A low-power power supply monitoring method and system based on artificial intelligence

Abstract

The present invention relates to a low-power power monitoring based on artificial intelligence, including step S1: acquiring power circuit sensor data to form a port parameter set X; step S2: embedding a piecewise linear interpolation function into a convolutional neural network structure; step S3: training the historical port parameter set X and the corresponding historical solution to obtain a power self-regulation model, wherein the power self-regulation model includes a plurality of deep neural networks with the same number of battery cells; step S4: determining the weights of different battery cell positions through training to obtain a final power self-regulation model; step S5: the power self-regulation model is transmitted to a display screen in real time through a communication circuit according to the real-time parameters of the circuit, and the abnormal position is displayed and a solution is given. The power self-regulation model of the present invention includes the setting of a plurality of deep neural networks with the same number of battery cells, which greatly enhances the accuracy of power monitoring and enhances the user experience.

Description

Technical Field

[0001] The present invention relates to the technical field of power supply monitoring, and in particular to a low-power consumption power supply monitoring method and system based on artificial intelligence. Background Art

[0002] At present, in the process of modern social and economic development, it is inseparable from the provision of electricity at any time. It is not difficult to see the importance of power supply. Conversely, the development of social economy and science and technology also provides economic and technical resources to support the development of power supply applications. Therefore, power supply applications and social economy promote each other and develop in a coordinated manner. At present, under the social development environment, new and higher requirements are put forward for power supply. It is necessary to pay attention to the application of power supply monitoring system in power supply to ensure the healthy, stable and sustainable development of the power supply industry. From a macro perspective, this also has important practical significance and value for the development of social economy.

[0003] In the prior art, although there are many power monitoring systems and methods, with the vigorous growth of power demand for new energy vehicles and the changes in multiple power application scenarios, the data accuracy of the existing power monitoring system is low, the data transmission rate is slow, and it cannot match the historical data. In particular, the battery data of existing new energy vehicles needs to be monitored in real time, and there are high requirements for how to realize automated monitoring of various batteries, lithium batteries, and lead-acid batteries, which does not match the existing large demand for timely updating and training of power data. How to make power monitoring more intelligent and humanized, improve its operating efficiency and accuracy, and enhance user readability has become a new research topic, and the data transmission accuracy and efficiency of the existing power monitoring sensor are low; and it is mainly aimed at the processing of existing power supplies, and does not involve the method of predicting current and future power data through historical power monitoring data. Therefore, a power monitoring technology that can increase the intelligence of power monitoring has become an urgent need to improve the user experience. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a low-power power supply monitoring system method and system based on artificial intelligence. The low-power power supply monitoring system method and system based on artificial intelligence of the present invention significantly improve the data accuracy and transmission efficiency of power supply monitoring, greatly enhance the accuracy, and enhance the user experience; the following technical scheme is adopted to achieve: a low-power power supply monitoring method based on artificial intelligence, comprising step S1: obtaining power circuit sensor data, including battery total voltage, battery cell voltage, battery cell temperature, DC distribution branch voltage, DC distribution branch current, rectifier three-phase input voltage, rectifier three-phase input current, AC distribution three-phase voltage, AC The three-phase current of the power distribution forms a port parameter set X; step S2: embed the piecewise linear interpolation function into the convolutional neural network structure to generate a self-regulating model; step S3: train the historical port parameter set X and the corresponding historical solution to obtain a power self-regulating model, wherein the power self-regulating model includes a plurality of deep neural networks with the same number of battery cells; step S4: determine the weights of different battery cell positions through training to obtain a final power self-regulating model; step S5: the power self-regulating model transmits the real-time parameters of the circuit to the display screen through the communication circuit in real time, displays the abnormal position and gives a solution, and then adjusts the data collected from different battery cells in real time;

[0005] The weights of different battery cell positions are obtained by the following steps: the influence values ​​of power participation when the battery cell works alone or all battery cells work at the same time are trained, and the influence value matrix S is set, where S(i,j) represents the power measurement data X when the battery cell i and battery j change equally. i With X j The difference between them is expressed as S(i,j)=-||X i -X j || 2 , then S is a×a matrix, S(d,d) represents the difference value of the same monomer under different power supply environments; the maximum number of iterations K max , the weight of the battery cell position is obtained as δ=[δ1,δ2,...,δ a ].

