Brain-on-chip intelligent complex control system and construction and training method therefor

Abstract

Disclosed in the present invention is a brain-on-a-chip intelligent complex control system, comprising a brain-on-a-chip basic module and a brain-on-a-chip information exchange and training module. The brain-on-a-chip information exchange and training module comprises a neural signal decoding unit, a reward and punishment control unit, a task control model and a mapping relationship model. The neural signal decoding unit is used for converting neural response data into a control instruction which can be recognized by an external device; the task control model is used for a control task for the external device, the task control model generates a target control instruction for the next time on the basis of task feedback information, and the task control model can obtain, on the basis of a correspondence, a neural response corresponding to the target control instruction; the mapping relationship model is used for constructing a mapping relationship between the stimulation sequence of a brain on a chip and neural responses; and the reward and punishment control unit is used for calculating, on the basis of the task feedback information, the degree of completion of a task executed on the basis of target control instruction, to generate a reward signal or a punishment signal for the brain on a chip, and applying the reward signal or the punishment signal to the brain-on-a-chip basic module.

Description

A brain-on-chip intelligent complex control system and its construction and training method Technical Field

[0001] The present invention relates to the field of artificial intelligence control technology, and in particular to a brain-on-chip intelligent complex control system and a construction and training method thereof. Background Art

[0002] The brain-on-chip intelligent complex is a fusion of biotechnology and information technology, integrating biological intelligence and machine intelligence to achieve information interaction and full interconnection. The brain-on-chip refers to the coupling of cortical or hippocampal neurons and brain organoids with electrodes to construct 2D and 3D neuron-electronic complexes. Through external neural regulation and learning training, a "brain-on-chip" electronic complex with a certain degree of intelligence is obtained. With the help of Micro Electrode Arrays (MEAs) technology, it is possible to simultaneously record and stimulate electrophysiological activities at multiple sites non-invasively and for a long time. The development of neuronal networks and the characteristics of discharge patterns under external stimulation have been studied and summarized at the mesoscopic scale, providing a new perspective for exploring the learning and memory mechanisms of the nervous system and improving the interpretability of biological intelligence. Therefore, the information interaction platform of the brain-on-chip intelligent complex is constructed based on MEAs, which can realize the culture of cells and the real-time recording and feedback of neural signals.

[0003] The establishment of an information interaction platform for an on-chip brain-intelligent complex is a necessary condition and an important experimental tool for the construction of an on-chip brain-intelligent complex. Existing studies have achieved the control of external devices by 2D neural networks in virtual or real environments to complete tasks, and characterized biological intelligence through closed-loop feedback information interaction. Due to the independent control of the two environments, there are problems such as high latency and poor control effect. In addition, as a complex nonlinear system, the neural response of the on-chip brain is highly dynamic and variable, making it difficult to form a stable induced response output mapping relationship. The present invention proposes an efficient, controllable, and flexible closed-loop control system for information interaction of an on-chip brain-intelligent complex, which can realize the learning and training process of the on-chip brain in a virtual environment, and intelligently control external devices to achieve dual-environment fusion closed-loop control agent tasks. At the same time, it integrates a deep learning optimization and regulation module to enable the neuron-induced response to reach the expected value, be autonomous and controllable, and reduce training costs. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies of the prior art and propose a brain-on-chip intelligent complex control system and a construction and training method thereof.

[0005] The present invention is achieved through the following technical solutions:

[0006] A brain-on-chip intelligent complex control system, comprising: a brain-on-chip basic module and a brain-on-chip information interaction and training module;

[0007] The brain-on-chip basic module includes a brain-on-chip and is capable of stimulating neurons of the brain-on-chip and collecting neural response signals of the brain-on-chip;

[0008] The on-chip brain information interaction and training module includes: a data preprocessing unit, a neural signal decoding unit, a reward and punishment control unit, a task control model, and a mapping relationship model;

[0009] Data preprocessing unit: used to process the collected raw neural response signals of the on-chip brain and extract effective spike potential signals as neural response data;

[0010] Neural signal decoding unit: used to map the pre-processed neural response data and convert it into control instructions that can be recognized by external devices;

[0011] Task control model: used for external device control tasks. This task control model can generate the next target control instruction based on task feedback information. The task control model integrates the correspondence between neural response data and control instructions, and can obtain the neural response corresponding to the target control instruction based on this correspondence.

