TWI306207B - - Google Patents

Description

1306207 IX. Invention: [Technical field of invention] The invention relates to a method and device for intelligently designing a heat dissipation opening of a computer casing, which is based on a neural network system and estimates the design of different heat dissipation openings. The impact of computer cooling performance will be able to make accurate and effective analysis and judgment for the heat dissipation solution of the widely used desktop electric chassis, and to quickly optimize the method of making the domain shell The upper computer as a whole achieves more efficient cooling performance. Φ [Prior Art] In recent years, due to the rapid development of warm electronic products (10), electronic products have gradually evolved toward high performance, high frequency, high speed and light and thin, and the chip size has been gradually miniaturized, resulting in the heat density of electronic components. Increasingly high, when Intel (INTEL) and American Supermicro (AMD) introduced micro-processing of pENT dirty 4 or above, the power of micro-processing f was found, and the pulse was increased with the operation of the microprocessor. And the internal added function lines have risen sharply; but the increase in consumption and the progress of the microprocessor process are getting smaller and smaller, and the microprocessor emits not only the increase in heat but also the appeal. 'There is also a rapid increase in heat flux, which causes the temperature of the microprocessor to rise rapidly. Because the microprocessor is _ crystal_domain, its temperature must be below the fixed temperature, and the connection of the line or The operation of the wafer will also be difficult due to the long exposure to the high temperature environment. Therefore, the reading environment in Fu must be maintained at an allowable temperature range. Above this temperature, the microprocessor cannot be cured or even burned. Possible Therefore, the heat dissipation of electronic products can be said to be an important factor for the current electronics industry to determine the stability of their products! It can also be seen that the heat-dissipation-related products are located in the electronic industry, in the middle and downstream links, and cannot be neglected. - Generally speaking, most of the current technical methods used in designing desktop computer cases are 1306207. For the second and second, the complex and detailed calculation design of the Singling, directly on the desktop computer case to make the heat dissipation opening' and then the heat dissipation opening device sensor to detect the position of the opening setting ten to the whole The degree of influence of the heat dissipation system, the further experimental and research tests to improve the overall heat dissipation performance, so as to repeat the above-mentioned complicated aspects, after trying to find the heat dissipation opening position that is the most mechanically improved overall heat dissipation performance, It is currently used in the design method and device of the heat dissipation opening of the desktop computer case; the current use of the technical street is lacking: L is not in line with the economy due to the large amount of research and development and manufacturing cost required for making the π contact for experimental inspection. Cost-effective; 2. Due to the need for experimentation and research testing, a large number of R&D Nippon copper and manpower are required to be used, so that it can not achieve maximum efficiency in the effective time; When reading the temperature measurement σ position, it is easy to cause errors due to human error; 4. The experimental equipment is easy to wear and cause experimental error; 5. The production of heat dissipation openings and the maintenance and repair of experimental equipment are expensive; 6. Due to the current technology used The workman _ 碰 碰 碰 碰 碰 碰 碰 造成 造成 造成 造成 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力 人力Happening. SUMMARY OF THE INVENTION The present invention will build a money H Hui Wei on the design surface of the scales based on the neural network, and simulate the resulting knots and devices, after the system, calculate the different heat dissipation opening design for the overall computer敎系,统子=路系统训练与学习1306207 The advantages of using this is the following: 1. Due to the R&D and peak benefits of this transfer, the cost of a lie is low, and the cost is high and low. Effectiveness 2. This method of making money is completely fast and precise, so it does not require a lot of research and development time and shortens the work time to achieve the highest efficiency in the time of picking up the new one; It has the aging work performance and reduces the error caused by the change of time. 3.Because this time, the county is responsible for the situation _ _ will not cause mistakes caused by people riding the loss; 4. The invention made by the invention _ Laisi (4) But the hair is not easy to lose; 5. This hair riding device has more Wei brain and mold body, _ transfer cost is low; thus reducing a lot of human resources. 6. This invention completely replaces the human being with the computer to do the required work, — Consumption; 7. This invention begins with simulation To the class of neural network to push the office ~ Workflow, with perfect continuity and self-two; brain carried out 'and thus in the entire development space. >, ,B is also more intelligent in the work [Embodiment] The intelligent theoretical design of the computer case of the present invention is divided into two parts: the H-shoe riding method mainly estimates the design of the heat-dissipating opening, The two parts of the physical object 5 show that the input layer of the road structure performs spindle transformation and reduction, and the p nsi〇n method operates on the neural network. The reduction step is as shown in Fig. 1, which greatly reduces the neural network. The first embodiment: A method and device for cooling the opening of a smart computer case, as shown in Fig. 1, the method flow includes a learning process and detection. Process and re-learning *% of its towel Wei is from the school library, 1306207 by the thermal opening design attribute vector as the inverse of the neural network of the private ^ θ,] scouring vector χ; The node facets should be s kinds of green fruit, (4) the neural network details of Yl are cpu temperature, and the neural network output node of % is the temperature monitoring point of 15 different positions, where; & is the second floor of the case The temperature of the square is monitored by the temperature of the upper left of the 3rd floor, the temperature of the upper right of the upper layer of the I domain, the temperature monitoring point of the upper right of the upper layer of the γ5 domain, and the temperature monitoring point of the center of the red layer of the % domain, γ7 It is the temperature monitoring point at the lower left of the middle layer of the casing, % is the temperature monitoring point at the upper left of the middle layer of the casing, Υ9 is the temperature monitoring point at the lower right of the middle layer of the casing, and YiG is the temperature monitoring point at the upper right of the middle layer of the casing. Temperature monitoring point in the center of the middle layer of the shell, Yi 2 is the temperature monitoring point at the lower left of the casing τ layer 'known as the temperature monitoring point at the upper left of the lower layer of the casing, Yw is the temperature monitoring point at the lower right of the lower layer of the casing, Y 〖5 is the temperature point at the upper right of the lower layer of the casing , Yi6 is the temperature monitoring point in the center of the lower layer of the casing, and then the inversion of the Weng, _ the value of the sample and the actual output value of the county is the minimum objective function, and the neural network is divided into the best To get the most accurate measurement of the results of the contact set and to obtain the optimal key value for each node of the inverted neural network. As shown in Figure 6, the following steps for the neural network learning process include : y ......X is the initial gift value %, w, and the deviation along the saturation of the water ((4)^ are small random numbers; the step from the stop condition is still shed 'the neurological scale output healthy and the sample actual output difference Less than the allowable value' is step 3 to step 10, otherwise the learning process ends; Step 3: For each training sample, step 4 to step 9; ν4. Transfer the input vector (the attribute vector of the heat dissipation opening design parameter to Each node of the hidden layer is fully connected; 13〇62〇7