[0006] Preferably, the piecewise linear interpolation function is embedded in the convolutional neural network structure to generate a self-regulating model, and the global pooling layer of the residual network is deleted, and the last fully connected layer is modified to a convolution layer with a convolution kernel size of 3*3 and a number of channels of D, and the output of the piecewise linear interpolation function is connected to the global pooling layer to generate a self-regulating model; the piecewise linear interpolation function is used to self-calibrate the sensor parameters.

[0007] Preferably, it also includes a port parameter set X data normalization processing module, which is used to normalize the convolution features extracted from the power circuit sensor data.

[0008] Preferably, the port parameter set X includes information filtering and denoising, feature extraction, and determination of battery orientation information.

[0009] Preferably, the communication circuit uses a carrier signal with a specific DC component and some AC low-frequency components removed for transmission, that is:

[0010] , where A is the amplitude, is the modulation adjustment value, Indicates the adjustment code signal, To increase the signal value range, t is time.

[0011] A low-power power monitoring system based on artificial intelligence, characterized in that it includes ADuC812, a voltage detection sensor, a current detection sensor, a display circuit, RS-232, a drive circuit, and an LCD display screen, wherein the ADuC812 includes a self-regulating model calculation module, and the specific calculation method is as follows: the parameter acquisition module is used to obtain power circuit sensor data, including battery total voltage, battery cell voltage, battery cell temperature, DC distribution branch voltage, DC distribution branch current, rectifier three-phase input voltage, rectifier three-phase input current, AC distribution three-phase voltage, and AC distribution three-phase current to form a port parameter set X; the self-regulating model generation module embeds a piecewise linear interpolation function into a convolutional neural network structure to generate a self-regulating model, and performs a self-regulating model generation on the historical port parameter set X. The power self-regulation model is obtained by training the corresponding historical solutions, and the power self-regulation model includes a plurality of deep neural networks with the same number of battery cells; a weight acquisition module, which determines the weights of different battery cell positions through training to obtain the final power self-regulation model; a real-time display module, the power self-regulation model transmits the real-time parameters of the circuit to the display screen through the communication circuit in real time, displays the abnormal position and gives a solution, and then adjusts the data collected from different battery cells in real time; the weights of different battery cell positions are obtained by the following steps: the influence value of the power participation when the battery cell works alone or all the battery cells work at the same time is trained, and the influence value matrix S is set, and S(i,j) represents the power measurement data X under the same change of battery cell i and battery j i With X j The difference between them is expressed as S(i,j)=-||X i -X j || 2 , then S is a×a matrix, S(d,d) represents the difference value of the same monomer under different power supply environments; set the maximum number of iterations Kmax , the weight of the battery cell position is obtained as

[0012] δ=[δ1,δ2,...,δ a ].

[0013] Preferably, the piecewise linear interpolation function is embedded in the convolutional neural network structure to generate a self-adjusting model, the global pooling layer of the residual network is deleted, the last fully connected layer is modified to a convolution layer with a convolution kernel size of 3*3 and a number of channels of D, and the output of the piecewise linear interpolation function is connected to the global pooling layer to generate a self-adjusting model; the piecewise linear interpolation function is used to self-calibrate the sensor parameters.

[0014] Preferably, it also includes a port parameter set X data normalization processing module, which is used to normalize the convolution features extracted from the power circuit sensor data.

[0015] Preferably, the port parameter set X includes information filtering and denoising, feature extraction, and determination of battery orientation information.

[0016] Preferably, the communication circuit uses a carrier signal with a specific DC component and some AC low-frequency components removed for transmission, that is:

[0017] , where A is the amplitude, is the modulation adjustment value, Indicates the adjustment code signal, To increase the signal value range, t is time.

