A cell culture and storage management method based on biological parameter monitoring

Abstract

The present application relates to the technical field of medical automation control, and particularly relates to a cell culture and storage management method based on biological parameter monitoring, comprising: accessing biological impedance spectrum and optical turbidity time series flow in the culture stage; extracting impedance imaginary part characteristics under specified polarization relaxation frequency to construct a decoupling matrix, and nonlinearly decoupling the optical turbidity time series flow to output a live cell turbidity matrix; using a dynamic time warping algorithm to perform time series nonlinear alignment on the matrix and the impedance spectrum to generate an aligned data flow; locking a flow transfer critical time according to a change rate trend and issuing a process flow transfer control word, while solidifying feature boundary parameters; extracting charge relaxation characteristics of the parameters and inversely solving real-time membrane permeability parameters based on a cell membrane stress dynamics model, and substituting the parameters into a non-homogeneous heat conduction partial differential equation set to solve to generate an adaptive control matrix; and finally generating a program cooling control sequence according to the adaptive control matrix and writing the sequence into a cooling step register.

Description

Technical Field

[0001] This invention relates to the field of medical automation control technology, and in particular to a cell culture and storage management method based on biological parameter monitoring. Background Technology

[0002] Continuous large-scale cell culture and programmed cryopreservation at ultra-low temperatures are crucial for ensuring cell safety and clinical efficacy. During cell culture, to accurately monitor real-time cell growth and determine optimal process flow, online monitoring devices are typically used to track cell density, viability, and other vital signs. Currently, the mainstream online monitoring method relies primarily on optical absorbance measurement (i.e., optical turbidimetry). This method indirectly estimates the total cell count by measuring the attenuation of transmitted light intensity after a collimated monitoring beam penetrates the cell suspension, and uses this as the basis for subsequent programmed cryopreservation control.

[0003] Traditional online optical detection methods cannot effectively identify and remove non-living particles such as dead cells, culture medium debris, and microbubbles. The background scattering components generated by these non-living particles overlap significantly with the scattering signals from living cells, making it impossible for the system to specifically extract the density index of living cells from the mixed optical turbidity signal. This results in low physical purity and data reliability in online monitoring. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a cell culture and storage management method based on biological parameter monitoring, which aims to improve the problem that the density index of live cells cannot be specifically extracted from mixed optical turbidity signals, resulting in low physical purity and data reliability in online monitoring.

[0005] This invention provides the following technical solution: a cell culture and storage management method based on biological parameter monitoring, comprising:

[0006] S1, the bioimpedance spectrum time-series flow and optical turbidity time-series flow during the inoculation and culture phases;

[0007] S2. Extract the imaginary part features of the bioimpedance spectrum time-series flow at the specified polarization relaxation frequency and construct a decoupling matrix. Use the decoupling matrix to perform nonlinear property decoupling on the optical turbidity time-series flow and output the live cell turbidity matrix.

[0008] S3. Input the live cell turbidity matrix and bioimpedance spectrum time stream into the dynamic time warping algorithm for calculation, complete the time nonlinear alignment and generate the aligned data stream;

[0009] S4. Determine the critical time of flow based on the change rate trend of the aligned data stream, and issue a process flow control word when the critical time of flow is reached, while locking the bioimpedance spectrum time series corresponding to the critical time of flow as a characteristic boundary parameter.

[0010] S5. Extract the charge relaxation features of the feature boundary parameters, invert and decouple the real-time membrane permeability parameters based on the cell membrane stress dynamics model, and input the real-time membrane permeability parameters and the obtained environmental thermodynamic parameters into a set of non-homogeneous heat conduction partial differential equations to solve simultaneously and generate an adaptive control matrix.

[0011] S6. Generate a programmed cooling control sequence based on the adaptive control matrix, and write the programmed cooling control sequence into the cooling step register.

[0012] Preferably, in S1, the bioimpedance spectrum time-series flow and optical turbidity time-series flow during the inoculation phase include:

[0013] A multi-frequency AC signal is injected into the culture system, and the complex impedance response of the culture system at each characteristic frequency point is collected and aligned to generate a bioimpedance spectrum time-series flow.

[0014] The collimated monitoring beam is driven to penetrate the culture system, the transmitted light intensity attenuation coefficient is calculated and converted into absorbance characteristics, and an optical turbidity time-series flow is generated.

[0015] Preferably, in S2, extracting the imaginary part features of the bioimpedance spectrum temporal current at a specified polarization relaxation frequency and constructing the decoupling matrix includes:

[0016] The imaginary part response of the time-series bioimpedance spectrum is separated to obtain the imaginary part spectral line of the complex impedance as a function of frequency.

[0017] The first derivative zero-crossing point of the imaginary part spectral line is detected, the frequency corresponding to the zero-crossing point is determined as the specified polarization relaxation frequency, and the corresponding imaginary part amplitude is extracted as the imaginary part feature.

[0018] The imaginary features are normalized and mapped, and the mapped correction coefficients are used as diagonal elements to construct a diagonal feature matrix, thereby generating a decoupling matrix.

[0019] Preferably, in S2, the nonlinear property decoupling of the optical turbidity time-series flow using the decoupling matrix includes:

[0020] The optical turbidity time-series stream is divided into frames according to a time sliding window to construct an initial turbidity feature matrix;

[0021] The Hadamard product operator is used to perform element-wise multiplication of the decoupling matrix and the initial turbidity characteristic matrix, so as to use the correction coefficient in the decoupling matrix to suppress the medium background scattering component in the initial turbidity characteristic matrix.

[0022] The initial turbidity feature matrix after calculation is subjected to inverse frame fusion to output the live cell turbidity matrix.

[0023] Preferably, in S3, inputting the live cell turbidity matrix and bioimpedance spectrum time-series stream into the dynamic time warping algorithm for calculation includes:

[0024] The turbidity matrix of the living cells and the time-series flow of the bioimpedance spectrum are used as the time-series vectors of the first and second dimensions.

[0025] Calculate the Euclidean distance between the sampling points in the first-dimensional temporal vector and the second-dimensional temporal vector to generate a local distance matrix;

[0026] The local distance matrix is ​​subjected to boundary constraint summation and iteration to construct a global dynamic programming matrix.

[0027] Preferably, in S3, completing the temporal nonlinear alignment and generating the aligned data stream includes:

[0028] Search for the inverse path along the direction of minimum gradient along the global dynamic programming matrix to determine the optimal regularized path with the minimum global cumulative distance;

[0029] Based on the optimal normalization path, the time series vectors of the first dimension and the second dimension are nonlinearly resampled and aligned.

[0030] The aligned two-dimensional time-series vectors are fused with multimodal features to generate an aligned data stream.

[0031] Preferably, in S4, determining the critical transition time based on the rate of change trend of the aligned data stream includes:

[0032] The aligned data stream is subjected to first-order temporal difference operation along the time axis according to a preset step size to calculate the time-series signal of the rate of change of each biological parameter in the aligned data stream within adjacent time windows.

[0033] The rate of change time-series signal is smoothed and filtered to eliminate noise ripple, and the maximum peak inflection point of the smoothed and filtered rate of change time-series signal on the overall time-series axis is extracted.

