Neural network energy storage battery operating status monitoring method based on RRAM array

Abstract

The present invention discloses a neural network energy storage battery operating status monitoring method based on an RRAM array. The method specifically includes the following steps: Step 1: Using an energy storage battery characteristic quantity measurement module, the characteristic quantities of the energy storage battery module are collected. The characteristic quantities include characteristic gases, battery temperature, and characteristic sounds generated by the safety valve; Step 2: Converting the characteristic quantities collected in Step 1 into optical signals, which are then converted into electrical signals using an optical fiber demodulator and transmitted to a signal processing module; Step 3: The signal processing module processes the collected characteristic quantities and transmits them to a neural network module; Step 4: After receiving the processed signal from the signal processing module, the neural network module identifies abnormal energy storage battery status based on the circuit triggering status of the trained neural network and transmits the information to a monitoring center. The present invention can predict the health status of energy storage batteries and quickly and accurately determine the health status of energy storage batteries.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery health status monitoring in energy storage power stations, and relates to a method for monitoring the operating status of energy storage batteries using a neural network based on an RRAM array. Background Art

[0002] With the proposed "dual carbon" goals, the future will be towards building a new power system dominated by new energy. The intermittent and volatile nature of new energy sources such as photovoltaics and wind power can impact the stability and reliability of power supply. Energy storage can effectively promote the absorption of new energy sources and become a key strategic support for their large-scale utilization in the future. Energy storage supporting new energy sources has become a major trend. However, currently commercialized energy storage battery modules can, for various reasons, cause thermal runaway or even fire during operation. These safety hazards pose significant challenges to the safe and stable operation of energy storage systems supporting new energy sources.

[0003] To ensure the safe and stable operation of energy storage power stations, it is necessary to detect and provide early warnings about the health status of the battery modules in electrochemical energy storage power stations. Current BMS (Battery Management Systems) often suffer from delayed warnings and an inability to fully reflect the health status of energy storage batteries, leading to lagging management of energy storage power stations. Summary of the Invention

[0004] The purpose of the present invention is to provide a neural network energy storage battery operating status monitoring method based on RRAM array, which can predict the health status of the energy storage battery and quickly and accurately judge the health status of the energy storage battery.

[0005] The technical solution adopted by the present invention is a method for monitoring the operating status of a neural network energy storage battery based on an RRAM array, which specifically includes the following steps:

[0006] Step 1: Collect characteristic quantities of the energy storage battery module through the energy storage battery characteristic quantity measurement module, where the characteristic quantities include characteristic gas, battery temperature, and characteristic sound generated by the safety valve;

[0007] Step 2: convert the feature quantity collected in step 1 into an optical signal, and then convert the optical signal into an electrical signal through an optical fiber demodulator and transmit it to the signal processing module;

[0008] Step 3: The signal processing module processes the collected feature values ​​and transmits them to the neural network module;

[0009] In step 4, after the neural network module receives the signal processed by the signal processing module, it loads the signal onto the trained RRAM array, which will trigger the corresponding RRAM circuit. Based on the circuit triggering status of the trained neural network, the abnormal state of the energy storage battery can be identified and the information can be sent to the monitoring center.

[0010] The present invention is also characterized in that:

[0011] In step 1, the characteristic gas is measured by a fiber optic gas sensor; the battery temperature is measured by a fiber optic temperature sensor; and the characteristic sound generated by the safety valve is measured by a fiber optic sound pressure sensor.

[0012] In step 2, the optical fiber demodulator converts the optical signal into an electrical signal and transmits it to the signal processing module through the 5G network.

[0013] In step 3, the neural network module includes an input layer and an output layer, wherein the input layer includes a first neuron circuit and a first weight RRAM array; and the output layer includes a second weight RRAM array and a second neuron circuit.

[0014] In the input layer of the neural network, the function of the first neuron circuit is to determine whether the input data exceeds the set threshold. If it does, the first neuron is activated and the set-reset operation is completed. After multiple set-reset operations, the first weight RRAM array will form a fixed resistance.