[0012] Mapping relationship model: used to construct the mapping relationship between the stimulation sequence and neural response of the brain-on-chip;

[0013] Reward and punishment control unit: used to calculate the completion degree of the target control instruction execution task based on the task feedback information to generate a reward signal or punishment signal for the on-chip brain and apply it to the on-chip brain basic module.

[0014] In the above technical solution, the basic module of the brain-on-chip includes: an on-chip brain, a data acquisition unit and a stimulation unit, wherein the on-chip brain is a coupling body of neurons and MEAs chips, the data acquisition unit is used to read the neural response signals of the on-chip brain in real time; the stimulation unit is used to apply corresponding stimulation to the neurons of the on-chip brain according to the stimulation sequence of the on-chip brain information interaction and training module.

[0015] The construction and training methods of the above-mentioned brain-on-chip intelligent complex control system are as follows:

[0016] Step 1: Prepare and culture a mature brain-on-a-chip;

[0017] Step 2: Conduct a preliminary experiment to determine the stimulation input electrode sites and stimulation response electrode sites of the brain-on-chip;

[0018] Step 3: training a mapping relationship model between stimulus sequence and neural response;

[0019] Step 4: constructing a correspondence between the neural response data and the external device control instructions, and embedding the correspondence into the neural signal decoding unit so that the neural signal decoding unit can convert the neural response data into control instructions for the external device;

[0020] Step 5: Construct a task control model for the external device. This task control model can generate the target control instruction for the next time based on the task feedback information. The task control model also integrates the correspondence between the neural response data constructed in step 4 and the external device control instruction. Based on this correspondence, the neural response corresponding to the target control instruction is obtained. Then, the stimulation sequence required for the neural response is obtained based on the mapping relationship model in step 3. Finally, this stimulation sequence is sent to the on-chip brain basic module to stimulate the on-chip brain, prompting the on-chip brain to produce the expected neural response. The neural response is then converted into the target control instruction for the external device by the neural signal decoding unit.

[0021] Step 6: Build a reward and punishment control unit

[0022] The target control instructions generated by the task control model are sent not only to the mapping relationship model but also to the reward and punishment control unit, and the reward and punishment control unit is capable of receiving task feedback information from external devices; the reward and punishment control unit calculates the degree of completion of the target control instruction execution task based on the task feedback information to generate a reward signal or a punishment signal for the on-chip brain, and the reward signal or the punishment signal is applied to the on-chip brain to train the on-chip brain to produce more accurate neural responses according to control needs.

[0023] In the above technical solution, in step 1, the cortex and hippocampal neurons of 18-day embryonic rats are extracted to construct a 2D neuronal network or an artificially cultured 3D biological tissue culture similar to the human brain - brain organoids, which are attached to the surface of the MEAs chip for coupling and cultured until they mature.

[0024] In the above technical solution, step 2 includes the following steps:

[0025] Step 2.1, selecting candidate stimulation input electrode sites and stimulation response electrode sites;

[0026] Design a stimulation sequence consisting of bidirectional pulses and apply stimulation to all electrode sites on the brain-on-chip in a random order. During each stimulation, preliminary selection of candidate stimulation input electrode sites is based on two indicators: the total number of electrode sites that respond to the stimulation and the response strength of the remaining electrode sites.