5: Calculate the round-out value of the hidden layer v w & w),

I where the action function is defined as /(6)=”^ ; Step 6: Calculate the step of each round-out node. VII 4* f ^ • δΐ t--the return error of the output node ΑΆ , 0^ = 1,2 , ~, call and key correction amount, = slightly %, 岣 =%, where the learning rate 0 < α < ΐ 'actual turn-out value ^ ; -M_ *^· 7 · β 十-hidden layer per-node key The knot correction value n« and the return error Κ=(|^Ά(1-/0,(~,/ = 1,2,.·., corpse); Step IX·Bu new parent key value You know You), ~((10)(from w), 4("(four)), where ~("ew" = w)A(〇'£/) + ·龙,w"wew) = ~(〇w) +△~ 0k(new)-ek(〇ld) + A0k > 0J(new) = 9J(〇ld) + Aej ; ^10: Whether the test condition is true or not; Μ process, as shown in Figure 7 The steps include 'step __ . . Enter the attribute vector of the sample design parameter of the heat sink opening to be tested; Step 2: God, the network evaluates the heat dissipation opening design, and uses the reduced-order spindle as the neural network input vector. And calculate the output value of each output node of the neural network; the actual evaluation error of the network is large. Step 3: Check whether the gorge is misjudged Sample (where the sample is misjudged to represent the nerve at the allowable value), if any, the misjudged sample data is stored in the learning sample database for the re-learning of the data; Step 4: _ whether the county, if it is spicy - if it is The job detection process, · 10 1306207 Re-learning process, as shown in Figure 8, the steps include, step one · add all misjudged samples to the learning sample database; Step two: face the above learning f touch, Make the oracle _ _ _ _ _ _ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ ♦ "The item belongs to the pottery as a reverse pass [miscellaneous, into the heart = 1, 2, ~ a _ point to 丨, 2, shoulder to the side to see the silk, curry is secret, _ touch (four) Na, shot in the middle, fine The neural network uses the known rounding value and output value of the training sample (ie, the training sample attribute in the learning sample database to the corresponding rounding result) to adjust the weight of each node, so that the neural network output value is away from the actual round. The error of the value is the smallest The standard function, adjust each section _ knot value to optimize 'small reading 纟 职 职 职 , , , , , , , , , , , , , 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习 学习The vector is used as the wheel vector, and the heat dissipation opening design is evaluated via the neural network. If there is a misjudgment sample in the evaluation process, the sample material is misjudged and stored in the learning sample database to facilitate the learning of the data; the learning process is determined by the decision. The sample is added to the learning sample database, so that the neural network adjusts the weights of each node, so that the subsequent detection process can no longer cause misjudgment for the similar or identical misjudged samples, and the method of improving the wisdom of the computer case heat dissipation opening is Estimation accuracy of the device. The invention relates to a method for restoring the opening of a smart electric housing, and the intelligent electric housing heat-dissipating device comprises a method for designing a sample database and a heat dissipation opening of a computer casing. 'The learning sample database stores the genus of the scattered sample design of the training sample. 11 1306207 .. Sexual vector' This attribute vector is obtained after computer simulation analysis. Each training material contains 26 items of data, the top 10 The item (or the previous "dimensional" data is the attribute vector of the heat dissipation opening design, in the order of Χι to Χι〇, - where: \ is the coordinates of the X-axis of the heat sink inlet of the computer case, and χ 2 is the cooling inlet of the computer case Position is the coordinate of the Y-axis, X3 is the coordinate of the cooling hole of the computer case in the Z-axis, χ4 is the diameter of the cooling air inlet of the computer case, and X5 is the coordinate of the cooling air outlet of the computer case at the axis of the , axis, which is a computer machine. The position of the heat dissipation air outlet of the shell is in the coordinate axis of the γ axis 'X7 is the coordinate of the cooling air outlet of the computer case in the 2 axis, & the diameter of the cooling air outlet of the computer case 'X9 is the number of holes in the heat dissipation of the computer case, Xig is the computer ® Shell heat dissipation side hole diameter, item 11 (or item) is the neural network rotation node Yi is the CPU temperature, the brother 12 items (or "+2 items" is the neural network output point Y2 is the lower left of the upper layer of the case Temperature monitoring point, item 13 (or "+3 item" is the neural network output node Y3 is the temperature monitoring point on the upper left side of the upper layer of the chassis, and the 14th item (or "+4 item" is the neural network output node I is The temperature monitoring point on the lower right side of the upper layer of the casing, item 15 (or "+5 items" is the temperature monitoring point of the upper part of the upper layer of the upper layer of the neural network output node Ys, and the 16th item (or "+6 items") is the nerve The network output node γ6 is the temperature monitoring point in the center of the upper layer of the casing, and the 17th item (or “+7 item” is the neural network output node Υ7 is the temperature monitoring point at the lower left of the middle layer of the casing, • Item I8 (or Item) is the neural network output node Υ8 is the temperature monitoring point at the upper left of the middle layer of the casing, and the 19th item (or “+9 item” is the neural network output node Υ9 is the temperature monitoring point at the lower right of the middle layer of the casing, the 20th The item (or Chuan item) is the neural network output node Υ10 is the temperature monitoring point on the upper right side of the middle layer of the casing, and the 21st item (or item) is the neural network output node Yu. The temperature monitoring point in the middle of the middle layer of the casing, item 22 (or w+72) is the temperature monitoring point of the neural network output node Y12 for the lower left side of the casing, and the 23rd item (or item) is the neural network output. Node Υ 13 is the temperature monitoring point at the upper left of the lower layer of the casing, and item 24 (or item) is the temperature monitoring point of the lower right side of the lower layer of the chassis.