[0015] The beneficial effects of the present invention are as follows:

[0016] 1. This invention improves the tediousness of manual inspections and the misjudgment caused by human error through online real-time monitoring, which helps improve the power supply quality of energy storage power stations and ensure safety and reliability.

[0017] 2. The present invention can monitor the operating status of energy storage batteries in real time and provide early warning of energy storage battery failures through historical data;

[0018] 3. The RRAM neural network can self-learn the health status data of the energy storage battery, form memories in the neural network through training, and monitor the health status of the energy storage battery in real time.

[0019] 4. The optical fiber sensor based on the FP principle used in the present invention has the advantages of high sensitivity, fast response, and high precision, and can quickly, timely and accurately detect data on multiple characteristic values ​​of the measured energy storage battery. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a schematic diagram of the energy storage battery operating status system structure of the RRAM array-based neural network energy storage battery operating status monitoring method of the present invention;

[0021] Figure 2 Schematic diagram of an optical fiber gas sensor in the RRAM array-based neural network energy storage battery operating status monitoring method of the present invention;

[0022] Figure 3This is a diagram of the bayonet structure of the cavity top of the optical fiber gas sensor in the RRAM array-based neural network energy storage battery operating status monitoring method of the present invention;

[0023] Figure 4 This is a structural diagram of the stepped slot at the bottom of the cavity of the optical fiber gas sensor in the RRAM array-based neural network energy storage battery operating status monitoring method of the present invention;

[0024] Figure 5 This is a structural diagram of the optical fiber ceramic pin in the optical fiber gas sensor in the RRAM array-based neural network energy storage battery operating status monitoring method of the present invention;

[0025] Figure 6 It is a schematic diagram of the application system structure of the optical fiber gas sensor in the neural network energy storage battery operating status monitoring method based on the RRAM array of the present invention.

[0026] In the figure, 1. Fabry-Perot cavity, 2. Fiber optic ceramic pin, 3. Grid hole A, 4. Grid hole B, 5. Slot, 6. Sensitive end reflective surface, 7. Top fixing piece, 8. Bayonet, 9. Fiber optic ceramic pin socket, 10. Fiber optic ceramic pin signal transmitting end, 11. Fiber optic ceramic pin tail, 12. Fiber optic cable, 13. Fiber optic gas sensor, 14. Fiber optic sensor demodulator, 15. Computer unit. DETAILED DESCRIPTION

[0027] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] The present invention is based on the neural network energy storage battery operation status monitoring method of RRAM array, such as Figure 1 As shown, it includes a measurement module E1, a signal processing module E2, a neural network module E3 and an energy storage battery monitoring center E4.

[0029] The measurement module E1 represents all measurement units in the energy storage battery cabinet, covering characteristic quantity monitoring modules E11 to E1n for 1 to n energy storage batteries. The physical quantities detected by each monitoring module include the characteristic gas released by the energy storage battery, the temperature of the energy storage battery, and the characteristic sound emitted when the safety valve is activated.

[0030] The signal processing module E2 processes the characteristic quantities of the state of each energy storage battery through a filtering unit, a signal amplifying unit and an analog-to-digital conversion unit.

[0031] The submodules E21 to E2n included in the signal processing module E2 process the information from the corresponding 1 to n energy storage battery state characteristic quantities respectively; the processed information is fed into the neural network module E3 for analysis and organization.

[0032] Neural network module E3 organizes and analyzes the information processed by the processing module in real time, ultimately categorizing and summarizing it in different areas and sending it to the energy storage battery monitoring center E4. It also reports any abnormal battery conditions to the monitoring center E4. Based on this aggregated real-time and historical information, monitoring center E4 can assess the battery's health and schedule appropriate repairs or replacements for faulty or potentially hazardous batteries.

[0033] The communication between the above different modules is carried out through the 5G communication network with ultra-high reliability and low latency communication to transmit information.