[0027] Determine the electrode sites whose response intensities corresponding to the candidate stimulation input electrode sites are greater than a set threshold as candidate stimulation response electrode sites;

[0028] Step 2.2, determining the final stimulation input electrode site and stimulation response electrode site from the candidate stimulation input electrode sites and stimulation response electrode sites;

[0029] First, the number of final stimulation input electrode sites is determined according to the control requirements of the external device. Then, based on the determined number of final stimulation input electrode sites, the same number of stimulation input electrode sites that are farthest apart from each other are selected from the candidate stimulation input electrode sites in step 2.1 as the final stimulation input electrode sites. Then, based on the determined final stimulation input electrode sites, the corresponding stimulation response electrode sites that are closest to them are selected as the final stimulation response electrode sites.

[0030] In the above technical solution, the step 3 includes the following steps:

[0031] Step 3.1: First, collect the data set: Stimulus sequences are applied to the final stimulation input electrode sites and stimulation response electrode sites determined in Step 2, while simultaneously acquiring and recording the actual neural response signals of the on-chip brain units. These neural response signals are then further processed by the data preprocessing unit to extract valid spike sequence data, which are used as the neural response data corresponding to the stimulation sequence.

[0032] In step 3.2, the artificial neural network model is trained using the data set from step 3.1, so that the model learns the input-output relationship between the stimulus sequence and the neural response, and finally obtains a mapping relationship model between the stimulus sequence and the neural response.

[0033] In the above technical solution, in the process of constructing and training the on-chip brain-intelligent complex control system, real external devices are not used, but a virtual external device environment is used to facilitate the development and debugging of the on-chip brain-intelligent complex control system, as well as the learning and efficient training of the on-chip brain.

[0034] The advantages and beneficial effects of the present invention are:

[0035] The present invention not only uses a 2D brain-on-chip, but also a 3D brain-on-chip that is closer to the human brain's organizational structure and neural pathways. It breaks through the limitations of low cell type richness and simple network structure in a single brain region, solves the problem of few cell types, and greatly improves the biological intelligence of the brain-on-chip.

[0036] This paper constructs a mapping model based on artificial neural networks to assist in regulating the on-chip brain to achieve the desired response. It also develops an innovative hybrid intelligence optimization strategy based on in vitro biological neural networks to generate stimulation parameters more appropriate for the on-chip brain's current activity state. This approach overturns the current open-loop experimental model of human intervention and fixed training, overcomes the difficulty of regulating the expected stable response characteristics of the on-chip brain over a long period of time, and addresses the poor robustness of closed-loop control strategies.

[0037] The present invention uses the on-chip brain information interaction and training module to enable the on-chip brain to perform control tasks in virtual environments and real external scenes. At the same time, combined with the reward and punishment stimulation mechanism, it realizes precise closed-loop control training of the on-chip brain in vitro, shortens the training time, and completes efficient control of external tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] FIG1 is a framework diagram of the control system of the brain-on-chip intelligent complex of the present invention.

[0039] FIG2 is a schematic diagram of the on-chip brain intelligent complex control system of the present invention connected to a virtual external device environment.

[0040] FIG3 is a schematic diagram of a virtual external device environment for a robot obstacle avoidance task designed in the present invention.

[0041] For ordinary technicians in this field, other relevant drawings can be obtained based on the above drawings without any creative work. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention are further described below with reference to specific embodiments.

[0043] Example 1

[0044] A brain-on-chip intelligent complex control system, see Figure 1, includes: a brain-on-chip basic module and an on-chip brain information interaction and training module.

[0045] The on-chip brain basic module includes an on-chip brain and is capable of applying stimulation to neurons of the on-chip brain and collecting neural response signals of the on-chip brain, that is, it is capable of realizing stimulation input and neural response output of the on-chip brain.

[0046] Specifically, the brain-on-chip basic module consists of three components: the brain-on-chip, the data acquisition unit, and the stimulation unit. The brain-on-chip is the coupling of neurons and MEAs chips. Specifically, it can be a 2D cortical or hippocampal neuronal network or a 3D brain organoid coupled to the MEAs chip, which has been cultured to maturity. The data acquisition unit is used to read the neural response signals of the brain-on-chip in real time, laying the foundation for further signal decoding. The stimulation unit is used to apply corresponding voltage stimulation to the neurons of the brain-on-chip according to the stimulation sequence of the brain-on-chip information interaction and training module.

