189results about How to "With self-learning function" patented technology

The invention discloses a fire disaster detection system and a method thereof. The fire disaster detection system at least comprises two changeable cameras as a first camera of colorful or near infrared mode and a second camera of colorful or white-black mode, for obtaining the video images of detected region; an image collection module connected with two changeable cameras for converting and filtering the video images collected by the two changeable cameras into digital images, to obtain time sequence images; a fire disaster recognition module connected with the image collection module for analyzing the time sequence images, to obtain flame, smoke and fire disaster occurrence probability; and an alarm module connected with the fire disaster recognition module for comparing the flame, smoke and fire disaster occurrence probability with a preset threshold value, and emitting alarm message according to the comparison result. The invention can effectively obtain the fire disaster video images under various backlight conditions, to realize early and reliable fire disaster alarm.

The invention provides a fire detection system and a method thereof. The system comprises a video collection module, a flame judging module that extracts flame characteristic parameters from video images collected and carries out statistical analysis according to the flame characteristic parameters so as to calculate the flame probability of occurrence of a detection area, a smoke judging module used for extracting smoke characteristic parameters and carrying out statistical analysis according to the smoke characteristic parameters so as to calculate the smoke probability of occurrence of a detection area, a fire judging module that carries out fusion calculation according to the flame probability of occurrence and the smoke probability of occurrence so as to determine fire probability of occurrence, a fault judging module used for identifying fault information and an alerting module used for making a comparison between the probability of occurrence of flame, smoke and fire and a preset threshold value and sending out alerting information according to a comparison result. The fire detection system and the method of the invention can simultaneously realize the detection and alerting of flame and smoke, especially has very good identifying capability of the smoke at the initial period of fire and has the function of early alerting.

The invention belongs to the metallurgy technical field and in particular relates to an intermediate blank cooling system and cooling control technology. The intermediate blank cooling system is characterized in that an aerial fog cooling device body is arranged between a rough mill and a finishing mill. The aerial fog cooling device body is composed of an upper cooling header and a lower coolingheader, and an aerial fog nozzle and a compression air pipe are arranged on the cooling water pipe of the upper cooling header and the lower cooling header; and the on/off control valve and flow control valve of the cooling water pipe and the compression air pipe are connected with a control valve station. The intermediate blank aerial fog cooling technology is controlled by the aerial fog controltechnology sprayed on the cooling header arranged near roller ways at the front and rear of the mills. The invention has fast cooling speed without changing metal performance, water yield regulatingrange is wide and the water yield can be regulated continuously, the cooling speed of the intermediate blank can be precisely controlled, and the invention can be widely applicable to middle and thickplate production line.

An intelligent dialectics based on the probabilistic reasoning model of Bayes network and computing technology for traditional Chinese medicine to diagnose diseases is disclosed. The signal processing technique and image recognizing technique are used to automatically recognize the pulse condition and the tongue picture, analyze them and provide the result. In its interrogation interface, after the disease signs of a patient are input, the dialectic result can be output. It has also self-learning, storage and print function.

The invention discloses an intelligent automatic spraying unit based on a digital bus and a usage method thereof. The intelligent automatic spraying unit comprises a mechanical arm (1), a spraying device (2), a control valve bank (3) and an electric control device (4). The spraying device comprises a body (21) and a spraying head (23). The interior of the body is at least provided with two inner channels (211). Outer channels (212) which are identical to the inner channels in number and are communicated with the inner channels respectively are arranged outside the body. The body is further provided with a pattern recognition sensor, a position sensor and a range sensor. Spraying orifices (231) communicated with the inner channels are formed in the positions, corresponding to the inner channels, of the interior of the spraying head. The control valve bank comprises a positive-pressure control solenoid valve, a paint control solenoid valve and a negative-pressure control solenoid valve which are connected with the inner channels and the outer channels through pipelines. The electric control device comprises a pattern recognition circuit, a spraying parameter analyzing and planning circuit and the like. The spraying unit can enhance the recovery rate of paint waste gas and achieve intelligent spraying, thereby being suitable for digital bus factories.

The invention discloses a prisoner psychological health state assessment method and a system based on multi-modal information. The method comprises the following steps: obtaining physiological signals, facial expression images and voice signals of reformed prisoners and prisoners to be tested after they experience virtual reality scenes, and extracting physiological signal features, facial expression image features and voice signal features from the obtained signals; inputting the physiological signal features, facial expression image features and voice signal features of the reformed prisonerinto a pre-trained neural network model, and outputting a psychological state assessment vector of the reformed prisoners; inputting the physiological signal features, the facial expression featuresand the voice signal features of the prisoners to be tested into a pre-trained neural network model, and outputting a psychological state assessment vector of the prisoners to be tested; calculating the psychological state assessment vector distance between the prisoners to be tested and the reformed prisoners; and assessing the mental health state of the prisoners according to the distance.