[0034] The sampling timestamp corresponding to the maximum peak inflection point is locked as the critical time of circulation.

[0035] Preferably, in S5, the real-time membrane permeability parameters decoupled from the cell membrane stress dynamics model include:

[0036] Based on the charge relaxation characteristics of the feature boundary parameters, the time constant and the peak value of the imaginary part of the complex permittivity of the corresponding charge relaxation curve are extracted.

[0037] The time constant and the peak value of the imaginary part of the complex permittivity are used as boundary conditions and imported into the preset Maxwell-Wagner interface polarization dynamics equation to establish a multi-shell complex impedance topological equivalent mapping.

[0038] The transmembrane conductivity and membrane capacitance change rate are solved along the multi-shell complex impedance topological equivalent mapping and combined to obtain the real-time membrane permeability parameters.

[0039] Preferably, in S5, the simultaneous solution of the real-time membrane permeability parameters and the acquired environmental thermodynamic parameters into a system of non-homogeneous heat conduction partial differential equations includes:

[0040] The environmental thermodynamic parameters are transformed into the initial temperature field spatial boundary and discretized into a three-dimensional finite element space mesh.

[0041] The real-time membrane permeability parameters are converted into non-homogeneous energy source terms and substituted into the non-homogeneous heat conduction partial differential equations as the heat source input within the three-dimensional finite element space grid.

[0042] The non-homogeneous partial differential equations of heat conduction are iteratively solved along the three-dimensional finite element space mesh to calculate the time evolution of the target temperature that satisfies the cell ice crystal growth, and the control weights are extracted to construct an adaptive control matrix.

[0043] Preferably, in S6, generating the programmed cooling control sequence based on the adaptive control matrix includes:

[0044] The adaptive control matrix is ​​decoupled by dimension to separate the target temperature value and the hold time value at each time step;

[0045] The slope of instantaneous temperature change is calculated based on the difference between the target temperature values ​​between adjacent time steps and the time span.

[0046] The target temperature value, holding time value, and instantaneous temperature change slope within each time step are concatenated using time-series logic to generate a program-based cooling control sequence.

[0047] The present invention has the following beneficial effects:

[0048] 1. In this invention, a decoupling matrix is ​​constructed by extracting the imaginary part features under the impedance spectrum polarization relaxation frequency, and multiplied with the initial turbidity feature matrix by the Hadamard product. This achieves the specific reduction of background scattering components caused by non-living particles using microscopic electrical features, and solves the problem of low purity in live cell monitoring caused by the inability to remove background noise such as dead cells and culture medium fragments in traditional detection.

[0049] 2. In this invention, the asynchronous photoelectric timing vector is aligned by a dynamic time warping algorithm, and first-order timing difference and smoothing filtering are performed on the aligned data stream under a unified time reference. This achieves adaptive and high-precision capture of the maximum peak inflection point of cell growth curvature to lock the critical time of flow, solving the problem of judgment lag and poor control accuracy caused by traditional switching relying on experience timing or manual absolute threshold.

[0050] 3. In this invention, critical characteristics are inverted into real-time membrane permeability parameters through kinetic equations, and these parameters are used as internal heat sources to solve simultaneously with the finite element mesh based on environmental thermodynamic parameters and the set of partial differential equations for heat conduction. This enables the calculation of a control sequence without ice crystal growth based on the feedforward solution of cell microscopic characteristics and writes it into the cooling step register. This solves the problem of high cell cryopreservation loss rate caused by the disconnect between traditional temperature control and cell properties and the inability to dynamically adjust power. Attached Figure Description

[0051] Figure 1 This is a flowchart of a cell culture and storage management method based on biological parameter monitoring proposed in this invention;

[0052] Figure 2 This is a flowchart illustrating the multimodal data property decoupling and temporal nonlinear alignment of a cell culture and storage management method based on biological parameter monitoring proposed in this invention.

[0053] Figure 3 This is a flowchart illustrating the generation of a cross-scale cooling control sequence for a cell culture and storage management method based on biological parameter monitoring, as proposed in this invention. Detailed Implementation

[0054] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Example 1:

[0056] In a first embodiment of the present invention, the present invention provides a cell culture and storage management method based on biological parameter monitoring, such as... Figures 1-3 As shown, it includes:

[0057] S1, the bioimpedance spectrum time-series flow and optical turbidity time-series flow during the inoculation and culture phases;

[0058] Furthermore, in S1, the bioimpedance spectrum temporal current and optical turbidity temporal current during the inoculation phase include:

[0059] Multi-frequency AC signals are injected into the culture system, and the complex impedance response of the culture system at each characteristic frequency point is collected and aligned to generate a bioimpedance spectrum time-series flow.

[0060] The driving collimation monitoring beam penetrates the culture system, calculates the attenuation coefficient of transmitted light intensity and converts it into absorbance characteristics, generating an optical turbidity time-series flow.

[0061] Specifically, in the large-scale or continuous culture of biological samples, non-invasive online monitoring of cell state is the physical basis for ensuring consistent cell quality. As the data input source for the overall control process, the multimodal parameter access process during the culture stage is completed through parallel calculation of electrical and optical dual pathways. Its core lies in using differentiated physical fields to synchronously discretize and characterize the electrical polarization and optical scattering properties of the cell suspension, providing a benchmark time-series data stream for subsequent decoupling of live cell properties.

[0062] The data processing flow of the electrical channel begins with the physical modulation and injection of a weak AC signal. The system controls a high-frequency direct digital frequency synthesizer to generate a continuously step-modulated sinusoidal AC current signal. This signal, as a multi-frequency AC signal, is introduced into the cell culture system through an array of inert electrodes symmetrically distributed on the outer wall or inside the culture carrier. To capture the dielectric relaxation response of cells at different electric field frequencies, a frequency sweep sequence is cyclically switched between preset characteristic frequencies.

[0063] When a multi-frequency AC signal flows through a complex impedance network containing individual cells and culture medium, a phase-shifted response voltage is generated between the excitation and response electrodes due to the capacitive barrier effect of the cell membrane and the conductivity difference between electrolytes inside and outside the cell. The system uses a high-speed analog-to-digital converter to perform high-frequency synchronous sampling of this response voltage, obtaining a discrete digital voltage sequence. The acquired digital voltage sequence is sent to a quadrature demodulator, where it undergoes phase-sensitive detection and integral filtering with a reference signal of the same frequency. The in-phase and out-of-phase components of the culture system at various specific characteristic frequencies are calculated, and then converted into the real and imaginary parts of the complex impedance, thus forming the complex impedance response.

[0064] Since the frequency sweep sampling at different frequencies has a natural sequential order on the time axis, the data processing kernel calls the clock step alignment operator to perform linear interpolation and time resampling on the complex impedance response at each characteristic frequency point based on the start timestamp of the frequency sweep cycle, thereby eliminating the phase lag introduced by the frequency sweep process and finally generating a bioimpedance spectrum time-series stream aligned at a unified time node.