[0047] The on-chip brain information interaction and training module is used for information exchange between the on-chip brain and external devices, enabling closed-loop control of external devices by the on-chip brain. It can also train the on-chip brain to better complete tasks while controlling external devices. The on-chip brain information interaction and training module consists of five components: a data preprocessing unit, a neural signal decoding unit, a reward and punishment control unit, a task control model, and a mapping relationship model. These components are described below:

[0048] Data preprocessing unit: used to process the collected on-chip brain raw neural response signals, remove noise and disturbance signals through bandpass filtering, and extract effective spike signals from background noise based on the threshold method.

[0049] Neural signal decoding unit: used to perform linear or nonlinear mapping on the preprocessed spike potential signal and convert it into a control signal that can be recognized by external equipment.

[0050] Task control model: used for control tasks of external devices. This task control model can generate the target control instructions for the next time based on the task feedback information. The task control model integrates the correspondence between neural response data and control instructions, and can obtain the neural response corresponding to the target control instruction based on the correspondence.

[0051] Mapping relationship model: used to construct the mapping relationship between the stimulation sequence of the on-chip brain and the neural response of the on-chip brain.

[0052] Reward and punishment control unit: used to compare the difference between the target control instruction and the task feedback information (i.e., the degree of completion of the task controlled by the target control instruction to the external device) to generate a reward signal or a punishment signal for the on-chip brain.

[0053] Example 2

[0054] This implementation introduces the construction and training methods of the aforementioned brain-on-chip intelligent complex control system. Its working principles and steps are as follows:

[0055] Step 1: Prepare and culture a mature brain-on-a-chip.

[0056] The cortical and hippocampal neurons of 18-day embryonic rats are extracted to construct 2D neuronal networks or artificially cultured 3D biological tissue cultures similar to the human brain - brain organoids, which are attached to the surface of MEAs chips for coupling and cultured until they mature.

[0057] Step 2, Pre-experimental stage: Determine the stimulation input electrode sites and stimulation response electrode sites of the brain-on-a-chip. Specifically, step 2 includes the following steps:

[0058] Step 2.1: Select candidate stimulation input electrode sites and stimulation response electrode sites.

[0059] A stimulation sequence consisting of bidirectional pulses (for example, 10 1 Hz pulses) is designed, and stimulation is applied to all electrode sites of the brain-on-chip in a random order. During each stimulation, candidate stimulation input electrode sites are preliminarily selected based on two indicators: the total number of electrode sites that respond to the stimulation and the response intensity of the remaining electrode sites of the electrode site currently being stimulated (that is, the higher the response intensity and the greater the number of electrode sites that respond, the stronger the response will be to the electrode site applied with the stimulation. Therefore, stimulation input electrode sites whose response intensity meets the set response intensity threshold m and the number of electrode sites that meet the set response intensity threshold m also meets the set number threshold n can be selected as candidate stimulation input electrode sites).

[0060] The evoked response histogram (PSTH) is an indicator that quantifies the strength of a neuron's response to stimulation (the larger the PSTH value, the stronger the neuron's response to the input stimulation). Based on this indicator, the electrode site with a stronger response intensity (the response intensity is greater than the set threshold) corresponding to the above-mentioned candidate stimulation input electrode site is determined as the candidate stimulation response electrode site.

[0061] Step 2.2: Determine the final stimulation input electrode site and stimulation response electrode site from the candidate stimulation input electrode sites and stimulation response electrode sites.