[0018] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0019] The problem that power supply monitoring in traditional technology is not associated with historical surveying and mapping data is solved. The power supply monitoring method of the present application can update the historical monitoring data in real time and analyze the current data to find out the abnormal points; the self-regulating model is generated by embedding the piecewise linear interpolation function into the convolutional neural network structure, and then the battery cell calculation method with different weight scales is obtained, which solves the shortcomings of single data prediction and low accuracy in the prior art; the communication circuit adopts the carrier signal that removes the specific DC component and part of the AC low-frequency component for transmission, and sets In order to increase the signal value range, low-value signals are filtered out and high-value signals are retained, thereby improving the accuracy of power supply monitoring; through the setting of the matching process, the data accuracy and comparison with historical power supply monitoring data are greatly increased; through the setting of multiple deep neural networks in the power supply self-regulation model, the accuracy of power supply monitoring is greatly enhanced, the accuracy of power supply monitoring and transmission efficiency are improved, and the user experience is enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a diagram of a low-power power supply monitoring system based on artificial intelligence of the present invention; DETAILED DESCRIPTION

[0021] Those skilled in the art understand that, as mentioned in the background technology, the traditional power monitoring has low accuracy, slow data transmission rate, and cannot match historical data. Although there are many power monitoring systems and methods, with the vigorous growth of power demand for new energy vehicles and the changes in multiple power application scenarios, the existing power monitoring system data has low accuracy, slow data transmission rate, and cannot match historical data. In particular, the battery data of existing new energy vehicles needs to be monitored in real time. There are high requirements for how to realize automatic monitoring of various batteries, lithium batteries, and lead-acid batteries, which does not match the existing large demand for timely updating and training of power data. How to make power monitoring more intelligent and humanized, improve its operating efficiency and accuracy, and enhance user readability has become a new research topic, and the data transmission accuracy and efficiency of existing power monitoring sensors are low; and it is mainly aimed at the processing of existing power supplies, and does not involve the method of predicting current and future power data through historical power monitoring data. Therefore, a power monitoring technology that can increase the intelligence of power monitoring has become an urgent need, thereby improving the user experience. How to make power monitoring more intelligent and humanized, improve its operating efficiency and accuracy, and enhance user readability has become a new research topic. In order to make the above-mentioned objects, features and beneficial effects of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0022] Embodiment 1:

[0023] Figure 1 A system diagram of a low-power power monitoring method based on artificial intelligence of the present application is shown, including step S1: obtaining power circuit sensor data, including battery total voltage, battery cell voltage, battery cell temperature, DC power distribution shunt voltage, DC power distribution shunt current, rectifier three-phase input voltage, rectifier three-phase input current, AC power distribution three-phase voltage, AC power distribution three-phase current to form a port parameter set X;

[0024]

[0025] Step S2: embed the piecewise linear interpolation function into the convolutional neural network structure to generate a self-regulating model; Step S3: train the historical port parameter set X and the corresponding historical solution to obtain a power self-regulating model, wherein the power self-regulating model includes a plurality of deep neural networks having the same number as the battery cells; Step S4: determine the weights of different battery cell positions through training to obtain a final power self-regulating model; Step S5: the power self-regulating model transmits the circuit real-time parameters to the display screen through the communication circuit in real time, displays the abnormal position and gives a solution, and then adjusts the data collected from different battery cells in real time;

[0026] In some embodiments, the voltage and current are detected by some methods: ① online detection of voltage and current: the voltage measurement range is 0V~1000V, and the accuracy is 1V; the current measurement range is OA~50A, and the accuracy is 10mA; ② online display of voltage and current: display the current voltage and current values; Set voltage and current thresholds: The voltage and current alarm thresholds can be set through the keyboard or serial port; ④ Over (under) voltage and over (under) current protection and alarm: When the voltage or current exceeds (or is lower than) the set threshold, the system will sound an alarm and automatically cut off the power supply until the fault is eliminated; ⑤ Communication function: It has master-slave communication and LCD and PC communication functions; ⑥ Anti-interference function: It has good anti-interference ability and can be used in most occasions.

[0027] A / D conversion is a device that converts the measured analog quantity into a digital quantity. The conversion accuracy of this device directly affects the measurement accuracy of the system. The conversion accuracy of the 12-bit high-performance ADC integrated in the ADuc812 microcontroller is 0.0006V.