[0034] The measurement module E1 can collect information on multiple physical quantities that reflect the health status of the energy storage battery, such as characteristic gases, battery temperature, and the characteristic sound produced by the safety valve. Among them, the sensor is the main component of the measurement module E1. The characteristic gases of the energy storage battery are measured by the fiber optic gas sensor;

[0035] The battery temperature of the energy storage battery is measured by a fiber optic temperature sensor;

[0036] The characteristic sound generated by the safety valve in the energy storage battery is measured by a fiber optic sound pressure sensor;

[0037] The fiber optic gas sensor and fiber optic temperature sensor are installed at the center above the energy storage battery module, and the fiber optic sound pressure sensor is installed near the safety valve of the energy storage battery.

[0038] The signal processing module E2 includes a filtering unit, a signal amplifying unit, and an analog-to-digital conversion unit. The data processed by each signal processing module E2 is summarized in the signal analysis module.

[0039] The neural network module E3 is connected to each signal processing module E2 through a 5G communication network. The neural network module E3 is an RRAM neural network, and its function is to identify the status of the energy storage battery through a neural network trained with historical data.

[0040] The energy storage battery monitoring center manages the battery status of the entire energy storage power station and classifies and stores the real-time data of the energy storage batteries.

[0041] This invention monitors the operating status of each energy storage battery online and uses RRAM to aggregate and calculate multiple characteristic parameters, including gas release, temperature, and sound, from each battery in the battery cluster. This aggregate calculates the characteristic parameter values ​​for the entire battery cluster, and then provides early safety warnings for the battery modules based on the characteristic values ​​of each battery. This facilitates the precise identification of faulty batteries, facilitates the timely detection of safety hazards, and significantly improves the operational performance and grid-connected safety of energy storage power stations.

[0042] The optical fiber sensor made by the FP (Fabry-Perot) interferometry principle in the present invention has the advantages of high sensitivity, passivity and rapid response. It can accurately and real-timely detect the state characteristics of the energy storage battery. At the same time, the use of multiple eigenvalues ​​online detection also increases the accuracy of early warning and improves the operational safety of the energy storage power station.

[0043] Fiber optic gas sensors based on Fabry-Perot cavity, such as Figures 2-4 As shown, it includes a Fabry-Perot cavity body 1, a fiber optic ceramic pin 2, a top fixing part 7, and a sensitive end reflective surface 6. The fiber optic ceramic pin 2 is inserted into a cylindrical fiber optic ceramic pin socket 9 designed on the top of the Fabry-Perot cavity body 1. A bayonet 8 is designed at the top of the Fabry-Perot cavity body 1. Figure 5 As shown, the bayonet 8 is used to fix the fiber optic ceramic pin tail 11. The fiber optic ceramic pin tail 11 is connected to the fiber optic ceramic pin signal transmitting end 10; the fiber optic ceramic pin 2 is connected to the optical fiber line 12.

[0044] The Fabry-Perot chamber 1 is designed with three groups of quarter-circular grid holes A3 and three groups of quarter-circular grid holes B4 near the bottom side, and each group of circular grid holes is designed to be aligned. The three groups of quarter-circular grid holes A3 and the three groups of quarter-circular grid holes B4 are symmetrically arranged one by one.

[0045] A slot 5 is designed at the bottom of the Fabry-Perot cavity 1 , and the sensitive end reflective surface 6 is placed in the designed slot 5 .

[0046] The quarter-circular aperture A3 at the bottom of the Fabry-Perot chamber 1 is 0.5 mm from the bottom surface of the chamber 1. Each aperture is 0.5 mm high, and the spacing between adjacent apertures is also 0.5 mm. The horizontal length of the apertures is 4 mm. The structures of the quarter-circular aperture A3 and the quarter-circular aperture B4 are identical.