[0065] While the electrical channel operates, the optical channel simultaneously constructs a turbidity data stream through dynamic calculation of transmitted light intensity attenuation. A constant current source drives a laser diode to emit highly stable monochromatic parallel light. This beam is shaped by a collimating lens group to form a collimated monitoring beam, which penetrates the culture system along a direction perpendicular to the light-transmitting window of the culture carrier. When the collimated monitoring beam passes through the culture system, suspended cell particles, microbubbles, and macromolecules in the culture medium scatter and absorb photons, causing the energy of the beam to attenuate in the direction of spatial propagation.

[0066] The transmitted light beam, after penetrating the culture system, is received by a photodetector on the opposite side. The photodetector converts the received photon energy into a weak photocurrent signal. This photocurrent signal is amplified transimpedance by a preamplifier and biased by a reference voltage, then converted into an RF digital signal. The processing kernel calculates the transmitted light intensity attenuation coefficient in real time. The transmitted light intensity attenuation coefficient is calculated using the following absorbance conversion formula:

[0067] ;

[0068] In this formula, This represents the calculated absorbance characteristics. This represents the digital quantity of transmitted light intensity demodulated by the photodetector within the current sampling period. This represents the digital value of the transmitted light intensity of the culture system blank medium reference, which is pre-stored in the configuration register. In a set of feasible implementation data, if the digital value of the transmitted light intensity of the blank medium reference... The digital value of transmitted light intensity demodulated during the current sampling period is 4095 full-scale count. If the count is 2048, then the calculated absorbance characteristic is... Approximately 0.301. The computational kernel uses a fixed hardware timer interrupt as the sampling trigger source, and arranges the absorbance features calculated in each interrupt cycle in a circular data buffer according to the sampling order, thus constructing an optical turbidity time-series stream that continuously reflects the dynamic evolution of the macroscopic scattering density of the culture system.

[0069] The generated bioimpedance spectrum and optical turbidity time-series flows are written as cascaded intermediate variables into the shared main memory region of the system. At this point, the bioimpedance spectrum time-series flow manifests in the time domain as a matrix flow of multidimensional characteristic frequency response vectors advancing over time; its microscopic physical essence lies in mapping the polarization charge relaxation state of the cell membrane interface. Conversely, the optical turbidity time-series flow manifests as a time-series scalar flow of one-dimensional absorbance; its macroscopic physical essence lies in mapping the overall light-blocking effect of the entire suspension system. These two time-series flows maintain strict step-synchronization on the time base and directly participate in the decoupling operations of subsequent stages, thus serving as multimodal input sources to the subsequent nonlinear property decoupling process.

[0070] S2. Extract the imaginary part of the bioimpedance spectrum time-series current at the specified polarization relaxation frequency and construct a decoupling matrix. Use the decoupling matrix to perform nonlinear property decoupling on the optical turbidity time-series current and output the live cell turbidity matrix.

[0071] Furthermore, in S2, the extraction of the imaginary part of the bioimpedance spectrum temporal current at a specified polarization relaxation frequency and the construction of the decoupling matrix include:

[0072] The imaginary part response of the time-series current of the bioimpedance spectrum is separated to obtain the imaginary part spectral line of the complex impedance as a function of frequency.

[0073] The first derivative zero-crossing point of the imaginary part spectral line is detected, the frequency corresponding to the zero-crossing point is determined as the specified polarization relaxation frequency, and the corresponding imaginary part amplitude is extracted as the imaginary part feature.

[0074] The imaginary features are normalized and mapped, and the mapped correction coefficients are used as diagonal elements to construct a diagonal feature matrix, thus generating a decoupling matrix.

[0075] Furthermore, in S2, the nonlinear property decoupling of the optical turbidity time-series flow using the decoupling matrix includes:

[0076] The optical turbidity time-series stream is divided into frames according to a time sliding window to construct an initial turbidity feature matrix;

[0077] The Hadamard product operator is used to perform element-wise multiplication of the decoupling matrix and the initial turbidity characteristic matrix, so as to use the correction coefficient in the decoupling matrix to suppress the medium background scattering component in the initial turbidity characteristic matrix.

[0078] The initial turbidity feature matrix after calculation is subjected to inverse frame fusion to output the live cell turbidity matrix.

[0079] Specifically, in the nonlinear property decoupling stage of cellular multimodal parameters, accurately extracting the specific physical indicators of living cells from mixed time-series signals is the core technical support for eliminating background interference and achieving precise state monitoring. The data processing flow of this nonlinear property decoupling process begins with spectral separation of the bioimpedance spectrum time-series stream accessed and stored in the system memory. The data processing kernel extracts the imaginary response component from the complex impedance time-series data and arranges the imaginary amplitudes at each frequency point in ascending order of frequency, mapping them to an imaginary spectral line representing the dynamic evolution of the imaginary part of the complex impedance with frequency distribution over time.

[0080] To adaptively locate the most sensitive response feature points of live cells in a dynamically fluctuating culture environment, the system invokes a differential operator to calculate the first derivative of the imaginary spectral line. The differential calculation logic employs the difference quotient between adjacent frequency points to solve for the rate of change curve of the imaginary amplitude with frequency. Subsequently, a zero-crossing detector searches along the frequency axis for the sign transition position of this rate of change curve, precisely locking the frequency corresponding to the zero-crossing point where the first derivative changes from positive to negative and becomes zero as the specified polarization relaxation frequency. At this specified polarization relaxation frequency, the system simultaneously reads the absolute amplitude of the imaginary spectral line as the imaginary feature characterizing the polarization intensity of the live cell membrane.

[0081] The extracted imaginary features are then fed into a spatial mapping process to construct a decoupling operator. The system transforms these imaginary features into a dimensionless control scalar using a preset normalization mapping function. The normalization mapping logic divides the imaginary amplitude in the current sampling period by the maximum reference amplitude in the historical calibration period to output dynamic property correction coefficients. To matrixify these one-dimensional scalar coefficients for streaming dot product operations with multi-dimensional optical features, the data processing kernel calls a diagonal matrix construction operator. This operator fills the dynamic property correction coefficients onto the main diagonal of a square matrix filled with all zeros, leaving all other positions zero. This generates a diagonal feature matrix with specific physical culling capabilities, which serves as the decoupling matrix in subsequent processing.

[0082] Meanwhile, the optical turbidity time-series stream generated in the previous stage is matrix-expanded in an independent data buffer channel. Since the original optical turbidity data is a one-dimensional scalar time series, in order to match the spatial decoupling operation and introduce temporally localized features, the system calls the sliding window slicing operator to dynamically frame the optical turbidity time-series stream according to the preset time window span and sliding step size. The fixed-length absorbance sampling points extracted within each time window are converted into row vectors of a matrix. As the sliding window advances, the row vectors of multiple consecutive frames are concatenated vertically to construct the initial turbidity feature matrix.