The invention relates to the technical field of a concrete mixing ratio design, in particular to a concrete mixing ratio design method based on artificial intelligence, which comprises the following steps of : (1) adopting an artificial neural network method to establish a relation between concrete performance index and concrete mixing ratio so as to obtain a concrete performance prediction artificial neural network model; (2) based on the concrete performance prediction artificial neural network model, adopting genetic algorithm, aiming at meeting all performance requirements and leading the cost to be lowest, designing and optimizing to the concrete mixing ratio; and (3) retraining the concrete performance prediction artificial neural network model if the concrete mixing ratio obtained by optimization is not in the learned range of the artificial neural network. The concrete mixing ratio design method based on artificial intelligence is beneficial to quickly and accurately designing a concrete mixing ratio which can meet a plurality of performance requirements and has lower cost.

The invention relates to a pure electric vehicle cooling system control method based on deep reinforcement learning, which comprises the following steps: first, obtaining the temperature of the powerbattery of the pure electric vehicle, the working current of the power battery, and the ambient temperature information; Construct DDPG algorithm model, carry on reinforcement learning training, get agroup of optimal PID control parameters of electronic pump. The PID control quantity is obtained by the PID input quantity, and the electronic pump is controlled based on the PID control quantity. The work of the electronic pump changes the flow rate of coolant in the power battery cooling system to achieve the purpose of cooling the power battery. At the same time, the information of the power battery is transmitted to the environment sensing module, and the first step is returned to circulate the whole process. The invention introduces the depth reinforcement learning into the PID control algorithm, the depth reinforcement learning can better interact with the environment, has the self-learning function, adapts to the dynamic characteristics of the uncertain system, and therefore can adapt to the complex and changeable characteristics of the pure electric vehicle running environment, and realizes on-line control under different actual scenes.

The invention relates to an ontology-based anti-money laundering early-warning method, comprising the following steps: (1) building an anti-money laundering ontology composed of a client information ontology, a financial institution information ontology and a transaction information ontology, and linking the client information ontology, the financial institution information ontology and the transaction information ontology into the anti-money laundering ontology by constraint conditions; (2) building databases including a transaction flow database, a CBR case database, a law and regulation rule database and an auxiliary database; and (3) retrieving the ontology, discriminating the large-sum transactions and the suspicious transactions through operation, classifying the levels of risk by CBR and generating the report files. The method has the beneficial effect of providing convenience for anti-money laundering work, thus not only increasing the correct rate of anti-money laundering work but also greatly improving the working efficiency of the financial institutions.

The invention provides a PID locomotive automatic driving optimal control method based on reinforcement learning. The PID locomotive automatic driving optimal control method comprises the steps of: regarding current route information and a speed difference between a locomotive actual running speed and an optimal speed as input of reinforcement PID at first, acquiring a group of optimal PID controlparameters through reinforcement learning, realizing PID control according to the optimal PID control parameters provided by a reinforcement learning module, providing a control quantity, and controlling operation of a locomotive. The PID locomotive automatic driving optimal control method applies reinforcement learning to PID parameter adjustment, can interact with the environment by means of reinforcement learning, has the self-learning capability, can realize the PID control having the optimal parameter group, improves the optimization effect, and reduces manual design difficulty.

The invention discloses an intelligent automatic grinding and polishing unit based on a digital bus and a usage method thereof. The intelligent automatic grinding and polishing unit comprises a grinding and polishing device (2), a control valve bank (3) and the like. A body (21) of the grinding and polishing device is at least provided with two inner channels (211) and outer channels (212) which are identical to the inner channels in number and are communicated with the inner channels respectively. The body is further provided with a pattern recognition sensor, a position sensor and a range sensor. A working device (23) is internally provided with spraying orifices (231) communicated with the inner channels. The intelligent automatic grinding and polishing unit further comprises a grinding head drive mechanism (232) and a grinding head (234). The control valve bank comprises a positive-pressure gas control solenoid valve, a negative-pressure gas control solenoid valve and a brightener control solenoid valve which are connected with the inner channels and the outer channels. The intelligent automatic grinding and polishing unit can decrease the air pollution degree of a working site, lower the probability of secondary pollution and achieve intelligent grinding and polishing, thereby being suitable for grinding and polishing procedures in digital bus factories.