[0062] Specifically: First, determine the number of final stimulation input electrode sites based on the control requirements of the external device. For example, if the controlled external device is a car, it is necessary to control the left wheel (i.e., the left wheel motor) and the right wheel (i.e., the right wheel motor) of the car to achieve the control of the car's steering and movement. The number of final stimulation input electrode sites required is 2, corresponding to the control of the left wheel and the right wheel respectively; then, based on the determined number of final stimulation input electrode sites, select the same number of stimulation input electrode sites that are the farthest apart from each other from the candidate stimulation input electrode sites in step 2.1 as the final stimulation input electrode sites; then, based on the determined final stimulation input electrode sites, select the stimulation response electrode site that is closest to it as the final stimulation response electrode site.

[0063] Step 3: Train the mapping relationship model between the stimulus sequence and the neural response

[0064] Step 3.1, first collect the data set: stimulate the final stimulation input electrode site and stimulation response electrode site determined in step 2 and collect the neural response signals corresponding to the stimulation, that is, input the stimulation sequence data to the on-chip brain unit through the stimulation unit, and at the same time collect and record the real neural response signals of the on-chip brain unit through the data acquisition unit; these neural response signals are further processed by the data preprocessing unit to extract valid spike potential sequence data, and the obtained spike potential sequence data is used as the neural response data corresponding to the stimulation sequence.

[0065] Step 3.2, then use the above data set to train the artificial neural network model, where a single input is the neural response at a certain moment and the stimulation sequence for a period of time, and the output of the artificial neural network model is the predicted neural response for a period of time thereafter. The weights in the artificial neural network model are adjusted by constructing an objective function to minimize the error between the predicted neural response and the true neural response to train the model, so that the model learns the input-output relationship between the stimulation sequence and the neural response, and finally obtains a mapping relationship model between the stimulation sequence and the neural response.

[0066] Step 4: Construct a neural signal decoding unit

[0067] Based on the neural response data collected in step 3.1, a correspondence between the neural response data (i.e., spike potential sequence data) and the external device control instructions is constructed, and this correspondence is embedded in the neural signal decoding unit, so that the neural signal decoding unit can convert (decode) the neural response data into control instructions for the external device.

[0068] Step 5: Build a task control model for external devices

[0069] Construct a task control model suitable for external device control tasks, such as a car obstacle avoidance task control model, a car trajectory tracking task control model, a manipulator grasping task control model, etc. The task control model can generate the target control instructions for the next time based on the task feedback information, and the task control model also integrates the correspondence between the neural response data constructed in step 4 and the external device control instructions, and then the neural response corresponding to the target control instruction can be obtained according to the correspondence. Then, according to the mapping relationship model in step 3, the stimulation sequence required for the neural response can be obtained. Finally, the stimulation sequence will be sent to the stimulation unit of the on-chip brain basic module to stimulate the on-chip brain, prompting the on-chip brain to produce the desired neural response, and the neural response will be converted into the target control instruction for the external device through the neural signal decoding unit.

[0070] Step 6: Build a reward and punishment control unit

[0071] The target control instructions generated by the task control model are sent not only to the mapping relationship model but also to the reward and punishment control unit, and the reward and punishment control unit can also receive task feedback information from external devices; then the reward and punishment control unit can generate a reward signal or a punishment signal for the on-chip brain by comparing the difference between the target control instructions and the task feedback information (that is, the degree of completion of the task performed by the external device controlled by the target control instructions) (if the degree of completion meets the set threshold, a reward signal is generated for the on-chip brain; if the degree of completion is less than the set threshold, a punishment signal is generated for the on-chip brain). The reward signal or the punishment signal will be applied to the on-chip brain through the stimulation unit to train the on-chip brain to produce more accurate neural responses according to control needs.

[0072] Example 3

[0073] On the basis of the above embodiments, further speaking, referring to FIG2 , in the process of constructing and training the on-chip brain intelligent complex control system, real external devices are not used, but a virtual external device environment is used to facilitate the development and debugging of the on-chip brain intelligent complex control system, as well as the learning and efficient training of the on-chip brain.

[0074] The necessity of using a virtual external device environment: Due to the special characteristics of brain-on-chips, such as electrophysiological activity, and the limitations of the actual physical environment of the control task, brain-on-chips have the following problems:

[0075] 1) Brain-on-a-chip culture cycles are long and costly, prone to fatigue under prolonged stimulation, highly dependent on the culture environment, and inherently fragile, with improper handling easily leading to cell death.