[0083] After generating the initial turbidity characteristic matrix, the system synchronously inputs the decoupling matrix and the initial turbidity characteristic matrix into the matrix streaming processor to perform nonlinear multiplication. Since the decoupling matrix is ​​a diagonal matrix, the multiplication operation between the two matrices is represented by calling the Hadamard product operator to perform the product of the row elements. The computational logic of this nonlinear property decoupling process follows the matrix decoupling formula:

[0084] ;

[0085] In this formula, This represents the turbidity matrix of living cells after property decoupling. This represents the diagonal feature matrix generated after the imaginary part feature normalization mapping, which is the decoupling matrix. This represents the initial turbidity feature matrix constructed after framing based on optical absorbance. This represents the Hadamard product operator, which performs a nonlinear multiplication of corresponding elements of two matrices one by one.

[0086] In a set of feasible implementation data, if the initial turbidity feature matrix The absorbance eigenvalue of a specific spatiotemporal node is 0.500, and the corresponding decoupling matrix is... The diagonal dynamic property correction coefficient in the matrix is ​​0.800, so the positional decoupling eigenvalue obtained by the Hadamard product operation is approximately 0.400. Through this dot product mechanism, the correction coefficients in the decoupling matrix dynamically reduce the amplitude of each element in the initial turbidity characteristic matrix. Since dead cells and culture medium debris do not possess complete electrical polarization relaxation characteristics, their imaginary response at a specified polarization relaxation frequency approaches zero, enabling the calculated correction coefficients to specifically suppress the medium background scattering component caused by non-living particles in the initial turbidity characteristic matrix.

[0087] After background scattering suppression, the initial turbidity feature matrix then enters the signal reconstruction stage. The system calls the inverse frame fusion operator, and according to the overlapping and addition rule that is the opposite of the aforementioned time sliding window, restores each row vector of the feature matrix to a continuous scalar on the time axis. Mean smoothing and denoising are then performed within the overlapping time period, finally outputting a high-purity live cell turbidity matrix. This live cell turbidity matrix accurately represents the actual temporal evolution of live cell density in the culture system and is directly written as backbone data into the storage area of ​​the next-level dynamic time warping algorithm to participate in the temporal nonlinear alignment process.

[0088] S3. Input the live cell turbidity matrix and bioimpedance spectrum time series stream into the dynamic time warping algorithm for calculation, complete the time nonlinear alignment and generate the aligned data stream;

[0089] Furthermore, in S3, the live cell turbidity matrix and bioimpedance spectrum time-series flow are input into the dynamic time warping algorithm for calculation, including:

[0090] The turbidity matrix of live cells and the time-series flow of bioimpedance spectrum are used as the time-series vectors of the first and second dimensions.

[0091] Calculate the Euclidean distance between the sampling points in the first-dimensional time vector and the second-dimensional time vector, and generate a local distance matrix;

[0092] By performing boundary constraint summation and iteration on the local distance matrix, a global dynamic programming matrix is ​​constructed.

[0093] Furthermore, in S3, completing the temporal nonlinear alignment and generating the aligned data stream includes:

[0094] Search for the inverse path along the direction of minimum gradient along the global dynamic programming matrix to determine the optimal regularized path with the minimum global cumulative distance;

[0095] The time series vectors of the first dimension and the second dimension are nonlinearly resampled and aligned based on the optimal normalization path.

[0096] The aligned two-dimensional time-series vectors are fused with multimodal features to generate an aligned data stream.

[0097] Specifically, after acquiring the pre-decoupled and reconstructed live cell turbidity matrix and the synchronously accessed bioimpedance spectrum time-series stream, a dynamic time warping algorithm is used to nonlinearly stretch and warp the time axis of these two heterogeneous data streams to eliminate the temporal lag caused by the difference in response mechanisms between optical and electrical physical fields. The data processing flow of this nonlinear alignment process begins with the vectorized recombination of the two multimodal data streams. The computational kernel extracts one-dimensional features from the live cell turbidity matrix along the time axis as the first-dimensional time-series vector, while simultaneously reducing the dimensionality of the bioimpedance spectrum time-series stream or extracting characteristic frequency responses as the second-dimensional time-series vector.

[0098] To quantitatively characterize the modal differences between two time series vectors at any time point pair, the system constructs a two-dimensional topological space and sequentially calculates the distances between sampling points in the first-dimensional time series vector and sampling points in the second-dimensional time series vector. The specific calculation logic is based on the following Euclidean distance calculation formula:

[0099] ;

[0100] In this formula, Represents the first dimension of the time series vector. The nth sampling point and the second-dimensional time vector in the second dimension Local physical distance between sampling points Represents the first dimension of the time series vector. The amplitude characteristics of each sampling point Represents the second dimension of the time series vector. The amplitude characteristics of each sampling point are analyzed. The computational kernel traverses all combinations of sampling points and calculates a two-dimensional matrix containing the physical difference density between each spatiotemporal node, thereby generating a local distance matrix.

[0101] After generating the local distance matrix, the system invokes the iterative accumulation operator to perform boundary constraint summation iterations along the time-aligned direction to construct the dynamic programming space for optimal path backtracking.

[0102] ;

[0103] In this formula, This represents the global cumulative distance value at the current grid node. This represents the local physical distance at the corresponding position in the local distance matrix. This represents the cumulative vertical distance at the previous time step node. Indicates the cumulative horizontal distance. This represents the cumulative alignment distance in the diagonal direction. This represents extracting the smallest cumulative scalar value from the three adjacent nodes in the three directions.

[0104] In a set of feasible implementation data, if the local physical distance of the current node... The calculated value is 0.050. The cumulative distance to its left neighboring node is 0.320, the cumulative distance to its lower neighboring node is 0.280, and the cumulative distance to its lower left diagonal node is 0.210. Therefore, the extracted minimum cumulative scalar value is 0.210. Summing these two values, the global cumulative distance value of the current grid node is output. The value is 0.260. The computational kernel starts from the grid boundary and is strictly subject to the physical constraints of the starting boundary, the ending boundary, and the monotonic continuity boundary. It iteratively updates the cumulative value of each grid element row by row, and finally constructs a global dynamic programming matrix that fully reflects the topology of multimodal temporal evolution loss.

[0105] After constructing the global dynamic programming matrix, the system transitions to the path optimization and data flow reconstruction phase. The path optimization kernel first locates the upper right boundary element of the global dynamic programming matrix, using it as the trigger source for the backtracking search. The zero-crossing point and gradient searcher backtracks from this boundary element to the starting boundary element. In each step, the cumulative distance difference in the three directions within the neighborhood of the current node is calculated, and the path is selected along the direction with the minimum gradient. Through continuous closed-loop backtracking search, the optimal regularized path that minimizes the sum of the global cumulative distances is identified.

[0106] After obtaining the optimal normalized path, the resampling operator establishes a many-to-many or one-to-many time axis projective mapping relationship. Based on the time index chain contained in the optimal normalized path, nonlinear resampling stretching and time reconstruction are performed on the first-dimensional time series vector and the second-dimensional time series vector, respectively. This operation, through point interpolation or time-domain compression, aligns the inflection points of the two time series signals, eliminating the phase lag differences between electrical and optical indicators caused by phase transition lag or transmission impedance. The two-dimensional time series vector with the time axis aligned is synchronously input into the multimodal feature fusion unit, where matrix column concatenation and dimension alignment are performed to generate an aligned data stream with synchronous spatial dimensions under a unified time reference. This aligned data stream is then directly sent to the next-level rate of change trend analysis unit as a continuous state feature stream to participate in the calculation of the critical transition time.