[0047] The present invention is based on an optical fiber gas sensor of a Fabry-Perot cavity. The light source signal adopts a broadband laser light source with a wavelength of 40nm. The light signal is emitted by the ceramic pin to the reflective surface of the sensitive end. When reaching the reflective surface of the sensitive end, the light signal will be refracted. Since the refractive index n of the light signal in different gases is different, it can be shown from formula (1) that the intensity of the light will change. Air is initially introduced into the Fabry-Perot cavity 1, and the refractive index n of air is 1. By changing the gas composition in the Fabry-Perot cavity 1, the refractive index n of the light signal will change with different gas compositions. At this time, the intensity of the incident light will change accordingly. The light signal is passed into the optical fiber coupler, and then the optical signal is demodulated and calculated by the optical fiber sensor demodulator.

[0048] The manufacturing method and steps of the fiber optic gas sensor based on the Fabry-Perot cavity of the present invention are as follows:

[0049] Step 1: Design the fiber-optic Fabry-Perot cavity gas sensor cavity. The overall structure is cylindrical, with a bayonet fixture at the top of the cylinder corresponding to the tail of the fiber-optic ceramic ferrule, which allows for stable and horizontal insertion of the ferrule into the cavity. The cylindrical cavity is divided into two sections. The upper section is designed with a cylindrical through-hole with a diameter of 2.6 mm in the center for inserting the fiber-optic ceramic ferrule. The lower section has a slightly larger cylindrical hole in the middle for filling the gas to be measured. A quarter-circular ring-shaped aperture is designed near the bottom of the lower section to allow the gas to be measured to enter the Fabry-Perot cavity. The bottom of the cylindrical cavity is designed as a step to accommodate the sensitive end reflector 6.

[0050] Step 2: Insert the optical fiber ceramic pin 2 into the designed Fabry-Perot cavity 1. A top fixing piece 7 is provided on the top of the Fabry-Perot cavity 1. The design of the top fixing piece 7 allows the tail of the optical fiber ceramic pin 2 to be firmly inserted into the Fabry-Perot cavity 1.

[0051] Step 3: Place the sensitive end reflective surface 6 horizontally in the stepped bayonet 8 designed at the bottom of the cylindrical Fabry-Perot cavity 1, and then use glue to bond the sensitive end reflective surface 6 to the surrounding Fabry-Perot cavity 1.

[0052] Through the above three steps, the production of the optical fiber gas sensor is completed.

[0053] The beneficial effects of the design of the top fixing piece 7 and the bottom stepped bayonet 8 of the Fabry-Perot chamber body 1 in steps 2 and 3 are as follows:

[0054] The beneficial effect of the design of the top fixing member 7 is that the horizontal end surface of the optical fiber ceramic pin 2 emitting the optical signal can be kept level with the end surface of the cylindrical Fabry-Perot cavity 1 .

[0055] The beneficial effect of the design of the stepped bayonet 8 is that the sensitive end reflective surface 6 can be kept level with the end surface of the cylindrical Fabry-Perot cavity 1 .

[0056] The above two beneficial effects can keep the optical signal emitting end face of the optical fiber ceramic ferrule 2 and the sensitive end reflective surface 6 on different but parallel horizontal planes, making the round-trip reflection of light more horizontal, reducing optical signal loss, and improving the sensitivity of the sensor, making the measurement results more accurate.

[0057] Figure 2 The disc-shaped sensitive end reflective surface 6 is shown as only half, in order to make it easier to see the internal structure of the gas sensor. The actual sensitive end reflective surface 6 is a full circle.

[0058] like Figure 6As shown, the application system of the fiber optic gas sensor based on the Fabry-Perot cavity of the present invention includes a fiber optic gas sensor based on the Fabry-Perot cavity 13, a fiber optic sensor demodulator 14, and a computer unit 15. By connecting the fiber optic gas sensor 13 based on the Fabry-Perot cavity and the fiber optic sensor demodulator 14, the amount of the gas to be measured can be completed.