12 1306207 Item 25 (or ΜΛ5) is the neural network output node Y1S is the temperature monitoring point on the upper right side of the lower layer of the chassis, and the 26th item (or item) is the temperature of the neural network output node γ1ό is the center of the lower layer of the chassis. The monitoring point is based on the evaluation standard values obtained by the existing known standard methods. The method and device for the heat dissipation opening of the smart computer case as described above, wherein the neural network for evaluating the heat dissipation opening of the computer case in the first step of the detection process is as follows: Step -: the learning process will be completed in the training The neural network key value is fixed ~~, H·Step 2· For each test sample, perform step 3 to step 5; Step 3: Transfer the wheel vector (the attribute vector of the heat dissipation opening design to the hidden layer) Each node is fully connected; Step 4. The output value of the hidden layer is ~ 7 , where the action function is defined as step 5. Calculate the output of each output layer ^ yk~f(Z^Jkhj~9k) ^An embodiment: The method of designing a computer shell heat dissipation method includes a device and a device, as shown in Fig. 1, the steps of the S and the step-down process and the detection process are included, Re-learning process 'Where to learn and downgrade step one: «3 / Ben: _, with all training samples (training sample 4 / - take 4, before "item heat dissipation opening attribute vector book of the month ϋ «仃 data ), · matrix'» , 〇丨力'孑目关Array i? (autocorrelation ~ Ρ(φ,) , . a category / appearance of suspicion, '疋 疋 疋 之 之 属性 属性 属性 咖 咖 咖 咖 咖 咖 咖 咖 咖 咖 咖 咖 咖 咖 咖 咖 咖 咖 咖 咖 机 机 机 机 机 机 机 机 机 机 机 机 DS DS DS DS Is the property vector of the third solid class, expecting the property of the familiar opening to 13 1306207. Step 2: The feature value of the object (such as the job and the feature vector (four) sang s), and formalizing the domain vector ( N_aiize), set the initial state (4); V3···Cursing the reduced-order matrix φ, as shown in the figure, the reduced-order matrix Φ consists of w eigenvectors corresponding to the largest pre-w-solid eigenvalues of the correlation matrix Λ m Step 4. Spindle conversion and reduction, as shown in Figure 2, by ==Φ'χ, 'The input vector is reduced from χ, the η dimension to 4%, where is the principal axis; Step 5: Nerve _Evaluate the slanting mouth of the singularity as shown in Figure 3. The "rear-in vector" is used to evaluate the training sample by the inverse-transmission neural network. If the estimation error is less than the allowable value, then Go back to step three, repeat the above steps by inputting the vector and then lowering the first step. (or at least) the spindle dimension until 'the estimated error is greater than the material value, the minimum spindle dimension required for the crusting is W = W + 1 and the learning process is stopped; the steps of the detection process are as follows, Step 1: Enter the design conditions of the heat sink opening to be tested Attribute vector; Step 2: Convert the attribute vector of the heat dissipation opening design condition to the main vehicle conversion and reduce the order, and use η expansion green to convert the attribute vector of the heat dissipation opening design condition to the spindle conversion: the opening is set to 'the main axis is reduced Dimensions are the minimum number of spindle vectors required to maintain the accuracy of the thermal stimuli condition during learning and reduction; through the network input to step 3: neural network evaluation of the heat dissipation opening design, the spindle after the reduction is used as the god I And the output of each output node of the three-dimensional neural network * value from the output node of the large-wheel output value corresponding to the heat-dissipation opening design condition represented by the von test sample heat dissipation opening 14 1306207

Step 4: 疋No, there is a sample of Wu Jue, if there is a mistakenly stored in the learning sample database for the re-learning of the data; V. V. Test 疋 no end, if no step back - if Yes, the detection process is stopped; the learning process, as shown in the fourth paragraph, includes the steps, and the sample is added to the learning county database;

Step 2: The above-mentioned learning and descending process is carried out, so that the nodes of the nerves and ugliness can be re-adjusted to optimize the 'financial effect', and the false positives will not be misjudged, and the estimation of the present invention will be improved. Accuracy; by the combination of the above processes, the heat dissipation opening "item attribute vector is used as an input vector of the inverse neural network,, / = 1, 2, ..., ~ output node # = 1, 2,, ^ correspond to the output of the type 2, respectively. Therefore, the neural network output node of 1 is CPU temperature, and the neural network output node of γ2 to ^ is monitored by the temperature of 15 points. γ2 is the casing. Temperature monitoring at the lower left of the upper level