[0028] The weights of different battery cell positions are obtained by the following steps: the influence values ​​of power participation when the battery cell works alone or all battery cells work at the same time are trained, and the influence value matrix S is set, where S(i,j) represents the power measurement data X when the battery cell i and battery j change equally. i With X j The difference between them is expressed as S(i,j)=-||X i -X j || 2 , then S is a×a matrix, S(d,d) represents the difference value of the same monomer under different power supply environments; the maximum number of iterations K max , the weight of the battery cell position is obtained as δ=[δ1,δ2,...,δ a ].

[0029] In some embodiments, K maxIt is the number of calculation iterations set manually based on the experience value or when the square difference between the weight value after training and the historical experience value is less than the set threshold;

[0030] In some embodiments, the influence value of power participation when the battery cells work individually or all the battery cells work simultaneously is trained, including training with a convolutional neural network, wherein the input of the convolutional neural network is the parameters of the battery cells in various historical states and their changes, and the output is the optimized weight value.

[0031] The piecewise linear interpolation function is embedded in the convolutional neural network structure to generate a self-regulating model. The PLC programming software uses STEP 7-Micro / WIN SMART. Through concise program design, the real-time monitoring of the system is completed. The RMP power supply monitoring system designs the main program and the corresponding subprograms. The required subprograms can be called when controlling. The PLC main program flow of the RMP power supply. Before the program runs, temporary fault detection is included as follows: first initialize the blocking pulse, determine whether the power supply has a phase failure, determine whether the set value is zero, and then make a high-voltage closing judgment. After the high-voltage closing is detected, the blocking pulse is unlocked, the current is adjusted according to the given input, and the actual current and voltage are detected to determine whether there is a fault signal. If there is a fault signal, the fault detection subroutine is called to detect the corresponding fault, the fault subroutine, and the fault alarm is issued and the pulse is blocked; if there is no fault signal, then after the detection, it is determined whether there is a shutdown information. If there is a shutdown information, the pulse is blocked.

[0032] In some embodiments, the piecewise linear interpolation function is embedded in the convolutional neural network structure to generate a self-adjusting model, the global pooling layer of the residual network is deleted, the last fully connected layer is modified to a convolution layer with a convolution kernel size of 3*3 and a number of channels of D, and the output of the piecewise linear interpolation function is connected to the global pooling layer to generate a self-adjusting model; the piecewise linear interpolation function is used to self-calibrate the sensor parameters.

[0033] In some embodiments, a port parameter set X data normalization processing module is also included, which is used to normalize the convolution features extracted from the power circuit sensor data.

[0034] In some embodiments, the port parameter set X includes filtering and denoising of information, feature extraction, and determination of battery orientation information.

[0035] In some embodiments, the communication circuit uses a carrier signal with a specific DC component and some AC low-frequency components removed for transmission, that is:

[0036] , where A is the amplitude, is the modulation adjustment value, Indicates the adjustment code signal, To increase the signal value range, t is time.

[0037] Embodiment 2:

[0038] A low-power power monitoring system based on artificial intelligence includes ADuC812, a voltage detection sensor, a current detection sensor, a display circuit, RS-232, a drive circuit, and an LCD display screen, wherein the ADuC812 includes a self-regulating model calculation module, and the specific calculation method is as follows: the parameter acquisition module is used to obtain power circuit sensor data, including battery total voltage, battery cell voltage, battery cell temperature, DC distribution branch voltage, DC distribution branch current, rectifier three-phase input voltage, rectifier three-phase input current, AC distribution three-phase voltage, and AC distribution three-phase current to form a port parameter set X; the self-regulating model generation module embeds a piecewise linear interpolation function into a convolutional neural network structure to generate a self-regulating model, and performs a self-regulating model generation on the historical port parameter set X and the corresponding The power self-regulation model is obtained by training with historical solutions, and the power self-regulation model includes a plurality of deep neural networks with the same number of battery cells; a weight acquisition module is used to determine the weights of different battery cell positions through training to obtain the final power self-regulation model; a real-time display module is used, and the power self-regulation model is transmitted to the display screen through the communication circuit in real time according to the real-time parameters of the circuit, and the abnormal position is displayed and a solution is given, so as to adjust the data collected from different battery cells in real time; the weights of different battery cell positions are obtained by the following steps: the influence value of the power participation when the battery cell works alone or all the battery cells work at the same time is trained, and the influence value matrix S is set, and S(i,j) represents the power measurement data X when the battery cell i and the battery j change equally i With X j The difference between them is expressed as S(i,j)=-||X i -X j || 2 , then S is a×a matrix, S(d,d) represents the difference value of the same monomer under different power supply environments; the maximum number of iterations K max , the weight of the battery cell position is obtained as δ=[δ1,δ2,...,δ a ].