[0076] 2) There are certain limitations on the physical environment for implementing different control tasks: building a real physical environment is time-consuming and labor-intensive; the real physical environment has system noise and errors, which are not friendly to the learning and training of neurons; the accuracy of sensors in the real physical environment is limited by different devices, etc.

[0077] Taking all the above factors into consideration, in order to facilitate the development and debugging of the on-chip brain intelligent complex control system, as well as the learning and training of the on-chip brain, it is preferred to adopt a virtual external device environment.

[0078] Figure 3 shows a schematic diagram of the virtual external device environment for the robot obstacle avoidance task designed in this embodiment. It mainly includes a virtual robot (i.e., a small car), a visual workspace for the obstacle avoidance task (i.e., a virtual obstacle scene), a robot status information panel, a learning and training and control operation area, and modules such as stimulus signal setting, data recording, and export. In addition, it includes three types of obstacle distribution: regular, random, and manually defined obstacle layout. When the robot moves in the field, real-time signal display, trajectory synchronization, and left and right wheel response deviation curves can be performed, providing intuitive and convenient display indicators for judging the robot's obstacle avoidance status.

[0079] The virtual robot mainly has two movable wheels, namely the left wheel and the right wheel. The robot controls its movement direction through the speed difference between the left and right wheels. The visual workspace of the obstacle avoidance task is used to visualize the obstacle scene and the robot's movement trajectory; the robot status information panel mainly presents the robot's current movement status information and the distance information to the nearest obstacle; the learning training and control operation area is mainly used to set learning parameters and control the learning process of the on-chip brain.

[0080] The above is an exemplary description of the present invention. It should be noted that, without departing from the core of the present invention, any simple deformation, modification or other equivalent replacement that can be made by other skilled in the art without expending creative labor falls within the scope of protection of the present invention.

Claims

1. A brain-on-chip intelligent complex control system, characterized in that: Includes on-chip brain basic module and on-chip brain information interaction and training module; The brain-on-chip basic module includes a brain-on-chip, and is capable of applying stimulation to neurons of the brain-on-chip and collecting neural response signals of the brain-on-chip; The on-chip brain information interaction and training module includes: a data preprocessing unit, a neural signal decoding unit, a reward and punishment control unit, a task control model and a mapping relationship model; Data preprocessing unit: used to process the collected raw neural response signals of the on-chip brain and extract effective spike potential signals as neural response data; Neural signal decoding unit: used to map the pre-processed neural response data and convert it into control instructions that can be recognized by external devices; Task control model: used for control tasks of external devices. The task control model can generate the target control instruction for the next time according to the task feedback information. The task control model integrates the correspondence between neural response data and control instructions, and can obtain the neural response corresponding to the target control instruction according to the correspondence. Mapping relationship model: used to construct the mapping relationship between the stimulation sequence and neural response of the brain-on-chip; Reward and punishment control unit: used to calculate the completion degree of the target control instruction execution task based on the task feedback information, and then generate a reward signal or punishment signal for the on-chip brain, and apply it to the on-chip brain basic module.

2. The on-chip brain-intelligence complex control system according to claim 1, characterized in that: The basic module of the brain-on-chip includes: a brain-on-chip, a data acquisition unit and a stimulation unit. The brain-on-chip is a coupling body of neurons and MEAs chips. The data acquisition unit is used to read the neural response signals of the brain-on-chip in real time; the stimulation unit is used to apply corresponding stimulation to the neurons of the brain-on-chip according to the stimulation sequence of the brain-on-chip information interaction and training module.

