pessimistic about.

Even if it is organic and has no ammunition, it cannot form combat effectiveness.

If it can be realized, it may be of great help in solving the problem of insufficient target test accuracy that has plagued it for a long time.

Just give me a pillow when you feel sleepy.

"Perhaps the weapon system of Project 83 can also see hope within this year..."

This was Chang Haonan's last thought before falling asleep.

…

The next morning, Chang Haonan left the dormitory after washing up and prepared to go straight to the library, borrow some materials and continue to study the idea proposed by Xu Yang.

Especially the second half related to Kalman filtering.

But when I went out with full of expectations, I happened to meet Yao Mengna who had also just opened the door to the room.

The two looked at each other.

"Why is Team Leader Chang leaving so late? Don't you need to prepare meeting materials in advance?"

Yao Mengna locked the door with a hint of teasing in her tone.

After hearing these words, Chang Haonan was stunned for a moment, and then slapped his forehead heavily.

Being a manager requires getting used to. Having been a screwdriver for decades, he already has little experience in this area. In addition, after reading letters all night, his mind was filled with algorithm-related things, which made him almost forget about today. About the morning meeting.

He even proposed to hold this meeting last night.

Fortunately, I was reminded by Yao Mengna when I went out today, and the entire task allocation process was in Chang Haonan's mind, so it would not delay the important matter.

However, he still controlled himself and did not show any annoyance. He just took a breath and pointed at his head.

"No need, I have all the knowledge in my head. Just start assigning tasks when I get there."

…

However, the duration of the final morning meeting exceeded Chang Haonan's estimate.

It was not until after lunch time that he found the information he wanted from the library and returned to a computer room specially set aside for him in the computing center.

Now that he knew clearly that the idea was feasible, Chang Haonan didn't even need to use the system. He quickly found the part he wanted from the pile of information he received.

In fact, with the current system level, there is no way to carry out two projects at the same time, so he can only rely on himself this time.

For almost a whole afternoon, Chang Haonan almost kept analyzing and calculating ideas on draft paper.

"often……"

When Yao Mengna opened the door and walked in to ask him to have dinner, she only saw a figure writing furiously from behind.

After hesitating for a moment, she pursed her lips and gently closed the door again.

At this time, Chang Haonan had completed the entire calculation process and took out a stack of letter paper from the side to write a reply.

Even if he were to read it by himself the next day, he might not be able to understand the rather scrawled content on those draft papers.

On the original timeline, the unscented Kalman filter algorithm (UKF) was developed to solve the problem of large linear truncation errors in the EKF algorithm.

Chang Haonan's original idea was to use this algorithm combined with the noise variance optimization proposed by Xu Yang, which should be a very important breakthrough in the field of information extraction.

However, as his understanding of the algorithm deepened, he discovered that the theoretical mathematical derivation process of the UKF algorithm was not rigorous, and after several calculations and simulations, it was proved that the numerical value in a high-dimensional state was still not stable enough.

In the process of learning and researching the Kalman filter method, a flash of inspiration gave him a more subtle idea:

"Huishu appreciates that the idea of ​​using the improved differential evolution algorithm to automatically adjust the variance matrix of the process noise matrix is ​​quite novel. After numerical simulation calculations, it has been proved that it has good estimation accuracy, but there are still some problems in the EKF filtering algorithm itself. question……"

“Since EKF needs to obtain the Jacobian matrix, the specific form of the nonlinear function must be known exactly, and black box encapsulation cannot be achieved.”

"Since the derivatives of nonlinear functions of the state equation and observation equation cannot be obtained in many practical problems, it is difficult to obtain the Jacobian matrix."

"..."

"According to Bayesian theory and Spherical-Radial Cubature rules, the integral in the Cartesian coordinate system can be converted into a spherical-radial integral, and a set of deterministic sampling points can be used to perform state estimation through the propagation of nonlinear equations. Specific steps are as follows:"

"The first step: determine the initial value of the filter: X0=E(X), P0=E[(X0-X0)(X0-X0)^T]..."

"The second step is to calculate the volume point at time k-1, P(k-1|k-1)=S(k-1)S(k-1)^T, where S(k-1) passes through P( k-1|k-1) Choleski decomposition is obtained. For the volumetric Kalman filter algorithm, the point set ξ and the corresponding weight w are..."

Time passed by minute by second. At the beginning, Chang Haonan still needed to write for a while and then stopped thinking for a while. However, as the calculation process progressed, his entire idea became more and more smooth, and he seemed to have entered a certain stage. A state similar to "flow".

So when he finally wrote "Reply now, leave the details later" at the end of the letter and signed his name, he saw the dinner packed by someone on the table next to him.

"I always thought that forgetting about food and sleep was a figure of speech..."

Chang Haonan thought of himself who forgot to sleep last night and forgot to eat tonight. He smiled and shook his head, then folded the letter and put it aside.

When opening the lunch box, he glanced at his system panel and then froze on the spot.

[Scientific research points: 110

Theoretical level: LV2 (500/10000)

Engineering experience: LV1 (270/1000)

Management ability: LV1 (100/1000)

Comprehensive ability: LV1

Projects in progress: 1

Possible projects: 0]

Because this is the first time since his rebirth that his [theoretical level] experience has increased.

And it’s 500 points at a time!

You know, the things he wrote have not even been read by the second person, and they have not had any real impact yet.

"It seems that I have done a big job this time..."

Chang Haonan leaned back on the chair, opened the disposable chopsticks, rubbed the sore back of his neck and muttered to himself.

I was driving late on the road just now. I always thought that the UKF algorithm was proposed in 2000, but today I saw that it was proposed in 1991, so the subsequent plot about the improvement of the Kalman filter method has to be adjusted, but now the details have been Straighten it out again, no problem.