Measuring point, Υ3 is the temperature monitoring lion at the upper left of the upper layer of the casing, γ4 is the temperature monitoring point at the lower right of the upper layer of the casing, Υ5 is the temperature monitoring point at the upper right of the machine rainbow layer, and Υ6 is the temperature monitoring point at the center of the upper layer of the casing. Υ7 is the temperature point at the lower left of the casing layer, γ8 is the temperature monitoring point at the upper left of the casing layer Υ9 is the temperature monitoring point at the lower right of the middle layer of the casing, and YiG is the temperature monitoring point at the upper right of the middle layer of the casing. Yn is the temperature monitoring point of the machine layer, and Yi2 is the temperature monitoring point at the lower left of the lower layer of the casing. 'Υ13 is the temperature point on the upper left side of the lower layer of the casing, and the temperature monitoring point on the lower right side of the lower layer of the % domain. !5 is the temperature monitoring point on the upper right side of the lower layer of the casing, and Υΐ6 is the temperature monitoring point in the center of the lower layer of the casing, and then the learning and reduction process, the detection process and the re-learning process are sequentially performed, among which 15

1306207 and the reduced-order process towel's conversion of the attribute vector of the heat-dissipation design parameters to the orthogonal record by the K_L expansiGn method, and the minimum number of axial miles needed to reduce the estimated precision The complexity of the God 35 network estimation 'simultaneous neural network paste training sample known round human value and input iij Jing Xue miscellaneous dragon, scale Yang (four) i and its corresponding heat dissipation opening condition) adjust the weight of each node ' The error between the neural network output value and the actual output value of the sample is the minimum objective function, and the _ key value is adjusted to be optimized to improve the neural network. After the end of the process, the weights of the nodes are fixed. Eli detection process estimation; in the detection process will wait for the job attribute vector axis KL which method - the method of spindle rotation and reduction and the reduced spindle as an input vector through the neural network for cooling openings Estimate the 'miscellaneous process if there is a mistaken successor, the occupational error female capital is stored in the study sample database to facilitate the acquisition of the ff material; the learning is through the misunderstanding of this person in the learning sample database to make KL expansio nRe-adjust the spindle orientation and the neural network to adjust the weights of each node, so as to achieve the method and device for improving the cooling opening of the smart computer case by the subsequent detection of the previous error or the erroneous succession of the system. Estimation accuracy. * As above, the smart dragon shell heats off the green and the device, its cap smart computer machine #',,, open the device of the alpha contains the personal electric skill learning sample Laiku and intelligent wisdom The computer machine breaks the heat and opens the D design method. The new quotation is the attribute vector of the heat dissipation opening design parameter of the data storage age. The recording and setting property vector is obtained after the simulation analysis, and each -'s I practice The billet contains 26 items of data. The data of the first 1 item (or the front side dimension) is the design parameter of the heat dissipation opening. The order of the data is XWXi〇'. The position of the heat sink inlet of the computer case is at the coordinates of the axis. For the computer machine salty heat dissipation σ position in the chain of γ #, & for the electric housing heat sink position at 16 !3〇6207

The coordinate of the Z axis, X4 is the diameter of the air inlet of the computer case, and the X5 is the coordinate of the X X pumping of the computer case. The X6 is the coordinate of the heat dissipation air outlet of the computer case on the γ axis, and X is the electric w. Positioned in the coordinate of the Z axis, X8 is the heat dissipation air outlet of the computer case, ... k X9 is the number of holes in the heat dissipation side of the computer case, X1G is the diameter of the heat dissipation side hole of the computer case, item 11 (or /, 乂7 items)疋 arsenic through the network output node Yi is the CPU temperature, the 12th item (or "+2 item" is the neural network rotation node γ is the temperature monitoring point at the lower left, the 13th item (or "+3 items") ) is the neural network output node "h is the temperature monitoring point on the upper left side of the upper layer of the casing, and the 14th item (or "+4 item" is the temperature monitoring point of the lower layer of the upper layer of the upper part of the casing, the first 15 L+5 items) is the nerve_output node & is the machine=the upper temperature monitoring point on the upper layer, and the 16th item (or “+<5 item is the temperature monitoring point in the center of the neural network” item 17 ( Or π+7 terms) is the neural network output wheel γ6 is the upper layer output node Y of the casing? It is the temperature monitoring point at the lower left of the middle layer of the casing, and the 18th item (or item) is the neural network. The output node ° 8 horse machine set the temperature monitoring point in the upper left of the middle layer, the 19th item (or "+9 item" is the temperature monitoring point of the neural network output node to the lower right of the middle layer, the 20th item (or ?2+) Chuan Xiang) is the neural network output node γι〇 for the sound in the machine

The temperature monitoring point on the upper right is '21st item (or „+77 items) is the temperature monitoring point of the neural network output node% is the middle of the middle layer of the casing. m) is the nerve transmission (four) point I々 the lower layer of the lower case Fangzhi temperature monitoring point, 帛23 items (or items) is the temperature monitoring point 'the 24th item (or „+) of the lower left part of the lower layer of the γ 机 机 机 是 是 是 是 神经 神经 神经 神经 神经The temperature monitoring point '2S item (or „+λ5 item) at the lower right of the lower layer is the neural network output node ^ is the temperature monitoring point on the upper right side of the lower layer of the chassis' The % item (or item) is the neural network output node Υΐ6 For the temperature monitoring in the center of the lower layer of the casing, it is based on the evaluation standard value of the existing known fresh hybrid feeding. The method and device for the heat dissipation opening of the smart computer casing as described above, wherein the learning and reduction step 17 1306207 The neural network of step 5 evaluates the different design conditions of the heat-dissipating opening, and the round-out nodes correspond to the reduced-order spindle vector as the input vector of the inverse-transition neural network, and the output vector, by the inverse-transfer neural network. The output value and the actual round-out value of the sample The point key value is adjusted to the optimal value, and the zero is the most Φ is the number of the target'. The neural network is per-recorded, and the results are as follows. The following steps are described below. Step-: The subordinate value of the slave reduction "...and the material reading and the secret return (four) θ