The invention provides an energy-saving control method and system for an air-source heat pump system. The method comprises the following steps that S1, a building standard cooling and heating load prediction model is constructed; S2, building refinement partitioning is conducted according to the building standard cooling and heating load prediction model, a cooling and heating energy-using demanddynamic curve of all subzones of a building is generated, and the dynamic curve is updated once at intervals; and S3, an air-source heat pump unit and an indoor tail end thereof, a freezing water pump, a valve unit, a fan and a heat pump host respond to an instruction sent by a central host in time, and according to the energy-using demand dynamic curve and combination of a time delay caused by the adjustable segment length and fluid velocity, operation stages of the air-source heat pump system is divided into an equipment starting stage, a steady state operation stage and an equipment shutdown stage for staging control. The invention further comprises the energy-saving control system of the air-source heat pump system. The energy-saving control system has the advantages of being efficient, stable, free of pollution, low in energy consumption, significant in energy-saving effect and the like, and therefore the total energy efficiency of the system can be improved by 5% to 25% on the basis of the original system efficiency.

The embodiment of the invention discloses a PCA-BP-based emotion recognition method and system. The method comprises the steps that a first emotional sound spectrum feature vector of training audio isobtained by blending a first acoustic feature vector and a first sample entropy feature of the training audio; a PCA algorithm is adopted for conducting dimensionality reduction on the emotional sound spectrum feature vector, and a second emotional sound spectrum feature vector is obtained; the second emotional sound spectrum feature vectors of the training audio corresponding to various emotionsare input into a preset BP neural network model separately for training, and sound spectrum feature vector emotion models corresponding to various emotions are obtained; a second acoustic feature vector and a second sample entropy feature of to-be-detected audio are blended to obtain a third emotional sound spectrum feature vector of the to-be-detected audio, the third emotional sound spectrum feature vector is compared with all the acoustic feature vector emotion models, the emotion model matching degree is calculated, and the acoustic feature vector emotion model corresponding to the maximum emotion model matching degree is output. The technical problems that the current voice emotion recognition process is complex, the implement difficulty is high, the accuracy is low, and the efficiency is low are solved.

The invention discloses a support vector machine and multi-body dynamics simulation-based high-voltage circuit breaker identification method. The method comprises the steps of building a three-dimensional model of a high-voltage circuit breaker, and calibrating and verifying the three-dimensional model; by changing size, material and property parameters of the three-dimensional model, performing analogue simulation on various faults of the high-voltage circuit breaker to obtain various corresponding mechanical property waveforms; performing waveform interception, mathematic morphological filtering and characteristic extraction processing on the mechanical property waveforms corresponding to the various faults to obtain an eigenvector sample set; based on a support vector machine and a one-to-many classification method thereof, training a classifier, and calculating the credibility of each classification; and inputting actually measured unknown fault eigenvectors to the training support vector machine, performing fault type identification to realize mechanical fault classification of the high-voltage circuit breaker, and finishing self-learning. According to the method, a large amount of physical simulation experiments do not need to be carried out; the faults of the high-voltage circuit breaker can be accurately identified; and the method has a self-learning function and a wide application prospect.

The invention relates to a transmission grid fault detection system and method based on an internet of thing technology. The method is characterized by: monitoring online and in real time whether changes of the voltage and current of a high-voltage transmission line three-phase fixed point are balanced by utilizing a three-phase zero sequence voltage and a current balance measurement method; and judging whether the fixed point of a high-voltage overhead line is abnormal based on the change features of the voltage and the current under various conditions. The system comprises grounding, open circuit and short circuit fault diagnoses. The system and method of the invention enable fault monitoring level to be improved, wherein the short circuit fault alarm accuracy can reach more than 98.5%, and the grounding fault alarm accuracy can reach more than 97%. A detection device possesses a self-learning function, namely, a plurality of threshold values do not need to be set according to the transmission lines with different loads, so that the workload before putting the detection device into use is reduced, the universality is improved, the power cut range is reduced, the power cut time is shortened, and the power supply reliability is improved substantially.

The invention relates to a fault diagnosis method based on an electronic equipment testability model, which is applicable to field fault diagnosis of all electronic equipment and belongs to the technical field of measurement and control. The main technical points of the method are as follows: the corresponding relationship of equipment fault mode, signal and test is established based on an electronic equipment testability model; a computer analyses out the test logic contained in the testability model according to the corresponding relationship; automatic reasoning is carried out and corresponding test is performed on a diagnosis node according to the fault phenomenon; and before next test diagnosis, the test process is optimized according to the current test result to realize dynamic fault diagnosis logic. The technical effect is that problems such as difficult and high-cost manual troubleshooting and poor flexibility, versatility and expansibility of binary tree-based troubleshooting are overcome fundamentally, and automatic and intelligent field fault diagnosis of electronic equipment can be realized.

The related traction circuit fault diagnosis system based on waveform recognition for underground locomotive comprises: a target circuit slot, a semi-physical analog simulation normal load, a data collector, an ADC, a data buffer, a CPU, a target fault database, a record label buffer, an algorithm software package based on waveform recognition to diagnose and position the fault, and a display. This invention can position fault accurately.