[0059] like Figure 2 As shown, when the gas to be measured flows into the quarter-circular gate holes A3 and B4 at the bottom of the optical fiber gas sensor 13, the gas composition in the Fabry-Perot cavity 1 changes, that is, the refractive index of the light changes, and the intensity of the light reflected back by the optical fiber sensor will change. The light intensity change caused by the change in the gas composition in the Fabry-Perot cavity I R It can be expressed as:

[0060]

[0061] Where R represents the reflectivity of the optical signal, I0 represents the incident light intensity, λ represents the wavelength of the optical signal, L represents the cavity length of the Fabry-Perot cavity, and n represents the refractive index of the FP cavity, typically air, where n = 1. The optical signal intensity is varied by introducing gases of varying concentrations. The optical fiber sensor interrogator can then calculate the gas composition corresponding to this varying optical signal intensity.

[0062] The present invention provides a method for monitoring the operating status of a neural network energy storage battery based on an RRAM array, which specifically includes the following steps:

[0063] In step 1, three fiber optic sensors within the energy storage battery characteristic quantity measurement module E1 measure different characteristic physical quantities. This information is collected and converted into optical signals. A fiber optic demodulator then converts the optical signals into electrical signals, which are then wirelessly transmitted to the signal processing module E2 via the 5G communication network. A fiber optic gas sensor and a fiber optic temperature sensor are installed at the center of the energy storage battery module to provide real-time measurements of characteristic gases and temperatures. A fiber optic sound pressure sensor is installed near the energy storage battery safety valve to provide real-time measurements of characteristic sound.

[0064] In step 2, the 5G communication network transmits the electrical signal demodulated by the fiber demodulator to the signal processing module E2 for processing of characteristic physical quantities. Each signal processing submodule E21 through E2n processes the corresponding energy storage battery status information and classifies the collected information according to different characteristic physical quantities before transmitting it to the neural network module E3 for data aggregation and calculation.

[0065] Step 3: The neural network module E3 (neural network module E3 is an RRAM neural network module) receives the data processed by the signal processing module E2. Since each characteristic signal will cause the voltage signal loaded on the RRAM device to be different, different current signals will be generated. These signals are loaded onto the trained RRAM array. The training process is to extract the data parameters of the monitoring characteristic quantity and input them into the RRAM. The three characteristic values ​​are stored by adjusting the corresponding resistance values ​​in the RRAM array. The first layer of the neural network transmits the electrical signal to the second layer of the neural network through the synapse. The second layer of neurons performs summation calculation on the signal, and the result of the calculation is recorded as U. A , and the threshold voltage U B Compare. When U A Greater than U B When the neural network module receives processed data from E2, it triggers the corresponding RRAM circuit. Based on the circuit triggering behavior of the trained neural network, abnormal energy storage battery conditions are identified and transmitted to the monitoring center. For temperature signals, the trained neural network's resistance range is set to R1-R2. If the resistance value triggered by the real-time temperature signal is outside the trained resistance range, an alarm signal is sent to the monitoring center. If it is within the trained resistance range, no alarm signal is issued. For gas signals, the trained neural network's resistance range is set to R3-R4. If the resistance value triggered by the real-time gas signal is outside the trained resistance range, an alarm signal is sent to the monitoring center. If it is within the trained resistance range, no alarm signal is issued. For sound signals: Assume that the resistance range of the trained neural network is R5-R6. If the resistance value triggered by the real-time sound signal is not within the trained resistance range, an alarm signal will be sent to the monitoring center. If it is within the trained resistance range, no alarm signal will be issued.

[0066] In step 4, the circuit triggering status and stored real-time data on the RRAM device are transmitted to the energy storage battery monitoring center E4 and displayed. Based on the abnormal status signals from the neural network module E3 and the real-time and historical operating status data of the energy storage battery, personnel can promptly arrange a reasonable repair or replacement plan for the energy storage battery experiencing abnormal conditions, saving troubleshooting time and improving the probability and efficiency of repair.