For a small random number; J * turn two: when the stop secret has not arrived, the difference between the neural network output value and the training sample actual rotation is less than the valley value, step three to step oblique, close the learning process; : For each training sample, do step 4 to step 9; Step 4: Pass the input vector (the main axis is privately transmitted to each node of the hidden layer for full connection; Step 5: Calculate the output value of the hidden layer\, i where The action function is defined as / (about = _l^; • Step 6: Calculate each output node eight, eight; j Step seven: Calculate the return error %, 乂 plus (i") of each output node (eight, yes = 1,2,"., and the key correction amount Δη^=<\, suffix=, where the learning rate 〇<«<1, the actual output value L; Step 8: Calculate the hidden layer each Node bond correction value, △^ and return error, (/iy,y = i,2,.··,;?); k=\ 18 1306207 wjk (new) = wjlc (old) + Awjk 5 Ui} {new) = (o/J) + Auy, 0k {new) = 0k {old) A0k J Θ} (new) = 9J {old) + ! Step 10: Test if the stop condition is true; Wisdom The method and device for designing the heat dissipation opening of the computer case, wherein the neural network evaluation heat dissipation opening design step of the third step of the detection process is as follows: Step 1: The neural network key that is trained and completed during the learning and reduction process The value is fixed; Step 2: For each test sample, perform steps 3 to 5; Step 3: Transfer the input vector (spindle vector) to each node of the hidden layer for full-join operation; Step 4: Calculate the hidden layer The output value \, &=/(J>yx,D, where the action function is defined as i = ; \ + ep Step 5: Calculate the output value of each output layer.

19 1306207 [Simple description of the diagram] The i-th image KL Expansion combined with the neural network: the re-learning of the brain cavity heat dissipation opening Figure 2 The intelligent design of the electric occupational shell _ σ machine miscellaneous mouth composition (four) = one (four) learning 4th Figure KL Expansion combined with neural network intelligent design eDonkey Figure 5 inverted neural network architecture Figure 6 intelligent design computer chassis cooling opening neural network training Figure 7 smart design computer case cooling opening The neural network test silk diagram 8th figure smart design computer case heat dissipation opening nerve network Figure 9 computer case internal temperature quantity sense #+3⁄43⁄4 grasp fine [main component symbol description]

Xl is the cooling air inlet position of the computer case. The coordinate K of the X-axis is the cooling air inlet position of the computer case. The coordinate X3 of the γ-axis is the cooling air inlet position of the computer case. The coordinate X4 of the Z-axis is the cooling air inlet of the computer case. Diameter\ is the computer case heat dissipation air outlet position in the X-axis coordinate Χό is the computer case heat dissipation air outlet position in the γ-axis coordinate I is the computer case heat dissipation air outlet position in the x-axis coordinate X» for the computer case heat dissipation The diameter of the air outlet is the number of holes in the heat dissipation of the computer case. Χίο is the heat dissipation side hole diameter of the computer case. 为 is the temperature of the CPU. Y2 is the temperature monitoring point on the lower left side of the upper layer of the case. The temperature monitoring point on the upper left side of the upper case is Υ4. The temperature monitoring point YS at the lower right of the upper layer of the casing is the temperature monitoring point at the upper right of the upper layer of the casing. The temperature monitoring point at the center of the upper layer of the casing is the temperature monitoring point at the lower left of the middle layer of the casing.

20 D 1306207 γ8 is the temperature monitoring point γ9 at the upper left of the middle layer of the casing is the temperature monitoring point at the lower right of the middle layer of the casing Υΐο is the temperature monitoring point at the upper right of the middle layer of the casing Yll is the temperature monitoring point Yl2 at the center of the middle layer of the casing is the machine The temperature monitoring point Yl3 at the lower left of the lower layer of the shell is the temperature monitoring point at the upper left of the lower layer of the casing. Yl4 is the temperature monitoring point at the lower right of the lower layer of the casing. The temperature monitoring point Yl5 at the upper right of the lower layer of the casing is the temperature at the center of the lower layer of the casing. Monitoring points

twenty one

Claims (1)