3. The method for constructing and training a brain-on-chip intelligent complex control system according to claim 1, characterized in that: The following steps are involved: Step 1: Prepare and culture a mature brain-on-chip; Step 2: Conduct a preliminary experiment to determine the stimulation input electrode sites and stimulation response electrode sites of the brain-on-chip; Step 3, training to obtain a mapping relationship model between the stimulus sequence and the neural response; Step 4, constructing a correspondence between the neural response data and the external device control instruction, and embedding the correspondence into the neural signal decoding unit so that the neural signal decoding unit can convert the neural response data into the control instruction for the external device; Step 5, constructing a task control model for the external device, the task control model can generate the target control instruction for the next time according to the task feedback information, and the task control model also integrates the correspondence between the neural response data constructed in step 4 and the external device control instruction, and obtains the neural response corresponding to the target control instruction according to the correspondence, and then obtains the stimulation sequence required for the neural response according to the mapping relationship model of step 3, and finally sends the stimulation sequence to the on-chip brain basic module to stimulate the on-chip brain to induce the on-chip brain to produce the expected neural response, and the neural response is then converted into the target control instruction for the external device by the neural signal decoding unit; Step 6: Build a reward and punishment control unit The target control instructions generated by the task control model are sent to the reward and punishment control unit in addition to the mapping relationship model, and the reward and punishment control unit can receive task feedback information sent by external devices; the reward and punishment control unit calculates the completion degree of the target control instruction execution task based on the task feedback information, and then generates a reward signal or punishment signal for the on-chip brain, and the reward signal or punishment signal is applied to the on-chip brain to train the on-chip brain to produce more accurate neural responses according to control needs.

4. The method for constructing and training a brain-on-chip intelligent complex control system according to claim 3, characterized in that: In step 1, cortical and hippocampal neurons of 18-day embryonic rats are extracted to construct 2D neuronal networks or artificially cultured 3D biological tissue cultures similar to the human brain - brain organoids, attached to the surface of MEAs chips for coupling, and cultured until they mature.

5. The method for constructing and training a brain-on-chip intelligent complex control system according to claim 3, characterized in that: The step 2 comprises the following steps: Step 2.1, selecting candidate stimulation input electrode sites and stimulation response electrode sites; Design a stimulation sequence consisting of bidirectional pulses, and apply stimulation to all electrode sites on the brain-on-chip in a random order. During each stimulation, the candidate stimulation input electrode sites are preliminarily selected based on the total number of electrode sites that respond to the stimulation and the response intensity of the remaining electrode sites of the currently stimulated electrode site; Determine the electrode sites whose response strengths corresponding to the candidate stimulation input electrode sites are greater than a set threshold as candidate stimulation response electrode sites; Step 2.2, determining the final stimulation input electrode site and stimulation response electrode site from the candidate stimulation input electrode sites and stimulation response electrode sites; First, the number of final stimulation input electrode sites is determined according to the control requirements of the external device, and then, according to the determined number of final stimulation input electrode sites, the same number of stimulation input electrode sites that are farthest apart from each other are selected from the candidate stimulation input electrode sites in step 2.1 as the final stimulation input electrode sites; then, according to the determined final stimulation input electrode sites, the corresponding stimulation response electrode sites that are closest to them are selected as the final stimulation response electrode sites.

6. The method for constructing and training a brain-on-chip intelligent complex control system according to claim 3, characterized in that: The step 3 comprises the following steps: Step 3.1, first collect the data set: input the stimulation sequence to the final stimulation input electrode site and stimulation response electrode site determined in step 2, and collect and record the real neural response signals of the on-chip brain unit; these neural response signals are further processed by the data preprocessing unit to extract the effective spike potential sequence data, and the obtained spike potential sequence data is used as the neural response data corresponding to the stimulation sequence; Step 3.2, using the data set in step 3.1 to train the artificial neural network model, so that the model learns the input-output relationship between the stimulus sequence and the neural response, and finally obtains the mapping relationship model between the stimulus sequence and the neural response.

7. The method for constructing and training a brain-on-chip intelligent complex control system according to claim 3, characterized in that: In the process of constructing and training the control system of the brain-on-chip intelligent complex, a virtual external device environment is used.

Images