[0107] S4. Determine the critical time of flow based on the trend of the rate of change of the aligned data stream, and issue the process flow control word when the critical time of flow is reached. At the same time, lock the bioimpedance spectrum time series flow corresponding to the critical time of flow as the characteristic boundary parameter.

[0108] Furthermore, in S4, determining the critical transition time based on the trend of the rate of change of the aligned data stream includes:

[0109] The aligned data stream is subjected to first-order temporal difference operation along the time axis according to a preset step size to calculate the time series signal of the rate of change of each biological parameter in the aligned data stream within adjacent time windows.

[0110] The rate of change time series signal is smoothed and filtered to eliminate noise ripple, and the maximum peak inflection point of the smoothed and filtered rate of change time series signal on the overall time axis is extracted.

[0111] The sampling timestamp corresponding to the inflection point of the maximum peak is locked as the critical time of circulation.

[0112] Specifically, the data processing flow, which determines the critical time of transition based on the rate of change trend of the aligned data stream, begins with time-domain discretization and analysis of the aligned data stream generated in the previous stage. The processing kernel reads the aligned data stream from the system's shared memory, which contains the live cell turbidity matrix components and multi-frequency bioimpedance spectrum components aligned to the time reference.

[0113] To extract the dynamic evolution trend of each biological parameter over time and eliminate system drift, the computational kernel calls a first-order temporal difference operator to perform numerical differentiation operations along the time axis on the aligned data stream according to a preset step size. The computational logic of the first-order temporal difference operation follows the following first-order numerical difference formula:

[0114] ;

[0115] In this formula, Indicates the sampling point at the current time. The calculated rate of change characteristic value, This indicates that a specific biological parameter in the aligned data stream is advanced forward at the time step. Modal amplitude after sampling points, This indicates the current sampling point for this biological parameter. Modal amplitude at that location, This indicates the preset step size. The computational kernel traverses the multi-dimensional parameter channels in the aligned data stream, and by performing the aforementioned sliding window difference quotient calculation on the feature data within adjacent time windows, it outputs multiple independent time-series signals that reflect the rate of change of the instantaneous curvature of various biological parameters.

[0116] Because differential operations inherently amplify high-frequency noise, the generated original rate-of-change time-series signal typically contains random ripples introduced by sensor disturbances or fluid fluctuations. To avoid interfering with the abrupt change detection of the control system, the system inputs the rate-of-change time-series signal into a digital low-pass smoothing filter for noise reduction. The smoothing filter's operational logic employs a moving average sliding window filter formula:

[0117] ;

[0118] In this formula, This represents the smoothed rate of change signal component after noise and ripple removal. This indicates the preset filter window width. This represents the original rate of change feature value corresponding to the current sampling point and its historical sequence. In a set of feasible implementation data, if the filter window width... Assuming a count of 5, the original differential rate of change signal values ​​for the current and four historical periods are 0.120, 0.130, 0.110, 0.140, and 0.100, respectively. The smoothed rate of change signal component is then calculated by summing and averaging. Approximately 0.120. The system continuously arranges the smoothed rate of change signal components after multiple denoising along the time axis to obtain the smoothed and filtered rate of change time-series signal.

[0119] After generating the smoothed and filtered rate-of-change time-series signal, the mutation identification kernel calls the global extremum retrieval operator to scan the amplitude trajectory of the signal along the overall time-series axis. When cell growth transitions from the logarithmic phase to the plateau phase, the cell division rate reaches the maximum saturation point of the logarithmic curvature, which corresponds to the peak point of the first derivative of the smoothed and filtered rate-of-change time-series signal on the time-series axis. The retrieval operator locks the geometric inflection point at this maximum amplitude, identifies it as the maximum peak inflection point, and simultaneously captures the absolute sampling timestamp corresponding to this inflection point in the hardware timer, locking this sampling timestamp as the critical transition time.

[0120] After determining the critical time for pipeline transition, the system enters the state machine control transition and parameter solidification stage. The clock comparison unit within the microprocessor monitors the current total operating timeline in real time. The moment the accumulated system operating base time reaches the critical time, the interrupt controller is triggered, and the system immediately issues a process flow control word to the hardware driver register on the bus. This process flow control word consists of a set of machine instructions containing specific binary control bits, used to cut off the preceding environmental temperature and humidity maintenance loop and simultaneously send a pipeline start signal to the subsequent cooling step register. Within the same frequency sampling period when the process flow control word is issued, the system calls the data latch operator to instantaneously freeze the original bioimpedance spectrum time-series stream received from the preceding stage. The system writes the real and imaginary parts of the multi-frequency complex impedance characteristic data corresponding to the critical time into the non-volatile characteristic storage area, latching it as a characteristic boundary parameter. This characteristic boundary parameter serves as a static physical slice of the current cell community's vital signs integrity, directly participating in subsequent real-time membrane permeability inversion calculations based on the cell membrane stress dynamics model.

[0121] S5. Extract the charge relaxation features of the characteristic boundary parameters, invert and decouple the real-time membrane permeability parameters based on the cell membrane stress dynamics model, and input the real-time membrane permeability parameters and the acquired environmental thermodynamic parameters into the non-homogeneous heat conduction partial differential equation system to solve simultaneously and generate an adaptive control matrix.

[0122] Furthermore, in S5, the real-time membrane permeability parameters are decoupled and inverted based on the cell membrane stress dynamics model, including:

[0123] Based on the charge relaxation characteristics of the characteristic boundary parameters, the time constant and the peak value of the imaginary part of the complex permittivity of the corresponding charge relaxation curve are extracted.

[0124] By incorporating the time constant and the peak value of the imaginary part of the complex permittivity as boundary conditions into the pre-defined Maxwell-Wagner interface polarization dynamics equation, a topological equivalent mapping of multi-shell complex impedance is established.

[0125] The transmembrane conductivity and membrane capacitance change rate are solved by the topological equivalent mapping along the multi-shell complex impedance, and the solutions are combined to obtain real-time membrane permeability parameters.

[0126] Furthermore, in S5, the real-time membrane permeability parameters and the acquired environmental thermodynamic parameters are input together into a system of non-homogeneous heat conduction partial differential equations for simultaneous solution, including:

[0127] The environmental thermodynamic parameters are transformed into the initial temperature field spatial boundary and discretized into a three-dimensional finite element spatial mesh.

[0128] The real-time membrane permeability parameters are converted into non-homogeneous energy source terms and substituted into the non-homogeneous heat conduction partial differential equations as the heat source input within the three-dimensional finite element space mesh.

[0129] The non-homogeneous partial differential equations of heat conduction are iteratively solved along the three-dimensional finite element space mesh to calculate the time evolution of the target temperature that satisfies the cell ice crystal growth, and the control weights are extracted to construct an adaptive control matrix.