306 306207 X. Applying for a patent Fan Park·· = kind=(4) Collecting dirty sighs, which include Liu and the reduction process, the detection process and the re-learning process. The steps of the learning and reduction process are included, steps: Then the county axis, the age round county 'instrument (four) attribute vector (4 data before the training sample), calculate the relevant moment called face, elation ^) R system (four)}, where the force is the attribute vector of the heat dissipation opening ' ♦,) is seen The probability density function, E{XjX;} is the expected value of the attribute vector in the first category, "the number of attribute vectors for the heat dissipation opening; the eigenvector normalization (_alize), set the initial state...; Step 3: Find the input The car φ, the lower part of the W eigenvalue corresponding to the sacred sacred value is composed of H_1; Step 4: the main axis (four), -~ er - the utmost to the heart dimension, where % is the principal axis vector; The neural network evaluates the design conditions of the heat dissipation opening, and takes the input vector as the input vector. The training sample is evaluated by the inverse neural network to evaluate the design conditions of the heat dissipation opening of the training sample. If the estimation error is less than the allowable value, then go back to step 3 and input the vector. Lower one Repeat the above step t until it is found that the minimum (or minimum) spindle dimension can be estimated. If the difference is greater than the allowable value, then the minimum spindle dimension of the estimated shrinkage is determined as the learning process; The neural network for the white and the step-down process step 5 evaluates the design conditions of the heat dissipation opening, which is to reduce the order 22 1306207 = as the input vector of the reverse transmission god _ road, and the "round-out nodes correspond to the increase and decrease, respectively The error of the inverted neural network rotation value and the actual output value of the sample is minimized as a function of the objective, and the neural network sections, ', and the threshold value of the indicator are adjusted to be optimized to obtain the most accurate round-out. As a result, the 焱 焱 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递 传递The deviation of the saturation function of the horizontal axis (bias) & slices are small random numbers; Step -: When the stop condition has not arrived, the neural network output value and the actual output of the training sample are different I; Valley 4 value, step three Go to step ten, otherwise end Learning process; exhaust __' step to step IX; Step 4: Transfer the input vector (spindle vector) 1 to each node of the hidden layer for full-join operation; 'Step 5: Calculate the hidden layer The output value v, where the action function is defined as / (wide) = 1; Step 6: say ten counts each output node eight,; Step seven: calculate the return error of each output node < 鸣痛丨会(hbu. '·, call and key correction amount, Δκ, where the learning rate 0 <«<1, the actual output value ^; Step 8: Calculate the hidden layer per-node bond correction value △ ~ = 吟, and back error Step IX. Update each key value ^(_), ~—风州卿), where 23 1306207 Wjk<JleM,^wjk(〇ld) + hM>jk > uij(new) = uij(〇 Ld) + Auij . ek(new)=,〇k(〇ld) + h9k > 0j(new) = θ.(〇ι^ +Αθ^ . Step 10: Test if the stop condition is true; As follows, Step 1: The property vector X/ of the design conditions of the heat sink opening to be tested by the wheel: Step 2: Perform the spindle transformation and reduction of the attribute vector of the heat dissipation opening design condition. The minimum number of spindle vectors required to estimate the accuracy of the σ 鸠 (4) _ material conversion ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; (2) Samples again) 'If there is', the sample data will be stored in the learning sample database for the re-learning data acquisition; Step 5: Whether the detection is over, if no, go back to the step--, if yes, stop the detection process; 'The steps are included, Step--Add all the misjudgment samples to the learning sample database. · Step Township··==Speaking and lowering the call to make the weight of each node of the neural network re-judgment 'Calling to the sorcerer of the sorcerer's sorrows will occur again.' and the estimation accuracy of the invention is soaring; 24 1306207 By the combination of the above processes, the attribute vector of divergence *, , and Khan is passed through KL. The expansion is converted into the principal vector as the inherited vector of the inverse neural network (4), ·,...the output nodes Λ,Α=1,2,···, β correspond to the seeding out, respectively, so the nerve of Yl Network output cpu temperature I to the god of I6, the temperature monitoring point of different positions of the point _ point, 1 γ2 is the temperature monitoring _ at the lower left of the upper layer of the casing, I is the temperature monitoring point at the upper left of the upper layer of the casing, I is the machine, and the upper layer At the lower right temperature monitoring point, I is the temperature monitoring point on the upper right side of the upper layer of the casing, I is the temperature monitoring in the center of the upper layer of the casing, I is the temperature monitoring point at the lower left of the casing cap, and the % is the upper left of the casing cap. The temperature monitoring _,% is the temperature monitoring point at the lower right of the machine layer, % 12 is the temperature monitoring point at the upper right of the middle layer of the casing, % is the temperature monitoring point in the center of the middle layer of the casing, and γ 14 is the lower left of the lower layer of the lining Temperature monitoring _, Yu is the temperature monitoring point on the upper left side of the lower layer of the casing, and the 'degree of the right side of the lower layer of the γ 16 y yoke's lower layer [measuring point, % is the temperature monitoring point on the upper right side of the lower layer of the casing, I. For the machine; vX lower layer of the temperature of the towel is difficult to monitor, and then according to the handle line learning and reduction process, the detection process and re-mixing 'the towel f and the reduced order process towel, by κ_[which coffee method will heat Open. The attribute vector of the 30 parameters is converted to the orthogonal main axis to avoid the attribute vectors from interfering with each other, and to maintain the minimum number of spindle vectors required for estimation accuracy, so as to reduce the complexity of neural network estimation, and the neural network is complete. The sample is known as the input and loser _ _ _ learning stomach sample database, training sample attribute vector and its corresponding thermal opening design conditions) adjust the weight of each node, so that the error between the neural network output value and the sample actual output value is the smallest The objective function adjusts the node key values to optimize the neural network estimation accuracy, and fixes the weights of each node after the end of the learning and reduction process to facilitate