[0039] In some embodiments, the piecewise linear interpolation function is embedded in the convolutional neural network structure to generate a self-adjusting model, the global pooling layer of the residual network is deleted, the last fully connected layer is modified to a convolution layer with a convolution kernel size of 3*3 and a number of channels of D, and the output of the piecewise linear interpolation function is connected to the global pooling layer to generate a self-adjusting model; the piecewise linear interpolation function is used to self-calibrate the sensor parameters.

[0040] In some embodiments, a port parameter set X data normalization processing module is also included, which is used to normalize the convolution features extracted from the power circuit sensor data.

[0041] In some embodiments, the port parameter set X includes filtering and denoising of information, feature extraction, and determination of battery orientation information.

[0042] In some embodiments, the communication circuit uses a carrier signal with a specific DC component and some AC low-frequency components removed for transmission, that is:

[0043] , where A is the amplitude, is the modulation adjustment value, Indicates the adjustment code signal, To increase the signal value range, t is time.

[0044] The method and system of the present invention solve the problem that power supply monitoring in traditional technology is not associated with historical surveying and mapping data. The power supply monitoring method of the present application can realize real-time updating of historical monitoring data and analyze current data to find abnormal points; a self-regulating model is generated by embedding a piecewise linear interpolation function into a convolutional neural network structure, and then a battery cell calculation method with different weight scales is obtained, which solves the shortcomings of single data prediction and low accuracy in the prior art; the communication circuit adopts a carrier signal that removes a specific DC component and some AC low-frequency components for transmission, and sets In order to increase the signal value range, low-value signals are filtered out and high-value signals are retained, thereby improving the accuracy of power supply monitoring; through the setting of the matching process, the data accuracy and comparison with historical power supply monitoring data are greatly increased. The microprocessor-based power supply monitoring method and system of the present invention significantly improve the power supply monitoring accuracy and transmission efficiency, greatly enhance the accuracy, and enhance the user experience.

[0045] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products, and thus the present application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0046] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the scope defined by the claims.

Claims

1. A low-power power monitoring method based on artificial intelligence, characterized in that: The method comprises the steps of: obtaining power circuit sensor data, including battery total voltage, battery cell voltage, battery cell temperature, DC power distribution shunt voltage, DC power distribution shunt current, rectifier three-phase input voltage, rectifier three-phase input current, AC power distribution three-phase voltage, and AC power distribution three-phase current to form a port parameter set X; step S2: embedding a piecewise linear interpolation function into a convolutional neural network structure to generate a self-regulating model; step S3: training the historical port parameter set X and the corresponding historical solutions to obtain a power self-regulating model, wherein the power self-regulating model includes a plurality of deep neural networks having the same number as the battery cells; step S4: determining the weights of different battery cell positions through training to obtain a final power self-regulating model; step S5: the power self-regulating model transmits the data to a display screen through a communication circuit in real time according to the real-time parameters of the circuit, displays the abnormal position and gives a solution, and then adjusts the collected data of different battery cells in real time; The weights of different battery cell positions are obtained by the following steps: the influence values ​​of power participation when the battery cell works alone or all battery cells work at the same time are trained, and the influence value matrix S is set, where S(i,j) represents the power measurement data X when the battery cell i and battery j change equally. i With X j The difference between them is expressed as S(i,j)=-||X i -X j || 2 , then S is a×a matrix, S(d,d) represents the difference value of the same monomer under different power supply environments; the maximum number of iterations K max , the weight of the battery cell position is obtained as δ=[δ1,δ2,...,δ a ]; The piecewise linear interpolation function is embedded in the convolutional neural network structure to generate a self-regulating model, and the global pooling layer of the residual network is deleted, and the last fully connected layer is modified to a convolution layer with a convolution kernel size of 3*3 and a number of channels of D. The output of the piecewise linear interpolation function is connected to the global pooling layer to generate a self-regulating model; the piecewise linear interpolation function is used to self-calibrate the sensor parameters.