The invention discloses a weight learning based stack shape planning method. The concrete steps include: step 1, calculating stacking priority [Ri] for all to-be-stacked cargo tanks, wherein Ri=PvVi+PsSi+PwWi; step 2, ranking Ri in sequence from large to small; step 3, taking the current [Ri]max cargo container and trying to stack it on L1 to Lk in sequence until a stackable stack shape layer Lm is found; step 4, update the current stack shape state; step 5, removing the cargo containers counted to the stack shape from Ri; step 6, checking whether residual cargo containers exist or not, if residual cargo containers exist, repeating steps from step 3 to step 5; step 6, implementing stack shape planning. The method is simple in use and parameters are clear and easy to understand. Automatic stack shape planning can be implemented in a comparatively short period and dynamic planning can even be made for each stack order; high efficiency is achieved. The method provided by the invention isfully automatic and no manual intervention is required. The method provided by the invention has a self learning function and the quality of planned stack shapes is improved continuously towards the optimum.

The invention discloses a gear shifting manipulator of an automobile synchronizer testbed and a control device of the gear shifting manipulator. The gear shifting manipulator comprises a mounting bottom plate and a four-bar mechanism which is formed by hinge joint of a gear selecting shaft, a gear shifting rocker arm, a middle connecting bar and a gear shifting support bar, the gear selecting shaft is rotatably mounted on the mounting plate, an output shaft of a gear selecting motor is in transmission connection with the gear selecting shaft, a gear shifting air cylinder is hinged between the gear selecting shaft and the gear shifting rocker arm, a gear shifting connection bar is hinged to the upper end of the gear shifting support bar, and a tensile pressure sensor and angle sensors are arranged on the four-bar mechanism respectively. The control device comprises an MCU (micro control unit) and a human-computer interface LCD (liquid crystal display). The gear shifting manipulator is electrically connected with the MCU via the tensile pressure sensor and the angle sensors. The MCU is electrically connected with a gear shafting electromagnetic valve of the gear shifting air cylinder and the gear shifting motor. The human-computer interface LCD is in communication connection with the MCU. The gear shifting manipulator of the automobile synchronizer testbed and the control device of the gear shifting manipulator have the advantages of simplicity in operation, high positioning precision, precise controllability, stability in operation and capability of simulating a gear shifting effect of a human hand better.

The present invention discloses an electronic watch automatic assembling mechanism. The electronic watch automatic assembling mechanism comprises a first assembling transmission band, a second assembling transmission band and a third assembling transmission band which are respectively arranged at the left portion, the middle portion and the right portion of the electronic watch automatic assembling mechanism, the first assembling transmission transmits a movement, the second assembling transmission band transmits a watch housing, and the third assembling transmission band transmits a cover. The second assembling transmission band comprises a left support side plate, a right support side plate, a reinforcement seat and a guide sleeve, there are four telescoping cylinders at the bottom portion of the second assembling transmission band, first connection bars are respectively installed at left and right sides of the extension bar of the second assembling transmission band, the first connection bars are connected with a clamp block. The mechanism also comprises a left support frame and a right support frame, a gear post, a second guide post, a left movement case and a right movement case, wherein a first guide post is arranged between the left support frame and the second support frame. The electronic watch automatic assembling mechanism can assemble four electronic watches at one time, after the shaped dimension of the electronic watch is replaced, a location die is not needed to be made once again, and the electronic watch automatic assembling mechanism has self learning function to optimize the assembling process, realize unmanned assembling of the electronic watch and reduce the production cost.

The invention discloses an atmospheric pressure altimeter for measurement by use of a multilayer perception neural network, which includes static press and a temperature measurement unit, the static press and the temperature measurement unit is orderly connected to an A/D transverter unit, a center process unit, a communication interface unit, which are respectively connected to a power module, the static press and the temperature measurement unit include a static sensor and a temperature sensor, the center process includes microprocessor and data process unit of the multilayer perception neural network. The benefit effect of the invention is that time shift and a temperature time are avoided, learning function is provided and automatically learned through the neural network, difference of real-time precision is under 0.5%, volume is small, weight is light, and reliability is high.

The invention discloses a resource reserving method based on service volume, which includes following steps: establishes at least one set according to the prior D day`s service volume before the nth time span of the (D+1)th day to be forecasted and the prior (n-1)th time span`s service volume of the (D+1)th day; and then divides all the service volume in each set into at least one category; then forecasts the resource needed in the nth time span of the (D+1)th day to be forecasted according to the category; and finally, reserves the resource according to the forecasted result. The invention also discloses a resource reserving device based on service volume, which includes a construction module, a classification module and a forecast module. Through the method and the device, the invention can realize reasonable resource reservation, therefore can reduce the session switch dropping rate and farthest improve the throughput of the system.