the estimation of the detection process; the detection process will wait for the detection sample The attribute vector is converted and reduced by the k_l expansion method, and the reduced-order spindle is used as the input vector, and the design is based on the 25 1306207 If there is a mistake, the county misjudges the samples and stores the sample materials in the study sample database to facilitate the acquisition of the learning materials; the learning process is based on misjudgment; this is added to the learning sample database to re-adjust KL exqingi〇n The spindle orientation and neural network adjust the weight of each node. 2. The method of the heat dissipation opening of the smart computer case according to the patent scope of the Shenqing patent scope, wherein the learning sample database stores the attribute vector of the heat dissipation opening design parameter of the training sample, and the attribute vector of the heat dissipation opening design is via Obtained after the simulation analysis, each training data contains 26 items of data, before and after (or before, the data is the attribute vector of the heat dissipation opening design parameter is Χι to Χι, where 乂 is the current housing In the X standard, & is the coordinate of the _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ When the wind is out, it is placed at the _ coordinate, & is the coordinates of the γ axis of the cooling air outlet of the computer case, χ7 is the coordinate of the cooling air outlet of the computer case at the axis of the Ζ axis, and Χ8 is the heat dissipation σ of the casing. Yu, & is the number of oil holes in the computer case, Χ1〇 is the heat sink side hole diameter of the computer case, the Uth item (or W item) is the neural network output node I • ^CPU temperature, item 12 (or W Item) is the neural network output node γ2 is the temperature monitoring at the lower left of the upper layer of the casing Point, item 13 (or (4)) is the neural network output node ^ is the temperature monitoring point on the upper left side of the upper layer of the chassis 'the second item (or (4)) is the neural network input _ point 'the lower right side of the upper case The temperature monitoring point, item I5 (or (4)) is the neural network wheeling node, which is the temperature upper right of the upper layer of the casing, and the 16th item (or (4) pole god _ road wheel (10) point is on the casing The temperature monitoring point of the Guangzheng Central 'the πth term (or „+7 term) is the neural network output node γ7 is the temperature monitoring point at the lower left of the middle layer of the casing' Item 18 (or (4)) is the neural network output node % is the temperature monitoring point at the upper left of the middle layer of the machine 26 I3062Q7, the 19th item (or "+9 item" is the temperature monitoring point of the neural network output node 丫9 is the lower right of the middle layer of the casing, item 20 (or "+ Chuan Xiang) is the temperature monitoring point on the upper right side of the middle layer of the casing, and the 21st item (or item) is the temperature monitoring point of the center of the middle layer of the casing. The item (or "+/2") is the neural network output node Ylz is the temperature monitoring point at the lower left of the lower layer of the chassis, and the 23rd item (or „+73 items) is the neural network. The road output node Yl3 is the temperature monitoring point at the upper left of the lower layer of the casing, and the 24th item (or "+74 item" is the neural network output node Υ14 is the temperature point of the lower right side of the casing, the 25th item (or the state 5) Item) is the temperature monitoring point on the upper right side of the lower layer of the chassis, and the 26th item (or 6 items). The 输出神红 network output node Υΐ6 is the temperature monitoring point in the center of the lower layer of the chassis. The correspondence between the above output and the round-in value can be obtained by the experimental process (or the input and round-off evaluation obtained by the known standard method). 3. The Zhidao machine and the heat-dissipating opening mentioned in the first paragraph of the patent application scope The method of design, in which 兮"~team step, Khan julu and stare < detection process step three of the neural network opening material steps Wei as follows: f 细 fine _ 丨 丨 _ _ fine steps 2: for each - For each sample, do step 3 to step 5; μ ^ ' Step 3: Turn the people to the bottom (record to private transmission _ hide step four: calculate _ layer of the input ^ ^ partial 卩. tree full connection operation; m - 丄^ 4), its read function is defined as step five: calculate the output of each output layer - ㈣Σν ^). 4. A method of mentally ill computer Gu _ _ mouth, private law axis and re-learning satirical @ # _u 5 sub-process, detected 'study f process is from the f sample f library, borrowed domain design 27 1306207 The input direction is used as the input vector of the inverted neural network, and the "input nodes correspond to the five kinds of collapsed Z fruits respectively." Therefore, the neural network output node of Yl is the CPU temperature, and the neural network input of I to Yl6 2 is the temperature monitoring point of 15 different positions, where 1 is the temperature at the lower left of the upper layer of the casing and 丨^ Υ3 is the temperature monitoring point at the upper left of the upper layer of the casing, and 丫4 is the temperature Υ5 at the lower right of the upper layer of the casing. It is the temperature monitoring point on the upper right side of the upper layer of the casing, Υ6 is the temperature monitoring point in the center of the upper layer of the casing, Υ7 is the temperature monitoring point of the left τ side of the middle layer of the casing, and Υ8 is the temperature L at the upper left of the machine layer, then the point is ''9 The temperature is engraved on the lower right side of the middle layer of the casing, Υι. is the temperature red measuring point on the upper right side of the machine layer, Yll is the temperature monitoring of the financial layer of the casing layer, and the % is the temperature measuring point at the lower left of the lower layer, Υ13 It is the temperature monitoring point at the upper left of the lower layer of the casing, and L is the temperature I measuring point at the lower right of the lower layer of the casing, γ1 5 is the temperature monitoring point on the upper right side of the lower layer of the casing, and Υΐ6 is the temperature red measurement point in the center of the lower layer of the casing. The error of the output value of the money transfer network and the actual touch value of the sample is minimized 2 as the target secret view _ Each section of the road _ is light to the most touch, giving the most accurate results of the scatter, , and PU boxes, and obtaining the optimal key value for each node of the inverted neural network. Including: Step 1: Set the first she, (4) Shui Weiqier (bias) " are complex small chaos; J step - 'When the item has not arrived, the neural network output value and the actual output of the training sample difference Less than the allowable value, step L needle, no listening beam