[0130] Specifically, in the programmed cooling and cryopreservation control system for cells, the core computation for achieving adaptive and precise temperature control lies in solving the cross-scale simultaneous equations of microscopic biological characteristics and macroscopic thermodynamic evolution. The data processing flow for this multimodal parameter property inversion and flow field solution begins with feature extraction of the characteristic boundary parameters written to the non-volatile storage region in the previous stage. The processing kernel reads the real and imaginary parts of the multi-frequency complex impedance feature data solidified at the critical moment, and calculates the corresponding relaxation response curve of the complex permittivity as a function of frequency using a frequency-domain dielectric spectrum transformation algorithm. The system calls the curve fitting operator to extract the time constant of the corresponding charge relaxation curve, and simultaneously captures the maximum point in the trajectory of the imaginary part of the complex permittivity as a function of frequency in the spectrum, locking it as the peak value of the imaginary part of the complex permittivity.

[0131] The extracted time constant and the peak value of the imaginary part of the complex permittivity are used as boundary conditions and are simultaneously imported into the pre-defined Maxwell-Wagner interfacial polarization kinetic equation. This kinetic equation is constructed based on the dielectric boundary conditions of the cell nucleus, membrane layer, and external medium. The computational kernel uses these two electrical relaxation characteristic parameters to analytically constrain the unknown coefficients in the kinetic equation, establishing a multi-shell complex impedance topological equivalent mapping that maps the electrical impedance characteristics of the cell's double membrane structure.

[0132] After generating this mapping, the inversion solver invokes a nonlinear numerical iterative operator to perform inverse parameter approximation along the multi-shell complex impedance topological equivalent mapping, calculating the transmembrane conductivity characterizing the opening and closing state of cell membrane ion channels, and the membrane capacitance change rate characterizing the mechanical deformation and compressive state of the cell membrane. Next, the system invokes a property merging operator to perform multidimensional weighted scaling and merging of the transmembrane conductivity and membrane capacitance change rate, ultimately generating real-time membrane permeability parameters that reflect the current overall damaged barrier state of the cell.

[0133] Subsequently, the system enters a cascaded computational flow of microscopic physical properties and macroscopic thermal field meshes. The processing kernel acquires the inherent physical parameters of the external cryopreservation environment through sensor data interfaces, obtaining environmental thermodynamic parameters, specifically including the wall thickness of the cryopreservation container, the thermal conductivity of the material, and the specific heat capacity of the cell suspension. The microprocessor calls the finite element preprocessing operator to mesh the three-dimensional geometric and physical space of the solid and fluid corresponding to the environmental thermodynamic parameters, converting it into a discrete three-dimensional finite element space mesh.

[0134] Simultaneously, the real-time membrane permeability parameters generated in the previous stage are converted into a non-homogeneous energy source term through property transformation equations. This non-homogeneous energy source term essentially represents the dynamic heat flux generated by the cell during cooling and freezing due to internal water seepage and phase change heat release. The system uses this non-homogeneous energy source term as an internal heat source input variable, substituting it into each discrete node of the three-dimensional finite element space mesh, thereby completing the parameter boundary coupling of the non-homogeneous heat conduction partial differential equation system.

[0135] After the internal heat source injection is completed, the finite element numerical solver is started. The system calls the time-domain step-size adaptive solver to perform a spatiotemporally coupled iterative solution of the non-homogeneous heat conduction partial differential equations containing the internal heat source input along the three-dimensional finite element space mesh. This solution process follows the following spatiotemporal evolution governing differential equations of the temperature field:

[0136] ;

[0137] In this formula, This indicates the macroscopic density of the cell suspension. This indicates the isobaric specific heat capacity of a cell suspension. This represents the temperature value of the spatial nodes to be solved. Indicates the time of cooling evolution. Represents the Hamiltonian operator. This represents the thermal conductivity coefficient matrix of the cryopreservation system. This represents the heat source term within the grid node, derived from real-time membrane permeability parameters. In a set of feasible implementation data, if the macroscopic density of the cell suspension... Counting 1050, isobaric specific heat capacity With a count of 4180, the current spacetime grid's conducted heat dissipation energy term has a count of -1500, while the calculated internal heat source term... With a count of 500, the simultaneous iteration of the equations can solve for the temperature change rate at that instant. The solver continuously advances in the time domain with an adaptive time step until it traverses the entire preset cooling and cryopreservation cycle, thereby calculating the target temperature time evolution process that satisfies the condition of no internal ice crystal growth barrier within the cell. Finally, the matrix construction operator extracts the control weight coefficients of the target temperature time evolution process at each discrete time step and concatenates them as row and column vectors into a spatial tensor to generate an adaptive control matrix. This adaptive control matrix is ​​directly written into the bus-shared memory control register as the skeleton data for generating the executable sequence of the underlying temperature control hardware.

[0138] S6. Generate a program cooling control sequence based on the adaptive control matrix, and write the program cooling control sequence into the cooling step register.

[0139] Furthermore, in S6, the generated programmed cooling control sequence based on the adaptive control matrix includes:

[0140] The adaptive control matrix is ​​decoupled by dimension to separate the target temperature value and the hold time value at each time step;

[0141] The slope of instantaneous temperature change is calculated based on the difference between the target temperature values ​​between adjacent time steps and the time span.

[0142] The target temperature value, holding time value, and instantaneous temperature change slope within each time step are concatenated using time-series logic to generate a program-based cooling control sequence.

[0143] Specifically, the data processing flow for generating the cooling control sequence based on the adaptive control matrix begins with multi-dimensional spatial decoupling of the adaptive control matrix stored in the shared register of main memory. This adaptive control matrix embeds the optimal temperature field spatiotemporal evolution state weights calculated by the simultaneous operation of the previous multi-shell property inversion and the heat conduction differential equation system.

[0144] To transform the tensor matrix into a one-dimensional process linked list that can be executed sequentially by a hardware timer, the system control instructions invoke the control matrix dimension splitting operator. The splitting operation logic performs orthogonal projection and dimension reduction decoupling along the time dimension axis of the matrix, stripping away the spatial control layer in the higher-order matrix and extracting the temperature target values ​​corresponding to each continuous time step on the discrete time axis, along with the corresponding hold-time values. The separated temperature target values ​​are represented as a digital temperature scalar sequence in degrees Celsius or Kelvin, and the hold-time values ​​are represented as a duration scalar sequence in clock reference counts or seconds. A one-to-one mapping is established between the two in the data structure through time step indexes.

[0145] After dimensional decoupling, the computational kernel calls the differential slope solver to dynamically reconstruct the cooling rate between adjacent steps. Since the underlying refrigeration temperature control mechanism must rely on a defined rate of temperature change to adjust the refrigerant flow or the pulse width modulation duty cycle of the heating element when performing cooling actions, the system needs to dynamically calculate the process slope within each time segment. The specific calculation logic follows the instantaneous slope solution formula:

[0146] ;

[0147] In this formula, Indicates the first The slope of the instantaneous temperature change within each time step. Indicates the first The target temperature setpoint for each time step. Indicates the first The initial temperature setpoint for each time step. This represents the discrete time step span between two sets of temperature setpoints. In a set of feasible implementation data, if the... Initial temperature setpoint of the step -40.00 degrees Celsius, advance one time step. The temperature setpoint for the next critical node after 60 seconds. If the temperature evolves to -43.00 degrees Celsius, then the slope of the instantaneous temperature change calculated using the above difference quotient operation is... The temperature is approximately -0.05 degrees Celsius per second. The computational kernel traverses all discrete clock nodes, sequentially solving for a process slope vector table covering the entire cryopreservation cycle.