2. The low-power power monitoring method based on artificial intelligence according to claim 1 is characterized in that: It also includes a port parameter set X data normalization processing module, which is used to normalize the convolution features extracted from the power circuit sensor data.

3. The low-power consumption power supply monitoring method based on artificial intelligence according to claim 1 is characterized in that: The port parameter set X includes information filtering and denoising, feature extraction, and determination of battery orientation information.

4. The low-power power monitoring method based on artificial intelligence according to claim 1 is characterized in that: The communication circuit uses a carrier signal with a specific DC component and some AC low-frequency components removed for transmission, namely: f(t)=Acos[ωt+θc(t)]+γ, where A is the amplitude, θ is the modulation adjustment value, c(t) represents the adjustment code signal, γ is the range value of the increased signal value, and t is time.

5. A low-power power monitoring system based on artificial intelligence, characterized in that: It includes ADuC812, voltage detection sensor, current detection sensor, display circuit, RS-232, drive circuit, LCD display screen, wherein ADuC812 includes self-regulating model calculation module, and the specific calculation method is as follows: the parameter acquisition module is used to obtain the power circuit sensor data, including the total battery voltage, battery cell voltage, battery cell temperature, DC distribution branch voltage, DC distribution branch current, rectifier three-phase input voltage, rectifier three-phase input current, AC distribution three-phase voltage, AC distribution three-phase current to form a port parameter set X; the self-regulating model generation module embeds the piecewise linear interpolation function into the convolutional neural network structure to generate a self-regulating model, trains the historical port parameter set X and the corresponding historical solution, and obtains the power self-regulating model, wherein the power self-regulating model includes a plurality of deep neural networks with the same number of battery cells; The weight acquisition module determines the weights of different battery cell positions through training to obtain the final power self-regulation model; The real-time display module, the power self-adjustment model transmits the real-time parameters of the circuit to the display screen through the communication circuit in real time, displays the abnormal position and gives a solution, and then adjusts the data collected by different battery cells in real time; the weights of different battery cell positions are obtained by the following steps: the influence value of the power participation when the battery cell works alone or all battery cells work at the same time is trained, and the influence value matrix S is set, S(i,j) represents the power measurement data X under the same change of battery cell i and battery j i With X j The difference between them is expressed as S(i,j)=-||X i -X j || 2 , then S is a×a matrix, S(d,d) represents the difference value of the same monomer under different power supply environments; the maximum number of iterations K max , the weight of the battery cell position is obtained as δ=[δ1,δ2,...,δ a ]; The piecewise linear interpolation function is embedded in the convolutional neural network structure to generate a self-adjusting model, the global pooling layer of the residual network is deleted, the last fully connected layer is modified to a convolution layer with a convolution kernel size of 3*3 and a number of channels of D, and the output of the piecewise linear interpolation function is connected to the global pooling layer to generate a self-adjusting model; the piecewise linear interpolation function is used to self-calibrate the sensor parameters.

6. The low-power consumption power supply monitoring system based on artificial intelligence according to claim 5 is characterized in that: It also includes a port parameter set X data normalization processing module, which is used to normalize the convolution features extracted from the power circuit sensor data.

7. The low-power consumption power supply monitoring system based on artificial intelligence according to claim 5 is characterized in that: The port parameter set X includes information filtering and denoising, feature extraction, and determination of battery orientation information.

8. The low-power consumption power monitoring system based on artificial intelligence according to claim 5 is characterized in that: The communication circuit uses a carrier signal with a specific DC component and some AC low-frequency components removed for transmission, namely: f(t)=Acos[ωt+θc(t)]+γ, where A is the amplitude, θ is the modulation adjustment value, c(t) represents the adjustment code signal, γ is the range value of the increased signal value, and t is time.

Images