learning process; V-- for each training sample, step four to step nine; step four human vector _ opening design parameters attributes ♦ transmitted to the hidden layer _ _ _ points for full-join operation; 28 I3062Q7 Step 5: Calculate the output value of the hidden layer v~=/(stone, where the action function is defined as /(妁=_11^ ; , 1 + θ Step 6··Calculate each— Output nodes π / (ΣΆ,; J step seven · s ten count each - round out nodes Return error __fan) heart (11), (eight, /4: = 1, 2, '.., speak) and key correction amount / ^ = tons, ., then 4 = £ ^, where the learning rate ο<α<ι 'actual output value G Step 8: Calculate the key correction value and return error of each node of the hidden layer, (V: 1, 2, ..., / 7); Step 9: Update each - The key value is ~(_), the %(four) is (4) calendar, where '"qing" = '".W) + —,,~(卿)=«〆.”) + △%, W = 〇k (old) + A0k , eMew) = ej{〇ld)^9j ; Step 10: Test whether the stop condition is true; The detection process, including its steps, Step -: Enter the attribute vector of the sample design parameter of the heat sink opening to be tested Network input to step 2: fine _ 嶋 post (four), input (four) for her quantity, and calculate the output value of each output node of the neural network; data stored in the study sample capital step three: detect whether there is a misjudgment sample, if any, will Mistake the sample library to facilitate the acquisition of the re-learning data; if yes, stop the detection process; Step 4: Whether the detection is over, if not go back to step 1, then the learning process, the steps include, 29 1306 207 Steps—·· Add all the mistakes to the learning sample database; Step 2: Re-execute the above learning process to optimize the weights of the neural network nodes to effectively approximate or identically misjudge the samples. New adjustments are made to improve the estimation accuracy of the present invention; the stomach is misjudged, and the combination of the above-mentioned various processes, with the design of the heat dissipation opening, the "network" of the "network" In the process of the corresponding nodes, the process of the process of learning, the process of detection and the process of re-learning, and the learning of m by the _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ In the sample database, the output of the training "u target adjusts the weight of each node, so that the value is compared with the sample axis, and the her value is adjusted to = to improve the accuracy of the neural network estimation. After the end, and fixed the weight of each node, in order to check the estimation of the Z process; _ over the financial materials to be checked _ this hybrid vector as the input vector, via God = _ _ _, tear, Cong Shigang Na I, mistake _ shell material Store in the sample database for further study Made of material; false relearning process via the learning sample into the sample database, adjust the neural network node weights, in order to achieve subsequent detection process for the $ t similarities or her mistake and then ship the forest misjudgment case. The method of the smart chassis gap contact according to claim 4, wherein the learning sample database stores an attribute vector of a heat dissipation opening design of the training sample, and the attribute vector is replaced by a computer. After analysis, the 'per training data contains % item data, and the first W item (or before) 1 is the attribute vector of the heat dissipation opening design, in the order of & to &, where: Χι is the computer case' , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , Shell heat sink inlet diameter, X5 is the computer case heat dissipation air outlet position in the Υ & 直 straight in the X-scale, χ 6 is the chassis cooling heat outlet position in the Υ axis coordinates, X7 is the computer case 散政熟 (4) σ Positioned on the axis of the Ζ axis, Χ8 is the computer machine to record the heat out diameter, χ9 is the number of holes in the computer case heat dissipation, knowing that the computer case is cooling money, the first U Na (four) ship nerve ring is secret ι is cpu temperature ' Item 12 (or (5)) is the god Lai Road output node Y2 is the upper left of the chassis Side of the temperature monitoring point, of η item (or items ...) is a neural network node Y3 wheel monitoring temperature of the upper left upper housing Measuring point, item 14 (or (4)) is the neural network rounding and saving 4 is the temperature monitoring point on the lower right side of the upper layer of the casing, and the 15th item (or η+5 items) is the neural network output node & The temperature monitoring point on the upper right side of the upper case 'Item 16 (or (4)) is the temperature monitoring point of the neural network rotation node % is the center of the upper layer of the chassis 'the 17th item (or "+7 item" is the neural network The output node of the road ¥? is the temperature monitoring point at the lower left of the middle layer of the casing. Item 18 (or (4)) is the temperature monitoring point of the neural network output node % at the upper left of the middle layer of the casing '19th item (or (4)) The neural network output node % is the temperature monitoring point at the lower right of the middle layer of the casing, and the 20th item (or "+chuan" is the neural network output node Υι〇 is the temperature monitoring point on the upper right side of the middle layer of the casing, item 21 (or "+" is the temperature monitoring point of the neural network output node γη is the center of the middle layer of the machine. The 22nd item (or π+υ) is the temperature of the neural network output node γΐ2 is the lower left of the lower layer of the chassis. Monitoring point, item 23 (or η+73) is the neural network output node Υΐ3 is the temperature monitoring point at the upper left of the lower layer of the chassis, item 24 (or π+ Item) is the neural network output node Yu is the temperature monitoring point on the lower right side of the lower layer of the chassis, and the 25th item (or „+75 items) is the temperature monitoring point of the upper right layer of the lower layer of the chassis of the neural network output node Y15, the 26th Item (or "+7 (5 items) is the neural network output node Υ16 is the temperature monitoring point in the center of the lower layer of the chassis. The relationship between the above output and the input value 31 1306207 can be obtained by the experimental process (or by the known standard method). Obtaining the evaluation value of the input and output. 6. The method for applying the heat dissipation opening of the smart computer case according to the fourth aspect of the patent application, wherein the neural network of the second step of the detecting process evaluates the heat dissipation opening of the computer case Including: Step 1: Fix the neural network key value of the training completed in the learning process; Step 2: For each test sample, perform steps 3 to 5; Step 3: Input the vector (heat dissipation opening design) Attribute vector), transfer to each node of the hidden layer for full-join operation; Step 4: Calculate the output value of the hidden layer\ where; Step 5: Calculate the output value of each output layer, eight = / (^) ; ϋ)). 32