[0148] After generating a vector table composed of instantaneous slopes for each segment, the system enters the pipeline reassembly stage of the underlying machine control word. The timing logic splicing operator invokes the data packet encapsulation protocol, using the target temperature value, hold time length, and the just-calculated instantaneous temperature change slope contained in each independent time step as the underlying control fields. The splicing operator concatenates these discrete control fields at the tail and splices them in binary serial order according to the ascending time axis of the time steps. The control field of each step is encapsulated into a standard machine instruction frame containing a feature start bit, temperature control word, time count control word, slope control word, and check bit. The instruction frames of multiple consecutive time steps are spliced ​​and reconstructed sequentially to generate a program cooling control sequence composed of multiple segments of stepped cooling process control words.

[0149] After generating the programmed cooling control sequence, the microprocessor sends the data stream to the hardware core of the cryogenic controller via the internal peripheral interconnect bus. The bus controller sends an overwrite enable signal, directly overwriting the programmed cooling control sequence into the cooling step register as streaming data. The write process forcibly clears the original historical static cooling parameters in the register, realizing dynamic refreshing of the underlying hardware control parameters. The cooling step register is directly cascaded with the underlying timer interrupt service routine and pulse width modulation generator hardware, and the programmed cooling control sequence latched in the register serves as the reference comparison value for the hardware timer interrupt. As the cryogenic system's overall clock advances, the temperature control hardware dynamically drives the linear refrigeration valve to adjust the liquid nitrogen spray rate based on the instantaneous change slope and temperature target value within the current register frame. This ensures that the actual heat dissipation curve inside the cryogenic chamber strictly tracks the optimal evolution process mapped by the adaptive control matrix, completing the full physical closed loop from cell multimodal dynamic monitoring to adaptive programmed cooling storage.

[0150] Example 2:

[0151] During large-scale adherent or suspension culture of stem cells, the long culture period inevitably leads to the accumulation of a large number of dead cell fragments, and the respiration of high-density cell communities causes frequent microbubbles to form in the culture medium. The presence of these non-living particles prevents traditional online optical turbidity monitoring devices from specifically extracting the actual density of living cells. The background scattering components caused by these non-living particles severely overlap in the mixed optical signal, resulting in significant lag in the system's determination of the flow switching point at the end of logarithmic growth, and even triggering erroneous flow control actions. Ultimately, because the cooling and temperature control parameters are completely decoupled from the actual microscopic properties of living cells, the cell loss rate after cryopreservation and thawing is extremely high. To solve these problems, this invention provides a cell culture and storage management method based on biological parameter monitoring, the structure of which is as follows: Figure 1 As shown. The specific implementation process of this method is as follows:

[0152] First, the bioimpedance spectrum temporal current and optical turbidity temporal current of the culture stage are connected in parallel. The system injects multi-frequency weak AC signals from 10Hz to 10MHz into the stem cell culture system, dynamically acquires the complex impedance response at each characteristic frequency point and performs time-phase alignment to generate the bioimpedance spectrum temporal current; simultaneously, it drives the collimation monitoring beam to penetrate the system, calculates the transmitted light intensity attenuation coefficient and converts it into absorbance characteristics to generate the optical turbidity temporal current.

[0153] The system performs joint denoising and property decoupling on two time-series streams. The system separates the imaginary spectral lines of the bioimpedance spectrum time-series stream, locks the specified polarization relaxation frequency corresponding to stem cells by detecting the zero-crossing point of the first derivative, and extracts the imaginary amplitude at this frequency as the imaginary feature. This imaginary feature is normalized and mapped, and the generated dynamic property correction coefficients are used as diagonal elements to construct a decoupling matrix. Simultaneously, the one-dimensional optical turbidity time-series stream is framed in 10-minute time-sliding frames to construct an initial turbidity feature matrix. The matrix operation kernel calls the Hadamard product operator to perform element-wise multiplication of the decoupling matrix and the initial turbidity feature matrix, dynamically suppressing background scattering components caused by debris and bubbles using polarization features. After inverse frame fusion and reconstruction, a high-purity live-cell turbidity matrix is ​​output.

[0154] To eliminate the temporal lag of photoelectric indicators, the system inputs the generated live cell turbidity matrix and the bioimpedance spectrum time series into the dynamic time warping algorithm. A local distance matrix is ​​constructed by calculating the Euclidean distance between the sampling points of the two vectors, and a global dynamic programming matrix is ​​generated through boundary constraint summation and iteration. The computational kernel backtracks along the direction of minimum gradient to lock the optimal warping path with the minimum global cumulative distance. Based on this, nonlinear resampling and alignment of the two time series vectors are performed, and multimodal feature fusion is executed to generate an aligned data stream under a unified time reference.

[0155] During the flow control phase, the system performs first-order temporal difference and moving average filtering on the aligned data stream, extracts the maximum peak inflection point of the denoised rate of change time-series signal distribution over time (i.e., the stem cell division curvature saturation point), and locks its sampling timestamp as the flow critical time. When the total running time axis reaches this flow critical time, the system issues a process flow control word to switch processes, and simultaneously freezes and locks the corresponding bioimpedance spectrum time-series stream as a static characteristic boundary parameter in the non-volatile storage area.

[0156] The system then shifts to cross-scale cooling control. The processing kernel extracts the time constant and the peak value of the imaginary part of the complex permittivity of the characteristic boundary parameters, imports them into the Maxwell-Wagner interface polarization kinetic equations, and inversely calculates the transmembrane conductivity and membrane capacitance change rate of the stem cells, merging them into real-time membrane permeability parameters. The system acquires environmental thermodynamic parameters such as the cavity conductivity coefficient and discretizes them into a three-dimensional finite element space mesh. The real-time membrane permeability parameters are transformed into non-homogeneous energy source terms and substituted into the mesh nodes. A spatiotemporally coupled iterative solution is performed using a set of non-homogeneous heat conduction partial differential equations to calculate the time evolution of the target temperature that satisfies the condition of no internal ice crystal growth, and an adaptive control matrix is ​​constructed. Subsequently, orthogonal dimensionality reduction is performed on this matrix to separate the target temperature value and holding time for each step, and the instantaneous change slope between adjacent steps is calculated. The timing logic splicing process concatenates the target value, holding time and slope using binary logic, reconstructs and generates the program cooling control sequence, and overwrites it into the underlying cooling step register. This dynamically drives the linear liquid nitrogen valve to change the spray duty cycle, so that the actual heat dissipation trajectory of the freezing chamber accurately tracks the optimal temperature process, and safely completes the ultra-low temperature cryopreservation of stem cells.