[0067] The neural network module E3 needs to be constructed in layers, with two layers in total;

[0068] The first neural network layer is the input layer, which receives data processed by the signal processing module E2. The input layer includes a first neuron circuit and a first weighted RRAM array. The first neuron circuit is used to determine whether the input data exceeds a set threshold. If so, the first neuron is activated, completing the set-reset operation. After multiple set-reset operations, the first weighted RRAM array will form a fixed resistance.

[0069] The second layer is an output layer, which includes a second weight RRAM array and a second neuron circuit;

[0070] The second weight RRAM array is located between the first neuron circuit and the second neuron circuit, and is used to determine the weight of output data of the first neuron circuit.

[0071] The second neuron circuit acts the same as the first neuron circuit.

[0072] Different positions in the output layer correspond to the areas of three different characteristic physical quantities of the energy storage battery operating status (characteristic gas released by the energy storage battery, the temperature of the energy storage battery, and the characteristic sound emitted when the safety valve is activated).

[0073] The data in these three different areas will be loaded with voltage signals of different sizes. These signals will trigger the pre-trained RRAM circuit. Depending on whether the triggered resistance value is within the resistance range corresponding to the trained neural network, it can be judged whether the real-time stored data is within the normal range.

[0074] Finally, the output result of the second neuron circuit is sent to the monitoring center E4 through the 5G communication network for display.

[0075] The monitoring center E4 can obtain real-time data on the health status of each energy storage battery in the energy storage power station. Through real-time data and signals transmitted by neural networks, the service center can formulate maintenance plans in a timely manner.

Claims

1. A neural network energy storage battery operating status monitoring method based on RRAM array, characterized by: The specific process includes the following: Step 1: Collect characteristic quantities of the energy storage battery module through the energy storage battery characteristic quantity measurement module, where the characteristic quantities include characteristic gas, battery temperature, and characteristic sound generated by the safety valve; In step 1, the characteristic gas is measured by a fiber optic gas sensor; the battery temperature is measured by a fiber optic temperature sensor; and the characteristic sound generated by the safety valve is measured by a fiber optic sound pressure sensor. The fiber optic gas sensor includes a fiber optic ceramic ferrule, which is inserted into a cylindrical fiber optic ceramic ferrule socket above a Fabry-Perot cavity. A bayonet is designed at the top of the Fabry-Perot cavity to secure the tail of the fiber optic ceramic ferrule, which is connected to the signal transmitting end of the fiber optic ceramic ferrule, and the fiber optic ceramic ferrule is connected to the optical fiber line. Near the bottom side of the Fabry-Perot cavity, there are three groups of quarter-circular grid holes A and quarter-circular grid holes B, and each group of circular grid holes is designed to be aligned. The three groups of quarter-circular grid holes A and the three groups of quarter-circular grid holes B are symmetrically arranged one by one. A slot is designed at the bottom of the Fabry-Perot cavity, and the reflective surface of the sensitive end is placed in the slot. The structures of the quarter-circular grid holes A and the quarter-circular grid holes B are exactly the same. Step 2: convert the feature quantity collected in step 1 into an optical signal, and then convert the optical signal into an electrical signal through an optical fiber demodulator and transmit it to the signal processing module; In step 2, the optical fiber demodulator converts the optical signal into an electrical signal and transmits it to the signal processing module via the 5G network; Step 3: The signal processing module processes the collected feature values ​​and transmits them to the neural network module; In step 3, the neural network module includes an input layer and an output layer, wherein the input layer includes a first neuron circuit and a first weight RRAM array; the output layer includes a second weight RRAM array and a second neuron circuit; in the input layer of the neural network, the first neuron circuit is used to determine whether the input data exceeds a set threshold. If so, the first neuron is activated, completing a set-reset operation. After multiple set-reset operations, the first weight RRAM array forms a fixed resistance. In step 4, after the neural network module receives the signal processed by the signal processing module, it loads the signal onto the trained RRAM array, which will trigger the corresponding RRAM circuit. Based on the circuit triggering status of the trained neural network, the abnormal state of the energy storage battery can be identified and the information can be sent to the monitoring center.

Images