[0157] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for cell culture and storage management based on biological parameter monitoring, characterized in that, include: S1, the bioimpedance spectrum time-series flow and optical turbidity time-series flow during the inoculation and culture phases; S2. Extract the imaginary part features of the bioimpedance spectrum time-series flow at the specified polarization relaxation frequency and construct a decoupling matrix. Use the decoupling matrix to perform nonlinear property decoupling on the optical turbidity time-series flow and output the live cell turbidity matrix. S3. Input the live cell turbidity matrix and bioimpedance spectrum time stream into the dynamic time warping algorithm for calculation, complete the time nonlinear alignment and generate the aligned data stream; S4. Determine the critical time of flow based on the change rate trend of the aligned data stream, and issue a process flow control word when the critical time of flow is reached, while locking the bioimpedance spectrum time series corresponding to the critical time of flow as a characteristic boundary parameter. S5. Extract the charge relaxation features of the feature boundary parameters, invert and decouple the real-time membrane permeability parameters based on the cell membrane stress dynamics model, and input the real-time membrane permeability parameters and the obtained environmental thermodynamic parameters into a set of non-homogeneous heat conduction partial differential equations to solve simultaneously and generate an adaptive control matrix. S6. Generate a programmed cooling control sequence based on the adaptive control matrix, and overwrite the programmed cooling control sequence into the cooling step register.

2. The cell culture and storage management method based on biological parameter monitoring according to claim 1, characterized in that, In S1, the bioimpedance spectrum time-series flow and optical turbidity time-series flow during the inoculation and culture phase include: A multi-frequency AC signal is injected into the culture system, and the complex impedance response of the culture system at each characteristic frequency point is collected and aligned to generate a bioimpedance spectrum time-series flow. The collimated monitoring beam is driven to penetrate the culture system, the transmitted light intensity attenuation coefficient is calculated and converted into absorbance characteristics, and an optical turbidity time-series flow is generated.

3. The cell culture and storage management method based on biological parameter monitoring according to claim 1, characterized in that, In S2, extracting the imaginary part features of the bioimpedance spectrum time-series current at a specified polarization relaxation frequency and constructing the decoupling matrix includes: The imaginary part response of the time-series bioimpedance spectrum is separated to obtain the imaginary part spectral line of the complex impedance as a function of frequency. The first derivative zero-crossing point of the imaginary part spectral line is detected, the frequency corresponding to the zero-crossing point is determined as the specified polarization relaxation frequency, and the corresponding imaginary part amplitude is extracted as the imaginary part feature. The imaginary features are normalized and mapped, and the mapped correction coefficients are used as diagonal elements to construct a diagonal feature matrix, thereby generating a decoupling matrix.

4. The cell culture and storage management method based on biological parameter monitoring according to claim 1, characterized in that, In S2, the nonlinear property decoupling of the optical turbidity time-series flow using the decoupling matrix includes: The optical turbidity time-series stream is divided into frames according to a time sliding window to construct an initial turbidity feature matrix; The Hadamard product operator is used to perform element-wise multiplication of the decoupling matrix and the initial turbidity characteristic matrix, so as to suppress the medium background scattering component in the initial turbidity characteristic matrix by using the correction coefficient in the decoupling matrix. The initial turbidity feature matrix after calculation is subjected to inverse frame fusion to output the live cell turbidity matrix.

5. The cell culture and storage management method based on biological parameter monitoring according to claim 1, characterized in that, In S3, the calculation of the live cell turbidity matrix and bioimpedance spectrum time-series flow by inputting them into the dynamic time warping algorithm includes: The turbidity matrix of the living cells and the time-series flow of the bioimpedance spectrum are used as the time-series vectors of the first and second dimensions. Calculate the Euclidean distance between the sampling points in the first-dimensional temporal vector and the second-dimensional temporal vector to generate a local distance matrix; The local distance matrix is ​​subjected to boundary constraint summation and iteration to construct a global dynamic programming matrix.

6. The cell culture and storage management method based on biological parameter monitoring according to claim 5, characterized in that, In S3, completing the timing nonlinear alignment and generating the aligned data stream includes: Search for the inverse path along the direction of minimum gradient along the global dynamic programming matrix to determine the optimal regularized path with the minimum global cumulative distance; Based on the optimal normalization path, the time series vectors of the first dimension and the second dimension are nonlinearly resampled and aligned. The aligned two-dimensional time-series vectors are fused with multimodal features to generate an aligned data stream.

7. The cell culture and storage management method based on biological parameter monitoring according to claim 1, characterized in that, In S4, determining the critical transition time based on the rate of change trend of the aligned data stream includes: The aligned data stream is subjected to first-order temporal difference operation along the time axis according to a preset step size to calculate the time-series signal of the rate of change of each biological parameter in the aligned data stream within adjacent time windows. The rate of change time-series signal is smoothed and filtered to eliminate noise ripple, and the maximum peak inflection point of the smoothed and filtered rate of change time-series signal on the overall time-series axis is extracted. The sampling timestamp corresponding to the maximum peak inflection point is locked as the critical time of circulation.

8. The cell culture and storage management method based on biological parameter monitoring according to claim 1, characterized in that, In S5, the real-time membrane permeability parameters decoupled from the cell membrane stress dynamics model include: Based on the charge relaxation characteristics of the feature boundary parameters, the time constant and the peak value of the imaginary part of the complex permittivity of the corresponding charge relaxation curve are extracted. The time constant and the peak value of the imaginary part of the complex permittivity are used as boundary conditions and imported into the preset Maxwell-Wagner interface polarization dynamics equation to establish a multi-shell complex impedance topological equivalent mapping. The transmembrane conductivity and membrane capacitance change rate are solved along the multi-shell complex impedance topological equivalent mapping and combined to obtain the real-time membrane permeability parameters.

9. The cell culture and storage management method based on biological parameter monitoring according to claim 1, characterized in that, In S5, the real-time membrane permeability parameters and the acquired environmental thermodynamic parameters are input together into a set of non-homogeneous heat conduction partial differential equations for simultaneous solution, including: The environmental thermodynamic parameters are transformed into the initial temperature field spatial boundary and discretized into a three-dimensional finite element space mesh. The real-time membrane permeability parameters are converted into non-homogeneous energy source terms and substituted into the non-homogeneous heat conduction partial differential equations as the heat source input within the three-dimensional finite element space grid. The non-homogeneous partial differential equations of heat conduction are iteratively solved along the three-dimensional finite element space mesh to calculate the time evolution of the target temperature that satisfies the cell ice crystal growth, and the control weights are extracted to construct an adaptive control matrix.

10. A cell culture and storage management method based on biological parameter monitoring according to claim 1, characterized in that, In S6, generating the programmed cooling control sequence based on the adaptive control matrix includes: The adaptive control matrix is ​​decoupled by dimension to separate the target temperature value and the hold time value at each time step; The slope of instantaneous temperature change is calculated based on the difference between the target temperature values ​​between adjacent time steps and the time span. The target temperature value, holding time value, and instantaneous temperature change slope within each time step are concatenated using time-series logic to generate a program-based cooling control sequence.

Images